linux-stable/drivers/virtio/virtio_mem.c

3009 lines
81 KiB
C
Raw Normal View History

virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
// SPDX-License-Identifier: GPL-2.0-or-later
/*
* Virtio-mem device driver.
*
* Copyright Red Hat, Inc. 2020
*
* Author(s): David Hildenbrand <david@redhat.com>
*/
#include <linux/virtio.h>
#include <linux/virtio_mem.h>
#include <linux/workqueue.h>
#include <linux/slab.h>
#include <linux/module.h>
#include <linux/mm.h>
#include <linux/memory_hotplug.h>
#include <linux/memory.h>
#include <linux/hrtimer.h>
#include <linux/crash_dump.h>
#include <linux/mutex.h>
#include <linux/bitmap.h>
#include <linux/lockdep.h>
#include <linux/log2.h>
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
#include <acpi/acpi_numa.h>
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
static bool unplug_online = true;
module_param(unplug_online, bool, 0644);
MODULE_PARM_DESC(unplug_online, "Try to unplug online memory");
static bool force_bbm;
module_param(force_bbm, bool, 0444);
MODULE_PARM_DESC(force_bbm,
"Force Big Block Mode. Default is 0 (auto-selection)");
static unsigned long bbm_block_size;
module_param(bbm_block_size, ulong, 0444);
MODULE_PARM_DESC(bbm_block_size,
"Big Block size in bytes. Default is 0 (auto-detection).");
/*
* virtio-mem currently supports the following modes of operation:
*
* * Sub Block Mode (SBM): A Linux memory block spans 2..X subblocks (SB). The
* size of a Sub Block (SB) is determined based on the device block size, the
* pageblock size, and the maximum allocation granularity of the buddy.
* Subblocks within a Linux memory block might either be plugged or unplugged.
* Memory is added/removed to Linux MM in Linux memory block granularity.
*
* * Big Block Mode (BBM): A Big Block (BB) spans 1..X Linux memory blocks.
* Memory is added/removed to Linux MM in Big Block granularity.
*
* The mode is determined automatically based on the Linux memory block size
* and the device block size.
*
* User space / core MM (auto onlining) is responsible for onlining added
* Linux memory blocks - and for selecting a zone. Linux Memory Blocks are
* always onlined separately, and all memory within a Linux memory block is
* onlined to the same zone - virtio-mem relies on this behavior.
*/
/*
* State of a Linux memory block in SBM.
*/
enum virtio_mem_sbm_mb_state {
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Unplugged, not added to Linux. Can be reused later. */
VIRTIO_MEM_SBM_MB_UNUSED = 0,
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* (Partially) plugged, not added to Linux. Error on add_memory(). */
VIRTIO_MEM_SBM_MB_PLUGGED,
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Fully plugged, fully added to Linux, offline. */
VIRTIO_MEM_SBM_MB_OFFLINE,
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Partially plugged, fully added to Linux, offline. */
VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
/* Fully plugged, fully added to Linux, onlined to a kernel zone. */
VIRTIO_MEM_SBM_MB_KERNEL,
/* Partially plugged, fully added to Linux, online to a kernel zone */
VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
/* Fully plugged, fully added to Linux, onlined to ZONE_MOVABLE. */
VIRTIO_MEM_SBM_MB_MOVABLE,
/* Partially plugged, fully added to Linux, onlined to ZONE_MOVABLE. */
VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
VIRTIO_MEM_SBM_MB_COUNT
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
};
/*
* State of a Big Block (BB) in BBM, covering 1..X Linux memory blocks.
*/
enum virtio_mem_bbm_bb_state {
/* Unplugged, not added to Linux. Can be reused later. */
VIRTIO_MEM_BBM_BB_UNUSED = 0,
/* Plugged, not added to Linux. Error on add_memory(). */
VIRTIO_MEM_BBM_BB_PLUGGED,
/* Plugged and added to Linux. */
VIRTIO_MEM_BBM_BB_ADDED,
/* All online parts are fake-offline, ready to remove. */
VIRTIO_MEM_BBM_BB_FAKE_OFFLINE,
VIRTIO_MEM_BBM_BB_COUNT
};
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
struct virtio_mem {
struct virtio_device *vdev;
/* We might first have to unplug all memory when starting up. */
bool unplug_all_required;
/* Workqueue that processes the plug/unplug requests. */
struct work_struct wq;
atomic_t wq_active;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
atomic_t config_changed;
/* Virtqueue for guest->host requests. */
struct virtqueue *vq;
/* Wait for a host response to a guest request. */
wait_queue_head_t host_resp;
/* Space for one guest request and the host response. */
struct virtio_mem_req req;
struct virtio_mem_resp resp;
/* The current size of the device. */
uint64_t plugged_size;
/* The requested size of the device. */
uint64_t requested_size;
/* The device block size (for communicating with the device). */
uint64_t device_block_size;
/* The determined node id for all memory of the device. */
int nid;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Physical start address of the memory region. */
uint64_t addr;
/* Maximum region size in bytes. */
uint64_t region_size;
/* The parent resource for all memory added via this device. */
struct resource *parent_resource;
virtio-mem: add memory via add_memory_driver_managed() Virtio-mem managed memory is always detected and added by the virtio-mem driver, never using something like the firmware-provided memory map. This is the case after an ordinary system reboot, and has to be guaranteed after kexec. Especially, virtio-mem added memory resources can contain inaccessible parts ("unblocked memory blocks"), blindly forwarding them to a kexec kernel is dangerous, as unplugged memory will get accessed (esp. written). Let's use the new way of adding special driver-managed memory introduced in commit 7b7b27214bba ("mm/memory_hotplug: introduce add_memory_driver_managed()"). This will result in no entries in /sys/firmware/memmap ("raw firmware- provided memory map"), the memory resource will be flagged IORESOURCE_MEM_DRIVER_MANAGED (esp., kexec_file_load() will not place kexec images on this memory), and it is exposed as "System RAM (virtio_mem)" in /proc/iomem, so esp. kexec-tools can properly handle it. Example /proc/iomem before this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM 340000000-347ffffff : System RAM 348000000-34fffffff : System RAM [...] Example /proc/iomem after this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM (virtio_mem) 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM (virtio_mem) 340000000-347ffffff : System RAM (virtio_mem) 348000000-34fffffff : System RAM (virtio_mem) [...] Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: teawater <teawaterz@linux.alibaba.com> Fixes: 5f1f79bbc9e26 ("virtio-mem: Paravirtualized memory hotplug") Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200611093518.5737-1-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
2020-06-11 09:35:18 +00:00
/*
* Copy of "System RAM (virtio_mem)" to be used for
* add_memory_driver_managed().
*/
const char *resource_name;
/* Memory group identification. */
int mgid;
/*
* We don't want to add too much memory if it's not getting onlined,
* to avoid running OOM. Besides this threshold, we allow to have at
* least two offline blocks at a time (whatever is bigger).
*/
#define VIRTIO_MEM_DEFAULT_OFFLINE_THRESHOLD (1024 * 1024 * 1024)
atomic64_t offline_size;
uint64_t offline_threshold;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* If set, the driver is in SBM, otherwise in BBM. */
bool in_sbm;
union {
struct {
/* Id of the first memory block of this device. */
unsigned long first_mb_id;
/* Id of the last usable memory block of this device. */
unsigned long last_usable_mb_id;
/* Id of the next memory bock to prepare when needed. */
unsigned long next_mb_id;
/* The subblock size. */
uint64_t sb_size;
/* The number of subblocks per Linux memory block. */
uint32_t sbs_per_mb;
/*
* Some of the Linux memory blocks tracked as "partially
* plugged" are completely unplugged and can be offlined
* and removed -- which previously failed.
*/
bool have_unplugged_mb;
/* Summary of all memory block states. */
unsigned long mb_count[VIRTIO_MEM_SBM_MB_COUNT];
/*
* One byte state per memory block. Allocated via
* vmalloc(). Resized (alloc+copy+free) on demand.
*
* With 128 MiB memory blocks, we have states for 512
* GiB of memory in one 4 KiB page.
*/
uint8_t *mb_states;
/*
* Bitmap: one bit per subblock. Allocated similar to
* sbm.mb_states.
*
* A set bit means the corresponding subblock is
* plugged, otherwise it's unblocked.
*
* With 4 MiB subblocks, we manage 128 GiB of memory
* in one 4 KiB page.
*/
unsigned long *sb_states;
} sbm;
struct {
/* Id of the first big block of this device. */
unsigned long first_bb_id;
/* Id of the last usable big block of this device. */
unsigned long last_usable_bb_id;
/* Id of the next device bock to prepare when needed. */
unsigned long next_bb_id;
/* Summary of all big block states. */
unsigned long bb_count[VIRTIO_MEM_BBM_BB_COUNT];
/* One byte state per big block. See sbm.mb_states. */
uint8_t *bb_states;
/* The block size used for plugging/adding/removing. */
uint64_t bb_size;
} bbm;
};
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Mutex that protects the sbm.mb_count, sbm.mb_states,
* sbm.sb_states, bbm.bb_count, and bbm.bb_states
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*
* When this lock is held the pointers can't change, ONLINE and
* OFFLINE blocks can't change the state and no subblocks will get
* plugged/unplugged.
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
*
* In kdump mode, used to serialize requests, last_block_addr and
* last_block_plugged.
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*/
struct mutex hotplug_mutex;
bool hotplug_active;
/* An error occurred we cannot handle - stop processing requests. */
bool broken;
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
/* Cached valued of is_kdump_kernel() when the device was probed. */
bool in_kdump;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* The driver is being removed. */
spinlock_t removal_lock;
bool removing;
/* Timer for retrying to plug/unplug memory. */
struct hrtimer retry_timer;
unsigned int retry_timer_ms;
#define VIRTIO_MEM_RETRY_TIMER_MIN_MS 50000
#define VIRTIO_MEM_RETRY_TIMER_MAX_MS 300000
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Memory notifier (online/offline events). */
struct notifier_block memory_notifier;
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
#ifdef CONFIG_PROC_VMCORE
/* vmcore callback for /proc/vmcore handling in kdump mode */
struct vmcore_cb vmcore_cb;
uint64_t last_block_addr;
bool last_block_plugged;
#endif /* CONFIG_PROC_VMCORE */
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Next device in the list of virtio-mem devices. */
struct list_head next;
};
/*
* We have to share a single online_page callback among all virtio-mem
* devices. We use RCU to iterate the list in the callback.
*/
static DEFINE_MUTEX(virtio_mem_mutex);
static LIST_HEAD(virtio_mem_devices);
static void virtio_mem_online_page_cb(struct page *page, unsigned int order);
static void virtio_mem_fake_offline_going_offline(unsigned long pfn,
unsigned long nr_pages);
static void virtio_mem_fake_offline_cancel_offline(unsigned long pfn,
unsigned long nr_pages);
virtio-mem: don't always trigger the workqueue when offlining memory Let's trigger from offlining code only when we're not allowed to unplug online memory. Handle the other case (memmap possibly freeing up another memory block) when actually removing memory. We now also properly handle the case when removing already offline memory blocks via virtio_mem_mb_remove(). When removing via virtio_mem_remove(), when unloading the driver, virtio_mem_retry() is a NOP and safe to use. While at it, move retry handling when offlining out of virtio_mem_notify_offline(), to share it with Big Block Mode (BBM) soon. This is a preparation for Big Block Mode (BBM), whereby we can see some temporary offlining of memory blocks without actually making progress. Imagine you have a Big Block that spans to Linux memory blocks. Assume the first Linux memory blocks has no unmovable data on it. When we would call offline_and_remove_memory() on the big block, we would 1. Try to offline the first block. Works, notifiers triggered. virtio_mem_retry() called. 2. Try to offline the second block. Does not work. 3. Re-online first block. 4. Exit to main loop, exit workqueue. 5. Retry immediately (due to virtio_mem_retry()), go to 1. The result are endless retries. Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20201112133815.13332-16-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-11-12 13:38:01 +00:00
static void virtio_mem_retry(struct virtio_mem *vm);
static int virtio_mem_create_resource(struct virtio_mem *vm);
static void virtio_mem_delete_resource(struct virtio_mem *vm);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Register a virtio-mem device so it will be considered for the online_page
* callback.
*/
static int register_virtio_mem_device(struct virtio_mem *vm)
{
int rc = 0;
/* First device registers the callback. */
mutex_lock(&virtio_mem_mutex);
if (list_empty(&virtio_mem_devices))
rc = set_online_page_callback(&virtio_mem_online_page_cb);
if (!rc)
list_add_rcu(&vm->next, &virtio_mem_devices);
mutex_unlock(&virtio_mem_mutex);
return rc;
}
/*
* Unregister a virtio-mem device so it will no longer be considered for the
* online_page callback.
*/
static void unregister_virtio_mem_device(struct virtio_mem *vm)
{
/* Last device unregisters the callback. */
mutex_lock(&virtio_mem_mutex);
list_del_rcu(&vm->next);
if (list_empty(&virtio_mem_devices))
restore_online_page_callback(&virtio_mem_online_page_cb);
mutex_unlock(&virtio_mem_mutex);
synchronize_rcu();
}
/*
* Calculate the memory block id of a given address.
*/
static unsigned long virtio_mem_phys_to_mb_id(unsigned long addr)
{
return addr / memory_block_size_bytes();
}
/*
* Calculate the physical start address of a given memory block id.
*/
static unsigned long virtio_mem_mb_id_to_phys(unsigned long mb_id)
{
return mb_id * memory_block_size_bytes();
}
/*
* Calculate the big block id of a given address.
*/
static unsigned long virtio_mem_phys_to_bb_id(struct virtio_mem *vm,
uint64_t addr)
{
return addr / vm->bbm.bb_size;
}
/*
* Calculate the physical start address of a given big block id.
*/
static uint64_t virtio_mem_bb_id_to_phys(struct virtio_mem *vm,
unsigned long bb_id)
{
return bb_id * vm->bbm.bb_size;
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Calculate the subblock id of a given address.
*/
static unsigned long virtio_mem_phys_to_sb_id(struct virtio_mem *vm,
unsigned long addr)
{
const unsigned long mb_id = virtio_mem_phys_to_mb_id(addr);
const unsigned long mb_addr = virtio_mem_mb_id_to_phys(mb_id);
return (addr - mb_addr) / vm->sbm.sb_size;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Set the state of a big block, taking care of the state counter.
*/
static void virtio_mem_bbm_set_bb_state(struct virtio_mem *vm,
unsigned long bb_id,
enum virtio_mem_bbm_bb_state state)
{
const unsigned long idx = bb_id - vm->bbm.first_bb_id;
enum virtio_mem_bbm_bb_state old_state;
old_state = vm->bbm.bb_states[idx];
vm->bbm.bb_states[idx] = state;
BUG_ON(vm->bbm.bb_count[old_state] == 0);
vm->bbm.bb_count[old_state]--;
vm->bbm.bb_count[state]++;
}
/*
* Get the state of a big block.
*/
static enum virtio_mem_bbm_bb_state virtio_mem_bbm_get_bb_state(struct virtio_mem *vm,
unsigned long bb_id)
{
return vm->bbm.bb_states[bb_id - vm->bbm.first_bb_id];
}
/*
* Prepare the big block state array for the next big block.
*/
static int virtio_mem_bbm_bb_states_prepare_next_bb(struct virtio_mem *vm)
{
unsigned long old_bytes = vm->bbm.next_bb_id - vm->bbm.first_bb_id;
unsigned long new_bytes = old_bytes + 1;
int old_pages = PFN_UP(old_bytes);
int new_pages = PFN_UP(new_bytes);
uint8_t *new_array;
if (vm->bbm.bb_states && old_pages == new_pages)
return 0;
new_array = vzalloc(new_pages * PAGE_SIZE);
if (!new_array)
return -ENOMEM;
mutex_lock(&vm->hotplug_mutex);
if (vm->bbm.bb_states)
memcpy(new_array, vm->bbm.bb_states, old_pages * PAGE_SIZE);
vfree(vm->bbm.bb_states);
vm->bbm.bb_states = new_array;
mutex_unlock(&vm->hotplug_mutex);
return 0;
}
#define virtio_mem_bbm_for_each_bb(_vm, _bb_id, _state) \
for (_bb_id = vm->bbm.first_bb_id; \
_bb_id < vm->bbm.next_bb_id && _vm->bbm.bb_count[_state]; \
_bb_id++) \
if (virtio_mem_bbm_get_bb_state(_vm, _bb_id) == _state)
#define virtio_mem_bbm_for_each_bb_rev(_vm, _bb_id, _state) \
for (_bb_id = vm->bbm.next_bb_id - 1; \
_bb_id >= vm->bbm.first_bb_id && _vm->bbm.bb_count[_state]; \
_bb_id--) \
if (virtio_mem_bbm_get_bb_state(_vm, _bb_id) == _state)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Set the state of a memory block, taking care of the state counter.
*/
static void virtio_mem_sbm_set_mb_state(struct virtio_mem *vm,
unsigned long mb_id, uint8_t state)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const unsigned long idx = mb_id - vm->sbm.first_mb_id;
uint8_t old_state;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
old_state = vm->sbm.mb_states[idx];
vm->sbm.mb_states[idx] = state;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
BUG_ON(vm->sbm.mb_count[old_state] == 0);
vm->sbm.mb_count[old_state]--;
vm->sbm.mb_count[state]++;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Get the state of a memory block.
*/
static uint8_t virtio_mem_sbm_get_mb_state(struct virtio_mem *vm,
unsigned long mb_id)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const unsigned long idx = mb_id - vm->sbm.first_mb_id;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return vm->sbm.mb_states[idx];
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Prepare the state array for the next memory block.
*/
static int virtio_mem_sbm_mb_states_prepare_next_mb(struct virtio_mem *vm)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
int old_pages = PFN_UP(vm->sbm.next_mb_id - vm->sbm.first_mb_id);
int new_pages = PFN_UP(vm->sbm.next_mb_id - vm->sbm.first_mb_id + 1);
uint8_t *new_array;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (vm->sbm.mb_states && old_pages == new_pages)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return 0;
new_array = vzalloc(new_pages * PAGE_SIZE);
if (!new_array)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return -ENOMEM;
mutex_lock(&vm->hotplug_mutex);
if (vm->sbm.mb_states)
memcpy(new_array, vm->sbm.mb_states, old_pages * PAGE_SIZE);
vfree(vm->sbm.mb_states);
vm->sbm.mb_states = new_array;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
mutex_unlock(&vm->hotplug_mutex);
return 0;
}
#define virtio_mem_sbm_for_each_mb(_vm, _mb_id, _state) \
for (_mb_id = _vm->sbm.first_mb_id; \
_mb_id < _vm->sbm.next_mb_id && _vm->sbm.mb_count[_state]; \
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
_mb_id++) \
if (virtio_mem_sbm_get_mb_state(_vm, _mb_id) == _state)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
#define virtio_mem_sbm_for_each_mb_rev(_vm, _mb_id, _state) \
for (_mb_id = _vm->sbm.next_mb_id - 1; \
_mb_id >= _vm->sbm.first_mb_id && _vm->sbm.mb_count[_state]; \
_mb_id--) \
if (virtio_mem_sbm_get_mb_state(_vm, _mb_id) == _state)
/*
* Calculate the bit number in the subblock bitmap for the given subblock
* inside the given memory block.
*/
static int virtio_mem_sbm_sb_state_bit_nr(struct virtio_mem *vm,
unsigned long mb_id, int sb_id)
{
return (mb_id - vm->sbm.first_mb_id) * vm->sbm.sbs_per_mb + sb_id;
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Mark all selected subblocks plugged.
*
* Will not modify the state of the memory block.
*/
static void virtio_mem_sbm_set_sb_plugged(struct virtio_mem *vm,
unsigned long mb_id, int sb_id,
int count)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
__bitmap_set(vm->sbm.sb_states, bit, count);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Mark all selected subblocks unplugged.
*
* Will not modify the state of the memory block.
*/
static void virtio_mem_sbm_set_sb_unplugged(struct virtio_mem *vm,
unsigned long mb_id, int sb_id,
int count)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
__bitmap_clear(vm->sbm.sb_states, bit, count);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Test if all selected subblocks are plugged.
*/
static bool virtio_mem_sbm_test_sb_plugged(struct virtio_mem *vm,
unsigned long mb_id, int sb_id,
int count)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (count == 1)
return test_bit(bit, vm->sbm.sb_states);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* TODO: Helper similar to bitmap_set() */
return find_next_zero_bit(vm->sbm.sb_states, bit + count, bit) >=
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
bit + count;
}
/*
* Test if all selected subblocks are unplugged.
*/
static bool virtio_mem_sbm_test_sb_unplugged(struct virtio_mem *vm,
unsigned long mb_id, int sb_id,
int count)
{
const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
/* TODO: Helper similar to bitmap_set() */
return find_next_bit(vm->sbm.sb_states, bit + count, bit) >=
bit + count;
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Find the first unplugged subblock. Returns vm->sbm.sbs_per_mb in case there is
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
* none.
*/
static int virtio_mem_sbm_first_unplugged_sb(struct virtio_mem *vm,
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
unsigned long mb_id)
{
const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, 0);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return find_next_zero_bit(vm->sbm.sb_states,
bit + vm->sbm.sbs_per_mb, bit) - bit;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Prepare the subblock bitmap for the next memory block.
*/
static int virtio_mem_sbm_sb_states_prepare_next_mb(struct virtio_mem *vm)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const unsigned long old_nb_mb = vm->sbm.next_mb_id - vm->sbm.first_mb_id;
const unsigned long old_nb_bits = old_nb_mb * vm->sbm.sbs_per_mb;
const unsigned long new_nb_bits = (old_nb_mb + 1) * vm->sbm.sbs_per_mb;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
int old_pages = PFN_UP(BITS_TO_LONGS(old_nb_bits) * sizeof(long));
int new_pages = PFN_UP(BITS_TO_LONGS(new_nb_bits) * sizeof(long));
unsigned long *new_bitmap, *old_bitmap;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (vm->sbm.sb_states && old_pages == new_pages)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return 0;
new_bitmap = vzalloc(new_pages * PAGE_SIZE);
if (!new_bitmap)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return -ENOMEM;
mutex_lock(&vm->hotplug_mutex);
if (vm->sbm.sb_states)
memcpy(new_bitmap, vm->sbm.sb_states, old_pages * PAGE_SIZE);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
old_bitmap = vm->sbm.sb_states;
vm->sbm.sb_states = new_bitmap;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
mutex_unlock(&vm->hotplug_mutex);
vfree(old_bitmap);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return 0;
}
/*
* Test if we could add memory without creating too much offline memory -
* to avoid running OOM if memory is getting onlined deferred.
*/
static bool virtio_mem_could_add_memory(struct virtio_mem *vm, uint64_t size)
{
if (WARN_ON_ONCE(size > vm->offline_threshold))
return false;
return atomic64_read(&vm->offline_size) + size <= vm->offline_threshold;
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Try adding memory to Linux. Will usually only fail if out of memory.
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*
* Must not be called with the vm->hotplug_mutex held (possible deadlock with
* onlining code).
*
* Will not modify the state of memory blocks in virtio-mem.
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*/
static int virtio_mem_add_memory(struct virtio_mem *vm, uint64_t addr,
uint64_t size)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
int rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
virtio-mem: add memory via add_memory_driver_managed() Virtio-mem managed memory is always detected and added by the virtio-mem driver, never using something like the firmware-provided memory map. This is the case after an ordinary system reboot, and has to be guaranteed after kexec. Especially, virtio-mem added memory resources can contain inaccessible parts ("unblocked memory blocks"), blindly forwarding them to a kexec kernel is dangerous, as unplugged memory will get accessed (esp. written). Let's use the new way of adding special driver-managed memory introduced in commit 7b7b27214bba ("mm/memory_hotplug: introduce add_memory_driver_managed()"). This will result in no entries in /sys/firmware/memmap ("raw firmware- provided memory map"), the memory resource will be flagged IORESOURCE_MEM_DRIVER_MANAGED (esp., kexec_file_load() will not place kexec images on this memory), and it is exposed as "System RAM (virtio_mem)" in /proc/iomem, so esp. kexec-tools can properly handle it. Example /proc/iomem before this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM 340000000-347ffffff : System RAM 348000000-34fffffff : System RAM [...] Example /proc/iomem after this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM (virtio_mem) 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM (virtio_mem) 340000000-347ffffff : System RAM (virtio_mem) 348000000-34fffffff : System RAM (virtio_mem) [...] Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: teawater <teawaterz@linux.alibaba.com> Fixes: 5f1f79bbc9e26 ("virtio-mem: Paravirtualized memory hotplug") Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200611093518.5737-1-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
2020-06-11 09:35:18 +00:00
/*
* When force-unloading the driver and we still have memory added to
* Linux, the resource name has to stay.
*/
if (!vm->resource_name) {
vm->resource_name = kstrdup_const("System RAM (virtio_mem)",
GFP_KERNEL);
if (!vm->resource_name)
return -ENOMEM;
}
dev_dbg(&vm->vdev->dev, "adding memory: 0x%llx - 0x%llx\n", addr,
addr + size - 1);
/* Memory might get onlined immediately. */
atomic64_add(size, &vm->offline_size);
rc = add_memory_driver_managed(vm->mgid, addr, size, vm->resource_name,
MHP_MERGE_RESOURCE | MHP_NID_IS_MGID);
if (rc) {
atomic64_sub(size, &vm->offline_size);
dev_warn(&vm->vdev->dev, "adding memory failed: %d\n", rc);
/*
* TODO: Linux MM does not properly clean up yet in all cases
* where adding of memory failed - especially on -ENOMEM.
*/
}
return rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* See virtio_mem_add_memory(): Try adding a single Linux memory block.
*/
static int virtio_mem_sbm_add_mb(struct virtio_mem *vm, unsigned long mb_id)
{
const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
const uint64_t size = memory_block_size_bytes();
return virtio_mem_add_memory(vm, addr, size);
}
/*
* See virtio_mem_add_memory(): Try adding a big block.
*/
static int virtio_mem_bbm_add_bb(struct virtio_mem *vm, unsigned long bb_id)
{
const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
const uint64_t size = vm->bbm.bb_size;
return virtio_mem_add_memory(vm, addr, size);
}
/*
* Try removing memory from Linux. Will only fail if memory blocks aren't
* offline.
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*
* Must not be called with the vm->hotplug_mutex held (possible deadlock with
* onlining code).
*
* Will not modify the state of memory blocks in virtio-mem.
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*/
static int virtio_mem_remove_memory(struct virtio_mem *vm, uint64_t addr,
uint64_t size)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
virtio-mem: don't always trigger the workqueue when offlining memory Let's trigger from offlining code only when we're not allowed to unplug online memory. Handle the other case (memmap possibly freeing up another memory block) when actually removing memory. We now also properly handle the case when removing already offline memory blocks via virtio_mem_mb_remove(). When removing via virtio_mem_remove(), when unloading the driver, virtio_mem_retry() is a NOP and safe to use. While at it, move retry handling when offlining out of virtio_mem_notify_offline(), to share it with Big Block Mode (BBM) soon. This is a preparation for Big Block Mode (BBM), whereby we can see some temporary offlining of memory blocks without actually making progress. Imagine you have a Big Block that spans to Linux memory blocks. Assume the first Linux memory blocks has no unmovable data on it. When we would call offline_and_remove_memory() on the big block, we would 1. Try to offline the first block. Works, notifiers triggered. virtio_mem_retry() called. 2. Try to offline the second block. Does not work. 3. Re-online first block. 4. Exit to main loop, exit workqueue. 5. Retry immediately (due to virtio_mem_retry()), go to 1. The result are endless retries. Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20201112133815.13332-16-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-11-12 13:38:01 +00:00
int rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
dev_dbg(&vm->vdev->dev, "removing memory: 0x%llx - 0x%llx\n", addr,
addr + size - 1);
mm/memory_hotplug: remove nid parameter from remove_memory() and friends There is only a single user remaining. We can simply lookup the nid only used for node offlining purposes when walking our memory blocks. We don't expect to remove multi-nid ranges; and if we'd ever do, we most probably don't care about removing multi-nid ranges that actually result in empty nodes. If ever required, we can detect the "multi-nid" scenario and simply try offlining all online nodes. Link: https://lkml.kernel.org/r/20210712124052.26491-4-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@samba.org> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Vishal Verma <vishal.l.verma@intel.com> Cc: Dave Jiang <dave.jiang@intel.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Nathan Lynch <nathanl@linux.ibm.com> Cc: Laurent Dufour <ldufour@linux.ibm.com> Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com> Cc: Scott Cheloha <cheloha@linux.ibm.com> Cc: Anton Blanchard <anton@ozlabs.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Christophe Leroy <christophe.leroy@c-s.fr> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jia He <justin.he@arm.com> Cc: Joe Perches <joe@perches.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Michel Lespinasse <michel@lespinasse.org> Cc: Mike Rapoport <rppt@kernel.org> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Pankaj Gupta <pankaj.gupta@ionos.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Pierre Morel <pmorel@linux.ibm.com> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Rich Felker <dalias@libc.org> Cc: Sergei Trofimovich <slyfox@gentoo.org> Cc: Thiago Jung Bauermann <bauerman@linux.ibm.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Vitaly Kuznetsov <vkuznets@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: Will Deacon <will@kernel.org> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-08 02:55:09 +00:00
rc = remove_memory(addr, size);
if (!rc) {
atomic64_sub(size, &vm->offline_size);
virtio-mem: don't always trigger the workqueue when offlining memory Let's trigger from offlining code only when we're not allowed to unplug online memory. Handle the other case (memmap possibly freeing up another memory block) when actually removing memory. We now also properly handle the case when removing already offline memory blocks via virtio_mem_mb_remove(). When removing via virtio_mem_remove(), when unloading the driver, virtio_mem_retry() is a NOP and safe to use. While at it, move retry handling when offlining out of virtio_mem_notify_offline(), to share it with Big Block Mode (BBM) soon. This is a preparation for Big Block Mode (BBM), whereby we can see some temporary offlining of memory blocks without actually making progress. Imagine you have a Big Block that spans to Linux memory blocks. Assume the first Linux memory blocks has no unmovable data on it. When we would call offline_and_remove_memory() on the big block, we would 1. Try to offline the first block. Works, notifiers triggered. virtio_mem_retry() called. 2. Try to offline the second block. Does not work. 3. Re-online first block. 4. Exit to main loop, exit workqueue. 5. Retry immediately (due to virtio_mem_retry()), go to 1. The result are endless retries. Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20201112133815.13332-16-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-11-12 13:38:01 +00:00
/*
* We might have freed up memory we can now unplug, retry
* immediately instead of waiting.
*/
virtio_mem_retry(vm);
} else {
dev_dbg(&vm->vdev->dev, "removing memory failed: %d\n", rc);
}
virtio-mem: don't always trigger the workqueue when offlining memory Let's trigger from offlining code only when we're not allowed to unplug online memory. Handle the other case (memmap possibly freeing up another memory block) when actually removing memory. We now also properly handle the case when removing already offline memory blocks via virtio_mem_mb_remove(). When removing via virtio_mem_remove(), when unloading the driver, virtio_mem_retry() is a NOP and safe to use. While at it, move retry handling when offlining out of virtio_mem_notify_offline(), to share it with Big Block Mode (BBM) soon. This is a preparation for Big Block Mode (BBM), whereby we can see some temporary offlining of memory blocks without actually making progress. Imagine you have a Big Block that spans to Linux memory blocks. Assume the first Linux memory blocks has no unmovable data on it. When we would call offline_and_remove_memory() on the big block, we would 1. Try to offline the first block. Works, notifiers triggered. virtio_mem_retry() called. 2. Try to offline the second block. Does not work. 3. Re-online first block. 4. Exit to main loop, exit workqueue. 5. Retry immediately (due to virtio_mem_retry()), go to 1. The result are endless retries. Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20201112133815.13332-16-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-11-12 13:38:01 +00:00
return rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* See virtio_mem_remove_memory(): Try removing a single Linux memory block.
*/
static int virtio_mem_sbm_remove_mb(struct virtio_mem *vm, unsigned long mb_id)
{
const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
const uint64_t size = memory_block_size_bytes();
return virtio_mem_remove_memory(vm, addr, size);
}
/*
* Try offlining and removing memory from Linux.
*
* Must not be called with the vm->hotplug_mutex held (possible deadlock with
* onlining code).
*
* Will not modify the state of memory blocks in virtio-mem.
*/
static int virtio_mem_offline_and_remove_memory(struct virtio_mem *vm,
uint64_t addr,
uint64_t size)
{
virtio-mem: don't always trigger the workqueue when offlining memory Let's trigger from offlining code only when we're not allowed to unplug online memory. Handle the other case (memmap possibly freeing up another memory block) when actually removing memory. We now also properly handle the case when removing already offline memory blocks via virtio_mem_mb_remove(). When removing via virtio_mem_remove(), when unloading the driver, virtio_mem_retry() is a NOP and safe to use. While at it, move retry handling when offlining out of virtio_mem_notify_offline(), to share it with Big Block Mode (BBM) soon. This is a preparation for Big Block Mode (BBM), whereby we can see some temporary offlining of memory blocks without actually making progress. Imagine you have a Big Block that spans to Linux memory blocks. Assume the first Linux memory blocks has no unmovable data on it. When we would call offline_and_remove_memory() on the big block, we would 1. Try to offline the first block. Works, notifiers triggered. virtio_mem_retry() called. 2. Try to offline the second block. Does not work. 3. Re-online first block. 4. Exit to main loop, exit workqueue. 5. Retry immediately (due to virtio_mem_retry()), go to 1. The result are endless retries. Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20201112133815.13332-16-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-11-12 13:38:01 +00:00
int rc;
dev_dbg(&vm->vdev->dev,
"offlining and removing memory: 0x%llx - 0x%llx\n", addr,
addr + size - 1);
mm/memory_hotplug: remove nid parameter from remove_memory() and friends There is only a single user remaining. We can simply lookup the nid only used for node offlining purposes when walking our memory blocks. We don't expect to remove multi-nid ranges; and if we'd ever do, we most probably don't care about removing multi-nid ranges that actually result in empty nodes. If ever required, we can detect the "multi-nid" scenario and simply try offlining all online nodes. Link: https://lkml.kernel.org/r/20210712124052.26491-4-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@samba.org> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Vishal Verma <vishal.l.verma@intel.com> Cc: Dave Jiang <dave.jiang@intel.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Nathan Lynch <nathanl@linux.ibm.com> Cc: Laurent Dufour <ldufour@linux.ibm.com> Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com> Cc: Scott Cheloha <cheloha@linux.ibm.com> Cc: Anton Blanchard <anton@ozlabs.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Christophe Leroy <christophe.leroy@c-s.fr> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jia He <justin.he@arm.com> Cc: Joe Perches <joe@perches.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Michel Lespinasse <michel@lespinasse.org> Cc: Mike Rapoport <rppt@kernel.org> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Pankaj Gupta <pankaj.gupta@ionos.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Pierre Morel <pmorel@linux.ibm.com> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Rich Felker <dalias@libc.org> Cc: Sergei Trofimovich <slyfox@gentoo.org> Cc: Thiago Jung Bauermann <bauerman@linux.ibm.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Vitaly Kuznetsov <vkuznets@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: Will Deacon <will@kernel.org> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-09-08 02:55:09 +00:00
rc = offline_and_remove_memory(addr, size);
if (!rc) {
atomic64_sub(size, &vm->offline_size);
virtio-mem: don't always trigger the workqueue when offlining memory Let's trigger from offlining code only when we're not allowed to unplug online memory. Handle the other case (memmap possibly freeing up another memory block) when actually removing memory. We now also properly handle the case when removing already offline memory blocks via virtio_mem_mb_remove(). When removing via virtio_mem_remove(), when unloading the driver, virtio_mem_retry() is a NOP and safe to use. While at it, move retry handling when offlining out of virtio_mem_notify_offline(), to share it with Big Block Mode (BBM) soon. This is a preparation for Big Block Mode (BBM), whereby we can see some temporary offlining of memory blocks without actually making progress. Imagine you have a Big Block that spans to Linux memory blocks. Assume the first Linux memory blocks has no unmovable data on it. When we would call offline_and_remove_memory() on the big block, we would 1. Try to offline the first block. Works, notifiers triggered. virtio_mem_retry() called. 2. Try to offline the second block. Does not work. 3. Re-online first block. 4. Exit to main loop, exit workqueue. 5. Retry immediately (due to virtio_mem_retry()), go to 1. The result are endless retries. Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20201112133815.13332-16-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-11-12 13:38:01 +00:00
/*
* We might have freed up memory we can now unplug, retry
* immediately instead of waiting.
*/
virtio_mem_retry(vm);
return 0;
}
dev_dbg(&vm->vdev->dev, "offlining and removing memory failed: %d\n", rc);
/*
* We don't really expect this to fail, because we fake-offlined all
* memory already. But it could fail in corner cases.
*/
WARN_ON_ONCE(rc != -ENOMEM && rc != -EBUSY);
return rc == -ENOMEM ? -ENOMEM : -EBUSY;
}
/*
* See virtio_mem_offline_and_remove_memory(): Try offlining and removing
* a single Linux memory block.
*/
static int virtio_mem_sbm_offline_and_remove_mb(struct virtio_mem *vm,
unsigned long mb_id)
{
const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
const uint64_t size = memory_block_size_bytes();
return virtio_mem_offline_and_remove_memory(vm, addr, size);
}
/*
* Try (offlining and) removing memory from Linux in case all subblocks are
* unplugged. Can be called on online and offline memory blocks.
*
* May modify the state of memory blocks in virtio-mem.
*/
static int virtio_mem_sbm_try_remove_unplugged_mb(struct virtio_mem *vm,
unsigned long mb_id)
{
int rc;
/*
* Once all subblocks of a memory block were unplugged, offline and
* remove it.
*/
if (!virtio_mem_sbm_test_sb_unplugged(vm, mb_id, 0, vm->sbm.sbs_per_mb))
return 0;
/* offline_and_remove_memory() works for online and offline memory. */
mutex_unlock(&vm->hotplug_mutex);
rc = virtio_mem_sbm_offline_and_remove_mb(vm, mb_id);
mutex_lock(&vm->hotplug_mutex);
if (!rc)
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_UNUSED);
return rc;
}
/*
* See virtio_mem_offline_and_remove_memory(): Try to offline and remove a
* all Linux memory blocks covered by the big block.
*/
static int virtio_mem_bbm_offline_and_remove_bb(struct virtio_mem *vm,
unsigned long bb_id)
{
const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
const uint64_t size = vm->bbm.bb_size;
return virtio_mem_offline_and_remove_memory(vm, addr, size);
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Trigger the workqueue so the device can perform its magic.
*/
static void virtio_mem_retry(struct virtio_mem *vm)
{
unsigned long flags;
spin_lock_irqsave(&vm->removal_lock, flags);
if (!vm->removing)
queue_work(system_freezable_wq, &vm->wq);
spin_unlock_irqrestore(&vm->removal_lock, flags);
}
static int virtio_mem_translate_node_id(struct virtio_mem *vm, uint16_t node_id)
{
int node = NUMA_NO_NODE;
#if defined(CONFIG_ACPI_NUMA)
if (virtio_has_feature(vm->vdev, VIRTIO_MEM_F_ACPI_PXM))
node = pxm_to_node(node_id);
#endif
return node;
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Test if a virtio-mem device overlaps with the given range. Can be called
* from (notifier) callbacks lockless.
*/
static bool virtio_mem_overlaps_range(struct virtio_mem *vm, uint64_t start,
uint64_t size)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
return start < vm->addr + vm->region_size && vm->addr < start + size;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Test if a virtio-mem device contains a given range. Can be called from
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
* (notifier) callbacks lockless.
*/
static bool virtio_mem_contains_range(struct virtio_mem *vm, uint64_t start,
uint64_t size)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
return start >= vm->addr && start + size <= vm->addr + vm->region_size;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
static int virtio_mem_sbm_notify_going_online(struct virtio_mem *vm,
unsigned long mb_id)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
case VIRTIO_MEM_SBM_MB_OFFLINE:
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return NOTIFY_OK;
default:
break;
}
dev_warn_ratelimited(&vm->vdev->dev,
"memory block onlining denied\n");
return NOTIFY_BAD;
}
static void virtio_mem_sbm_notify_offline(struct virtio_mem *vm,
unsigned long mb_id)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
case VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL:
case VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL:
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
break;
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
case VIRTIO_MEM_SBM_MB_KERNEL:
case VIRTIO_MEM_SBM_MB_MOVABLE:
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_OFFLINE);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
break;
default:
BUG();
break;
}
}
static void virtio_mem_sbm_notify_online(struct virtio_mem *vm,
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
unsigned long mb_id,
unsigned long start_pfn)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const bool is_movable = is_zone_movable_page(pfn_to_page(start_pfn));
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
int new_state;
switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
new_state = VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL;
if (is_movable)
new_state = VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
break;
case VIRTIO_MEM_SBM_MB_OFFLINE:
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
new_state = VIRTIO_MEM_SBM_MB_KERNEL;
if (is_movable)
new_state = VIRTIO_MEM_SBM_MB_MOVABLE;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
break;
default:
BUG();
break;
}
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
virtio_mem_sbm_set_mb_state(vm, mb_id, new_state);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
static void virtio_mem_sbm_notify_going_offline(struct virtio_mem *vm,
unsigned long mb_id)
{
const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size);
unsigned long pfn;
int sb_id;
for (sb_id = 0; sb_id < vm->sbm.sbs_per_mb; sb_id++) {
if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, 1))
continue;
pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
sb_id * vm->sbm.sb_size);
virtio_mem_fake_offline_going_offline(pfn, nr_pages);
}
}
static void virtio_mem_sbm_notify_cancel_offline(struct virtio_mem *vm,
unsigned long mb_id)
{
const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size);
unsigned long pfn;
int sb_id;
for (sb_id = 0; sb_id < vm->sbm.sbs_per_mb; sb_id++) {
if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, 1))
continue;
pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
sb_id * vm->sbm.sb_size);
virtio_mem_fake_offline_cancel_offline(pfn, nr_pages);
}
}
static void virtio_mem_bbm_notify_going_offline(struct virtio_mem *vm,
unsigned long bb_id,
unsigned long pfn,
unsigned long nr_pages)
{
/*
* When marked as "fake-offline", all online memory of this device block
* is allocated by us. Otherwise, we don't have any memory allocated.
*/
if (virtio_mem_bbm_get_bb_state(vm, bb_id) !=
VIRTIO_MEM_BBM_BB_FAKE_OFFLINE)
return;
virtio_mem_fake_offline_going_offline(pfn, nr_pages);
}
static void virtio_mem_bbm_notify_cancel_offline(struct virtio_mem *vm,
unsigned long bb_id,
unsigned long pfn,
unsigned long nr_pages)
{
if (virtio_mem_bbm_get_bb_state(vm, bb_id) !=
VIRTIO_MEM_BBM_BB_FAKE_OFFLINE)
return;
virtio_mem_fake_offline_cancel_offline(pfn, nr_pages);
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* This callback will either be called synchronously from add_memory() or
* asynchronously (e.g., triggered via user space). We have to be careful
* with locking when calling add_memory().
*/
static int virtio_mem_memory_notifier_cb(struct notifier_block *nb,
unsigned long action, void *arg)
{
struct virtio_mem *vm = container_of(nb, struct virtio_mem,
memory_notifier);
struct memory_notify *mhp = arg;
const unsigned long start = PFN_PHYS(mhp->start_pfn);
const unsigned long size = PFN_PHYS(mhp->nr_pages);
int rc = NOTIFY_OK;
unsigned long id;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (!virtio_mem_overlaps_range(vm, start, size))
return NOTIFY_DONE;
if (vm->in_sbm) {
id = virtio_mem_phys_to_mb_id(start);
/*
* In SBM, we add memory in separate memory blocks - we expect
* it to be onlined/offlined in the same granularity. Bail out
* if this ever changes.
*/
if (WARN_ON_ONCE(size != memory_block_size_bytes() ||
!IS_ALIGNED(start, memory_block_size_bytes())))
return NOTIFY_BAD;
} else {
id = virtio_mem_phys_to_bb_id(vm, start);
/*
* In BBM, we only care about onlining/offlining happening
* within a single big block, we don't care about the
* actual granularity as we don't track individual Linux
* memory blocks.
*/
if (WARN_ON_ONCE(id != virtio_mem_phys_to_bb_id(vm, start + size - 1)))
return NOTIFY_BAD;
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Avoid circular locking lockdep warnings. We lock the mutex
* e.g., in MEM_GOING_ONLINE and unlock it in MEM_ONLINE. The
* blocking_notifier_call_chain() has it's own lock, which gets unlocked
* between both notifier calls and will bail out. False positive.
*/
lockdep_off();
switch (action) {
case MEM_GOING_OFFLINE:
mutex_lock(&vm->hotplug_mutex);
if (vm->removing) {
rc = notifier_from_errno(-EBUSY);
mutex_unlock(&vm->hotplug_mutex);
break;
}
vm->hotplug_active = true;
if (vm->in_sbm)
virtio_mem_sbm_notify_going_offline(vm, id);
else
virtio_mem_bbm_notify_going_offline(vm, id,
mhp->start_pfn,
mhp->nr_pages);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
break;
case MEM_GOING_ONLINE:
mutex_lock(&vm->hotplug_mutex);
if (vm->removing) {
rc = notifier_from_errno(-EBUSY);
mutex_unlock(&vm->hotplug_mutex);
break;
}
vm->hotplug_active = true;
if (vm->in_sbm)
rc = virtio_mem_sbm_notify_going_online(vm, id);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
break;
case MEM_OFFLINE:
if (vm->in_sbm)
virtio_mem_sbm_notify_offline(vm, id);
virtio-mem: don't always trigger the workqueue when offlining memory Let's trigger from offlining code only when we're not allowed to unplug online memory. Handle the other case (memmap possibly freeing up another memory block) when actually removing memory. We now also properly handle the case when removing already offline memory blocks via virtio_mem_mb_remove(). When removing via virtio_mem_remove(), when unloading the driver, virtio_mem_retry() is a NOP and safe to use. While at it, move retry handling when offlining out of virtio_mem_notify_offline(), to share it with Big Block Mode (BBM) soon. This is a preparation for Big Block Mode (BBM), whereby we can see some temporary offlining of memory blocks without actually making progress. Imagine you have a Big Block that spans to Linux memory blocks. Assume the first Linux memory blocks has no unmovable data on it. When we would call offline_and_remove_memory() on the big block, we would 1. Try to offline the first block. Works, notifiers triggered. virtio_mem_retry() called. 2. Try to offline the second block. Does not work. 3. Re-online first block. 4. Exit to main loop, exit workqueue. 5. Retry immediately (due to virtio_mem_retry()), go to 1. The result are endless retries. Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20201112133815.13332-16-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-11-12 13:38:01 +00:00
atomic64_add(size, &vm->offline_size);
virtio-mem: don't always trigger the workqueue when offlining memory Let's trigger from offlining code only when we're not allowed to unplug online memory. Handle the other case (memmap possibly freeing up another memory block) when actually removing memory. We now also properly handle the case when removing already offline memory blocks via virtio_mem_mb_remove(). When removing via virtio_mem_remove(), when unloading the driver, virtio_mem_retry() is a NOP and safe to use. While at it, move retry handling when offlining out of virtio_mem_notify_offline(), to share it with Big Block Mode (BBM) soon. This is a preparation for Big Block Mode (BBM), whereby we can see some temporary offlining of memory blocks without actually making progress. Imagine you have a Big Block that spans to Linux memory blocks. Assume the first Linux memory blocks has no unmovable data on it. When we would call offline_and_remove_memory() on the big block, we would 1. Try to offline the first block. Works, notifiers triggered. virtio_mem_retry() called. 2. Try to offline the second block. Does not work. 3. Re-online first block. 4. Exit to main loop, exit workqueue. 5. Retry immediately (due to virtio_mem_retry()), go to 1. The result are endless retries. Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20201112133815.13332-16-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-11-12 13:38:01 +00:00
/*
* Trigger the workqueue. Now that we have some offline memory,
* maybe we can handle pending unplug requests.
*/
if (!unplug_online)
virtio_mem_retry(vm);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
vm->hotplug_active = false;
mutex_unlock(&vm->hotplug_mutex);
break;
case MEM_ONLINE:
if (vm->in_sbm)
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
virtio_mem_sbm_notify_online(vm, id, mhp->start_pfn);
atomic64_sub(size, &vm->offline_size);
/*
* Start adding more memory once we onlined half of our
* threshold. Don't trigger if it's possibly due to our actipn
* (e.g., us adding memory which gets onlined immediately from
* the core).
*/
if (!atomic_read(&vm->wq_active) &&
virtio_mem_could_add_memory(vm, vm->offline_threshold / 2))
virtio_mem_retry(vm);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
vm->hotplug_active = false;
mutex_unlock(&vm->hotplug_mutex);
break;
case MEM_CANCEL_OFFLINE:
if (!vm->hotplug_active)
break;
if (vm->in_sbm)
virtio_mem_sbm_notify_cancel_offline(vm, id);
else
virtio_mem_bbm_notify_cancel_offline(vm, id,
mhp->start_pfn,
mhp->nr_pages);
vm->hotplug_active = false;
mutex_unlock(&vm->hotplug_mutex);
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
case MEM_CANCEL_ONLINE:
if (!vm->hotplug_active)
break;
vm->hotplug_active = false;
mutex_unlock(&vm->hotplug_mutex);
break;
default:
break;
}
lockdep_on();
return rc;
}
/*
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
* Set a range of pages PG_offline. Remember pages that were never onlined
* (via generic_online_page()) using PageDirty().
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*/
static void virtio_mem_set_fake_offline(unsigned long pfn,
unsigned long nr_pages, bool onlined)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
virtio-mem: use page_offline_(start|end) when setting PageOffline() Let's properly use page_offline_(start|end) to synchronize setting PageOffline(), so we won't have valid page access to unplugged memory regions from /proc/kcore. Existing balloon implementations usually allow reading inflated memory; doing so might result in unnecessary overhead in the hypervisor, which is currently the case with virtio-mem. For future virtio-mem use cases, it will be different when using shmem, huge pages, !anonymous private mappings, ... as backing storage for a VM. virtio-mem unplugged memory must no longer be accessed and access might result in undefined behavior. There will be a virtio spec extension to document this change, including a new feature flag indicating the changed behavior. We really don't want to race against PFN walkers reading random page content. Link: https://lkml.kernel.org/r/20210526093041.8800-6-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Acked-by: Michael S. Tsirkin <mst@redhat.com> Acked-by: Mike Rapoport <rppt@linux.ibm.com> Reviewed-by: Oscar Salvador <osalvador@suse.de> Cc: Aili Yao <yaoaili@kingsoft.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Cc: Haiyang Zhang <haiyangz@microsoft.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Jiri Bohac <jbohac@suse.cz> Cc: "K. Y. Srinivasan" <kys@microsoft.com> Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Roman Gushchin <guro@fb.com> Cc: Stephen Hemminger <sthemmin@microsoft.com> Cc: Steven Price <steven.price@arm.com> Cc: Wei Liu <wei.liu@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-07-01 01:50:18 +00:00
page_offline_begin();
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
for (; nr_pages--; pfn++) {
struct page *page = pfn_to_page(pfn);
__SetPageOffline(page);
if (!onlined) {
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
SetPageDirty(page);
/* FIXME: remove after cleanups */
ClearPageReserved(page);
}
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
}
virtio-mem: use page_offline_(start|end) when setting PageOffline() Let's properly use page_offline_(start|end) to synchronize setting PageOffline(), so we won't have valid page access to unplugged memory regions from /proc/kcore. Existing balloon implementations usually allow reading inflated memory; doing so might result in unnecessary overhead in the hypervisor, which is currently the case with virtio-mem. For future virtio-mem use cases, it will be different when using shmem, huge pages, !anonymous private mappings, ... as backing storage for a VM. virtio-mem unplugged memory must no longer be accessed and access might result in undefined behavior. There will be a virtio spec extension to document this change, including a new feature flag indicating the changed behavior. We really don't want to race against PFN walkers reading random page content. Link: https://lkml.kernel.org/r/20210526093041.8800-6-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Acked-by: Michael S. Tsirkin <mst@redhat.com> Acked-by: Mike Rapoport <rppt@linux.ibm.com> Reviewed-by: Oscar Salvador <osalvador@suse.de> Cc: Aili Yao <yaoaili@kingsoft.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Cc: Haiyang Zhang <haiyangz@microsoft.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Jiri Bohac <jbohac@suse.cz> Cc: "K. Y. Srinivasan" <kys@microsoft.com> Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Roman Gushchin <guro@fb.com> Cc: Stephen Hemminger <sthemmin@microsoft.com> Cc: Steven Price <steven.price@arm.com> Cc: Wei Liu <wei.liu@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-07-01 01:50:18 +00:00
page_offline_end();
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
* Clear PG_offline from a range of pages. If the pages were never onlined,
* (via generic_online_page()), clear PageDirty().
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*/
static void virtio_mem_clear_fake_offline(unsigned long pfn,
unsigned long nr_pages, bool onlined)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
for (; nr_pages--; pfn++) {
struct page *page = pfn_to_page(pfn);
__ClearPageOffline(page);
if (!onlined)
ClearPageDirty(page);
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Release a range of fake-offline pages to the buddy, effectively
* fake-onlining them.
*/
static void virtio_mem_fake_online(unsigned long pfn, unsigned long nr_pages)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
unsigned long order = MAX_PAGE_ORDER;
unsigned long i;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* We might get called for ranges that don't cover properly aligned
* MAX_PAGE_ORDER pages; however, we can only online properly aligned
* pages with an order of MAX_PAGE_ORDER at maximum.
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*/
while (!IS_ALIGNED(pfn | nr_pages, 1 << order))
order--;
for (i = 0; i < nr_pages; i += 1 << order) {
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
struct page *page = pfn_to_page(pfn + i);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
/*
* If the page is PageDirty(), it was kept fake-offline when
* onlining the memory block. Otherwise, it was allocated
* using alloc_contig_range(). All pages in a subblock are
* alike.
*/
if (PageDirty(page)) {
virtio_mem_clear_fake_offline(pfn + i, 1 << order, false);
generic_online_page(page, order);
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
} else {
virtio_mem_clear_fake_offline(pfn + i, 1 << order, true);
free_contig_range(pfn + i, 1 << order);
adjust_managed_page_count(page, 1 << order);
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
}
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Try to allocate a range, marking pages fake-offline, effectively
* fake-offlining them.
*/
static int virtio_mem_fake_offline(struct virtio_mem *vm, unsigned long pfn,
unsigned long nr_pages)
{
const bool is_movable = is_zone_movable_page(pfn_to_page(pfn));
int rc, retry_count;
/*
* TODO: We want an alloc_contig_range() mode that tries to allocate
* harder (e.g., dealing with temporarily pinned pages, PCP), especially
* with ZONE_MOVABLE. So for now, retry a couple of times with
* ZONE_MOVABLE before giving up - because that zone is supposed to give
* some guarantees.
*/
for (retry_count = 0; retry_count < 5; retry_count++) {
/*
* If the config changed, stop immediately and go back to the
* main loop: avoid trying to keep unplugging if the device
* might have decided to not remove any more memory.
*/
if (atomic_read(&vm->config_changed))
return -EAGAIN;
rc = alloc_contig_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE,
GFP_KERNEL);
if (rc == -ENOMEM)
/* whoops, out of memory */
return rc;
else if (rc && !is_movable)
break;
else if (rc)
continue;
virtio_mem_set_fake_offline(pfn, nr_pages, true);
adjust_managed_page_count(pfn_to_page(pfn), -nr_pages);
return 0;
}
return -EBUSY;
}
/*
* Handle fake-offline pages when memory is going offline - such that the
* pages can be skipped by mm-core when offlining.
*/
static void virtio_mem_fake_offline_going_offline(unsigned long pfn,
unsigned long nr_pages)
{
struct page *page;
unsigned long i;
/*
* Drop our reference to the pages so the memory can get offlined
* and add the unplugged pages to the managed page counters (so
* offlining code can correctly subtract them again).
*/
adjust_managed_page_count(pfn_to_page(pfn), nr_pages);
/* Drop our reference to the pages so the memory can get offlined. */
for (i = 0; i < nr_pages; i++) {
page = pfn_to_page(pfn + i);
if (WARN_ON(!page_ref_dec_and_test(page)))
dump_page(page, "fake-offline page referenced");
}
}
/*
* Handle fake-offline pages when memory offlining is canceled - to undo
* what we did in virtio_mem_fake_offline_going_offline().
*/
static void virtio_mem_fake_offline_cancel_offline(unsigned long pfn,
unsigned long nr_pages)
{
unsigned long i;
/*
* Get the reference we dropped when going offline and subtract the
* unplugged pages from the managed page counters.
*/
adjust_managed_page_count(pfn_to_page(pfn), -nr_pages);
for (i = 0; i < nr_pages; i++)
page_ref_inc(pfn_to_page(pfn + i));
}
static void virtio_mem_online_page(struct virtio_mem *vm,
struct page *page, unsigned int order)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const unsigned long start = page_to_phys(page);
const unsigned long end = start + PFN_PHYS(1 << order);
unsigned long addr, next, id, sb_id, count;
bool do_online;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* We can get called with any order up to MAX_PAGE_ORDER. If our subblock
* size is smaller than that and we have a mixture of plugged and
* unplugged subblocks within such a page, we have to process in
* smaller granularity. In that case we'll adjust the order exactly once
* within the loop.
*/
for (addr = start; addr < end; ) {
next = addr + PFN_PHYS(1 << order);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (vm->in_sbm) {
id = virtio_mem_phys_to_mb_id(addr);
sb_id = virtio_mem_phys_to_sb_id(vm, addr);
count = virtio_mem_phys_to_sb_id(vm, next - 1) - sb_id + 1;
if (virtio_mem_sbm_test_sb_plugged(vm, id, sb_id, count)) {
/* Fully plugged. */
do_online = true;
} else if (count == 1 ||
virtio_mem_sbm_test_sb_unplugged(vm, id, sb_id, count)) {
/* Fully unplugged. */
do_online = false;
} else {
/*
* Mixture, process sub-blocks instead. This
* will be at least the size of a pageblock.
* We'll run into this case exactly once.
*/
order = ilog2(vm->sbm.sb_size) - PAGE_SHIFT;
do_online = virtio_mem_sbm_test_sb_plugged(vm, id, sb_id, 1);
continue;
}
} else {
/*
* If the whole block is marked fake offline, keep
* everything that way.
*/
id = virtio_mem_phys_to_bb_id(vm, addr);
do_online = virtio_mem_bbm_get_bb_state(vm, id) !=
VIRTIO_MEM_BBM_BB_FAKE_OFFLINE;
}
if (do_online)
generic_online_page(pfn_to_page(PFN_DOWN(addr)), order);
else
virtio_mem_set_fake_offline(PFN_DOWN(addr), 1 << order,
false);
addr = next;
}
}
static void virtio_mem_online_page_cb(struct page *page, unsigned int order)
{
const unsigned long addr = page_to_phys(page);
struct virtio_mem *vm;
rcu_read_lock();
list_for_each_entry_rcu(vm, &virtio_mem_devices, next) {
/*
* Pages we're onlining will never cross memory blocks and,
* therefore, not virtio-mem devices.
*/
if (!virtio_mem_contains_range(vm, addr, PFN_PHYS(1 << order)))
continue;
/*
* virtio_mem_set_fake_offline() might sleep. We can safely
* drop the RCU lock at this point because the device
* cannot go away. See virtio_mem_remove() how races
* between memory onlining and device removal are handled.
*/
rcu_read_unlock();
virtio_mem_online_page(vm, page, order);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return;
}
rcu_read_unlock();
/* not virtio-mem memory, but e.g., a DIMM. online it */
generic_online_page(page, order);
}
static uint64_t virtio_mem_send_request(struct virtio_mem *vm,
const struct virtio_mem_req *req)
{
struct scatterlist *sgs[2], sg_req, sg_resp;
unsigned int len;
int rc;
/* don't use the request residing on the stack (vaddr) */
vm->req = *req;
/* out: buffer for request */
sg_init_one(&sg_req, &vm->req, sizeof(vm->req));
sgs[0] = &sg_req;
/* in: buffer for response */
sg_init_one(&sg_resp, &vm->resp, sizeof(vm->resp));
sgs[1] = &sg_resp;
rc = virtqueue_add_sgs(vm->vq, sgs, 1, 1, vm, GFP_KERNEL);
if (rc < 0)
return rc;
virtqueue_kick(vm->vq);
/* wait for a response */
wait_event(vm->host_resp, virtqueue_get_buf(vm->vq, &len));
return virtio16_to_cpu(vm->vdev, vm->resp.type);
}
static int virtio_mem_send_plug_request(struct virtio_mem *vm, uint64_t addr,
uint64_t size)
{
const uint64_t nb_vm_blocks = size / vm->device_block_size;
const struct virtio_mem_req req = {
.type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_PLUG),
.u.plug.addr = cpu_to_virtio64(vm->vdev, addr),
.u.plug.nb_blocks = cpu_to_virtio16(vm->vdev, nb_vm_blocks),
};
int rc = -ENOMEM;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (atomic_read(&vm->config_changed))
return -EAGAIN;
dev_dbg(&vm->vdev->dev, "plugging memory: 0x%llx - 0x%llx\n", addr,
addr + size - 1);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
switch (virtio_mem_send_request(vm, &req)) {
case VIRTIO_MEM_RESP_ACK:
vm->plugged_size += size;
return 0;
case VIRTIO_MEM_RESP_NACK:
rc = -EAGAIN;
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
case VIRTIO_MEM_RESP_BUSY:
rc = -ETXTBSY;
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
case VIRTIO_MEM_RESP_ERROR:
rc = -EINVAL;
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
default:
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
dev_dbg(&vm->vdev->dev, "plugging memory failed: %d\n", rc);
return rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
static int virtio_mem_send_unplug_request(struct virtio_mem *vm, uint64_t addr,
uint64_t size)
{
const uint64_t nb_vm_blocks = size / vm->device_block_size;
const struct virtio_mem_req req = {
.type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_UNPLUG),
.u.unplug.addr = cpu_to_virtio64(vm->vdev, addr),
.u.unplug.nb_blocks = cpu_to_virtio16(vm->vdev, nb_vm_blocks),
};
int rc = -ENOMEM;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (atomic_read(&vm->config_changed))
return -EAGAIN;
dev_dbg(&vm->vdev->dev, "unplugging memory: 0x%llx - 0x%llx\n", addr,
addr + size - 1);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
switch (virtio_mem_send_request(vm, &req)) {
case VIRTIO_MEM_RESP_ACK:
vm->plugged_size -= size;
return 0;
case VIRTIO_MEM_RESP_BUSY:
rc = -ETXTBSY;
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
case VIRTIO_MEM_RESP_ERROR:
rc = -EINVAL;
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
default:
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
dev_dbg(&vm->vdev->dev, "unplugging memory failed: %d\n", rc);
return rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
static int virtio_mem_send_unplug_all_request(struct virtio_mem *vm)
{
const struct virtio_mem_req req = {
.type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_UNPLUG_ALL),
};
int rc = -ENOMEM;
dev_dbg(&vm->vdev->dev, "unplugging all memory");
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
switch (virtio_mem_send_request(vm, &req)) {
case VIRTIO_MEM_RESP_ACK:
vm->unplug_all_required = false;
vm->plugged_size = 0;
/* usable region might have shrunk */
atomic_set(&vm->config_changed, 1);
return 0;
case VIRTIO_MEM_RESP_BUSY:
rc = -ETXTBSY;
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
default:
break;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
dev_dbg(&vm->vdev->dev, "unplugging all memory failed: %d\n", rc);
return rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Plug selected subblocks. Updates the plugged state, but not the state
* of the memory block.
*/
static int virtio_mem_sbm_plug_sb(struct virtio_mem *vm, unsigned long mb_id,
int sb_id, int count)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id) +
sb_id * vm->sbm.sb_size;
const uint64_t size = count * vm->sbm.sb_size;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
int rc;
rc = virtio_mem_send_plug_request(vm, addr, size);
if (!rc)
virtio_mem_sbm_set_sb_plugged(vm, mb_id, sb_id, count);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return rc;
}
/*
* Unplug selected subblocks. Updates the plugged state, but not the state
* of the memory block.
*/
static int virtio_mem_sbm_unplug_sb(struct virtio_mem *vm, unsigned long mb_id,
int sb_id, int count)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id) +
sb_id * vm->sbm.sb_size;
const uint64_t size = count * vm->sbm.sb_size;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
int rc;
rc = virtio_mem_send_unplug_request(vm, addr, size);
if (!rc)
virtio_mem_sbm_set_sb_unplugged(vm, mb_id, sb_id, count);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return rc;
}
/*
* Request to unplug a big block.
*
* Will not modify the state of the big block.
*/
static int virtio_mem_bbm_unplug_bb(struct virtio_mem *vm, unsigned long bb_id)
{
const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
const uint64_t size = vm->bbm.bb_size;
return virtio_mem_send_unplug_request(vm, addr, size);
}
/*
* Request to plug a big block.
*
* Will not modify the state of the big block.
*/
static int virtio_mem_bbm_plug_bb(struct virtio_mem *vm, unsigned long bb_id)
{
const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
const uint64_t size = vm->bbm.bb_size;
return virtio_mem_send_plug_request(vm, addr, size);
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Unplug the desired number of plugged subblocks of a offline or not-added
* memory block. Will fail if any subblock cannot get unplugged (instead of
* skipping it).
*
* Will not modify the state of the memory block.
*
* Note: can fail after some subblocks were unplugged.
*/
static int virtio_mem_sbm_unplug_any_sb_raw(struct virtio_mem *vm,
unsigned long mb_id, uint64_t *nb_sb)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
int sb_id, count;
int rc;
sb_id = vm->sbm.sbs_per_mb - 1;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
while (*nb_sb) {
/* Find the next candidate subblock */
while (sb_id >= 0 &&
virtio_mem_sbm_test_sb_unplugged(vm, mb_id, sb_id, 1))
sb_id--;
if (sb_id < 0)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
break;
/* Try to unplug multiple subblocks at a time */
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
count = 1;
while (count < *nb_sb && sb_id > 0 &&
virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id - 1, 1)) {
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
count++;
sb_id--;
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
rc = virtio_mem_sbm_unplug_sb(vm, mb_id, sb_id, count);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc)
return rc;
*nb_sb -= count;
sb_id--;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
return 0;
}
/*
* Unplug all plugged subblocks of an offline or not-added memory block.
*
* Will not modify the state of the memory block.
*
* Note: can fail after some subblocks were unplugged.
*/
static int virtio_mem_sbm_unplug_mb(struct virtio_mem *vm, unsigned long mb_id)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
uint64_t nb_sb = vm->sbm.sbs_per_mb;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return virtio_mem_sbm_unplug_any_sb_raw(vm, mb_id, &nb_sb);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* Prepare tracking data for the next memory block.
*/
static int virtio_mem_sbm_prepare_next_mb(struct virtio_mem *vm,
unsigned long *mb_id)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
int rc;
if (vm->sbm.next_mb_id > vm->sbm.last_usable_mb_id)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return -ENOSPC;
/* Resize the state array if required. */
rc = virtio_mem_sbm_mb_states_prepare_next_mb(vm);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc)
return rc;
/* Resize the subblock bitmap if required. */
rc = virtio_mem_sbm_sb_states_prepare_next_mb(vm);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc)
return rc;
vm->sbm.mb_count[VIRTIO_MEM_SBM_MB_UNUSED]++;
*mb_id = vm->sbm.next_mb_id++;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return 0;
}
/*
* Try to plug the desired number of subblocks and add the memory block
* to Linux.
*
* Will modify the state of the memory block.
*/
static int virtio_mem_sbm_plug_and_add_mb(struct virtio_mem *vm,
unsigned long mb_id, uint64_t *nb_sb)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const int count = min_t(int, *nb_sb, vm->sbm.sbs_per_mb);
int rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (WARN_ON_ONCE(!count))
return -EINVAL;
/*
* Plug the requested number of subblocks before adding it to linux,
* so that onlining will directly online all plugged subblocks.
*/
rc = virtio_mem_sbm_plug_sb(vm, mb_id, 0, count);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc)
return rc;
/*
* Mark the block properly offline before adding it to Linux,
* so the memory notifiers will find the block in the right state.
*/
if (count == vm->sbm.sbs_per_mb)
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_OFFLINE);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
else
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Add the memory block to linux - if that fails, try to unplug. */
rc = virtio_mem_sbm_add_mb(vm, mb_id);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc) {
int new_state = VIRTIO_MEM_SBM_MB_UNUSED;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (virtio_mem_sbm_unplug_sb(vm, mb_id, 0, count))
new_state = VIRTIO_MEM_SBM_MB_PLUGGED;
virtio_mem_sbm_set_mb_state(vm, mb_id, new_state);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return rc;
}
*nb_sb -= count;
return 0;
}
/*
* Try to plug the desired number of subblocks of a memory block that
* is already added to Linux.
*
* Will modify the state of the memory block.
*
* Note: Can fail after some subblocks were successfully plugged.
*/
static int virtio_mem_sbm_plug_any_sb(struct virtio_mem *vm,
unsigned long mb_id, uint64_t *nb_sb)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
unsigned long pfn, nr_pages;
int sb_id, count;
int rc;
if (WARN_ON_ONCE(!*nb_sb))
return -EINVAL;
while (*nb_sb) {
sb_id = virtio_mem_sbm_first_unplugged_sb(vm, mb_id);
if (sb_id >= vm->sbm.sbs_per_mb)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
break;
count = 1;
while (count < *nb_sb &&
sb_id + count < vm->sbm.sbs_per_mb &&
!virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id + count, 1))
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
count++;
rc = virtio_mem_sbm_plug_sb(vm, mb_id, sb_id, count);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc)
return rc;
*nb_sb -= count;
if (old_state == VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
continue;
/* fake-online the pages if the memory block is online */
pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
sb_id * vm->sbm.sb_size);
nr_pages = PFN_DOWN(count * vm->sbm.sb_size);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
virtio_mem_fake_online(pfn, nr_pages);
}
if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, 0, vm->sbm.sbs_per_mb))
virtio_mem_sbm_set_mb_state(vm, mb_id, old_state - 1);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return 0;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
static int virtio_mem_sbm_plug_request(struct virtio_mem *vm, uint64_t diff)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
const int mb_states[] = {
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
};
uint64_t nb_sb = diff / vm->sbm.sb_size;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
unsigned long mb_id;
int rc, i;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (!nb_sb)
return 0;
/* Don't race with onlining/offlining */
mutex_lock(&vm->hotplug_mutex);
for (i = 0; i < ARRAY_SIZE(mb_states); i++) {
virtio_mem_sbm_for_each_mb(vm, mb_id, mb_states[i]) {
rc = virtio_mem_sbm_plug_any_sb(vm, mb_id, &nb_sb);
if (rc || !nb_sb)
goto out_unlock;
cond_resched();
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/*
* We won't be working on online/offline memory blocks from this point,
* so we can't race with memory onlining/offlining. Drop the mutex.
*/
mutex_unlock(&vm->hotplug_mutex);
/* Try to plug and add unused blocks */
virtio_mem_sbm_for_each_mb(vm, mb_id, VIRTIO_MEM_SBM_MB_UNUSED) {
if (!virtio_mem_could_add_memory(vm, memory_block_size_bytes()))
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return -ENOSPC;
rc = virtio_mem_sbm_plug_and_add_mb(vm, mb_id, &nb_sb);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc || !nb_sb)
return rc;
cond_resched();
}
/* Try to prepare, plug and add new blocks */
while (nb_sb) {
if (!virtio_mem_could_add_memory(vm, memory_block_size_bytes()))
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return -ENOSPC;
rc = virtio_mem_sbm_prepare_next_mb(vm, &mb_id);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc)
return rc;
rc = virtio_mem_sbm_plug_and_add_mb(vm, mb_id, &nb_sb);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc)
return rc;
cond_resched();
}
return 0;
out_unlock:
mutex_unlock(&vm->hotplug_mutex);
return rc;
}
/*
* Plug a big block and add it to Linux.
*
* Will modify the state of the big block.
*/
static int virtio_mem_bbm_plug_and_add_bb(struct virtio_mem *vm,
unsigned long bb_id)
{
int rc;
if (WARN_ON_ONCE(virtio_mem_bbm_get_bb_state(vm, bb_id) !=
VIRTIO_MEM_BBM_BB_UNUSED))
return -EINVAL;
rc = virtio_mem_bbm_plug_bb(vm, bb_id);
if (rc)
return rc;
virtio_mem_bbm_set_bb_state(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED);
rc = virtio_mem_bbm_add_bb(vm, bb_id);
if (rc) {
if (!virtio_mem_bbm_unplug_bb(vm, bb_id))
virtio_mem_bbm_set_bb_state(vm, bb_id,
VIRTIO_MEM_BBM_BB_UNUSED);
else
/* Retry from the main loop. */
virtio_mem_bbm_set_bb_state(vm, bb_id,
VIRTIO_MEM_BBM_BB_PLUGGED);
return rc;
}
return 0;
}
/*
* Prepare tracking data for the next big block.
*/
static int virtio_mem_bbm_prepare_next_bb(struct virtio_mem *vm,
unsigned long *bb_id)
{
int rc;
if (vm->bbm.next_bb_id > vm->bbm.last_usable_bb_id)
return -ENOSPC;
/* Resize the big block state array if required. */
rc = virtio_mem_bbm_bb_states_prepare_next_bb(vm);
if (rc)
return rc;
vm->bbm.bb_count[VIRTIO_MEM_BBM_BB_UNUSED]++;
*bb_id = vm->bbm.next_bb_id;
vm->bbm.next_bb_id++;
return 0;
}
static int virtio_mem_bbm_plug_request(struct virtio_mem *vm, uint64_t diff)
{
uint64_t nb_bb = diff / vm->bbm.bb_size;
unsigned long bb_id;
int rc;
if (!nb_bb)
return 0;
/* Try to plug and add unused big blocks */
virtio_mem_bbm_for_each_bb(vm, bb_id, VIRTIO_MEM_BBM_BB_UNUSED) {
if (!virtio_mem_could_add_memory(vm, vm->bbm.bb_size))
return -ENOSPC;
rc = virtio_mem_bbm_plug_and_add_bb(vm, bb_id);
if (!rc)
nb_bb--;
if (rc || !nb_bb)
return rc;
cond_resched();
}
/* Try to prepare, plug and add new big blocks */
while (nb_bb) {
if (!virtio_mem_could_add_memory(vm, vm->bbm.bb_size))
return -ENOSPC;
rc = virtio_mem_bbm_prepare_next_bb(vm, &bb_id);
if (rc)
return rc;
rc = virtio_mem_bbm_plug_and_add_bb(vm, bb_id);
if (!rc)
nb_bb--;
if (rc)
return rc;
cond_resched();
}
return 0;
}
/*
* Try to plug the requested amount of memory.
*/
static int virtio_mem_plug_request(struct virtio_mem *vm, uint64_t diff)
{
if (vm->in_sbm)
return virtio_mem_sbm_plug_request(vm, diff);
return virtio_mem_bbm_plug_request(vm, diff);
}
/*
* Unplug the desired number of plugged subblocks of an offline memory block.
* Will fail if any subblock cannot get unplugged (instead of skipping it).
*
* Will modify the state of the memory block. Might temporarily drop the
* hotplug_mutex.
*
* Note: Can fail after some subblocks were successfully unplugged.
*/
static int virtio_mem_sbm_unplug_any_sb_offline(struct virtio_mem *vm,
unsigned long mb_id,
uint64_t *nb_sb)
{
int rc;
rc = virtio_mem_sbm_unplug_any_sb_raw(vm, mb_id, nb_sb);
/* some subblocks might have been unplugged even on failure */
if (!virtio_mem_sbm_test_sb_plugged(vm, mb_id, 0, vm->sbm.sbs_per_mb))
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
if (rc)
return rc;
if (virtio_mem_sbm_test_sb_unplugged(vm, mb_id, 0, vm->sbm.sbs_per_mb)) {
/*
* Remove the block from Linux - this should never fail.
* Hinder the block from getting onlined by marking it
* unplugged. Temporarily drop the mutex, so
* any pending GOING_ONLINE requests can be serviced/rejected.
*/
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_UNUSED);
mutex_unlock(&vm->hotplug_mutex);
rc = virtio_mem_sbm_remove_mb(vm, mb_id);
BUG_ON(rc);
mutex_lock(&vm->hotplug_mutex);
}
return 0;
}
/*
* Unplug the given plugged subblocks of an online memory block.
*
* Will modify the state of the memory block.
*/
static int virtio_mem_sbm_unplug_sb_online(struct virtio_mem *vm,
unsigned long mb_id, int sb_id,
int count)
{
const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size) * count;
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
unsigned long start_pfn;
int rc;
start_pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
sb_id * vm->sbm.sb_size);
rc = virtio_mem_fake_offline(vm, start_pfn, nr_pages);
if (rc)
return rc;
/* Try to unplug the allocated memory */
rc = virtio_mem_sbm_unplug_sb(vm, mb_id, sb_id, count);
if (rc) {
/* Return the memory to the buddy. */
virtio_mem_fake_online(start_pfn, nr_pages);
return rc;
}
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
switch (old_state) {
case VIRTIO_MEM_SBM_MB_KERNEL:
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL);
break;
case VIRTIO_MEM_SBM_MB_MOVABLE:
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL);
break;
}
return 0;
}
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
/*
* Unplug the desired number of plugged subblocks of an online memory block.
* Will skip subblock that are busy.
*
* Will modify the state of the memory block. Might temporarily drop the
* hotplug_mutex.
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
*
* Note: Can fail after some subblocks were successfully unplugged. Can
* return 0 even if subblocks were busy and could not get unplugged.
*/
static int virtio_mem_sbm_unplug_any_sb_online(struct virtio_mem *vm,
unsigned long mb_id,
uint64_t *nb_sb)
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
{
int rc, sb_id;
/* If possible, try to unplug the complete block in one shot. */
if (*nb_sb >= vm->sbm.sbs_per_mb &&
virtio_mem_sbm_test_sb_plugged(vm, mb_id, 0, vm->sbm.sbs_per_mb)) {
rc = virtio_mem_sbm_unplug_sb_online(vm, mb_id, 0,
vm->sbm.sbs_per_mb);
if (!rc) {
*nb_sb -= vm->sbm.sbs_per_mb;
goto unplugged;
} else if (rc != -EBUSY)
return rc;
}
/* Fallback to single subblocks. */
for (sb_id = vm->sbm.sbs_per_mb - 1; sb_id >= 0 && *nb_sb; sb_id--) {
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
/* Find the next candidate subblock */
while (sb_id >= 0 &&
!virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, 1))
sb_id--;
if (sb_id < 0)
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
break;
rc = virtio_mem_sbm_unplug_sb_online(vm, mb_id, sb_id, 1);
if (rc == -EBUSY)
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
continue;
else if (rc)
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
return rc;
*nb_sb -= 1;
}
unplugged:
rc = virtio_mem_sbm_try_remove_unplugged_mb(vm, mb_id);
if (rc)
vm->sbm.have_unplugged_mb = 1;
/* Ignore errors, this is not critical. We'll retry later. */
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
return 0;
}
/*
* Unplug the desired number of plugged subblocks of a memory block that is
* already added to Linux. Will skip subblock of online memory blocks that are
* busy (by the OS). Will fail if any subblock that's not busy cannot get
* unplugged.
*
* Will modify the state of the memory block. Might temporarily drop the
* hotplug_mutex.
*
* Note: Can fail after some subblocks were successfully unplugged. Can
* return 0 even if subblocks were busy and could not get unplugged.
*/
static int virtio_mem_sbm_unplug_any_sb(struct virtio_mem *vm,
unsigned long mb_id,
uint64_t *nb_sb)
{
const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
switch (old_state) {
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
case VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL:
case VIRTIO_MEM_SBM_MB_KERNEL:
case VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL:
case VIRTIO_MEM_SBM_MB_MOVABLE:
return virtio_mem_sbm_unplug_any_sb_online(vm, mb_id, nb_sb);
case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
case VIRTIO_MEM_SBM_MB_OFFLINE:
return virtio_mem_sbm_unplug_any_sb_offline(vm, mb_id, nb_sb);
}
return -EINVAL;
}
static int virtio_mem_sbm_unplug_request(struct virtio_mem *vm, uint64_t diff)
{
const int mb_states[] = {
VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
VIRTIO_MEM_SBM_MB_OFFLINE,
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
VIRTIO_MEM_SBM_MB_MOVABLE,
VIRTIO_MEM_SBM_MB_KERNEL,
};
uint64_t nb_sb = diff / vm->sbm.sb_size;
unsigned long mb_id;
int rc, i;
if (!nb_sb)
return 0;
/*
* We'll drop the mutex a couple of times when it is safe to do so.
* This might result in some blocks switching the state (online/offline)
* and we could miss them in this run - we will retry again later.
*/
mutex_lock(&vm->hotplug_mutex);
/*
* We try unplug from partially plugged blocks first, to try removing
virtio-mem: prioritize unplug from ZONE_MOVABLE in Sub Block Mode Until now, memory provided by a single virtio-mem device was usually either onlined completely to ZONE_MOVABLE (online_movable) or to ZONE_NORMAL (online_kernel); however, that will change in the future. There are two reasons why we want to track to which zone a memory blocks belongs to and prioritize ZONE_MOVABLE blocks: 1) Memory managed by ZONE_MOVABLE can more likely get unplugged, therefore, resulting in a faster memory hotunplug process. Further, we can more reliably unplug and remove complete memory blocks, removing metadata allocated for the whole memory block. 2) We want to avoid corner cases where unplugging with the current scheme (highest to lowest address) could result in accidential zone imbalances, whereby we remove too much ZONE_NORMAL memory for ZONE_MOVABLE memory of the same device. Let's track the zone via memory block states and try unplug from ZONE_MOVABLE first. Rename VIRTIO_MEM_SBM_MB_ONLINE* to VIRTIO_MEM_SBM_MB_KERNEL* to avoid even longer state names. In commit 27f852795a06 ("virtio-mem: don't special-case ZONE_MOVABLE"), we removed slightly similar tracking for fully plugged memory blocks to support unplugging from ZONE_MOVABLE at all -- as we didn't allow partially plugged memory blocks in ZONE_MOVABLE before that. That commit already mentioned "In the future, we might want to remember the zone again and use the information when (un)plugging memory." Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20210602185720.31821-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2021-06-02 18:57:18 +00:00
* whole memory blocks along with metadata. We prioritize ZONE_MOVABLE
* as it's more reliable to unplug memory and remove whole memory
* blocks, and we don't want to trigger a zone imbalances by
* accidentially removing too much kernel memory.
*/
for (i = 0; i < ARRAY_SIZE(mb_states); i++) {
virtio_mem_sbm_for_each_mb_rev(vm, mb_id, mb_states[i]) {
rc = virtio_mem_sbm_unplug_any_sb(vm, mb_id, &nb_sb);
if (rc || !nb_sb)
goto out_unlock;
mutex_unlock(&vm->hotplug_mutex);
cond_resched();
mutex_lock(&vm->hotplug_mutex);
}
if (!unplug_online && i == 1) {
mutex_unlock(&vm->hotplug_mutex);
return 0;
}
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
}
mutex_unlock(&vm->hotplug_mutex);
virtio-mem: Paravirtualized memory hotunplug part 2 We also want to unplug online memory (contained in online memory blocks and, therefore, managed by the buddy), and eventually replug it later. When requested to unplug memory, we use alloc_contig_range() to allocate subblocks in online memory blocks (so we are the owner) and send them to our hypervisor. When requested to plug memory, we can replug such memory using free_contig_range() after asking our hypervisor. We also want to mark all allocated pages PG_offline, so nobody will touch them. To differentiate pages that were never onlined when onlining the memory block from pages allocated via alloc_contig_range(), we use PageDirty(). Based on this flag, virtio_mem_fake_online() can either online the pages for the first time or use free_contig_range(). It is worth noting that there are no guarantees on how much memory can actually get unplugged again. All device memory might completely be fragmented with unmovable data, such that no subblock can get unplugged. We are not touching the ZONE_MOVABLE. If memory is onlined to the ZONE_MOVABLE, it can only get unplugged after that memory was offlined manually by user space. In normal operation, virtio-mem memory is suggested to be onlined to ZONE_NORMAL. In the future, we will try to make unplug more likely to succeed. Add a module parameter to control if online memory shall be touched. As we want to access alloc_contig_range()/free_contig_range() from kernel module context, export the symbols. Note: Whenever virtio-mem uses alloc_contig_range(), all affected pages are on the same node, in the same zone, and contain no holes. Acked-by: Michal Hocko <mhocko@suse.com> # to export contig range allocator API Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Mike Rapoport <rppt@linux.ibm.com> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alexander Potapenko <glider@google.com> Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-6-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:29 +00:00
return nb_sb ? -EBUSY : 0;
out_unlock:
mutex_unlock(&vm->hotplug_mutex);
return rc;
}
/*
* Try to offline and remove a big block from Linux and unplug it. Will fail
* with -EBUSY if some memory is busy and cannot get unplugged.
*
* Will modify the state of the memory block. Might temporarily drop the
* hotplug_mutex.
*/
static int virtio_mem_bbm_offline_remove_and_unplug_bb(struct virtio_mem *vm,
unsigned long bb_id)
{
const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
unsigned long end_pfn = start_pfn + nr_pages;
unsigned long pfn;
struct page *page;
int rc;
if (WARN_ON_ONCE(virtio_mem_bbm_get_bb_state(vm, bb_id) !=
VIRTIO_MEM_BBM_BB_ADDED))
return -EINVAL;
/*
* Start by fake-offlining all memory. Once we marked the device
* block as fake-offline, all newly onlined memory will
* automatically be kept fake-offline. Protect from concurrent
* onlining/offlining until we have a consistent state.
*/
mutex_lock(&vm->hotplug_mutex);
virtio_mem_bbm_set_bb_state(vm, bb_id, VIRTIO_MEM_BBM_BB_FAKE_OFFLINE);
for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
page = pfn_to_online_page(pfn);
if (!page)
continue;
rc = virtio_mem_fake_offline(vm, pfn, PAGES_PER_SECTION);
if (rc) {
end_pfn = pfn;
goto rollback;
}
}
mutex_unlock(&vm->hotplug_mutex);
rc = virtio_mem_bbm_offline_and_remove_bb(vm, bb_id);
if (rc) {
mutex_lock(&vm->hotplug_mutex);
goto rollback;
}
rc = virtio_mem_bbm_unplug_bb(vm, bb_id);
if (rc)
virtio_mem_bbm_set_bb_state(vm, bb_id,
VIRTIO_MEM_BBM_BB_PLUGGED);
else
virtio_mem_bbm_set_bb_state(vm, bb_id,
VIRTIO_MEM_BBM_BB_UNUSED);
return rc;
rollback:
for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
page = pfn_to_online_page(pfn);
if (!page)
continue;
virtio_mem_fake_online(pfn, PAGES_PER_SECTION);
}
virtio_mem_bbm_set_bb_state(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED);
mutex_unlock(&vm->hotplug_mutex);
return rc;
}
/*
* Test if a big block is completely offline.
*/
static bool virtio_mem_bbm_bb_is_offline(struct virtio_mem *vm,
unsigned long bb_id)
{
const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
unsigned long pfn;
for (pfn = start_pfn; pfn < start_pfn + nr_pages;
pfn += PAGES_PER_SECTION) {
if (pfn_to_online_page(pfn))
return false;
}
return true;
}
/*
* Test if a big block is completely onlined to ZONE_MOVABLE (or offline).
*/
static bool virtio_mem_bbm_bb_is_movable(struct virtio_mem *vm,
unsigned long bb_id)
{
const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
struct page *page;
unsigned long pfn;
for (pfn = start_pfn; pfn < start_pfn + nr_pages;
pfn += PAGES_PER_SECTION) {
page = pfn_to_online_page(pfn);
if (!page)
continue;
if (page_zonenum(page) != ZONE_MOVABLE)
return false;
}
return true;
}
static int virtio_mem_bbm_unplug_request(struct virtio_mem *vm, uint64_t diff)
{
uint64_t nb_bb = diff / vm->bbm.bb_size;
uint64_t bb_id;
int rc, i;
if (!nb_bb)
return 0;
/*
* Try to unplug big blocks. Similar to SBM, start with offline
* big blocks.
*/
for (i = 0; i < 3; i++) {
virtio_mem_bbm_for_each_bb_rev(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED) {
cond_resched();
/*
* As we're holding no locks, these checks are racy,
* but we don't care.
*/
if (i == 0 && !virtio_mem_bbm_bb_is_offline(vm, bb_id))
continue;
if (i == 1 && !virtio_mem_bbm_bb_is_movable(vm, bb_id))
continue;
rc = virtio_mem_bbm_offline_remove_and_unplug_bb(vm, bb_id);
if (rc == -EBUSY)
continue;
if (!rc)
nb_bb--;
if (rc || !nb_bb)
return rc;
}
if (i == 0 && !unplug_online)
return 0;
}
return nb_bb ? -EBUSY : 0;
}
/*
* Try to unplug the requested amount of memory.
*/
static int virtio_mem_unplug_request(struct virtio_mem *vm, uint64_t diff)
{
if (vm->in_sbm)
return virtio_mem_sbm_unplug_request(vm, diff);
return virtio_mem_bbm_unplug_request(vm, diff);
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* Try to unplug all blocks that couldn't be unplugged before, for example,
* because the hypervisor was busy. Further, offline and remove any memory
* blocks where we previously failed.
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*/
static int virtio_mem_cleanup_pending_mb(struct virtio_mem *vm)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
unsigned long id;
int rc = 0;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (!vm->in_sbm) {
virtio_mem_bbm_for_each_bb(vm, id,
VIRTIO_MEM_BBM_BB_PLUGGED) {
rc = virtio_mem_bbm_unplug_bb(vm, id);
if (rc)
return rc;
virtio_mem_bbm_set_bb_state(vm, id,
VIRTIO_MEM_BBM_BB_UNUSED);
}
return 0;
}
virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_PLUGGED) {
rc = virtio_mem_sbm_unplug_mb(vm, id);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (rc)
return rc;
virtio_mem_sbm_set_mb_state(vm, id,
VIRTIO_MEM_SBM_MB_UNUSED);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
if (!vm->sbm.have_unplugged_mb)
return 0;
/*
* Let's retry (offlining and) removing completely unplugged Linux
* memory blocks.
*/
vm->sbm.have_unplugged_mb = false;
mutex_lock(&vm->hotplug_mutex);
virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL)
rc |= virtio_mem_sbm_try_remove_unplugged_mb(vm, id);
virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL)
rc |= virtio_mem_sbm_try_remove_unplugged_mb(vm, id);
virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL)
rc |= virtio_mem_sbm_try_remove_unplugged_mb(vm, id);
mutex_unlock(&vm->hotplug_mutex);
if (rc)
vm->sbm.have_unplugged_mb = true;
/* Ignore errors, this is not critical. We'll retry later. */
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return 0;
}
/*
* Update all parts of the config that could have changed.
*/
static void virtio_mem_refresh_config(struct virtio_mem *vm)
{
virtio-mem: check against mhp_get_pluggable_range() which memory we can hotplug Right now, we only check against MAX_PHYSMEM_BITS - but turns out there are more restrictions of which memory we can actually hotplug, especially om arm64 or s390x once we support them: we might receive something like -E2BIG or -ERANGE from add_memory_driver_managed(), stopping device operation. So, check right when initializing the device which memory we can add, warning the user. Try only adding actually pluggable ranges: in the worst case, no memory provided by our device is pluggable. In the usual case, we expect all device memory to be pluggable, and in corner cases only some memory at the end of the device-managed memory region to not be pluggable. Link: https://lkml.kernel.org/r/1612149902-7867-5-git-send-email-anshuman.khandual@arm.com Signed-off-by: David Hildenbrand <david@redhat.com> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: teawater <teawaterz@linux.alibaba.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Pankaj Gupta <pankaj.gupta@cloud.ionos.com> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-02-26 01:17:45 +00:00
const struct range pluggable_range = mhp_get_pluggable_range(true);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
uint64_t new_plugged_size, usable_region_size, end_addr;
/* the plugged_size is just a reflection of what _we_ did previously */
virtio_cread_le(vm->vdev, struct virtio_mem_config, plugged_size,
&new_plugged_size);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (WARN_ON_ONCE(new_plugged_size != vm->plugged_size))
vm->plugged_size = new_plugged_size;
/* calculate the last usable memory block id */
virtio_cread_le(vm->vdev, struct virtio_mem_config,
usable_region_size, &usable_region_size);
virtio-mem: check against mhp_get_pluggable_range() which memory we can hotplug Right now, we only check against MAX_PHYSMEM_BITS - but turns out there are more restrictions of which memory we can actually hotplug, especially om arm64 or s390x once we support them: we might receive something like -E2BIG or -ERANGE from add_memory_driver_managed(), stopping device operation. So, check right when initializing the device which memory we can add, warning the user. Try only adding actually pluggable ranges: in the worst case, no memory provided by our device is pluggable. In the usual case, we expect all device memory to be pluggable, and in corner cases only some memory at the end of the device-managed memory region to not be pluggable. Link: https://lkml.kernel.org/r/1612149902-7867-5-git-send-email-anshuman.khandual@arm.com Signed-off-by: David Hildenbrand <david@redhat.com> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: teawater <teawaterz@linux.alibaba.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Pankaj Gupta <pankaj.gupta@cloud.ionos.com> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-02-26 01:17:45 +00:00
end_addr = min(vm->addr + usable_region_size - 1,
pluggable_range.end);
virtio-mem: check against mhp_get_pluggable_range() which memory we can hotplug Right now, we only check against MAX_PHYSMEM_BITS - but turns out there are more restrictions of which memory we can actually hotplug, especially om arm64 or s390x once we support them: we might receive something like -E2BIG or -ERANGE from add_memory_driver_managed(), stopping device operation. So, check right when initializing the device which memory we can add, warning the user. Try only adding actually pluggable ranges: in the worst case, no memory provided by our device is pluggable. In the usual case, we expect all device memory to be pluggable, and in corner cases only some memory at the end of the device-managed memory region to not be pluggable. Link: https://lkml.kernel.org/r/1612149902-7867-5-git-send-email-anshuman.khandual@arm.com Signed-off-by: David Hildenbrand <david@redhat.com> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: teawater <teawaterz@linux.alibaba.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Pankaj Gupta <pankaj.gupta@cloud.ionos.com> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-02-26 01:17:45 +00:00
if (vm->in_sbm) {
vm->sbm.last_usable_mb_id = virtio_mem_phys_to_mb_id(end_addr);
if (!IS_ALIGNED(end_addr + 1, memory_block_size_bytes()))
vm->sbm.last_usable_mb_id--;
} else {
vm->bbm.last_usable_bb_id = virtio_mem_phys_to_bb_id(vm,
end_addr);
if (!IS_ALIGNED(end_addr + 1, vm->bbm.bb_size))
vm->bbm.last_usable_bb_id--;
}
/*
* If we cannot plug any of our device memory (e.g., nothing in the
* usable region is addressable), the last usable memory block id will
* be smaller than the first usable memory block id. We'll stop
* attempting to add memory with -ENOSPC from our main loop.
*/
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* see if there is a request to change the size */
virtio_cread_le(vm->vdev, struct virtio_mem_config, requested_size,
&vm->requested_size);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
dev_info(&vm->vdev->dev, "plugged size: 0x%llx", vm->plugged_size);
dev_info(&vm->vdev->dev, "requested size: 0x%llx", vm->requested_size);
}
/*
* Workqueue function for handling plug/unplug requests and config updates.
*/
static void virtio_mem_run_wq(struct work_struct *work)
{
struct virtio_mem *vm = container_of(work, struct virtio_mem, wq);
uint64_t diff;
int rc;
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
if (unlikely(vm->in_kdump)) {
dev_warn_once(&vm->vdev->dev,
"unexpected workqueue run in kdump kernel\n");
return;
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
hrtimer_cancel(&vm->retry_timer);
if (vm->broken)
return;
atomic_set(&vm->wq_active, 1);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
retry:
rc = 0;
/* Make sure we start with a clean state if there are leftovers. */
if (unlikely(vm->unplug_all_required))
rc = virtio_mem_send_unplug_all_request(vm);
if (atomic_read(&vm->config_changed)) {
atomic_set(&vm->config_changed, 0);
virtio_mem_refresh_config(vm);
}
/* Cleanup any leftovers from previous runs */
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (!rc)
rc = virtio_mem_cleanup_pending_mb(vm);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
if (!rc && vm->requested_size != vm->plugged_size) {
if (vm->requested_size > vm->plugged_size) {
diff = vm->requested_size - vm->plugged_size;
rc = virtio_mem_plug_request(vm, diff);
} else {
diff = vm->plugged_size - vm->requested_size;
rc = virtio_mem_unplug_request(vm, diff);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
}
/*
* Keep retrying to offline and remove completely unplugged Linux
* memory blocks.
*/
if (!rc && vm->in_sbm && vm->sbm.have_unplugged_mb)
rc = -EBUSY;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
switch (rc) {
case 0:
vm->retry_timer_ms = VIRTIO_MEM_RETRY_TIMER_MIN_MS;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
break;
case -ENOSPC:
/*
* We cannot add any more memory (alignment, physical limit)
* or we have too many offline memory blocks.
*/
break;
case -ETXTBSY:
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* The hypervisor cannot process our request right now
* (e.g., out of memory, migrating);
*/
case -EBUSY:
/*
* We cannot free up any memory to unplug it (all plugged memory
* is busy).
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*/
case -ENOMEM:
/* Out of memory, try again later. */
hrtimer_start(&vm->retry_timer, ms_to_ktime(vm->retry_timer_ms),
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
HRTIMER_MODE_REL);
break;
case -EAGAIN:
/* Retry immediately (e.g., the config changed). */
goto retry;
default:
/* Unknown error, mark as broken */
dev_err(&vm->vdev->dev,
"unknown error, marking device broken: %d\n", rc);
vm->broken = true;
}
atomic_set(&vm->wq_active, 0);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
static enum hrtimer_restart virtio_mem_timer_expired(struct hrtimer *timer)
{
struct virtio_mem *vm = container_of(timer, struct virtio_mem,
retry_timer);
virtio_mem_retry(vm);
vm->retry_timer_ms = min_t(unsigned int, vm->retry_timer_ms * 2,
VIRTIO_MEM_RETRY_TIMER_MAX_MS);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return HRTIMER_NORESTART;
}
static void virtio_mem_handle_response(struct virtqueue *vq)
{
struct virtio_mem *vm = vq->vdev->priv;
wake_up(&vm->host_resp);
}
static int virtio_mem_init_vq(struct virtio_mem *vm)
{
struct virtqueue *vq;
vq = virtio_find_single_vq(vm->vdev, virtio_mem_handle_response,
"guest-request");
if (IS_ERR(vq))
return PTR_ERR(vq);
vm->vq = vq;
return 0;
}
static int virtio_mem_init_hotplug(struct virtio_mem *vm)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
virtio-mem: check against mhp_get_pluggable_range() which memory we can hotplug Right now, we only check against MAX_PHYSMEM_BITS - but turns out there are more restrictions of which memory we can actually hotplug, especially om arm64 or s390x once we support them: we might receive something like -E2BIG or -ERANGE from add_memory_driver_managed(), stopping device operation. So, check right when initializing the device which memory we can add, warning the user. Try only adding actually pluggable ranges: in the worst case, no memory provided by our device is pluggable. In the usual case, we expect all device memory to be pluggable, and in corner cases only some memory at the end of the device-managed memory region to not be pluggable. Link: https://lkml.kernel.org/r/1612149902-7867-5-git-send-email-anshuman.khandual@arm.com Signed-off-by: David Hildenbrand <david@redhat.com> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: teawater <teawaterz@linux.alibaba.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Pankaj Gupta <pankaj.gupta@cloud.ionos.com> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-02-26 01:17:45 +00:00
const struct range pluggable_range = mhp_get_pluggable_range(true);
uint64_t unit_pages, sb_size, addr;
int rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* bad device setup - warn only */
if (!IS_ALIGNED(vm->addr, memory_block_size_bytes()))
dev_warn(&vm->vdev->dev,
"The alignment of the physical start address can make some memory unusable.\n");
if (!IS_ALIGNED(vm->addr + vm->region_size, memory_block_size_bytes()))
dev_warn(&vm->vdev->dev,
"The alignment of the physical end address can make some memory unusable.\n");
virtio-mem: check against mhp_get_pluggable_range() which memory we can hotplug Right now, we only check against MAX_PHYSMEM_BITS - but turns out there are more restrictions of which memory we can actually hotplug, especially om arm64 or s390x once we support them: we might receive something like -E2BIG or -ERANGE from add_memory_driver_managed(), stopping device operation. So, check right when initializing the device which memory we can add, warning the user. Try only adding actually pluggable ranges: in the worst case, no memory provided by our device is pluggable. In the usual case, we expect all device memory to be pluggable, and in corner cases only some memory at the end of the device-managed memory region to not be pluggable. Link: https://lkml.kernel.org/r/1612149902-7867-5-git-send-email-anshuman.khandual@arm.com Signed-off-by: David Hildenbrand <david@redhat.com> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: teawater <teawaterz@linux.alibaba.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Pankaj Gupta <pankaj.gupta@cloud.ionos.com> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-02-26 01:17:45 +00:00
if (vm->addr < pluggable_range.start ||
vm->addr + vm->region_size - 1 > pluggable_range.end)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
dev_warn(&vm->vdev->dev,
virtio-mem: check against mhp_get_pluggable_range() which memory we can hotplug Right now, we only check against MAX_PHYSMEM_BITS - but turns out there are more restrictions of which memory we can actually hotplug, especially om arm64 or s390x once we support them: we might receive something like -E2BIG or -ERANGE from add_memory_driver_managed(), stopping device operation. So, check right when initializing the device which memory we can add, warning the user. Try only adding actually pluggable ranges: in the worst case, no memory provided by our device is pluggable. In the usual case, we expect all device memory to be pluggable, and in corner cases only some memory at the end of the device-managed memory region to not be pluggable. Link: https://lkml.kernel.org/r/1612149902-7867-5-git-send-email-anshuman.khandual@arm.com Signed-off-by: David Hildenbrand <david@redhat.com> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: teawater <teawaterz@linux.alibaba.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Pankaj Gupta <pankaj.gupta@cloud.ionos.com> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-02-26 01:17:45 +00:00
"Some device memory is not addressable/pluggable. This can make some memory unusable.\n");
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Prepare the offline threshold - make sure we can add two blocks. */
vm->offline_threshold = max_t(uint64_t, 2 * memory_block_size_bytes(),
VIRTIO_MEM_DEFAULT_OFFLINE_THRESHOLD);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/*
* alloc_contig_range() works reliably with pageblock
* granularity on ZONE_NORMAL, use pageblock_nr_pages.
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
*/
sb_size = PAGE_SIZE * pageblock_nr_pages;
sb_size = max_t(uint64_t, vm->device_block_size, sb_size);
if (sb_size < memory_block_size_bytes() && !force_bbm) {
/* SBM: At least two subblocks per Linux memory block. */
vm->in_sbm = true;
vm->sbm.sb_size = sb_size;
vm->sbm.sbs_per_mb = memory_block_size_bytes() /
vm->sbm.sb_size;
/* Round up to the next full memory block */
virtio-mem: check against mhp_get_pluggable_range() which memory we can hotplug Right now, we only check against MAX_PHYSMEM_BITS - but turns out there are more restrictions of which memory we can actually hotplug, especially om arm64 or s390x once we support them: we might receive something like -E2BIG or -ERANGE from add_memory_driver_managed(), stopping device operation. So, check right when initializing the device which memory we can add, warning the user. Try only adding actually pluggable ranges: in the worst case, no memory provided by our device is pluggable. In the usual case, we expect all device memory to be pluggable, and in corner cases only some memory at the end of the device-managed memory region to not be pluggable. Link: https://lkml.kernel.org/r/1612149902-7867-5-git-send-email-anshuman.khandual@arm.com Signed-off-by: David Hildenbrand <david@redhat.com> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: teawater <teawaterz@linux.alibaba.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Pankaj Gupta <pankaj.gupta@cloud.ionos.com> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-02-26 01:17:45 +00:00
addr = max_t(uint64_t, vm->addr, pluggable_range.start) +
memory_block_size_bytes() - 1;
vm->sbm.first_mb_id = virtio_mem_phys_to_mb_id(addr);
vm->sbm.next_mb_id = vm->sbm.first_mb_id;
} else {
/* BBM: At least one Linux memory block. */
vm->bbm.bb_size = max_t(uint64_t, vm->device_block_size,
memory_block_size_bytes());
if (bbm_block_size) {
if (!is_power_of_2(bbm_block_size)) {
dev_warn(&vm->vdev->dev,
"bbm_block_size is not a power of 2");
} else if (bbm_block_size < vm->bbm.bb_size) {
dev_warn(&vm->vdev->dev,
"bbm_block_size is too small");
} else {
vm->bbm.bb_size = bbm_block_size;
}
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Round up to the next aligned big block */
virtio-mem: check against mhp_get_pluggable_range() which memory we can hotplug Right now, we only check against MAX_PHYSMEM_BITS - but turns out there are more restrictions of which memory we can actually hotplug, especially om arm64 or s390x once we support them: we might receive something like -E2BIG or -ERANGE from add_memory_driver_managed(), stopping device operation. So, check right when initializing the device which memory we can add, warning the user. Try only adding actually pluggable ranges: in the worst case, no memory provided by our device is pluggable. In the usual case, we expect all device memory to be pluggable, and in corner cases only some memory at the end of the device-managed memory region to not be pluggable. Link: https://lkml.kernel.org/r/1612149902-7867-5-git-send-email-anshuman.khandual@arm.com Signed-off-by: David Hildenbrand <david@redhat.com> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Cc: teawater <teawaterz@linux.alibaba.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Pankaj Gupta <pankaj.gupta@cloud.ionos.com> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-02-26 01:17:45 +00:00
addr = max_t(uint64_t, vm->addr, pluggable_range.start) +
vm->bbm.bb_size - 1;
vm->bbm.first_bb_id = virtio_mem_phys_to_bb_id(vm, addr);
vm->bbm.next_bb_id = vm->bbm.first_bb_id;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* Make sure we can add two big blocks. */
vm->offline_threshold = max_t(uint64_t, 2 * vm->bbm.bb_size,
vm->offline_threshold);
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
dev_info(&vm->vdev->dev, "memory block size: 0x%lx",
memory_block_size_bytes());
if (vm->in_sbm)
dev_info(&vm->vdev->dev, "subblock size: 0x%llx",
(unsigned long long)vm->sbm.sb_size);
else
dev_info(&vm->vdev->dev, "big block size: 0x%llx",
(unsigned long long)vm->bbm.bb_size);
/* create the parent resource for all memory */
rc = virtio_mem_create_resource(vm);
if (rc)
return rc;
/* use a single dynamic memory group to cover the whole memory device */
if (vm->in_sbm)
unit_pages = PHYS_PFN(memory_block_size_bytes());
else
unit_pages = PHYS_PFN(vm->bbm.bb_size);
rc = memory_group_register_dynamic(vm->nid, unit_pages);
if (rc < 0)
goto out_del_resource;
vm->mgid = rc;
/*
* If we still have memory plugged, we have to unplug all memory first.
* Registering our parent resource makes sure that this memory isn't
* actually in use (e.g., trying to reload the driver).
*/
if (vm->plugged_size) {
vm->unplug_all_required = true;
dev_info(&vm->vdev->dev, "unplugging all memory is required\n");
}
/* register callbacks */
vm->memory_notifier.notifier_call = virtio_mem_memory_notifier_cb;
rc = register_memory_notifier(&vm->memory_notifier);
if (rc)
goto out_unreg_group;
rc = register_virtio_mem_device(vm);
if (rc)
goto out_unreg_mem;
return 0;
out_unreg_mem:
unregister_memory_notifier(&vm->memory_notifier);
out_unreg_group:
memory_group_unregister(vm->mgid);
out_del_resource:
virtio_mem_delete_resource(vm);
return rc;
}
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
#ifdef CONFIG_PROC_VMCORE
static int virtio_mem_send_state_request(struct virtio_mem *vm, uint64_t addr,
uint64_t size)
{
const uint64_t nb_vm_blocks = size / vm->device_block_size;
const struct virtio_mem_req req = {
.type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_STATE),
.u.state.addr = cpu_to_virtio64(vm->vdev, addr),
.u.state.nb_blocks = cpu_to_virtio16(vm->vdev, nb_vm_blocks),
};
int rc = -ENOMEM;
dev_dbg(&vm->vdev->dev, "requesting state: 0x%llx - 0x%llx\n", addr,
addr + size - 1);
switch (virtio_mem_send_request(vm, &req)) {
case VIRTIO_MEM_RESP_ACK:
return virtio16_to_cpu(vm->vdev, vm->resp.u.state.state);
case VIRTIO_MEM_RESP_ERROR:
rc = -EINVAL;
break;
default:
break;
}
dev_dbg(&vm->vdev->dev, "requesting state failed: %d\n", rc);
return rc;
}
static bool virtio_mem_vmcore_pfn_is_ram(struct vmcore_cb *cb,
unsigned long pfn)
{
struct virtio_mem *vm = container_of(cb, struct virtio_mem,
vmcore_cb);
uint64_t addr = PFN_PHYS(pfn);
bool is_ram;
int rc;
if (!virtio_mem_contains_range(vm, addr, PAGE_SIZE))
return true;
if (!vm->plugged_size)
return false;
/*
* We have to serialize device requests and access to the information
* about the block queried last.
*/
mutex_lock(&vm->hotplug_mutex);
addr = ALIGN_DOWN(addr, vm->device_block_size);
if (addr != vm->last_block_addr) {
rc = virtio_mem_send_state_request(vm, addr,
vm->device_block_size);
/* On any kind of error, we're going to signal !ram. */
if (rc == VIRTIO_MEM_STATE_PLUGGED)
vm->last_block_plugged = true;
else
vm->last_block_plugged = false;
vm->last_block_addr = addr;
}
is_ram = vm->last_block_plugged;
mutex_unlock(&vm->hotplug_mutex);
return is_ram;
}
#endif /* CONFIG_PROC_VMCORE */
static int virtio_mem_init_kdump(struct virtio_mem *vm)
{
#ifdef CONFIG_PROC_VMCORE
dev_info(&vm->vdev->dev, "memory hot(un)plug disabled in kdump kernel\n");
vm->vmcore_cb.pfn_is_ram = virtio_mem_vmcore_pfn_is_ram;
register_vmcore_cb(&vm->vmcore_cb);
return 0;
#else /* CONFIG_PROC_VMCORE */
dev_warn(&vm->vdev->dev, "disabled in kdump kernel without vmcore\n");
return -EBUSY;
#endif /* CONFIG_PROC_VMCORE */
}
static int virtio_mem_init(struct virtio_mem *vm)
{
uint16_t node_id;
if (!vm->vdev->config->get) {
dev_err(&vm->vdev->dev, "config access disabled\n");
return -EINVAL;
}
/* Fetch all properties that can't change. */
virtio_cread_le(vm->vdev, struct virtio_mem_config, plugged_size,
&vm->plugged_size);
virtio_cread_le(vm->vdev, struct virtio_mem_config, block_size,
&vm->device_block_size);
virtio_cread_le(vm->vdev, struct virtio_mem_config, node_id,
&node_id);
vm->nid = virtio_mem_translate_node_id(vm, node_id);
virtio_cread_le(vm->vdev, struct virtio_mem_config, addr, &vm->addr);
virtio_cread_le(vm->vdev, struct virtio_mem_config, region_size,
&vm->region_size);
/* Determine the nid for the device based on the lowest address. */
if (vm->nid == NUMA_NO_NODE)
vm->nid = memory_add_physaddr_to_nid(vm->addr);
dev_info(&vm->vdev->dev, "start address: 0x%llx", vm->addr);
dev_info(&vm->vdev->dev, "region size: 0x%llx", vm->region_size);
dev_info(&vm->vdev->dev, "device block size: 0x%llx",
(unsigned long long)vm->device_block_size);
if (vm->nid != NUMA_NO_NODE && IS_ENABLED(CONFIG_NUMA))
dev_info(&vm->vdev->dev, "nid: %d", vm->nid);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
/*
* We don't want to (un)plug or reuse any memory when in kdump. The
* memory is still accessible (but not exposed to Linux).
*/
if (vm->in_kdump)
return virtio_mem_init_kdump(vm);
return virtio_mem_init_hotplug(vm);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
static int virtio_mem_create_resource(struct virtio_mem *vm)
{
/*
* When force-unloading the driver and removing the device, we
* could have a garbage pointer. Duplicate the string.
*/
const char *name = kstrdup(dev_name(&vm->vdev->dev), GFP_KERNEL);
if (!name)
return -ENOMEM;
virtio-mem: disallow mapping virtio-mem memory via /dev/mem We don't want user space to be able to map virtio-mem device memory directly (e.g., via /dev/mem) in order to have guarantees that in a sane setup we'll never accidentially access unplugged memory within the device-managed region of a virtio-mem device, just as required by the virtio-spec. As soon as the virtio-mem driver is loaded, the device region is visible in /proc/iomem via the parent device region. From that point on user space is aware of the device region and we want to disallow mapping anything inside that region (where we will dynamically (un)plug memory) until the driver has been unloaded cleanly and e.g., another driver might take over. By creating our parent IORESOURCE_SYSTEM_RAM resource with IORESOURCE_EXCLUSIVE, we will disallow any /dev/mem access to our device region until the driver was unloaded cleanly and removed the parent region. This will work even though only some memory blocks are actually currently added to Linux and appear as busy in the resource tree. So access to the region from user space is only possible a) if we don't load the virtio-mem driver. b) after unloading the virtio-mem driver cleanly. Don't build virtio-mem if access to /dev/mem cannot be restricticted -- if we have CONFIG_DEVMEM=y but CONFIG_STRICT_DEVMEM is not set. Link: https://lkml.kernel.org/r/20210920142856.17758-4-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Reviewed-by: Dan Williams <dan.j.williams@intel.com> Acked-by: Michael S. Tsirkin <mst@redhat.com> Cc: Andy Shevchenko <andy.shevchenko@gmail.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Hanjun Guo <guohanjun@huawei.com> Cc: Jason Wang <jasowang@redhat.com> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:35:53 +00:00
/* Disallow mapping device memory via /dev/mem completely. */
vm->parent_resource = __request_mem_region(vm->addr, vm->region_size,
virtio-mem: disallow mapping virtio-mem memory via /dev/mem We don't want user space to be able to map virtio-mem device memory directly (e.g., via /dev/mem) in order to have guarantees that in a sane setup we'll never accidentially access unplugged memory within the device-managed region of a virtio-mem device, just as required by the virtio-spec. As soon as the virtio-mem driver is loaded, the device region is visible in /proc/iomem via the parent device region. From that point on user space is aware of the device region and we want to disallow mapping anything inside that region (where we will dynamically (un)plug memory) until the driver has been unloaded cleanly and e.g., another driver might take over. By creating our parent IORESOURCE_SYSTEM_RAM resource with IORESOURCE_EXCLUSIVE, we will disallow any /dev/mem access to our device region until the driver was unloaded cleanly and removed the parent region. This will work even though only some memory blocks are actually currently added to Linux and appear as busy in the resource tree. So access to the region from user space is only possible a) if we don't load the virtio-mem driver. b) after unloading the virtio-mem driver cleanly. Don't build virtio-mem if access to /dev/mem cannot be restricticted -- if we have CONFIG_DEVMEM=y but CONFIG_STRICT_DEVMEM is not set. Link: https://lkml.kernel.org/r/20210920142856.17758-4-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Reviewed-by: Dan Williams <dan.j.williams@intel.com> Acked-by: Michael S. Tsirkin <mst@redhat.com> Cc: Andy Shevchenko <andy.shevchenko@gmail.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Hanjun Guo <guohanjun@huawei.com> Cc: Jason Wang <jasowang@redhat.com> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:35:53 +00:00
name, IORESOURCE_SYSTEM_RAM |
IORESOURCE_EXCLUSIVE);
if (!vm->parent_resource) {
kfree(name);
dev_warn(&vm->vdev->dev, "could not reserve device region\n");
dev_info(&vm->vdev->dev,
"reloading the driver is not supported\n");
return -EBUSY;
}
/* The memory is not actually busy - make add_memory() work. */
vm->parent_resource->flags &= ~IORESOURCE_BUSY;
return 0;
}
static void virtio_mem_delete_resource(struct virtio_mem *vm)
{
const char *name;
if (!vm->parent_resource)
return;
name = vm->parent_resource->name;
release_resource(vm->parent_resource);
kfree(vm->parent_resource);
kfree(name);
vm->parent_resource = NULL;
}
static int virtio_mem_range_has_system_ram(struct resource *res, void *arg)
{
return 1;
}
static bool virtio_mem_has_memory_added(struct virtio_mem *vm)
{
const unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
return walk_iomem_res_desc(IORES_DESC_NONE, flags, vm->addr,
vm->addr + vm->region_size, NULL,
virtio_mem_range_has_system_ram) == 1;
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
static int virtio_mem_probe(struct virtio_device *vdev)
{
struct virtio_mem *vm;
int rc;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
BUILD_BUG_ON(sizeof(struct virtio_mem_req) != 24);
BUILD_BUG_ON(sizeof(struct virtio_mem_resp) != 10);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
vdev->priv = vm = kzalloc(sizeof(*vm), GFP_KERNEL);
if (!vm)
return -ENOMEM;
init_waitqueue_head(&vm->host_resp);
vm->vdev = vdev;
INIT_WORK(&vm->wq, virtio_mem_run_wq);
mutex_init(&vm->hotplug_mutex);
INIT_LIST_HEAD(&vm->next);
spin_lock_init(&vm->removal_lock);
hrtimer_init(&vm->retry_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
vm->retry_timer.function = virtio_mem_timer_expired;
vm->retry_timer_ms = VIRTIO_MEM_RETRY_TIMER_MIN_MS;
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
vm->in_kdump = is_kdump_kernel();
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* register the virtqueue */
rc = virtio_mem_init_vq(vm);
if (rc)
goto out_free_vm;
/* initialize the device by querying the config */
rc = virtio_mem_init(vm);
if (rc)
goto out_del_vq;
virtio_device_ready(vdev);
/* trigger a config update to start processing the requested_size */
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
if (!vm->in_kdump) {
atomic_set(&vm->config_changed, 1);
queue_work(system_freezable_wq, &vm->wq);
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
return 0;
out_del_vq:
vdev->config->del_vqs(vdev);
out_free_vm:
kfree(vm);
vdev->priv = NULL;
return rc;
}
static void virtio_mem_deinit_hotplug(struct virtio_mem *vm)
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{
unsigned long mb_id;
int rc;
/*
* Make sure the workqueue won't be triggered anymore and no memory
* blocks can be onlined/offlined until we're finished here.
*/
mutex_lock(&vm->hotplug_mutex);
spin_lock_irq(&vm->removal_lock);
vm->removing = true;
spin_unlock_irq(&vm->removal_lock);
mutex_unlock(&vm->hotplug_mutex);
/* wait until the workqueue stopped */
cancel_work_sync(&vm->wq);
hrtimer_cancel(&vm->retry_timer);
if (vm->in_sbm) {
/*
* After we unregistered our callbacks, user space can online
* partially plugged offline blocks. Make sure to remove them.
*/
virtio_mem_sbm_for_each_mb(vm, mb_id,
VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL) {
rc = virtio_mem_sbm_remove_mb(vm, mb_id);
BUG_ON(rc);
virtio_mem_sbm_set_mb_state(vm, mb_id,
VIRTIO_MEM_SBM_MB_UNUSED);
}
/*
* After we unregistered our callbacks, user space can no longer
* offline partially plugged online memory blocks. No need to
* worry about them.
*/
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
}
/* unregister callbacks */
unregister_virtio_mem_device(vm);
unregister_memory_notifier(&vm->memory_notifier);
/*
* There is no way we could reliably remove all memory we have added to
* the system. And there is no way to stop the driver/device from going
* away. Warn at least.
*/
if (virtio_mem_has_memory_added(vm)) {
dev_warn(&vm->vdev->dev,
"device still has system memory added\n");
virtio-mem: add memory via add_memory_driver_managed() Virtio-mem managed memory is always detected and added by the virtio-mem driver, never using something like the firmware-provided memory map. This is the case after an ordinary system reboot, and has to be guaranteed after kexec. Especially, virtio-mem added memory resources can contain inaccessible parts ("unblocked memory blocks"), blindly forwarding them to a kexec kernel is dangerous, as unplugged memory will get accessed (esp. written). Let's use the new way of adding special driver-managed memory introduced in commit 7b7b27214bba ("mm/memory_hotplug: introduce add_memory_driver_managed()"). This will result in no entries in /sys/firmware/memmap ("raw firmware- provided memory map"), the memory resource will be flagged IORESOURCE_MEM_DRIVER_MANAGED (esp., kexec_file_load() will not place kexec images on this memory), and it is exposed as "System RAM (virtio_mem)" in /proc/iomem, so esp. kexec-tools can properly handle it. Example /proc/iomem before this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM 340000000-347ffffff : System RAM 348000000-34fffffff : System RAM [...] Example /proc/iomem after this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM (virtio_mem) 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM (virtio_mem) 340000000-347ffffff : System RAM (virtio_mem) 348000000-34fffffff : System RAM (virtio_mem) [...] Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: teawater <teawaterz@linux.alibaba.com> Fixes: 5f1f79bbc9e26 ("virtio-mem: Paravirtualized memory hotplug") Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200611093518.5737-1-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
2020-06-11 09:35:18 +00:00
} else {
virtio_mem_delete_resource(vm);
virtio-mem: add memory via add_memory_driver_managed() Virtio-mem managed memory is always detected and added by the virtio-mem driver, never using something like the firmware-provided memory map. This is the case after an ordinary system reboot, and has to be guaranteed after kexec. Especially, virtio-mem added memory resources can contain inaccessible parts ("unblocked memory blocks"), blindly forwarding them to a kexec kernel is dangerous, as unplugged memory will get accessed (esp. written). Let's use the new way of adding special driver-managed memory introduced in commit 7b7b27214bba ("mm/memory_hotplug: introduce add_memory_driver_managed()"). This will result in no entries in /sys/firmware/memmap ("raw firmware- provided memory map"), the memory resource will be flagged IORESOURCE_MEM_DRIVER_MANAGED (esp., kexec_file_load() will not place kexec images on this memory), and it is exposed as "System RAM (virtio_mem)" in /proc/iomem, so esp. kexec-tools can properly handle it. Example /proc/iomem before this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM 340000000-347ffffff : System RAM 348000000-34fffffff : System RAM [...] Example /proc/iomem after this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM (virtio_mem) 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM (virtio_mem) 340000000-347ffffff : System RAM (virtio_mem) 348000000-34fffffff : System RAM (virtio_mem) [...] Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: teawater <teawaterz@linux.alibaba.com> Fixes: 5f1f79bbc9e26 ("virtio-mem: Paravirtualized memory hotplug") Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200611093518.5737-1-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
2020-06-11 09:35:18 +00:00
kfree_const(vm->resource_name);
memory_group_unregister(vm->mgid);
virtio-mem: add memory via add_memory_driver_managed() Virtio-mem managed memory is always detected and added by the virtio-mem driver, never using something like the firmware-provided memory map. This is the case after an ordinary system reboot, and has to be guaranteed after kexec. Especially, virtio-mem added memory resources can contain inaccessible parts ("unblocked memory blocks"), blindly forwarding them to a kexec kernel is dangerous, as unplugged memory will get accessed (esp. written). Let's use the new way of adding special driver-managed memory introduced in commit 7b7b27214bba ("mm/memory_hotplug: introduce add_memory_driver_managed()"). This will result in no entries in /sys/firmware/memmap ("raw firmware- provided memory map"), the memory resource will be flagged IORESOURCE_MEM_DRIVER_MANAGED (esp., kexec_file_load() will not place kexec images on this memory), and it is exposed as "System RAM (virtio_mem)" in /proc/iomem, so esp. kexec-tools can properly handle it. Example /proc/iomem before this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM 340000000-347ffffff : System RAM 348000000-34fffffff : System RAM [...] Example /proc/iomem after this change: [...] 140000000-333ffffff : virtio0 140000000-147ffffff : System RAM (virtio_mem) 334000000-533ffffff : virtio1 338000000-33fffffff : System RAM (virtio_mem) 340000000-347ffffff : System RAM (virtio_mem) 348000000-34fffffff : System RAM (virtio_mem) [...] Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: teawater <teawaterz@linux.alibaba.com> Fixes: 5f1f79bbc9e26 ("virtio-mem: Paravirtualized memory hotplug") Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200611093518.5737-1-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
2020-06-11 09:35:18 +00:00
}
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* remove all tracking data - no locking needed */
if (vm->in_sbm) {
vfree(vm->sbm.mb_states);
vfree(vm->sbm.sb_states);
} else {
vfree(vm->bbm.bb_states);
}
}
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
static void virtio_mem_deinit_kdump(struct virtio_mem *vm)
{
#ifdef CONFIG_PROC_VMCORE
unregister_vmcore_cb(&vm->vmcore_cb);
#endif /* CONFIG_PROC_VMCORE */
}
static void virtio_mem_remove(struct virtio_device *vdev)
{
struct virtio_mem *vm = vdev->priv;
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
if (vm->in_kdump)
virtio_mem_deinit_kdump(vm);
else
virtio_mem_deinit_hotplug(vm);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
/* reset the device and cleanup the queues */
virtio_reset_device(vdev);
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
vdev->config->del_vqs(vdev);
kfree(vm);
vdev->priv = NULL;
}
static void virtio_mem_config_changed(struct virtio_device *vdev)
{
struct virtio_mem *vm = vdev->priv;
virtio-mem: kdump mode to sanitize /proc/vmcore access Although virtio-mem currently supports reading unplugged memory in the hypervisor, this will change in the future, indicated to the device via a new feature flag. We similarly sanitized /proc/kcore access recently. [1] Let's register a vmcore callback, to allow vmcore code to check if a PFN belonging to a virtio-mem device is either currently plugged and should be dumped or is currently unplugged and should not be accessed, instead mapping the shared zeropage or returning zeroes when reading. This is important when not capturing /proc/vmcore via tools like "makedumpfile" that can identify logically unplugged virtio-mem memory via PG_offline in the memmap, but simply by e.g., copying the file. Distributions that support virtio-mem+kdump have to make sure that the virtio_mem module will be part of the kdump kernel or the kdump initrd; dracut was recently [2] extended to include virtio-mem in the generated initrd. As long as no special kdump kernels are used, this will automatically make sure that virtio-mem will be around in the kdump initrd and sanitize /proc/vmcore access -- with dracut. With this series, we'll send one virtio-mem state request for every ~2 MiB chunk of virtio-mem memory indicated in the vmcore that we intend to read/map. In the future, we might want to allow building virtio-mem for kdump mode only, even without CONFIG_MEMORY_HOTPLUG and friends: this way, we could support special stripped-down kdump kernels that have many other config options disabled; we'll tackle that once required. Further, we might want to try sensing bigger blocks (e.g., memory sections) first before falling back to device blocks on demand. Tested with Fedora rawhide, which contains a recent kexec-tools version (considering "System RAM (virtio_mem)" when creating the vmcore header) and a recent dracut version (including the virtio_mem module in the kdump initrd). Link: https://lkml.kernel.org/r/20210526093041.8800-1-david@redhat.com [1] Link: https://github.com/dracutdevs/dracut/pull/1157 [2] Link: https://lkml.kernel.org/r/20211005121430.30136-10-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Cc: Baoquan He <bhe@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Dave Young <dyoung@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Rapoport <rppt@kernel.org> Cc: Oscar Salvador <osalvador@suse.de> Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vivek Goyal <vgoyal@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-11-09 02:32:02 +00:00
if (unlikely(vm->in_kdump))
return;
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
atomic_set(&vm->config_changed, 1);
virtio_mem_retry(vm);
}
#ifdef CONFIG_PM_SLEEP
static int virtio_mem_freeze(struct virtio_device *vdev)
{
/*
* When restarting the VM, all memory is usually unplugged. Don't
* allow to suspend/hibernate.
*/
dev_err(&vdev->dev, "save/restore not supported.\n");
return -EPERM;
}
static int virtio_mem_restore(struct virtio_device *vdev)
{
return -EPERM;
}
#endif
static unsigned int virtio_mem_features[] = {
#if defined(CONFIG_NUMA) && defined(CONFIG_ACPI_NUMA)
VIRTIO_MEM_F_ACPI_PXM,
#endif
virtio-mem: support VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE The initial virtio-mem spec states that while unplugged memory should not be read, the device still has to allow for reading unplugged memory inside the usable region. The primary motivation for this default handling was to simplify bringup of virtio-mem, because there were corner cases where Linux might have accidentially read unplugged memory inside added Linux memory blocks. In the meantime, we: 1. Removed /dev/kmem in commit bbcd53c96071 ("drivers/char: remove /dev/kmem for good") 2. Disallowed access to virtio-mem device memory via /dev/mem in commit 2128f4e21aa2 ("virtio-mem: disallow mapping virtio-mem memory via /dev/mem") 3. Sanitized access to virtio-mem device memory via /proc/kcore in commit 0daa322b8ff9 ("fs/proc/kcore: don't read offline sections, logically offline pages and hwpoisoned pages") 4. Sanitized access to virtio-mem device memory via /proc/vmcore in commit ce2814622e84 ("virtio-mem: kdump mode to sanitize /proc/vmcore access") "Accidential" access to unplugged memory is no longer possible; we can support the new VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE feature that will be required by some hypervisors implementing virtio-mem in the near future. Acked-by: Michael S. Tsirkin <mst@redhat.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Marek Kedzierski <mkedzier@redhat.com> Cc: Hui Zhu <teawater@gmail.com> Cc: Sebastien Boeuf <sebastien.boeuf@intel.com> Cc: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: Wei Yang <richard.weiyang@linux.alibaba.com> Signed-off-by: David Hildenbrand <david@redhat.com>
2021-03-15 15:57:13 +00:00
VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE,
};
static const struct virtio_device_id virtio_mem_id_table[] = {
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
{ VIRTIO_ID_MEM, VIRTIO_DEV_ANY_ID },
{ 0 },
};
static struct virtio_driver virtio_mem_driver = {
.feature_table = virtio_mem_features,
.feature_table_size = ARRAY_SIZE(virtio_mem_features),
virtio-mem: Paravirtualized memory hotplug Each virtio-mem device owns exactly one memory region. It is responsible for adding/removing memory from that memory region on request. When the device driver starts up, the requested amount of memory is queried and then plugged to Linux. On request, further memory can be plugged or unplugged. This patch only implements the plugging part. On x86-64, memory can currently be plugged in 4MB ("subblock") granularity. When required, a new memory block will be added (e.g., usually 128MB on x86-64) in order to plug more subblocks. Only x86-64 was tested for now. The online_page callback is used to keep unplugged subblocks offline when onlining memory - similar to the Hyper-V balloon driver. Unplugged pages are marked PG_offline, to tell dump tools (e.g., makedumpfile) to skip them. User space is usually responsible for onlining the added memory. The memory hotplug notifier is used to synchronize virtio-mem activity against memory onlining/offlining. Each virtio-mem device can belong to a NUMA node, which allows us to easily add/remove small chunks of memory to/from a specific NUMA node by using multiple virtio-mem devices. Something that works even when the guest has no idea about the NUMA topology. One way to view virtio-mem is as a "resizable DIMM" or a DIMM with many "sub-DIMMS". This patch directly introduces the basic infrastructure to implement memory unplug. Especially the memory block states and subblock bitmaps will be heavily used there. Notes: - In case memory is to be onlined by user space, we limit the amount of offline memory blocks, to not run out of memory. This is esp. an issue if memory is added faster than it is getting onlined. - Suspend/Hibernate is not supported due to the way virtio-mem devices behave. Limited support might be possible in the future. - Reloading the device driver is not supported. Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Tested-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Cc: Oscar Salvador <osalvador@suse.de> Cc: Michal Hocko <mhocko@kernel.org> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Dave Young <dyoung@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Stefan Hajnoczi <stefanha@redhat.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> Cc: linux-acpi@vger.kernel.org Signed-off-by: David Hildenbrand <david@redhat.com> Link: https://lore.kernel.org/r/20200507140139.17083-2-david@redhat.com Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
2020-05-07 14:01:25 +00:00
.driver.name = KBUILD_MODNAME,
.driver.owner = THIS_MODULE,
.id_table = virtio_mem_id_table,
.probe = virtio_mem_probe,
.remove = virtio_mem_remove,
.config_changed = virtio_mem_config_changed,
#ifdef CONFIG_PM_SLEEP
.freeze = virtio_mem_freeze,
.restore = virtio_mem_restore,
#endif
};
module_virtio_driver(virtio_mem_driver);
MODULE_DEVICE_TABLE(virtio, virtio_mem_id_table);
MODULE_AUTHOR("David Hildenbrand <david@redhat.com>");
MODULE_DESCRIPTION("Virtio-mem driver");
MODULE_LICENSE("GPL");