linux-stable/virt/kvm/vfio.c

383 lines
7.4 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
* VFIO-KVM bridge pseudo device
*
* Copyright (C) 2013 Red Hat, Inc. All rights reserved.
* Author: Alex Williamson <alex.williamson@redhat.com>
*/
#include <linux/errno.h>
#include <linux/file.h>
#include <linux/kvm_host.h>
#include <linux/list.h>
#include <linux/module.h>
#include <linux/mutex.h>
#include <linux/slab.h>
#include <linux/uaccess.h>
#include <linux/vfio.h>
#include "vfio.h"
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
#ifdef CONFIG_SPAPR_TCE_IOMMU
#include <asm/kvm_ppc.h>
#endif
struct kvm_vfio_group {
struct list_head node;
struct file *file;
};
struct kvm_vfio {
struct list_head group_list;
struct mutex lock;
bool noncoherent;
};
static void kvm_vfio_file_set_kvm(struct file *file, struct kvm *kvm)
{
void (*fn)(struct file *file, struct kvm *kvm);
fn = symbol_get(vfio_file_set_kvm);
if (!fn)
return;
fn(file, kvm);
symbol_put(vfio_file_set_kvm);
}
static bool kvm_vfio_file_enforced_coherent(struct file *file)
{
bool (*fn)(struct file *file);
bool ret;
fn = symbol_get(vfio_file_enforced_coherent);
if (!fn)
return false;
ret = fn(file);
symbol_put(vfio_file_enforced_coherent);
return ret;
}
static struct iommu_group *kvm_vfio_file_iommu_group(struct file *file)
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
{
struct iommu_group *(*fn)(struct file *file);
struct iommu_group *ret;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
fn = symbol_get(vfio_file_iommu_group);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
if (!fn)
return NULL;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
ret = fn(file);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
symbol_put(vfio_file_iommu_group);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
return ret;
}
#ifdef CONFIG_SPAPR_TCE_IOMMU
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
static void kvm_spapr_tce_release_vfio_group(struct kvm *kvm,
struct kvm_vfio_group *kvg)
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
{
struct iommu_group *grp = kvm_vfio_file_iommu_group(kvg->file);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
if (WARN_ON_ONCE(!grp))
return;
kvm_spapr_tce_release_iommu_group(kvm, grp);
}
#endif
/*
* Groups can use the same or different IOMMU domains. If the same then
* adding a new group may change the coherency of groups we've previously
* been told about. We don't want to care about any of that so we retest
* each group and bail as soon as we find one that's noncoherent. This
* means we only ever [un]register_noncoherent_dma once for the whole device.
*/
static void kvm_vfio_update_coherency(struct kvm_device *dev)
{
struct kvm_vfio *kv = dev->private;
bool noncoherent = false;
struct kvm_vfio_group *kvg;
mutex_lock(&kv->lock);
list_for_each_entry(kvg, &kv->group_list, node) {
if (!kvm_vfio_file_enforced_coherent(kvg->file)) {
noncoherent = true;
break;
}
}
if (noncoherent != kv->noncoherent) {
kv->noncoherent = noncoherent;
if (kv->noncoherent)
kvm_arch_register_noncoherent_dma(dev->kvm);
else
kvm_arch_unregister_noncoherent_dma(dev->kvm);
}
mutex_unlock(&kv->lock);
}
static int kvm_vfio_group_add(struct kvm_device *dev, unsigned int fd)
{
struct kvm_vfio *kv = dev->private;
struct kvm_vfio_group *kvg;
struct file *filp;
int ret;
filp = fget(fd);
if (!filp)
return -EBADF;
/* Ensure the FD is a vfio group FD.*/
if (!kvm_vfio_file_iommu_group(filp)) {
ret = -EINVAL;
goto err_fput;
}
mutex_lock(&kv->lock);
list_for_each_entry(kvg, &kv->group_list, node) {
if (kvg->file == filp) {
ret = -EEXIST;
goto err_unlock;
}
}
kvg = kzalloc(sizeof(*kvg), GFP_KERNEL_ACCOUNT);
if (!kvg) {
ret = -ENOMEM;
goto err_unlock;
}
kvg->file = filp;
list_add_tail(&kvg->node, &kv->group_list);
kvm_arch_start_assignment(dev->kvm);
mutex_unlock(&kv->lock);
kvm_vfio_file_set_kvm(kvg->file, dev->kvm);
kvm_vfio_update_coherency(dev);
return 0;
err_unlock:
mutex_unlock(&kv->lock);
err_fput:
fput(filp);
return ret;
}
static int kvm_vfio_group_del(struct kvm_device *dev, unsigned int fd)
{
struct kvm_vfio *kv = dev->private;
struct kvm_vfio_group *kvg;
struct fd f;
int ret;
f = fdget(fd);
if (!f.file)
return -EBADF;
ret = -ENOENT;
mutex_lock(&kv->lock);
list_for_each_entry(kvg, &kv->group_list, node) {
if (kvg->file != f.file)
continue;
list_del(&kvg->node);
kvm_arch_end_assignment(dev->kvm);
#ifdef CONFIG_SPAPR_TCE_IOMMU
kvm_spapr_tce_release_vfio_group(dev->kvm, kvg);
#endif
kvm_vfio_file_set_kvm(kvg->file, NULL);
fput(kvg->file);
kfree(kvg);
ret = 0;
break;
}
mutex_unlock(&kv->lock);
fdput(f);
kvm_vfio_update_coherency(dev);
return ret;
}
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
#ifdef CONFIG_SPAPR_TCE_IOMMU
static int kvm_vfio_group_set_spapr_tce(struct kvm_device *dev,
void __user *arg)
{
struct kvm_vfio_spapr_tce param;
struct kvm_vfio *kv = dev->private;
struct kvm_vfio_group *kvg;
struct fd f;
int ret;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
if (copy_from_user(&param, arg, sizeof(struct kvm_vfio_spapr_tce)))
return -EFAULT;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
f = fdget(param.groupfd);
if (!f.file)
return -EBADF;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
ret = -ENOENT;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
mutex_lock(&kv->lock);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
list_for_each_entry(kvg, &kv->group_list, node) {
struct iommu_group *grp;
if (kvg->file != f.file)
continue;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
grp = kvm_vfio_file_iommu_group(kvg->file);
if (WARN_ON_ONCE(!grp)) {
ret = -EIO;
goto err_fdput;
}
ret = kvm_spapr_tce_attach_iommu_group(dev->kvm, param.tablefd,
grp);
break;
}
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
mutex_unlock(&kv->lock);
err_fdput:
fdput(f);
return ret;
}
#endif
static int kvm_vfio_set_group(struct kvm_device *dev, long attr,
void __user *arg)
{
int32_t __user *argp = arg;
int32_t fd;
switch (attr) {
case KVM_DEV_VFIO_GROUP_ADD:
if (get_user(fd, argp))
return -EFAULT;
return kvm_vfio_group_add(dev, fd);
case KVM_DEV_VFIO_GROUP_DEL:
if (get_user(fd, argp))
return -EFAULT;
return kvm_vfio_group_del(dev, fd);
#ifdef CONFIG_SPAPR_TCE_IOMMU
case KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE:
return kvm_vfio_group_set_spapr_tce(dev, arg);
#endif
}
return -ENXIO;
}
static int kvm_vfio_set_attr(struct kvm_device *dev,
struct kvm_device_attr *attr)
{
switch (attr->group) {
case KVM_DEV_VFIO_GROUP:
return kvm_vfio_set_group(dev, attr->attr,
u64_to_user_ptr(attr->addr));
}
return -ENXIO;
}
static int kvm_vfio_has_attr(struct kvm_device *dev,
struct kvm_device_attr *attr)
{
switch (attr->group) {
case KVM_DEV_VFIO_GROUP:
switch (attr->attr) {
case KVM_DEV_VFIO_GROUP_ADD:
case KVM_DEV_VFIO_GROUP_DEL:
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
#ifdef CONFIG_SPAPR_TCE_IOMMU
case KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE:
#endif
return 0;
}
break;
}
return -ENXIO;
}
static void kvm_vfio_destroy(struct kvm_device *dev)
{
struct kvm_vfio *kv = dev->private;
struct kvm_vfio_group *kvg, *tmp;
list_for_each_entry_safe(kvg, tmp, &kv->group_list, node) {
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
#ifdef CONFIG_SPAPR_TCE_IOMMU
kvm_spapr_tce_release_vfio_group(dev->kvm, kvg);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 04:21:56 +00:00
#endif
kvm_vfio_file_set_kvm(kvg->file, NULL);
fput(kvg->file);
list_del(&kvg->node);
kfree(kvg);
kvm_arch_end_assignment(dev->kvm);
}
kvm_vfio_update_coherency(dev);
kfree(kv);
kfree(dev); /* alloc by kvm_ioctl_create_device, free by .destroy */
}
static int kvm_vfio_create(struct kvm_device *dev, u32 type);
static struct kvm_device_ops kvm_vfio_ops = {
.name = "kvm-vfio",
.create = kvm_vfio_create,
.destroy = kvm_vfio_destroy,
.set_attr = kvm_vfio_set_attr,
.has_attr = kvm_vfio_has_attr,
};
static int kvm_vfio_create(struct kvm_device *dev, u32 type)
{
struct kvm_device *tmp;
struct kvm_vfio *kv;
/* Only one VFIO "device" per VM */
list_for_each_entry(tmp, &dev->kvm->devices, vm_node)
if (tmp->ops == &kvm_vfio_ops)
return -EBUSY;
kv = kzalloc(sizeof(*kv), GFP_KERNEL_ACCOUNT);
if (!kv)
return -ENOMEM;
INIT_LIST_HEAD(&kv->group_list);
mutex_init(&kv->lock);
dev->private = kv;
return 0;
}
int kvm_vfio_ops_init(void)
{
return kvm_register_device_ops(&kvm_vfio_ops, KVM_DEV_TYPE_VFIO);
}
kvm: vfio: fix unregister kvm_device_ops of vfio After commit 80ce163 (KVM: VFIO: register kvm_device_ops dynamically), kvm_device_ops of vfio can be registered dynamically. Commit 3c3c29fd (kvm-vfio: do not use module_init) move the dynamic register invoked by kvm_init in order to fix broke unloading of the kvm module. However, kvm_device_ops of vfio is unregistered after rmmod kvm-intel module which lead to device type collision detection warning after kvm-intel module reinsmod. WARNING: CPU: 1 PID: 10358 at /root/cathy/kvm/arch/x86/kvm/../../../virt/kvm/kvm_main.c:3289 kvm_init+0x234/0x282 [kvm]() Modules linked in: kvm_intel(O+) kvm(O) nfsv3 nfs_acl auth_rpcgss oid_registry nfsv4 dns_resolver nfs fscache lockd sunrpc pci_stub bridge stp llc autofs4 8021q cpufreq_ondemand ipv6 joydev microcode pcspkr igb i2c_algo_bit ehci_pci ehci_hcd e1000e i2c_i801 ixgbe ptp pps_core hwmon mdio tpm_tis tpm ipmi_si ipmi_msghandler acpi_cpufreq isci libsas scsi_transport_sas button dm_mirror dm_region_hash dm_log dm_mod [last unloaded: kvm_intel] CPU: 1 PID: 10358 Comm: insmod Tainted: G W O 3.17.0-rc1 #2 Hardware name: Intel Corporation S2600CP/S2600CP, BIOS RMLSDP.86I.00.29.D696.1311111329 11/11/2013 0000000000000cd9 ffff880ff08cfd18 ffffffff814a61d9 0000000000000cd9 0000000000000000 ffff880ff08cfd58 ffffffff810417b7 ffff880ff08cfd48 ffffffffa045bcac ffffffffa049c420 0000000000000040 00000000000000ff Call Trace: [<ffffffff814a61d9>] dump_stack+0x49/0x60 [<ffffffff810417b7>] warn_slowpath_common+0x7c/0x96 [<ffffffffa045bcac>] ? kvm_init+0x234/0x282 [kvm] [<ffffffff810417e6>] warn_slowpath_null+0x15/0x17 [<ffffffffa045bcac>] kvm_init+0x234/0x282 [kvm] [<ffffffffa016e995>] vmx_init+0x1bf/0x42a [kvm_intel] [<ffffffffa016e7d6>] ? vmx_check_processor_compat+0x64/0x64 [kvm_intel] [<ffffffff810002ab>] do_one_initcall+0xe3/0x170 [<ffffffff811168a9>] ? __vunmap+0xad/0xb8 [<ffffffff8109c58f>] do_init_module+0x2b/0x174 [<ffffffff8109d414>] load_module+0x43e/0x569 [<ffffffff8109c6d8>] ? do_init_module+0x174/0x174 [<ffffffff8109c75a>] ? copy_module_from_user+0x39/0x82 [<ffffffff8109b7dd>] ? module_sect_show+0x20/0x20 [<ffffffff8109d65f>] SyS_init_module+0x54/0x81 [<ffffffff814a9a12>] system_call_fastpath+0x16/0x1b ---[ end trace 0626f4a3ddea56f3 ]--- The bug can be reproduced by: rmmod kvm_intel.ko insmod kvm_intel.ko without rmmod/insmod kvm.ko This patch fixes the bug by unregistering kvm_device_ops of vfio when the kvm-intel module is removed. Reported-by: Liu Rongrong <rongrongx.liu@intel.com> Fixes: 3c3c29fd0d7cddc32862c350d0700ce69953e3bd Signed-off-by: Wanpeng Li <wanpeng.li@linux.intel.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-10-09 10:30:08 +00:00
void kvm_vfio_ops_exit(void)
{
kvm_unregister_device_ops(KVM_DEV_TYPE_VFIO);
}