linux-stable.git/mm, branch v4.17.3 Linux kernel stable tree mm, page_alloc: do not break __GFP_THISNODE by zonelist reset 2018-06-25T23:51:32+00:00 Vlastimil Babka vbabka@suse.cz 2018-06-08T00:09:29+00:00 9ec1cadef3b8b0d380dca1fbb451b8bf67d67bf1 commit 7810e6781e0fcbca78b91cf65053f895bf59e85f upstream. In __alloc_pages_slowpath() we reset zonelist and preferred_zoneref for allocations that can ignore memory policies. The zonelist is obtained from current CPU's node. This is a problem for __GFP_THISNODE allocations that want to allocate on a different node, e.g. because the allocating thread has been migrated to a different CPU. This has been observed to break SLAB in our 4.4-based kernel, because there it relies on __GFP_THISNODE working as intended. If a slab page is put on wrong node's list, then further list manipulations may corrupt the list because page_to_nid() is used to determine which node's list_lock should be locked and thus we may take a wrong lock and race. Current SLAB implementation seems to be immune by luck thanks to commit 511e3a058812 ("mm/slab: make cache_grow() handle the page allocated on arbitrary node") but there may be others assuming that __GFP_THISNODE works as promised. We can fix it by simply removing the zonelist reset completely. There is actually no reason to reset it, because memory policies and cpusets don't affect the zonelist choice in the first place. This was different when commit 183f6371aac2 ("mm: ignore mempolicies when using ALLOC_NO_WATERMARK") introduced the code, as mempolicies provided their own restricted zonelists. We might consider this for 4.17 although I don't know if there's anything currently broken. SLAB is currently not affected, but in kernels older than 4.7 that don't yet have 511e3a058812 ("mm/slab: make cache_grow() handle the page allocated on arbitrary node") it is. That's at least 4.4 LTS. Older ones I'll have to check. So stable backports should be more important, but will have to be reviewed carefully, as the code went through many changes. BTW I think that also the ac->preferred_zoneref reset is currently useless if we don't also reset ac->nodemask from a mempolicy to NULL first (which we probably should for the OOM victims etc?), but I would leave that for a separate patch. Link: http://lkml.kernel.org/r/20180525130853.13915-1-vbabka@suse.cz Signed-off-by: Vlastimil Babka <vbabka@suse.cz> Fixes: 183f6371aac2 ("mm: ignore mempolicies when using ALLOC_NO_WATERMARK") Acked-by: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit 7810e6781e0fcbca78b91cf65053f895bf59e85f upstream.

In __alloc_pages_slowpath() we reset zonelist and preferred_zoneref for
allocations that can ignore memory policies.  The zonelist is obtained
from current CPU's node.  This is a problem for __GFP_THISNODE
allocations that want to allocate on a different node, e.g.  because the
allocating thread has been migrated to a different CPU.

This has been observed to break SLAB in our 4.4-based kernel, because
there it relies on __GFP_THISNODE working as intended.  If a slab page
is put on wrong node's list, then further list manipulations may corrupt
the list because page_to_nid() is used to determine which node's
list_lock should be locked and thus we may take a wrong lock and race.

Current SLAB implementation seems to be immune by luck thanks to commit
511e3a058812 ("mm/slab: make cache_grow() handle the page allocated on
arbitrary node") but there may be others assuming that __GFP_THISNODE
works as promised.

We can fix it by simply removing the zonelist reset completely.  There
is actually no reason to reset it, because memory policies and cpusets
don't affect the zonelist choice in the first place.  This was different
when commit 183f6371aac2 ("mm: ignore mempolicies when using
ALLOC_NO_WATERMARK") introduced the code, as mempolicies provided their
own restricted zonelists.

We might consider this for 4.17 although I don't know if there's
anything currently broken.

SLAB is currently not affected, but in kernels older than 4.7 that don't
yet have 511e3a058812 ("mm/slab: make cache_grow() handle the page
allocated on arbitrary node") it is.  That's at least 4.4 LTS.  Older
ones I'll have to check.

So stable backports should be more important, but will have to be
reviewed carefully, as the code went through many changes.  BTW I think
that also the ac->preferred_zoneref reset is currently useless if we
don't also reset ac->nodemask from a mempolicy to NULL first (which we
probably should for the OOM victims etc?), but I would leave that for a
separate patch.

Link: http://lkml.kernel.org/r/20180525130853.13915-1-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Fixes: 183f6371aac2 ("mm: ignore mempolicies when using ALLOC_NO_WATERMARK")
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
bdi: Move cgroup bdi_writeback to a dedicated low concurrency workqueue 2018-06-25T23:51:27+00:00 Tejun Heo tj@kernel.org 2018-05-23T17:56:32+00:00 1ca4ccb9f38197eb98ad222550ee143e17395386 commit f183464684190bacbfb14623bd3e4e51b7575b4c upstream. From 0aa2e9b921d6db71150633ff290199554f0842a8 Mon Sep 17 00:00:00 2001 From: Tejun Heo <tj@kernel.org> Date: Wed, 23 May 2018 10:29:00 -0700 cgwb_release() punts the actual release to cgwb_release_workfn() on system_wq. Depending on the number of cgroups or block devices, there can be a lot of cgwb_release_workfn() in flight at the same time. We're periodically seeing close to 256 kworkers getting stuck with the following stack trace and overtime the entire system gets stuck. [<ffffffff810ee40c>] _synchronize_rcu_expedited.constprop.72+0x2fc/0x330 [<ffffffff810ee634>] synchronize_rcu_expedited+0x24/0x30 [<ffffffff811ccf23>] bdi_unregister+0x53/0x290 [<ffffffff811cd1e9>] release_bdi+0x89/0xc0 [<ffffffff811cd645>] wb_exit+0x85/0xa0 [<ffffffff811cdc84>] cgwb_release_workfn+0x54/0xb0 [<ffffffff810a68d0>] process_one_work+0x150/0x410 [<ffffffff810a71fd>] worker_thread+0x6d/0x520 [<ffffffff810ad3dc>] kthread+0x12c/0x160 [<ffffffff81969019>] ret_from_fork+0x29/0x40 [<ffffffffffffffff>] 0xffffffffffffffff The events leading to the lockup are... 1. A lot of cgwb_release_workfn() is queued at the same time and all system_wq kworkers are assigned to execute them. 2. They all end up calling synchronize_rcu_expedited(). One of them wins and tries to perform the expedited synchronization. 3. However, that invovles queueing rcu_exp_work to system_wq and waiting for it. Because #1 is holding all available kworkers on system_wq, rcu_exp_work can't be executed. cgwb_release_workfn() is waiting for synchronize_rcu_expedited() which in turn is waiting for cgwb_release_workfn() to free up some of the kworkers. We shouldn't be scheduling hundreds of cgwb_release_workfn() at the same time. There's nothing to be gained from that. This patch updates cgwb release path to use a dedicated percpu workqueue with @max_active of 1. While this resolves the problem at hand, it might be a good idea to isolate rcu_exp_work to its own workqueue too as it can be used from various paths and is prone to this sort of indirect A-A deadlocks. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Cc: stable@vger.kernel.org Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit f183464684190bacbfb14623bd3e4e51b7575b4c upstream.

From 0aa2e9b921d6db71150633ff290199554f0842a8 Mon Sep 17 00:00:00 2001
From: Tejun Heo <tj@kernel.org>
Date: Wed, 23 May 2018 10:29:00 -0700

cgwb_release() punts the actual release to cgwb_release_workfn() on
system_wq.  Depending on the number of cgroups or block devices, there
can be a lot of cgwb_release_workfn() in flight at the same time.

We're periodically seeing close to 256 kworkers getting stuck with the
following stack trace and overtime the entire system gets stuck.

  [<ffffffff810ee40c>] _synchronize_rcu_expedited.constprop.72+0x2fc/0x330
  [<ffffffff810ee634>] synchronize_rcu_expedited+0x24/0x30
  [<ffffffff811ccf23>] bdi_unregister+0x53/0x290
  [<ffffffff811cd1e9>] release_bdi+0x89/0xc0
  [<ffffffff811cd645>] wb_exit+0x85/0xa0
  [<ffffffff811cdc84>] cgwb_release_workfn+0x54/0xb0
  [<ffffffff810a68d0>] process_one_work+0x150/0x410
  [<ffffffff810a71fd>] worker_thread+0x6d/0x520
  [<ffffffff810ad3dc>] kthread+0x12c/0x160
  [<ffffffff81969019>] ret_from_fork+0x29/0x40
  [<ffffffffffffffff>] 0xffffffffffffffff

The events leading to the lockup are...

1. A lot of cgwb_release_workfn() is queued at the same time and all
   system_wq kworkers are assigned to execute them.

2. They all end up calling synchronize_rcu_expedited().  One of them
   wins and tries to perform the expedited synchronization.

3. However, that invovles queueing rcu_exp_work to system_wq and
   waiting for it.  Because #1 is holding all available kworkers on
   system_wq, rcu_exp_work can't be executed.  cgwb_release_workfn()
   is waiting for synchronize_rcu_expedited() which in turn is waiting
   for cgwb_release_workfn() to free up some of the kworkers.

We shouldn't be scheduling hundreds of cgwb_release_workfn() at the
same time.  There's nothing to be gained from that.  This patch
updates cgwb release path to use a dedicated percpu workqueue with
@max_active of 1.

While this resolves the problem at hand, it might be a good idea to
isolate rcu_exp_work to its own workqueue too as it can be used from
various paths and is prone to this sort of indirect A-A deadlocks.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: stable@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mm: fix the NULL mapping case in __isolate_lru_page() 2018-06-02T16:33:47+00:00 Hugh Dickins hughd@google.com 2018-06-01T23:50:50+00:00 145e1a71e090575c74969e3daa8136d1e5b99fc8 George Boole would have noticed a slight error in 4.16 commit 69d763fc6d3a ("mm: pin address_space before dereferencing it while isolating an LRU page"). Fix it, to match both the comment above it, and the original behaviour. Although anonymous pages are not marked PageDirty at first, we have an old habit of calling SetPageDirty when a page is removed from swap cache: so there's a category of ex-swap pages that are easily migratable, but were inadvertently excluded from compaction's async migration in 4.16. Link: http://lkml.kernel.org/r/alpine.LSU.2.11.1805302014001.12558@eggly.anvils Fixes: 69d763fc6d3a ("mm: pin address_space before dereferencing it while isolating an LRU page") Signed-off-by: Hugh Dickins <hughd@google.com> Acked-by: Minchan Kim <minchan@kernel.org> Acked-by: Mel Gorman <mgorman@techsingularity.net> Reported-by: Ivan Kalvachev <ikalvachev@gmail.com> Cc: "Huang, Ying" <ying.huang@intel.com> Cc: Jan Kara <jack@suse.cz> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> 
George Boole would have noticed a slight error in 4.16 commit
69d763fc6d3a ("mm: pin address_space before dereferencing it while
isolating an LRU page").  Fix it, to match both the comment above it,
and the original behaviour.

Although anonymous pages are not marked PageDirty at first, we have an
old habit of calling SetPageDirty when a page is removed from swap
cache: so there's a category of ex-swap pages that are easily
migratable, but were inadvertently excluded from compaction's async
migration in 4.16.

Link: http://lkml.kernel.org/r/alpine.LSU.2.11.1805302014001.12558@eggly.anvils
Fixes: 69d763fc6d3a ("mm: pin address_space before dereferencing it while isolating an LRU page")
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Reported-by:  Ivan Kalvachev <ikalvachev@gmail.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/huge_memory.c: __split_huge_page() use atomic ClearPageDirty() 2018-06-02T16:33:47+00:00 Hugh Dickins hughd@google.com 2018-06-01T23:50:45+00:00 2d077d4b59924acd1f5180c6fb73b57f4771fde6 Swapping load on huge=always tmpfs (with khugepaged tuned up to be very eager, but I'm not sure that is relevant) soon hung uninterruptibly, waiting for page lock in shmem_getpage_gfp()'s find_lock_entry(), most often when "cp -a" was trying to write to a smallish file. Debug showed that the page in question was not locked, and page->mapping NULL by now, but page->index consistent with having been in a huge page before. Reproduced in minutes on a 4.15 kernel, even with 4.17's 605ca5ede764 ("mm/huge_memory.c: reorder operations in __split_huge_page_tail()") added in; but took hours to reproduce on a 4.17 kernel (no idea why). The culprit proved to be the __ClearPageDirty() on tails beyond i_size in __split_huge_page(): the non-atomic __bitoperation may have been safe when 4.8's baa355fd3314 ("thp: file pages support for split_huge_page()") introduced it, but liable to erase PageWaiters after 4.10's 62906027091f ("mm: add PageWaiters indicating tasks are waiting for a page bit"). Link: http://lkml.kernel.org/r/alpine.LSU.2.11.1805291841070.3197@eggly.anvils Fixes: 62906027091f ("mm: add PageWaiters indicating tasks are waiting for a page bit") Signed-off-by: Hugh Dickins <hughd@google.com> Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> 
Swapping load on huge=always tmpfs (with khugepaged tuned up to be very
eager, but I'm not sure that is relevant) soon hung uninterruptibly,
waiting for page lock in shmem_getpage_gfp()'s find_lock_entry(), most
often when "cp -a" was trying to write to a smallish file.  Debug showed
that the page in question was not locked, and page->mapping NULL by now,
but page->index consistent with having been in a huge page before.

Reproduced in minutes on a 4.15 kernel, even with 4.17's 605ca5ede764
("mm/huge_memory.c: reorder operations in __split_huge_page_tail()") added
in; but took hours to reproduce on a 4.17 kernel (no idea why).

The culprit proved to be the __ClearPageDirty() on tails beyond i_size in
__split_huge_page(): the non-atomic __bitoperation may have been safe when
4.8's baa355fd3314 ("thp: file pages support for split_huge_page()")
introduced it, but liable to erase PageWaiters after 4.10's 62906027091f
("mm: add PageWaiters indicating tasks are waiting for a page bit").

Link: http://lkml.kernel.org/r/alpine.LSU.2.11.1805291841070.3197@eggly.anvils
Fixes: 62906027091f ("mm: add PageWaiters indicating tasks are waiting for a page bit")
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kasan: fix memory hotplug during boot 2018-05-26T01:12:11+00:00 David Hildenbrand david@redhat.com 2018-05-25T21:48:11+00:00 3f1959721558a976aaf9c2024d5bc884e6411bf7 Using module_init() is wrong. E.g. ACPI adds and onlines memory before our memory notifier gets registered. This makes sure that ACPI memory detected during boot up will not result in a kernel crash. Easily reproducible with QEMU, just specify a DIMM when starting up. Link: http://lkml.kernel.org/r/20180522100756.18478-3-david@redhat.com Fixes: 786a8959912e ("kasan: disable memory hotplug") Signed-off-by: David Hildenbrand <david@redhat.com> Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Cc: Alexander Potapenko <glider@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> 
Using module_init() is wrong.  E.g.  ACPI adds and onlines memory before
our memory notifier gets registered.

This makes sure that ACPI memory detected during boot up will not result
in a kernel crash.

Easily reproducible with QEMU, just specify a DIMM when starting up.

Link: http://lkml.kernel.org/r/20180522100756.18478-3-david@redhat.com
Fixes: 786a8959912e ("kasan: disable memory hotplug")
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kasan: free allocated shadow memory on MEM_CANCEL_ONLINE 2018-05-26T01:12:11+00:00 David Hildenbrand david@redhat.com 2018-05-25T21:48:08+00:00 ed1596f9ab958dd156a66c9ff1029d3761c1786a We have to free memory again when we cancel onlining, otherwise a later onlining attempt will fail. Link: http://lkml.kernel.org/r/20180522100756.18478-2-david@redhat.com Fixes: fa69b5989bb0 ("mm/kasan: add support for memory hotplug") Signed-off-by: David Hildenbrand <david@redhat.com> Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Cc: Alexander Potapenko <glider@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> 
We have to free memory again when we cancel onlining, otherwise a later
onlining attempt will fail.

Link: http://lkml.kernel.org/r/20180522100756.18478-2-david@redhat.com
Fixes: fa69b5989bb0 ("mm/kasan: add support for memory hotplug")
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memory_hotplug: fix leftover use of struct page during hotplug 2018-05-26T01:12:11+00:00 Jonathan Cameron Jonathan.Cameron@huawei.com 2018-05-25T21:47:53+00:00 a21558618c5dfc55b6086743a88ce5a9c1588f0a The case of a new numa node got missed in avoiding using the node info from page_struct during hotplug. In this path we have a call to register_mem_sect_under_node (which allows us to specify it is hotplug so don't change the node), via link_mem_sections which unfortunately does not. Fix is to pass check_nid through link_mem_sections as well and disable it in the new numa node path. Note the bug only 'sometimes' manifests depending on what happens to be in the struct page structures - there are lots of them and it only needs to match one of them. The result of the bug is that (with a new memory only node) we never successfully call register_mem_sect_under_node so don't get the memory associated with the node in sysfs and meminfo for the node doesn't report it. It came up whilst testing some arm64 hotplug patches, but appears to be universal. Whilst I'm triggering it by removing then reinserting memory to a node with no other elements (thus making the node disappear then appear again), it appears it would happen on hotplugging memory where there was none before and it doesn't seem to be related the arm64 patches. These patches call __add_pages (where most of the issue was fixed by Pavel's patch). If there is a node at the time of the __add_pages call then all is well as it calls register_mem_sect_under_node from there with check_nid set to false. Without a node that function returns having not done the sysfs related stuff as there is no node to use. This is expected but it is the resulting path that fails... Exact path to the problem is as follows: mm/memory_hotplug.c: add_memory_resource() The node is not online so we enter the 'if (new_node)' twice, on the second such block there is a call to link_mem_sections which calls into drivers/node.c: link_mem_sections() which calls drivers/node.c: register_mem_sect_under_node() which calls get_nid_for_pfn and keeps trying until the output of that matches the expected node (passed all the way down from add_memory_resource) It is effectively the same fix as the one referred to in the fixes tag just in the code path for a new node where the comments point out we have to rerun the link creation because it will have failed in register_new_memory (as there was no node at the time). (actually that comment is wrong now as we don't have register_new_memory any more it got renamed to hotplug_memory_register in Pavel's patch). Link: http://lkml.kernel.org/r/20180504085311.1240-1-Jonathan.Cameron@huawei.com Fixes: fc44f7f9231a ("mm/memory_hotplug: don't read nid from struct page during hotplug") Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com> Reviewed-by: Pavel Tatashin <pasha.tatashin@oracle.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> 
The case of a new numa node got missed in avoiding using the node info
from page_struct during hotplug.  In this path we have a call to
register_mem_sect_under_node (which allows us to specify it is hotplug
so don't change the node), via link_mem_sections which unfortunately
does not.

Fix is to pass check_nid through link_mem_sections as well and disable
it in the new numa node path.

Note the bug only 'sometimes' manifests depending on what happens to be
in the struct page structures - there are lots of them and it only needs
to match one of them.

The result of the bug is that (with a new memory only node) we never
successfully call register_mem_sect_under_node so don't get the memory
associated with the node in sysfs and meminfo for the node doesn't
report it.

It came up whilst testing some arm64 hotplug patches, but appears to be
universal.  Whilst I'm triggering it by removing then reinserting memory
to a node with no other elements (thus making the node disappear then
appear again), it appears it would happen on hotplugging memory where
there was none before and it doesn't seem to be related the arm64
patches.

These patches call __add_pages (where most of the issue was fixed by
Pavel's patch).  If there is a node at the time of the __add_pages call
then all is well as it calls register_mem_sect_under_node from there
with check_nid set to false.  Without a node that function returns
having not done the sysfs related stuff as there is no node to use.
This is expected but it is the resulting path that fails...

Exact path to the problem is as follows:

 mm/memory_hotplug.c: add_memory_resource()

   The node is not online so we enter the 'if (new_node)' twice, on the
   second such block there is a call to link_mem_sections which calls
   into

  drivers/node.c: link_mem_sections() which calls

  drivers/node.c: register_mem_sect_under_node() which calls
     get_nid_for_pfn and keeps trying until the output of that matches
     the expected node (passed all the way down from
     add_memory_resource)

It is effectively the same fix as the one referred to in the fixes tag
just in the code path for a new node where the comments point out we
have to rerun the link creation because it will have failed in
register_new_memory (as there was no node at the time).  (actually that
comment is wrong now as we don't have register_new_memory any more it
got renamed to hotplug_memory_register in Pavel's patch).

Link: http://lkml.kernel.org/r/20180504085311.1240-1-Jonathan.Cameron@huawei.com
Fixes: fc44f7f9231a ("mm/memory_hotplug: don't read nid from struct page during hotplug")
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Reviewed-by: Pavel Tatashin <pasha.tatashin@oracle.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>
mm, memory_hotplug: make has_unmovable_pages more robust 2018-05-26T01:12:11+00:00 Michal Hocko mhocko@suse.com 2018-05-25T21:47:42+00:00 15c30bc0908514b4a72fadc3995d32313500393a Oscar has reported: : Due to an unfortunate setting with movablecore, memblocks containing bootmem : memory (pages marked by get_page_bootmem()) ended up marked in zone_movable. : So while trying to remove that memory, the system failed in do_migrate_range : and __offline_pages never returned. : : This can be reproduced by running : qemu-system-x86_64 -m 6G,slots=8,maxmem=8G -numa node,mem=4096M -numa node,mem=2048M : and movablecore=4G kernel command line : : linux kernel: BIOS-provided physical RAM map: : linux kernel: BIOS-e820: [mem 0x0000000000000000-0x000000000009fbff] usable : linux kernel: BIOS-e820: [mem 0x000000000009fc00-0x000000000009ffff] reserved : linux kernel: BIOS-e820: [mem 0x00000000000f0000-0x00000000000fffff] reserved : linux kernel: BIOS-e820: [mem 0x0000000000100000-0x00000000bffdffff] usable : linux kernel: BIOS-e820: [mem 0x00000000bffe0000-0x00000000bfffffff] reserved : linux kernel: BIOS-e820: [mem 0x00000000feffc000-0x00000000feffffff] reserved : linux kernel: BIOS-e820: [mem 0x00000000fffc0000-0x00000000ffffffff] reserved : linux kernel: BIOS-e820: [mem 0x0000000100000000-0x00000001bfffffff] usable : linux kernel: NX (Execute Disable) protection: active : linux kernel: SMBIOS 2.8 present. : linux kernel: DMI: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.0.0-prebuilt.qemu-project.org : linux kernel: Hypervisor detected: KVM : linux kernel: e820: update [mem 0x00000000-0x00000fff] usable ==> reserved : linux kernel: e820: remove [mem 0x000a0000-0x000fffff] usable : linux kernel: last_pfn = 0x1c0000 max_arch_pfn = 0x400000000 : : linux kernel: SRAT: PXM 0 -> APIC 0x00 -> Node 0 : linux kernel: SRAT: PXM 1 -> APIC 0x01 -> Node 1 : linux kernel: ACPI: SRAT: Node 0 PXM 0 [mem 0x00000000-0x0009ffff] : linux kernel: ACPI: SRAT: Node 0 PXM 0 [mem 0x00100000-0xbfffffff] : linux kernel: ACPI: SRAT: Node 0 PXM 0 [mem 0x100000000-0x13fffffff] : linux kernel: ACPI: SRAT: Node 1 PXM 1 [mem 0x140000000-0x1bfffffff] : linux kernel: ACPI: SRAT: Node 0 PXM 0 [mem 0x1c0000000-0x43fffffff] hotplug : linux kernel: NUMA: Node 0 [mem 0x00000000-0x0009ffff] + [mem 0x00100000-0xbfffffff] -> [mem 0x0 : linux kernel: NUMA: Node 0 [mem 0x00000000-0xbfffffff] + [mem 0x100000000-0x13fffffff] -> [mem 0 : linux kernel: NODE_DATA(0) allocated [mem 0x13ffd6000-0x13fffffff] : linux kernel: NODE_DATA(1) allocated [mem 0x1bffd3000-0x1bfffcfff] : : zoneinfo shows that the zone movable is placed into both numa nodes: : Node 0, zone Movable : pages free 160140 : min 1823 : low 2278 : high 2733 : spanned 262144 : present 262144 : managed 245670 : Node 1, zone Movable : pages free 448427 : min 3827 : low 4783 : high 5739 : spanned 524288 : present 524288 : managed 515766 Note how only Node 0 has a hutplugable memory region which would rule it out from the early memblock allocations (most likely memmap). Node1 will surely contain memmaps on the same node and those would prevent offlining to succeed. So this is arguably a configuration issue. Although one could argue that we should be more clever and rule early allocations from the zone movable. This would be correct but probably not worth the effort considering what a hack movablecore is. Anyway, We could do better for those cases though. We rely on start_isolate_page_range resp. has_unmovable_pages to do their job. The first one isolates the whole range to be offlined so that we do not allocate from it anymore and the later makes sure we are not stumbling over non-migrateable pages. has_unmovable_pages is overly optimistic, however. It doesn't check all the pages if we are withing zone_movable because we rely that those pages will be always migrateable. As it turns out we are still not perfect there. While bootmem pages in zonemovable sound like a clear bug which should be fixed let's remove the optimization for now and warn if we encounter unmovable pages in zone_movable in the meantime. That should help for now at least. Btw. this wasn't a real problem until commit 72b39cfc4d75 ("mm, memory_hotplug: do not fail offlining too early") because we used to have a small number of retries and then failed. This turned out to be too fragile though. Link: http://lkml.kernel.org/r/20180523125555.30039-2-mhocko@kernel.org Signed-off-by: Michal Hocko <mhocko@suse.com> Reported-by: Oscar Salvador <osalvador@techadventures.net> Tested-by: Oscar Salvador <osalvador@techadventures.net> Reviewed-by: Pavel Tatashin <pasha.tatashin@oracle.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Reza Arbab <arbab@linux.vnet.ibm.com> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Vitaly Kuznetsov <vkuznets@redhat.com> Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> 
Oscar has reported:
: Due to an unfortunate setting with movablecore, memblocks containing bootmem
: memory (pages marked by get_page_bootmem()) ended up marked in zone_movable.
: So while trying to remove that memory, the system failed in do_migrate_range
: and __offline_pages never returned.
:
: This can be reproduced by running
: qemu-system-x86_64 -m 6G,slots=8,maxmem=8G -numa node,mem=4096M -numa node,mem=2048M
: and movablecore=4G kernel command line
:
: linux kernel: BIOS-provided physical RAM map:
: linux kernel: BIOS-e820: [mem 0x0000000000000000-0x000000000009fbff] usable
: linux kernel: BIOS-e820: [mem 0x000000000009fc00-0x000000000009ffff] reserved
: linux kernel: BIOS-e820: [mem 0x00000000000f0000-0x00000000000fffff] reserved
: linux kernel: BIOS-e820: [mem 0x0000000000100000-0x00000000bffdffff] usable
: linux kernel: BIOS-e820: [mem 0x00000000bffe0000-0x00000000bfffffff] reserved
: linux kernel: BIOS-e820: [mem 0x00000000feffc000-0x00000000feffffff] reserved
: linux kernel: BIOS-e820: [mem 0x00000000fffc0000-0x00000000ffffffff] reserved
: linux kernel: BIOS-e820: [mem 0x0000000100000000-0x00000001bfffffff] usable
: linux kernel: NX (Execute Disable) protection: active
: linux kernel: SMBIOS 2.8 present.
: linux kernel: DMI: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.0.0-prebuilt.qemu-project.org
: linux kernel: Hypervisor detected: KVM
: linux kernel: e820: update [mem 0x00000000-0x00000fff] usable ==> reserved
: linux kernel: e820: remove [mem 0x000a0000-0x000fffff] usable
: linux kernel: last_pfn = 0x1c0000 max_arch_pfn = 0x400000000
:
: linux kernel: SRAT: PXM 0 -> APIC 0x00 -> Node 0
: linux kernel: SRAT: PXM 1 -> APIC 0x01 -> Node 1
: linux kernel: ACPI: SRAT: Node 0 PXM 0 [mem 0x00000000-0x0009ffff]
: linux kernel: ACPI: SRAT: Node 0 PXM 0 [mem 0x00100000-0xbfffffff]
: linux kernel: ACPI: SRAT: Node 0 PXM 0 [mem 0x100000000-0x13fffffff]
: linux kernel: ACPI: SRAT: Node 1 PXM 1 [mem 0x140000000-0x1bfffffff]
: linux kernel: ACPI: SRAT: Node 0 PXM 0 [mem 0x1c0000000-0x43fffffff] hotplug
: linux kernel: NUMA: Node 0 [mem 0x00000000-0x0009ffff] + [mem 0x00100000-0xbfffffff] -> [mem 0x0
: linux kernel: NUMA: Node 0 [mem 0x00000000-0xbfffffff] + [mem 0x100000000-0x13fffffff] -> [mem 0
: linux kernel: NODE_DATA(0) allocated [mem 0x13ffd6000-0x13fffffff]
: linux kernel: NODE_DATA(1) allocated [mem 0x1bffd3000-0x1bfffcfff]
:
: zoneinfo shows that the zone movable is placed into both numa nodes:
: Node 0, zone  Movable
:   pages free     160140
:         min      1823
:         low      2278
:         high     2733
:         spanned  262144
:         present  262144
:         managed  245670
: Node 1, zone  Movable
:   pages free     448427
:         min      3827
:         low      4783
:         high     5739
:         spanned  524288
:         present  524288
:         managed  515766

Note how only Node 0 has a hutplugable memory region which would rule it
out from the early memblock allocations (most likely memmap).  Node1
will surely contain memmaps on the same node and those would prevent
offlining to succeed.  So this is arguably a configuration issue.
Although one could argue that we should be more clever and rule early
allocations from the zone movable.  This would be correct but probably
not worth the effort considering what a hack movablecore is.

Anyway, We could do better for those cases though.  We rely on
start_isolate_page_range resp.  has_unmovable_pages to do their job.
The first one isolates the whole range to be offlined so that we do not
allocate from it anymore and the later makes sure we are not stumbling
over non-migrateable pages.

has_unmovable_pages is overly optimistic, however.  It doesn't check all
the pages if we are withing zone_movable because we rely that those
pages will be always migrateable.  As it turns out we are still not
perfect there.  While bootmem pages in zonemovable sound like a clear
bug which should be fixed let's remove the optimization for now and warn
if we encounter unmovable pages in zone_movable in the meantime.  That
should help for now at least.

Btw.  this wasn't a real problem until commit 72b39cfc4d75 ("mm,
memory_hotplug: do not fail offlining too early") because we used to
have a small number of retries and then failed.  This turned out to be
too fragile though.

Link: http://lkml.kernel.org/r/20180523125555.30039-2-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Oscar Salvador <osalvador@techadventures.net>
Tested-by: Oscar Salvador <osalvador@techadventures.net>
Reviewed-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Reza Arbab <arbab@linux.vnet.ibm.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/kasan: don't vfree() nonexistent vm_area 2018-05-26T01:12:11+00:00 Andrey Ryabinin aryabinin@virtuozzo.com 2018-05-25T21:47:38+00:00 0f901dcbc31f88ae41a2aaa365f7802b5d520a28 KASAN uses different routines to map shadow for hot added memory and memory obtained in boot process. Attempt to offline memory onlined by normal boot process leads to this: Trying to vfree() nonexistent vm area (000000005d3b34b9) WARNING: CPU: 2 PID: 13215 at mm/vmalloc.c:1525 __vunmap+0x147/0x190 Call Trace: kasan_mem_notifier+0xad/0xb9 notifier_call_chain+0x166/0x260 __blocking_notifier_call_chain+0xdb/0x140 __offline_pages+0x96a/0xb10 memory_subsys_offline+0x76/0xc0 device_offline+0xb8/0x120 store_mem_state+0xfa/0x120 kernfs_fop_write+0x1d5/0x320 __vfs_write+0xd4/0x530 vfs_write+0x105/0x340 SyS_write+0xb0/0x140 Obviously we can't call vfree() to free memory that wasn't allocated via vmalloc(). Use find_vm_area() to see if we can call vfree(). Unfortunately it's a bit tricky to properly unmap and free shadow allocated during boot, so we'll have to keep it. If memory will come online again that shadow will be reused. Matthew asked: how can you call vfree() on something that isn't a vmalloc address? vfree() is able to free any address returned by __vmalloc_node_range(). And __vmalloc_node_range() gives you any address you ask. It doesn't have to be an address in [VMALLOC_START, VMALLOC_END] range. That's also how the module_alloc()/module_memfree() works on architectures that have designated area for modules. [aryabinin@virtuozzo.com: improve comments] Link: http://lkml.kernel.org/r/dabee6ab-3a7a-51cd-3b86-5468718e0390@virtuozzo.com [akpm@linux-foundation.org: fix typos, reflow comment] Link: http://lkml.kernel.org/r/20180201163349.8700-1-aryabinin@virtuozzo.com Fixes: fa69b5989bb0 ("mm/kasan: add support for memory hotplug") Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Reported-by: Paul Menzel <pmenzel+linux-kasan-dev@molgen.mpg.de> Cc: Alexander Potapenko <glider@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> 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> 
KASAN uses different routines to map shadow for hot added memory and
memory obtained in boot process.  Attempt to offline memory onlined by
normal boot process leads to this:

    Trying to vfree() nonexistent vm area (000000005d3b34b9)
    WARNING: CPU: 2 PID: 13215 at mm/vmalloc.c:1525 __vunmap+0x147/0x190

    Call Trace:
     kasan_mem_notifier+0xad/0xb9
     notifier_call_chain+0x166/0x260
     __blocking_notifier_call_chain+0xdb/0x140
     __offline_pages+0x96a/0xb10
     memory_subsys_offline+0x76/0xc0
     device_offline+0xb8/0x120
     store_mem_state+0xfa/0x120
     kernfs_fop_write+0x1d5/0x320
     __vfs_write+0xd4/0x530
     vfs_write+0x105/0x340
     SyS_write+0xb0/0x140

Obviously we can't call vfree() to free memory that wasn't allocated via
vmalloc().  Use find_vm_area() to see if we can call vfree().

Unfortunately it's a bit tricky to properly unmap and free shadow
allocated during boot, so we'll have to keep it.  If memory will come
online again that shadow will be reused.

Matthew asked: how can you call vfree() on something that isn't a
vmalloc address?

  vfree() is able to free any address returned by
  __vmalloc_node_range().  And __vmalloc_node_range() gives you any
  address you ask.  It doesn't have to be an address in [VMALLOC_START,
  VMALLOC_END] range.

  That's also how the module_alloc()/module_memfree() works on
  architectures that have designated area for modules.

[aryabinin@virtuozzo.com: improve comments]
  Link: http://lkml.kernel.org/r/dabee6ab-3a7a-51cd-3b86-5468718e0390@virtuozzo.com
[akpm@linux-foundation.org: fix typos, reflow comment]
Link: http://lkml.kernel.org/r/20180201163349.8700-1-aryabinin@virtuozzo.com
Fixes: fa69b5989bb0 ("mm/kasan: add support for memory hotplug")
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Reported-by: Paul Menzel <pmenzel+linux-kasan-dev@molgen.mpg.de>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
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>
mm: fix nr_rotate_swap leak in swapon() error case 2018-05-26T01:12:10+00:00 Omar Sandoval osandov@fb.com 2018-05-25T21:47:17+00:00 7cbf319234adaa4518a28c8c523d3330e06638f0 If swapon() fails after incrementing nr_rotate_swap, we don't decrement it and thus effectively leak it. Make sure we decrement it if we incremented it. Link: http://lkml.kernel.org/r/b6fe6b879f17fa68eee6cbd876f459f6e5e33495.1526491581.git.osandov@fb.com Fixes: 81a0298bdfab ("mm, swap: don't use VMA based swap readahead if HDD is used as swap") Signed-off-by: Omar Sandoval <osandov@fb.com> Reviewed-by: Rik van Riel <riel@surriel.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> 
If swapon() fails after incrementing nr_rotate_swap, we don't decrement
it and thus effectively leak it.  Make sure we decrement it if we
incremented it.

Link: http://lkml.kernel.org/r/b6fe6b879f17fa68eee6cbd876f459f6e5e33495.1526491581.git.osandov@fb.com
Fixes: 81a0298bdfab ("mm, swap: don't use VMA based swap readahead if HDD is used as swap")
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Rik van Riel <riel@surriel.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>