We split the LRU lists into inactive and an active parts to maximize
workingset protection while allowing just enough inactive cache space to
faciltate readahead and writeback for one-off file accesses (e.g. a
linear scan through a file, or logging); or just enough inactive anon to
maintain recent reference information when reclaim needs to swap.
With cgroups and their nested LRU lists, we currently don't do this
correctly. While recursive cgroup reclaim establishes a relative LRU
order among the pages of all involved cgroups, inactive:active size
decisions are done on a per-cgroup level. As a result, we'll reclaim a
cgroup's workingset when it doesn't have cold pages, even when one of its
siblings has plenty of it that should be reclaimed first.
For example: workload A has 50M worth of hot cache but doesn't do any
one-off file accesses; meanwhile, parallel workload B scans files and
rarely accesses the same page twice.
If these workloads were to run in an uncgrouped system, A would be
protected from the high rate of cache faults from B. But if they were put
in parallel cgroups for memory accounting purposes, B's fast cache fault
rate would push out the hot cache pages of A. This is unexpected and
undesirable - the "scan resistance" of the page cache is broken.
This patch moves inactive:active size balancing decisions to the root of
reclaim - the same level where the LRU order is established.
It does this by looking at the recursive size of the inactive and the
active file sets of the cgroup subtree at the beginning of the reclaim
cycle, and then making a decision - scan or skip active pages - that
applies throughout the entire run and to every cgroup involved.
With that in place, in the test above, the VM will recognize that there
are plenty of inactive pages in the combined cache set of workloads A and
B and prefer the one-off cache in B over the hot pages in A. The scan
resistance of the cache is restored.
[cai@lca.pw: fix some -Wenum-conversion warnings]
Link: http://lkml.kernel.org/r/1573848697-29262-1-git-send-email-cai@lca.pw
Link: http://lkml.kernel.org/r/20191107205334.158354-4-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We use refault information to determine whether the cache workingset is
stable or transitioning, and dynamically adjust the inactive:active file
LRU ratio so as to maximize protection from one-off cache during stable
periods, and minimize IO during transitions.
With cgroups and their nested LRU lists, we currently don't do this
correctly. While recursive cgroup reclaim establishes a relative LRU
order among the pages of all involved cgroups, refaults only affect the
local LRU order in the cgroup in which they are occuring. As a result,
cache transitions can take longer in a cgrouped system as the active pages
of sibling cgroups aren't challenged when they should be.
[ Right now, this is somewhat theoretical, because the siblings, under
continued regular reclaim pressure, should eventually run out of
inactive pages - and since inactive:active *size* balancing is also
done on a cgroup-local level, we will challenge the active pages
eventually in most cases. But the next patch will move that relative
size enforcement to the reclaim root as well, and then this patch
here will be necessary to propagate refault pressure to siblings. ]
This patch moves refault detection to the root of reclaim. Instead of
remembering the cgroup owner of an evicted page, remember the cgroup that
caused the reclaim to happen. When refaults later occur, they'll
correctly influence the cross-cgroup LRU order that reclaim follows.
I.e. if global reclaim kicked out pages in some subgroup A/B/C, the
refault of those pages will challenge the global LRU order, and not just
the local order down inside C.
[hannes@cmpxchg.org: use page_memcg() instead of another lookup]
Link: http://lkml.kernel.org/r/20191115160722.GA309754@cmpxchg.org
Link: http://lkml.kernel.org/r/20191107205334.158354-3-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "mm: fix page aging across multiple cgroups".
When applications are put into unconfigured cgroups for memory accounting
purposes, the cgrouping itself should not change the behavior of the page
reclaim code. We expect the VM to reclaim the coldest pages in the
system. But right now the VM can reclaim hot pages in one cgroup while
there is eligible cold cache in others.
This is because one part of the reclaim algorithm isn't truly cgroup
hierarchy aware: the inactive/active list balancing. That is the part
that is supposed to protect hot cache data from one-off streaming IO.
The recursive cgroup reclaim scheme will scan and rotate the physical LRU
lists of each eligible cgroup at the same rate in a round-robin fashion,
thereby establishing a relative order among the pages of all those
cgroups. However, the inactive/active balancing decisions are made
locally within each cgroup, so when a cgroup is running low on cold pages,
its hot pages will get reclaimed - even when sibling cgroups have plenty
of cold cache eligible in the same reclaim run.
For example:
[root@ham ~]# head -n1 /proc/meminfo
MemTotal: 1016336 kB
[root@ham ~]# ./reclaimtest2.sh
Establishing 50M active files in cgroup A...
Hot pages cached: 12800/12800 workingset-a
Linearly scanning through 18G of file data in cgroup B:
real 0m4.269s
user 0m0.051s
sys 0m4.182s
Hot pages cached: 134/12800 workingset-a
The streaming IO in B, which doesn't benefit from caching at all, pushes
out most of the workingset in A.
Solution
This series fixes the problem by elevating inactive/active balancing
decisions to the toplevel of the reclaim run. This is either a cgroup
that hit its limit, or straight-up global reclaim if there is physical
memory pressure. From there, it takes a recursive view of the cgroup
subtree to decide whether page deactivation is necessary.
In the test above, the VM will then recognize that cgroup B has plenty of
eligible cold cache, and that the hot pages in A can be spared:
[root@ham ~]# ./reclaimtest2.sh
Establishing 50M active files in cgroup A...
Hot pages cached: 12800/12800 workingset-a
Linearly scanning through 18G of file data in cgroup B:
real 0m4.244s
user 0m0.064s
sys 0m4.177s
Hot pages cached: 12800/12800 workingset-a
Implementation
Whether active pages can be deactivated or not is influenced by two
factors: the inactive list dropping below a minimum size relative to the
active list, and the occurence of refaults.
This patch series first moves refault detection to the reclaim root, then
enforces the minimum inactive size based on a recursive view of the cgroup
tree's LRUs.
History
Note that this actually never worked correctly in Linux cgroups. In the
past it worked for global reclaim and leaf limit reclaim only (we used to
have two physical LRU linkages per page), but it never worked for
intermediate limit reclaim over multiple leaf cgroups.
We're noticing this now because 1) we're putting everything into cgroups
for accounting, not just the things we want to control and 2) we're moving
away from leaf limits that invoke reclaim on individual cgroups, toward
large tree reclaim, triggered by high-level limits, or physical memory
pressure that is influenced by local protections such as memory.low and
memory.min instead.
This patch (of 3):
When file pages are lower than the watermark on a node, we try to force
scan anonymous pages to counter-act the balancing algorithms preference
for new file pages when they are likely thrashing. This is a node-level
decision, but it's currently made each time we look at an lruvec. This is
unnecessarily expensive and also a layering violation that makes the code
harder to understand.
Clean this up by making the check once per node and setting a flag in the
scan_control.
Link: http://lkml.kernel.org/r/20191107205334.158354-2-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Rik van Riel <riel@surriel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The current writeback congestion tracking has separate flags for kswapd
reclaim (node level) and cgroup limit reclaim (memcg-node level). This is
unnecessarily complicated: the lruvec is an existing abstraction layer for
that node-memcg intersection.
Introduce lruvec->flags and LRUVEC_CONGESTED. Then track that at the
reclaim root level, which is either the NUMA node for global reclaim, or
the cgroup-node intersection for cgroup reclaim.
Link: http://lkml.kernel.org/r/20191022144803.302233-9-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This function is getting long and unwieldy, split out the memcg bits.
The updated shrink_node() handles the generic (node) reclaim aspects:
- global vmpressure notifications
- writeback and congestion throttling
- reclaim/compaction management
- kswapd giving up on unreclaimable nodes
It then calls a new shrink_node_memcgs() which handles cgroup specifics:
- the cgroup tree traversal
- memory.low considerations
- per-cgroup slab shrinking callbacks
- per-cgroup vmpressure notifications
[hannes@cmpxchg.org: rename "root" to "target_memcg", per Roman]
Link: http://lkml.kernel.org/r/20191025143640.GA386981@cmpxchg.org
Link: http://lkml.kernel.org/r/20191022144803.302233-8-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
An lruvec holds LRU pages owned by a certain NUMA node and cgroup.
Instead of awkwardly passing around a combination of a pgdat and a memcg
pointer, pass down the lruvec as soon as we can look it up.
Nested callers that need to access node or cgroup properties can look them
them up if necessary, but there are only a few cases.
Link: http://lkml.kernel.org/r/20191022144803.302233-7-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Most of the function body is inside a loop, which imposes an additional
indentation and scoping level that makes the code a bit hard to follow and
modify.
The looping only happens in case of reclaim-compaction, which isn't the
common case. So rather than adding yet another function level to the
reclaim path and have every reclaim invocation go through a level that
only exists for one specific cornercase, use a retry goto.
Link: http://lkml.kernel.org/r/20191022144803.302233-6-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Seven years after introducing the global_reclaim() function, I still have
to double take when reading a callsite. I don't know how others do it,
this is a terrible name.
Invert the meaning and rename it to cgroup_reclaim().
[ After all, "global reclaim" is just regular reclaim invoked from the
page allocator. It's reclaim on behalf of a cgroup limit that is a
special case of reclaim, and should be explicit - not the reverse. ]
sane_reclaim() isn't very descriptive either: it tests whether we can use
the regular writeback throttling - available during regular page reclaim
or cgroup2 limit reclaim - or need to use the broken
wait_on_page_writeback() method. Use "writeback_throttling_sane()".
Link: http://lkml.kernel.org/r/20191022144803.302233-5-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Roman Gushchin <guro@fb.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
inactive_list_is_low() should be about one thing: checking the ratio
between inactive and active list. Kitchensink checks like the one for
swap space makes the function hard to use and modify its callsites.
Luckly, most callers already have an understanding of the swap situation,
so it's easy to clean up.
get_scan_count() has its own, memcg-aware swap check, and doesn't even get
to the inactive_list_is_low() check on the anon list when there is no swap
space available.
shrink_list() is called on the results of get_scan_count(), so that check
is redundant too.
age_active_anon() has its own totalswap_pages check right before it checks
the list proportions.
The shrink_node_memcg() site is the only one that doesn't do its own swap
check. Add it there.
Then delete the swap check from inactive_list_is_low().
Link: http://lkml.kernel.org/r/20191022144803.302233-4-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There is a per-memcg lruvec and a NUMA node lruvec. Which one is being
used is somewhat confusing right now, and it's easy to make mistakes -
especially when it comes to global reclaim.
How it works: when memory cgroups are enabled, we always use the
root_mem_cgroup's per-node lruvecs. When memory cgroups are not compiled
in or disabled at runtime, we use pgdat->lruvec.
Document that in a comment.
Due to the way the reclaim code is generalized, all lookups use the
mem_cgroup_lruvec() helper function, and nobody should have to find the
right lruvec manually right now. But to avoid future mistakes, rename the
pgdat->lruvec member to pgdat->__lruvec and delete the convenience wrapper
that suggests it's a commonly accessed member.
While in this area, swap the mem_cgroup_lruvec() argument order. The name
suggests a memcg operation, yet it takes a pgdat first and a memcg second.
I have to double take every time I call this. Fix that.
Link: http://lkml.kernel.org/r/20191022144803.302233-3-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "mm: vmscan: cgroup-related cleanups".
Here are 8 patches that clean up the reclaim code's interaction with
cgroups a bit. They're not supposed to change any behavior, just make
the implementation easier to understand and work with.
This patch (of 8):
This function currently takes the node or lruvec size and subtracts the
zones that are excluded by the classzone index of the allocation. It uses
four different types of counters to do this.
Just add up the eligible zones.
[cai@lca.pw: fix an undefined behavior for zone id]
Link: http://lkml.kernel.org/r/20191108204407.1435-1-cai@lca.pw
[akpm@linux-foundation.org: deal with the MAX_NR_ZONES special case. per Qian Cai]
Link: http://lkml.kernel.org/r/64E60F6F-7582-427B-8DD5-EF97B1656F5A@lca.pw
Link: http://lkml.kernel.org/r/20191022144803.302233-2-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Since lumpy reclaim was removed in v3.5 scan_control is not used by
may_write_to_{queue|inode} and pageout() anymore, remove the unused
parameter.
Link: http://lkml.kernel.org/r/1570124498-19300-1-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Since 9092c71bb7 ("mm: use sc->priority for slab shrink targets") the
argument 'unsigned long *lru_pages' passed around with no purpose. Remove
it.
Link: http://lkml.kernel.org/r/20190228083329.31892-4-aryabinin@virtuozzo.com
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: William Kucharski <william.kucharski@oracle.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Print nr_reserved_highatomic in show_free_areas, because when alloc_harder
is false, this value will be subtracted from the free_pages in
__zone_watermark_ok. Printing this value can help analyze memory
allocaction failure issues.
Link: http://lkml.kernel.org/r/19515f3de2fb6abe66b52e03e4b676a21e82beda.1573634806.git.lijiazi@xiaomi.com
Signed-off-by: lijiazi <lijiazi@xiaomi.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Both file-backed pages and anonymous pages can be unmapped.
ISOLATE_UNMAPPED is not just for file-backed pages.
Link: http://lkml.kernel.org/r/20191024151621.GA20400@haolee.github.io
Signed-off-by: Hao Lee <haolee.swjtu@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Memory hotplug needs to be able to reset and reinit the pcpu allocator
batch and high limits but this action is internal to the VM. Move the
declaration to internal.h
Link: http://lkml.kernel.org/r/20191021094808.28824-4-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Qian Cai <cai@lca.pw>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Both the percpu_pagelist_fraction sysctl handler and memory hotplug have
a common requirement of updating the pcpu page allocation batch and high
values. Split the relevant helper to share common code.
No functional change.
Link: http://lkml.kernel.org/r/20191021094808.28824-3-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Qian Cai <cai@lca.pw>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
HugeTLB helper alloc_gigantic_page() implements fairly generic
allocation method where it scans over various zones looking for a large
contiguous pfn range before trying to allocate it with
alloc_contig_range().
Other than deriving the requested order from 'struct hstate', there is
nothing HugeTLB specific in there. This can be made available for
general use to allocate contiguous memory which could not have been
allocated through the buddy allocator.
alloc_gigantic_page() has been split carving out actual allocation
method which is then made available via new alloc_contig_pages() helper
wrapped under CONFIG_CONTIG_ALLOC. All references to 'gigantic' have
been replaced with more generic term 'contig'. Allocated pages here
should be freed with free_contig_range() or by calling __free_page() on
each allocated page.
Link: http://lkml.kernel.org/r/1571300646-32240-1-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@suse.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Pavel Tatashin <pavel.tatashin@microsoft.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In the case where KASAN directly allocates memory to back vmalloc space,
don't map the early shadow page over it.
We prepopulate pgds/p4ds for the range that would otherwise be empty.
This is required to get it synced to hardware on boot, allowing the
lower levels of the page tables to be filled dynamically.
Link: http://lkml.kernel.org/r/20191031093909.9228-5-dja@axtens.net
Signed-off-by: Daniel Axtens <dja@axtens.net>
Acked-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Supporting VMAP_STACK with KASAN_VMALLOC is straightforward:
- clear the shadow region of vmapped stacks when swapping them in
- tweak Kconfig to allow VMAP_STACK to be turned on with KASAN
Link: http://lkml.kernel.org/r/20191031093909.9228-4-dja@axtens.net
Signed-off-by: Daniel Axtens <dja@axtens.net>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Test kasan vmalloc support by adding a new test to the module.
Link: http://lkml.kernel.org/r/20191031093909.9228-3-dja@axtens.net
Signed-off-by: Daniel Axtens <dja@axtens.net>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "kasan: support backing vmalloc space with real shadow
memory", v11.
Currently, vmalloc space is backed by the early shadow page. This means
that kasan is incompatible with VMAP_STACK.
This series provides a mechanism to back vmalloc space with real,
dynamically allocated memory. I have only wired up x86, because that's
the only currently supported arch I can work with easily, but it's very
easy to wire up other architectures, and it appears that there is some
work-in-progress code to do this on arm64 and s390.
This has been discussed before in the context of VMAP_STACK:
- https://bugzilla.kernel.org/show_bug.cgi?id=202009
- https://lkml.org/lkml/2018/7/22/198
- https://lkml.org/lkml/2019/7/19/822
In terms of implementation details:
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=1)
This is unfortunate but given that this is a debug feature only, not the
end of the world. The benchmarks are also a stress-test for the vmalloc
subsystem: they're not indicative of an overall 2x slowdown!
This patch (of 4):
Hook into vmalloc and vmap, and dynamically allocate real shadow memory
to back the mappings.
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
To avoid the difficulties around swapping mappings around, this code
expects that the part of the shadow region that covers the vmalloc space
will not be covered by the early shadow page, but will be left unmapped.
This will require changes in arch-specific code.
This allows KASAN with VMAP_STACK, and may be helpful for architectures
that do not have a separate module space (e.g. powerpc64, which I am
currently working on). It also allows relaxing the module alignment
back to PAGE_SIZE.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=3D1)
This is unfortunate but given that this is a debug feature only, not the
end of the world.
The full benchmark results are:
Performance
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 662004 11404956 17.23 19144610 28.92 1.68
full_fit_alloc_test 710950 12029752 16.92 13184651 18.55 1.10
long_busy_list_alloc_test 9431875 43990172 4.66 82970178 8.80 1.89
random_size_alloc_test 5033626 23061762 4.58 47158834 9.37 2.04
fix_align_alloc_test 1252514 15276910 12.20 31266116 24.96 2.05
random_size_align_alloc_te 1648501 14578321 8.84 25560052 15.51 1.75
align_shift_alloc_test 147 830 5.65 5692 38.72 6.86
pcpu_alloc_test 80732 125520 1.55 140864 1.74 1.12
Total Cycles 119240774314 763211341128 6.40 1390338696894 11.66 1.82
Sequential, 2 cpus
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 1423150 14276550 10.03 27733022 19.49 1.94
full_fit_alloc_test 1754219 14722640 8.39 15030786 8.57 1.02
long_busy_list_alloc_test 11451858 52154973 4.55 107016027 9.34 2.05
random_size_alloc_test 5989020 26735276 4.46 68885923 11.50 2.58
fix_align_alloc_test 2050976 20166900 9.83 50491675 24.62 2.50
random_size_align_alloc_te 2858229 17971700 6.29 38730225 13.55 2.16
align_shift_alloc_test 405 6428 15.87 26253 64.82 4.08
pcpu_alloc_test 127183 151464 1.19 216263 1.70 1.43
Total Cycles 54181269392 308723699764 5.70 650772566394 12.01 2.11
fix_size_alloc_test 1420404 14289308 10.06 27790035 19.56 1.94
full_fit_alloc_test 1736145 14806234 8.53 15274301 8.80 1.03
long_busy_list_alloc_test 11404638 52270785 4.58 107550254 9.43 2.06
random_size_alloc_test 6017006 26650625 4.43 68696127 11.42 2.58
fix_align_alloc_test 2045504 20280985 9.91 50414862 24.65 2.49
random_size_align_alloc_te 2845338 17931018 6.30 38510276 13.53 2.15
align_shift_alloc_test 472 3760 7.97 9656 20.46 2.57
pcpu_alloc_test 118643 132732 1.12 146504 1.23 1.10
Total Cycles 54040011688 309102805492 5.72 651325675652 12.05 2.11
[dja@axtens.net: fixups]
Link: http://lkml.kernel.org/r/20191120052719.7201-1-dja@axtens.net
Link: https://bugzilla.kernel.org/show_bug.cgi?id=3D202009
Link: http://lkml.kernel.org/r/20191031093909.9228-2-dja@axtens.net
Signed-off-by: Mark Rutland <mark.rutland@arm.com> [shadow rework]
Signed-off-by: Daniel Axtens <dja@axtens.net>
Co-developed-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
With the new allocation approach introduced in the 5.2 kernel, it
becomes possible to get rid of one global spinlock. By doing that we
can further improve the KVA from the performance point of view.
Basically we can have two independent locks, one for allocation part and
another one for deallocation, because of two different entities: "free
data structures" and "busy data structures".
As a result, allocation/deallocation operations can still interfere
between each other in case of running simultaneously on different CPUs,
it means there is still dependency, but with two locks it becomes lower.
Summarizing:
- it reduces the high lock contention
- it allows to perform operations on "free" and "busy"
trees in parallel on different CPUs. Please note it
does not solve scalability issue.
Test results:
In order to evaluate this patch, we can run "vmalloc test driver" to see
how many CPU cycles it takes to complete all test cases running
sequentially. All online CPUs run it so it will cause a high lock
contention.
HiKey 960, ARM64, 8xCPUs, big.LITTLE:
<snip>
sudo ./test_vmalloc.sh sequential_test_order=1
<snip>
<default>
[ 390.950557] All test took CPU0=457126382 cycles
[ 391.046690] All test took CPU1=454763452 cycles
[ 391.128586] All test took CPU2=454539334 cycles
[ 391.222669] All test took CPU3=455649517 cycles
[ 391.313946] All test took CPU4=388272196 cycles
[ 391.410425] All test took CPU5=384036264 cycles
[ 391.492219] All test took CPU6=387432964 cycles
[ 391.578433] All test took CPU7=387201996 cycles
<default>
<patched>
[ 304.721224] All test took CPU0=391521310 cycles
[ 304.821219] All test took CPU1=393533002 cycles
[ 304.917120] All test took CPU2=392243032 cycles
[ 305.008986] All test took CPU3=392353853 cycles
[ 305.108944] All test took CPU4=297630721 cycles
[ 305.196406] All test took CPU5=297548736 cycles
[ 305.288602] All test took CPU6=297092392 cycles
[ 305.381088] All test took CPU7=297293597 cycles
<patched>
~14%-23% patched variant is better.
Link: http://lkml.kernel.org/r/20191022155800.20468-1-urezki@gmail.com
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oleksiy Avramchenko <oleksiy.avramchenko@sonymobile.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When running test_vmalloc.sh smoke the following print out states that
the fragment is missing.
# ./test_vmalloc.sh: You must have the following enabled in your kernel:
# CONFIG_TEST_VMALLOC=m
Rework to add the fragment 'CONFIG_TEST_VMALLOC=m' to the config file.
Link: http://lkml.kernel.org/r/20190916095217.19665-1-anders.roxell@linaro.org
Fixes: a05ef00c97 ("selftests/vm: add script helper for CONFIG_TEST_VMALLOC_MODULE")
Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: "Uladzislau Rezki (Sony)" <urezki@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When fit type is NE_FIT_TYPE there is a need in one extra object.
Usually the "ne_fit_preload_node" per-CPU variable has it and there is
no need in GFP_NOWAIT allocation, but there are exceptions.
This commit just adds more explanations, as a result giving answers on
questions like when it can occur, how often, under which conditions and
what happens if GFP_NOWAIT gets failed.
Link: http://lkml.kernel.org/r/20191016095438.12391-3-urezki@gmail.com
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Daniel Wagner <dwagner@suse.de>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oleksiy Avramchenko <oleksiy.avramchenko@sonymobile.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Allocation functions should comply with the given gfp_mask as much as
possible. The preallocation code in alloc_vmap_area doesn't follow that
pattern and it is using a hardcoded GFP_KERNEL. Although this doesn't
really make much difference because vmalloc is not GFP_NOWAIT compliant
in general (e.g. page table allocations are GFP_KERNEL) there is no
reason to spread that bad habit and it is good to fix the antipattern.
[mhocko@suse.com: rewrite changelog]
Link: http://lkml.kernel.org/r/20191016095438.12391-2-urezki@gmail.com
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Daniel Wagner <dwagner@suse.de>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oleksiy Avramchenko <oleksiy.avramchenko@sonymobile.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Some background. The preemption was disabled before to guarantee that a
preloaded object is available for a CPU, it was stored for. That was
achieved by combining the disabling the preemption and taking the spin
lock while the ne_fit_preload_node is checked.
The aim was to not allocate in atomic context when spinlock is taken
later, for regular vmap allocations. But that approach conflicts with
CONFIG_PREEMPT_RT philosophy. It means that calling spin_lock() with
disabled preemption is forbidden in the CONFIG_PREEMPT_RT kernel.
Therefore, get rid of preempt_disable() and preempt_enable() when the
preload is done for splitting purpose. As a result we do not guarantee
now that a CPU is preloaded, instead we minimize the case when it is
not, with this change, by populating the per cpu preload pointer under
the vmap_area_lock.
This implies that at least each caller that has done the preallocation
will not fallback to an atomic allocation later. It is possible that
the preallocation would be pointless or that no preallocation is done
because of the race but the data shows that this is really rare.
For example i run the special test case that follows the preload pattern
and path. 20 "unbind" threads run it and each does 1000000 allocations.
Only 3.5 times among 1000000 a CPU was not preloaded. So it can happen
but the number is negligible.
[mhocko@suse.com: changelog additions]
Link: http://lkml.kernel.org/r/20191016095438.12391-1-urezki@gmail.com
Fixes: 82dd23e84b ("mm/vmalloc.c: preload a CPU with one object for split purpose")
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Reviewed-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Acked-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Daniel Wagner <dwagner@suse.de>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oleksiy Avramchenko <oleksiy.avramchenko@sonymobile.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
gfpflags_allow_blocking() does not care about __GFP_HIGHMEM, so
highmem_mask can be removed.
Link: http://lkml.kernel.org/r/1568812319-3467-1-git-send-email-liuxiang_1999@126.com
Signed-off-by: Liu Xiang <liuxiang_1999@126.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Vincent has noticed [1] that there is something unusual with the memmap
allocations going on on his platform
: I noticed this because on my ARM64 platform, with 1 GiB of memory the
: first [and only] section is allocated from the zeroing path while with
: 2 GiB of memory the first 1 GiB section is allocated from the
: non-zeroing path.
The underlying problem is that although sparse_buffer_init allocates
enough memory for all sections on the node sparse_buffer_alloc is not
able to consume them due to mismatch in the expected allocation
alignement. While sparse_buffer_init preallocation uses the PAGE_SIZE
alignment the real memmap has to be aligned to section_map_size() this
results in a wasted initial chunk of the preallocated memmap and
unnecessary fallback allocation for a section.
While we are at it also change __populate_section_memmap to align to the
requested size because at least VMEMMAP has constrains to have memmap
properly aligned.
[1] http://lkml.kernel.org/r/20191030131122.8256-1-vincent.whitchurch@axis.com
[akpm@linux-foundation.org: tweak layout, per David]
Link: http://lkml.kernel.org/r/20191119092642.31799-1-mhocko@kernel.org
Fixes: 35fd1eb1e8 ("mm/sparse: abstract sparse buffer allocations")
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
Debugged-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Oscar Salvador <OSalvador@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Building the kernel on s390 with -Og produces the following warning:
WARNING: vmlinux.o(.text+0x28dabe): Section mismatch in reference from the function populate_section_memmap() to the function .meminit.text:__populate_section_memmap()
The function populate_section_memmap() references
the function __meminit __populate_section_memmap().
This is often because populate_section_memmap lacks a __meminit
annotation or the annotation of __populate_section_memmap is wrong.
While -Og is not supported, in theory this might still happen with
another compiler or on another architecture. So fix this by using the
correct section annotations.
[iii@linux.ibm.com: v2]
Link: http://lkml.kernel.org/r/20191030151639.41486-1-iii@linux.ibm.com
Link: http://lkml.kernel.org/r/20191028165549.14478-1-iii@linux.ibm.com
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Oscar Salvador <OSalvador@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
sparsemem without VMEMMAP has two allocation paths to allocate the
memory needed for its memmap (done in sparse_mem_map_populate()).
In one allocation path (sparse_buffer_alloc() succeeds), the memory is
not zeroed (since it was previously allocated with
memblock_alloc_try_nid_raw()).
In the other allocation path (sparse_buffer_alloc() fails and
sparse_mem_map_populate() falls back to memblock_alloc_try_nid()), the
memory is zeroed.
AFAICS this difference does not appear to be on purpose. If the code is
supposed to work with non-initialized memory (__init_single_page() takes
care of zeroing the struct pages which are actually used), we should
consistently not zero the memory, to avoid masking bugs.
( I noticed this because on my ARM64 platform, with 1 GiB of memory the
first [and only] section is allocated from the zeroing path while with
2 GiB of memory the first 1 GiB section is allocated from the
non-zeroing path. )
Michal:
"the main user visible problem is a memory wastage. The overal amount
of memory should be small. I wouldn't call it stable material."
Link: http://lkml.kernel.org/r/20191030131122.8256-1-vincent.whitchurch@axis.com
Signed-off-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Pavel Tatashin <pasha.tatashin@soleen.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Our onlining/offlining code is unnecessarily complicated. Only memory
blocks added during boot can have holes (a range that is not
IORESOURCE_SYSTEM_RAM). Hotplugged memory never has holes (e.g., see
add_memory_resource()). All memory blocks that belong to boot memory
are already online.
Note that boot memory can have holes and the memmap of the holes is
marked PG_reserved. However, also memory allocated early during boot is
PG_reserved - basically every page of boot memory that is not given to
the buddy is PG_reserved.
Therefore, when we stop allowing to offline memory blocks with holes, we
implicitly no longer have to deal with onlining memory blocks with
holes. E.g., online_pages() will do a walk_system_ram_range(...,
online_pages_range), whereby online_pages_range() will effectively only
free the memory holes not falling into a hole to the buddy. The other
pages (holes) are kept PG_reserved (via
move_pfn_range_to_zone()->memmap_init_zone()).
This allows to simplify the code. For example, we no longer have to
worry about marking pages that fall into memory holes PG_reserved when
onlining memory. We can stop setting pages PG_reserved completely in
memmap_init_zone().
Offlining memory blocks added during boot is usually not guaranteed to
work either way (unmovable data might have easily ended up on that
memory during boot). So stopping to do that should not really hurt.
Also, people are not even aware of a setup where onlining/offlining of
memory blocks with holes used to work reliably (see [1] and [2]
especially regarding the hotplug path) - I doubt it worked reliably.
For the use case of offlining memory to unplug DIMMs, we should see no
change. (holes on DIMMs would be weird).
Please note that hardware errors (PG_hwpoison) are not memory holes and
are not affected by this change when offlining.
[1] https://lkml.org/lkml/2019/10/22/135
[2] https://lkml.org/lkml/2019/8/14/1365
Link: http://lkml.kernel.org/r/20191119115237.6662-1-david@redhat.com
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The mem_sysfs_mutex isn't really helpful. Also, it's not really clear
what the mutex protects at all.
The device lists of the memory subsystem are protected separately. We
don't need that mutex when looking up. creating, or removing
independent devices. find_memory_block_by_id() will perform locking on
its own and grab a reference of the returned device.
At the time memory_dev_init() is called, we cannot have concurrent
hot(un)plug operations yet - we're still fairly early during boot. We
don't need any locking.
The creation/removal of memory block devices should be protected on a
higher level - especially using the device hotplug lock to avoid
documented issues (see Documentation/core-api/memory-hotplug.rst) - or
if that is reworked, using similar locking.
Protecting in the context of these functions only doesn't really make
sense. Especially, if we would have a situation where the same memory
blocks are created/deleted at the same time, there is something horribly
going wrong (imagining adding/removing a DIMM at the same time from two
call paths) - after the functions succeeded something else in the
callers would blow up (e.g., create_memory_block_devices() succeeded but
there are no memory block devices anymore).
All relevant call paths (except when adding memory early during boot via
ACPI, which is now documented) hold the device hotplug lock when adding
memory, and when removing memory. Let's document that instead.
Add a simple safety net to create_memory_block_devices() in case we
would actually remove memory blocks while adding them, so we'll never
dereference a NULL pointer. Simplify memory_dev_init() now that the
lock is gone.
Link: http://lkml.kernel.org/r/20190925082621.4927-1-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The {set,clear}_zone_contiguous are built whatever the configuratoon so
move the definitions outside the current ifdef to avoid the following
compiler warnings:
mm/page_alloc.c:1550:6: warning: no previous prototype for 'set_zone_contiguous' [-Wmissing-prototypes]
mm/page_alloc.c:1571:6: warning: no previous prototype for 'clear_zone_contiguous' [-Wmissing-prototypes]
Link: http://lkml.kernel.org/r/20191106123911.7435-1-ben.dooks@codethink.co.uk
Signed-off-by: Ben Dooks (Codethink) <ben.dooks@codethink.co.uk>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We have two types of users of page isolation:
1. Memory offlining: Offline memory so it can be unplugged. Memory
won't be touched.
2. Memory allocation: Allocate memory (e.g., alloc_contig_range()) to
become the owner of the memory and make use of
it.
For example, in case we want to offline memory, we can ignore (skip
over) PageHWPoison() pages, as the memory won't get used. We can allow
to offline memory. In contrast, we don't want to allow to allocate such
memory.
Let's generalize the approach so we can special case other types of
pages we want to skip over in case we offline memory. While at it, also
pass the same flags to test_pages_isolated().
Link: http://lkml.kernel.org/r/20191021172353.3056-3-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Suggested-by: Michal Hocko <mhocko@suse.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Pingfan Liu <kernelfans@gmail.com>
Cc: Qian Cai <cai@lca.pw>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Pavel Tatashin <pavel.tatashin@microsoft.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "mm: Memory offlining + page isolation cleanups", v2.
This patch (of 2):
We call __offline_isolated_pages() from __offline_pages() after all
pages were isolated and are either free (PageBuddy()) or PageHWPoison.
Nothing can stop us from offlining memory at this point.
In __offline_isolated_pages() we first set all affected memory sections
offline (offline_mem_sections(pfn, end_pfn)), to mark the memmap as
invalid (pfn_to_online_page() will no longer succeed), and then walk
over all pages to pull the free pages from the free lists (to the
isolated free lists, to be precise).
Note that re-onlining a memory block will result in the whole memmap
getting reinitialized, overwriting any old state. We already poision
the memmap when offlining is complete to find any access to
stale/uninitialized memmaps.
So, setting the pages PageReserved() is not helpful. The memap is
marked offline and all pageblocks are isolated. As soon as offline, the
memmap is stale either way.
This looks like a leftover from ancient times where we initialized the
memmap when adding memory and not when onlining it (the pages were set
PageReserved so re-onling would work as expected).
Link: http://lkml.kernel.org/r/20191021172353.3056-2-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Pavel Tatashin <pavel.tatashin@microsoft.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Pingfan Liu <kernelfans@gmail.com>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Let's drop the now unused functions.
Link: http://lkml.kernel.org/r/20190909114830.662-4-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Qian Cai <cai@lca.pw>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Sasha Levin <sashal@kernel.org>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Let's use the generic onlining function - which will now also take care
of calling kernel_map_pages().
Link: http://lkml.kernel.org/r/20190909114830.662-3-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Cc: Sasha Levin <sashal@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Oscar Salvador <osalvador@suse.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Qian Cai <cai@lca.pw>
Cc: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "mm/memory_hotplug: Export generic_online_page()".
Let's replace the __online_page...() functions by generic_online_page().
Hyper-V only wants to delay the actual onlining of un-backed pages, so
we can simpy re-use the generic function.
This patch (of 3):
Let's expose generic_online_page() so online_page_callback users can
simply fall back to the generic implementation when actually deciding to
online the pages.
Link: http://lkml.kernel.org/r/20190909114830.662-2-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Qian Cai <cai@lca.pw>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Sasha Levin <sashal@kernel.org>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
On PowerPC, the address ranges allocated to OpenCAPI LPC memory are
allocated from firmware. These address ranges may be higher than what
older kernels permit, as we increased the maximum permissable address in
commit 4ffe713b75 ("powerpc/mm: Increase the max addressable memory to
2PB"). It is possible that the addressable range may change again in
the future.
In this scenario, we end up with a bogus section returned from
__section_nr (see the discussion on the thread "mm: Trigger bug on if a
section is not found in __section_nr").
Adding a check here means that we fail early and have an opportunity to
handle the error gracefully, rather than rumbling on and potentially
accessing an incorrect section.
Further discussion is also on the thread ("powerpc: Perform a bounds
check in arch_add_memory")
http://lkml.kernel.org/r/20190827052047.31547-1-alastair@au1.ibm.com
Link: http://lkml.kernel.org/r/20191001004617.7536-2-alastair@au1.ibm.com
Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently during memory hot add procedure, memory gets into memblock
before calling arch_add_memory() which creates its linear mapping.
add_memory_resource() {
..................
memblock_add_node()
..................
arch_add_memory()
..................
}
But during memory hot remove procedure, removal from memblock happens
first before its linear mapping gets teared down with
arch_remove_memory() which is not consistent. Resource removal should
happen in reverse order as they were added. However this does not pose
any problem for now, unless there is an assumption regarding linear
mapping. One example was a subtle failure on arm64 platform [1].
Though this has now found a different solution.
try_remove_memory() {
..................
memblock_free()
memblock_remove()
..................
arch_remove_memory()
..................
}
This changes the sequence of resource removal including memblock and
linear mapping tear down during memory hot remove which will now be the
reverse order in which they were added during memory hot add. The
changed removal order looks like the following.
try_remove_memory() {
..................
arch_remove_memory()
..................
memblock_free()
memblock_remove()
..................
}
[1] https://patchwork.kernel.org/patch/11127623/
Memory hot remove now works on arm64 without this because a recent
commit 60bb462fc7ad ("drivers/base/node.c: simplify
unregister_memory_block_under_nodes()").
This does not fix a serious problem. It just removes an inconsistency
while freeing resources during memory hot remove which for now does not
pose a real problem.
David mentioned that re-ordering should still make sense for consistency
purpose (removing stuff in the reverse order they were added). This
patch is now detached from arm64 hot-remove series.
Michal:
: I would just a note that the inconsistency doesn't pose any problem now
: but if somebody makes any assumptions about linear mappings then it could
: get subtly broken like your example for arm64 which has found a different
: solution in the meantime.
Link: http://lkml.kernel.org/r/1569380273-7708-1-git-send-email-anshuman.khandual@arm.com
Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
page_shift() is supported after the commit 94ad933810 ("mm: introduce
page_shift()").
So replace with page_shift() in add_to_kill() for readability.
Link: http://lkml.kernel.org/r/543d8bc9-f2e7-3023-7c35-2e7ed67c0e82@huawei.com
Signed-off-by: Yunfeng Ye <yeyunfeng@huawei.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently soft_offline_page() receives struct page, and its sibling
memory_failure() receives pfn. This discrepancy looks weird and makes
precheck on pfn validity tricky. So let's align them.
Link: http://lkml.kernel.org/r/20191016234706.GA5493@www9186uo.sakura.ne.jp
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
add_to_kill() expects the first 'tk' to be pre-allocated, it makes
subsequent allocations on need basis, this makes the code a bit
difficult to read.
Move all the allocation internal to add_to_kill() and drop the **tk
argument.
Link: http://lkml.kernel.org/r/1565112345-28754-2-git-send-email-jane.chu@oracle.com
Signed-off-by: Jane Chu <jane.chu@oracle.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In this test, the parent and child both have writable private mappings.
The test shows that without the patch in this series, the parent and
child shared the same memory which is incorrect. In other words, COW
needs to be triggered so any writes to child's copy stays local to the
child.
Link: http://lkml.kernel.org/r/20191107195355.80608-2-joel@joelfernandes.org
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Nicolas Geoffray <ngeoffray@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
F_SEAL_FUTURE_WRITE has unexpected behavior when used with MAP_PRIVATE:
A private mapping created after the memfd file that gets sealed with
F_SEAL_FUTURE_WRITE loses the copy-on-write at fork behavior, meaning
children and parent share the same memory, even though the mapping is
private.
The reason for this is due to the code below:
static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
{
struct shmem_inode_info *info = SHMEM_I(file_inode(file));
if (info->seals & F_SEAL_FUTURE_WRITE) {
/*
* New PROT_WRITE and MAP_SHARED mmaps are not allowed when
* "future write" seal active.
*/
if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
return -EPERM;
/*
* Since the F_SEAL_FUTURE_WRITE seals allow for a MAP_SHARED
* read-only mapping, take care to not allow mprotect to revert
* protections.
*/
vma->vm_flags &= ~(VM_MAYWRITE);
}
...
}
And for the mm to know if a mapping is copy-on-write:
static inline bool is_cow_mapping(vm_flags_t flags)
{
return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
}
The patch fixes the issue by making the mprotect revert protection
happen only for shared mappings. For private mappings, using mprotect
will have no effect on the seal behavior.
The F_SEAL_FUTURE_WRITE feature was introduced in v5.1 so v5.3.x stable
kernels would need a backport.
[akpm@linux-foundation.org: reflow comment, per Christoph]
Link: http://lkml.kernel.org/r/20191107195355.80608-1-joel@joelfernandes.org
Fixes: ab3948f58f ("mm/memfd: add an F_SEAL_FUTURE_WRITE seal to memfd")
Signed-off-by: Nicolas Geoffray <ngeoffray@google.com>
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There's a bug in the new PAT code, the conversion of memtype_check_conflict()
is buggy:
8d04a5f97a5f: ("x86/mm/pat: Convert the PAT tree to a generic interval tree")
dprintk("Overlap at 0x%Lx-0x%Lx\n", match->start, match->end);
found_type = match->type;
- node = rb_next(&match->rb);
- while (node) {
- match = rb_entry(node, struct memtype, rb);
-
- if (match->start >= end) /* Checked all possible matches */
- goto success;
-
- if (is_node_overlap(match, start, end) &&
- match->type != found_type) {
+ match = memtype_interval_iter_next(match, start, end);
+ while (match) {
+ if (match->type != found_type)
goto failure;
- }
- node = rb_next(&match->rb);
+ match = memtype_interval_iter_next(match, start, end);
}
Note how the '>= end' condition to end the interval check, got converted
into:
+ match = memtype_interval_iter_next(match, start, end);
This is subtly off by one, because the interval trees interfaces require
closed interval parameters:
include/linux/interval_tree_generic.h
/* \
* Iterate over intervals intersecting [start;last] \
* \
* Note that a node's interval intersects [start;last] iff: \
* Cond1: ITSTART(node) <= last \
* and \
* Cond2: start <= ITLAST(node) \
*/ \
...
if (ITSTART(node) <= last) { /* Cond1 */ \
if (start <= ITLAST(node)) /* Cond2 */ \
return node; /* node is leftmost match */ \
[start;last] is a closed interval (note that '<= last' check) - while the
PAT 'end' parameter is 1 byte beyond the end of the range, because
ioremap() and the other mapping APIs usually use the [start,end)
half-open interval, derived from 'size'.
This is what ioremap() does for example:
/*
* Mappings have to be page-aligned
*/
offset = phys_addr & ~PAGE_MASK;
phys_addr &= PHYSICAL_PAGE_MASK;
size = PAGE_ALIGN(last_addr+1) - phys_addr;
retval = reserve_memtype(phys_addr, (u64)phys_addr + size,
pcm, &new_pcm);
phys_addr+size will be on a page boundary, after the last byte of the
mapped interval.
So the correct parameter to use in the interval tree searches is not
'end' but 'end-1'.
This could have relevance if conflicting PAT ranges are exactly adjacent,
for example a future WC region is followed immediately by an already
mapped UC- region - in this case memtype_check_conflict() would
incorrectly deny the WC memtype region and downgrade the memtype to UC-.
BTW., rather annoyingly this downgrading is done silently in
memtype_check_insert():
int memtype_check_insert(struct memtype *new,
enum page_cache_mode *ret_type)
{
int err = 0;
err = memtype_check_conflict(new->start, new->end, new->type, ret_type);
if (err)
return err;
if (ret_type)
new->type = *ret_type;
memtype_interval_insert(new, &memtype_rbroot);
return 0;
}
So on such a conflict we'd just silently get UC- in *ret_type, and write
it into the new region, never the wiser ...
So assuming that the patch below fixes the primary bug the diagnostics
side of ioremap() cache attribute downgrades would be another thing to
fix.
Anyway, I checked all the interval-tree iterations, and most of them are
off by one - but I think the one related to memtype_check_conflict() is
the one causing this particular performance regression.
The only correct interval-tree searches were these two:
arch/x86/mm/pat_interval.c: match = memtype_interval_iter_first(&memtype_rbroot, 0, ULONG_MAX);
arch/x86/mm/pat_interval.c: match = memtype_interval_iter_next(match, 0, ULONG_MAX);
The ULONG_MAX was hiding the off-by-one in plain sight. :-)
Note that the bug was probably benign in the sense of implementing a too
strict cache attribute conflict policy and downgrading cache attributes,
so AFAICS the worst outcome of this bug would be a performance regression,
not any instabilities.
Reported-by: kernel test robot <rong.a.chen@intel.com>
Reported-by: Kenneth R. Crudup <kenny@panix.com>
Reported-by: Mariusz Ceier <mceier+kernel@gmail.com>
Tested-by: Mariusz Ceier <mceier@gmail.com>
Tested-by: Kenneth R. Crudup <kenny@panix.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20191201144947.GA4167@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
For the case where the interpreter is compiled out or when the prog is jited
it is completely unnecessary to set the BPF insn pages as read-only. In fact,
on frequent churn of BPF programs, it could lead to performance degradation of
the system over time since it would break the direct map down to 4k pages when
calling set_memory_ro() for the insn buffer on x86-64 / arm64 and there is no
reverse operation. Thus, avoid breaking up large pages for data maps, and only
limit this to the module range used by the JIT where it is necessary to set
the image read-only and executable.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20191129222911.3710-1-daniel@iogearbox.net
A huge pud page can theoretically be faulted in racing with pmd_alloc()
in __handle_mm_fault(). That will lead to pmd_alloc() returning an
invalid pmd pointer.
Fix this by adding a pud_trans_unstable() function similar to
pmd_trans_unstable() and check whether the pud is really stable before
using the pmd pointer.
Race:
Thread 1: Thread 2: Comment
create_huge_pud() Fallback - not taken.
create_huge_pud() Taken.
pmd_alloc() Returns an invalid pointer.
This will result in user-visible huge page data corruption.
Note that this was caught during a code audit rather than a real
experienced problem. It looks to me like the only implementation that
currently creates huge pud pagetable entries is dev_dax_huge_fault()
which doesn't appear to care much about private (COW) mappings or
write-tracking which is, I believe, a prerequisite for create_huge_pud()
falling back on thread 1, but not in thread 2.
Link: http://lkml.kernel.org/r/20191115115808.21181-2-thomas_os@shipmail.org
Fixes: a00cc7d9dd ("mm, x86: add support for PUD-sized transparent hugepages")
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Xianting Tian