linux-toradex.git/mm, branch v3.12.14

mm, thp: fix infinite loop on memcg OOM

2014-03-05T16:13:54+00:00

commit 9845cbbd113fbb5b769a45d8e88dc47bc12df4e0 upstream.

Masayoshi Mizuma reported a bug with the hang of an application under
the memcg limit.  It happens on write-protection fault to huge zero page

If we successfully allocate a huge page to replace zero page but hit the
memcg limit we need to split the zero page with split_huge_page_pmd()
and fallback to small pages.

The other part of the problem is that VM_FAULT_OOM has special meaning
in do_huge_pmd_wp_page() context.  __handle_mm_fault() expects the page
to be split if it sees VM_FAULT_OOM and it will will retry page fault
handling.  This causes an infinite loop if the page was not split.

do_huge_pmd_wp_zero_page_fallback() can return VM_FAULT_OOM if it failed
to allocate one small page, so fallback to small pages will not help.

The solution for this part is to replace VM_FAULT_OOM with
VM_FAULT_FALLBACK is fallback required.

Signed-off-by: Kirill A. Shutemov 
Reported-by: Masayoshi Mizuma 
Reviewed-by: Michal Hocko 
Cc: Johannes Weiner 
Cc: Andrea Arcangeli 
Cc: David Rientjes 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Jiri Slaby

mm/memory-failure.c: move refcount only in !MF_COUNT_INCREASED

2014-02-22T21:32:24+00:00

commit 8d547ff4ac5927245e0833ac18528f939da0ee0e upstream.

mce-test detected a test failure when injecting error to a thp tail
page.  This is because we take page refcount of the tail page in
madvise_hwpoison() while the fix in commit a3e0f9e47d5e
("mm/memory-failure.c: transfer page count from head page to tail page
after split thp") assumes that we always take refcount on the head page.

When a real memory error happens we take refcount on the head page where
memory_failure() is called without MF_COUNT_INCREASED set, so it seems
to me that testing memory error on thp tail page using madvise makes
little sense.

This patch cancels moving refcount in !MF_COUNT_INCREASED for valid
testing.

[akpm@linux-foundation.org: s/&&/&/]
Signed-off-by: Naoya Horiguchi 
Cc: Andi Kleen 
Cc: Wanpeng Li 
Cc: Chen Gong 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm: __set_page_dirty_nobuffers() uses spin_lock_irqsave() instead of spin_lock_irq()

2014-02-20T19:08:00+00:00

commit a85d9df1ea1d23682a0ed1e100e6965006595d06 upstream.

During aio stress test, we observed the following lockdep warning.  This
mean AIO+numa_balancing is currently deadlockable.

The problem is, aio_migratepage disable interrupt, but
__set_page_dirty_nobuffers unintentionally enable it again.

Generally, all helper function should use spin_lock_irqsave() instead of
spin_lock_irq() because they don't know caller at all.

   other info that might help us debug this:
    Possible unsafe locking scenario:

          CPU0
          ----
     lock(&(&ctx->completion_lock)->rlock);
     
       lock(&(&ctx->completion_lock)->rlock);

    *** DEADLOCK ***

      dump_stack+0x19/0x1b
      print_usage_bug+0x1f7/0x208
      mark_lock+0x21d/0x2a0
      mark_held_locks+0xb9/0x140
      trace_hardirqs_on_caller+0x105/0x1d0
      trace_hardirqs_on+0xd/0x10
      _raw_spin_unlock_irq+0x2c/0x50
      __set_page_dirty_nobuffers+0x8c/0xf0
      migrate_page_copy+0x434/0x540
      aio_migratepage+0xb1/0x140
      move_to_new_page+0x7d/0x230
      migrate_pages+0x5e5/0x700
      migrate_misplaced_page+0xbc/0xf0
      do_numa_page+0x102/0x190
      handle_pte_fault+0x241/0x970
      handle_mm_fault+0x265/0x370
      __do_page_fault+0x172/0x5a0
      do_page_fault+0x1a/0x70
      page_fault+0x28/0x30

Signed-off-by: KOSAKI Motohiro 
Cc: Larry Woodman 
Cc: Rik van Riel 
Cc: Johannes Weiner 
Acked-by: David Rientjes 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm/swap: fix race on swap_info reuse between swapoff and swapon

2014-02-20T19:08:00+00:00

commit f893ab41e4dae2fe8991faf5d86d029068d1ef3a upstream.

swapoff clear swap_info's SWP_USED flag prematurely and free its
resources after that.  A concurrent swapon will reuse this swap_info
while its previous resources are not cleared completely.

These late freed resources are:
 - p->percpu_cluster
 - swap_cgroup_ctrl[type]
 - block_device setting
 - inode->i_flags &= ~S_SWAPFILE

This patch clears the SWP_USED flag after all its resources are freed,
so that swapon can reuse this swap_info by alloc_swap_info() safely.

[akpm@linux-foundation.org: tidy up code comment]
Signed-off-by: Weijie Yang 
Acked-by: Hugh Dickins 
Cc: Krzysztof Kozlowski 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm, oom: base root bonus on current usage

2014-02-13T21:50:19+00:00

commit 778c14affaf94a9e4953179d3e13a544ccce7707 upstream.

A 3% of system memory bonus is sometimes too excessive in comparison to
other processes.

With commit a63d83f427fb ("oom: badness heuristic rewrite"), the OOM
killer tries to avoid killing privileged tasks by subtracting 3% of
overall memory (system or cgroup) from their per-task consumption.  But
as a result, all root tasks that consume less than 3% of overall memory
are considered equal, and so it only takes 33+ privileged tasks pushing
the system out of memory for the OOM killer to do something stupid and
kill dhclient or other root-owned processes.  For example, on a 32G
machine it can't tell the difference between the 1M agetty and the 10G
fork bomb member.

The changelog describes this 3% boost as the equivalent to the global
overcommit limit being 3% higher for privileged tasks, but this is not
the same as discounting 3% of overall memory from _every privileged task
individually_ during OOM selection.

Replace the 3% of system memory bonus with a 3% of current memory usage
bonus.

By giving root tasks a bonus that is proportional to their actual size,
they remain comparable even when relatively small.  In the example
above, the OOM killer will discount the 1M agetty's 256 badness points
down to 179, and the 10G fork bomb's 262144 points down to 183500 points
and make the right choice, instead of discounting both to 0 and killing
agetty because it's first in the task list.

Signed-off-by: David Rientjes 
Reported-by: Johannes Weiner 
Acked-by: Johannes Weiner 
Cc: Michal Hocko 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

slub: Fix calculation of cpu slabs

2014-02-13T21:50:13+00:00

commit 8afb1474db4701d1ab80cd8251137a3260e6913e upstream.

  /sys/kernel/slab/:t-0000048 # cat cpu_slabs
  231 N0=16 N1=215
  /sys/kernel/slab/:t-0000048 # cat slabs
  145 N0=36 N1=109

See, the number of slabs is smaller than that of cpu slabs.

The bug was introduced by commit 49e2258586b423684f03c278149ab46d8f8b6700
("slub: per cpu cache for partial pages").

We should use page->pages instead of page->pobjects when calculating
the number of cpu partial slabs. This also fixes the mapping of slabs
and nodes.

As there's no variable storing the number of total/active objects in
cpu partial slabs, and we don't have user interfaces requiring those
statistics, I just add WARN_ON for those cases.

Acked-by: Christoph Lameter 
Reviewed-by: Wanpeng Li 
Signed-off-by: Li Zefan 
Signed-off-by: Pekka Enberg 
Signed-off-by: Greg Kroah-Hartman

mm: ignore VM_SOFTDIRTY on VMA merging

2014-02-13T21:50:12+00:00

commit 34228d473efe764d4db7c0536375f0c993e6e06a upstream.

The VM_SOFTDIRTY bit affects vma merge routine: if two VMAs has all bits
in vm_flags matched except dirty bit the kernel can't longer merge them
and this forces the kernel to generate new VMAs instead.

It finally may lead to the situation when userspace application reaches
vm.max_map_count limit and get crashed in worse case

 | (gimp:11768): GLib-ERROR **: gmem.c:110: failed to allocate 4096 bytes
 |
 | (file-tiff-load:12038): LibGimpBase-WARNING **: file-tiff-load: gimp_wire_read(): error
 | xinit: connection to X server lost
 |
 | waiting for X server to shut down
 | /usr/lib64/gimp/2.0/plug-ins/file-tiff-load terminated: Hangup
 | /usr/lib64/gimp/2.0/plug-ins/script-fu terminated: Hangup
 | /usr/lib64/gimp/2.0/plug-ins/script-fu terminated: Hangup

  https://bugzilla.kernel.org/show_bug.cgi?id=67651
  https://bugzilla.gnome.org/show_bug.cgi?id=719619#c0

Initial problem came from missed VM_SOFTDIRTY in do_brk() routine but
even if we would set up VM_SOFTDIRTY here, there is still a way to
prevent VMAs from merging: one can call

 | echo 4 > /proc/$PID/clear_refs

and clear all VM_SOFTDIRTY over all VMAs presented in memory map, then
new do_brk() will try to extend old VMA and finds that dirty bit doesn't
match thus new VMA will be generated.

As discussed with Pavel, the right approach should be to ignore
VM_SOFTDIRTY bit when we're trying to merge VMAs and if merge successed
we mark extended VMA with dirty bit where needed.

Signed-off-by: Cyrill Gorcunov 
Reported-by: Bastian Hougaard 
Reported-by: Mel Gorman 
Cc: Pavel Emelyanov 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

memcg: fix css reference leak and endless loop in mem_cgroup_iter

2014-02-13T21:50:11+00:00

commit 0eef615665ede1e0d603ea9ecca88c1da6f02234 upstream.

Commit 19f39402864e ("memcg: simplify mem_cgroup_iter") has reorganized
mem_cgroup_iter code in order to simplify it.  A part of that change was
dropping an optimization which didn't call css_tryget on the root of the
walked tree.  The patch however didn't change the css_put part in
mem_cgroup_iter which excludes root.

This wasn't an issue at the time because __mem_cgroup_iter_next bailed
out for root early without taking a reference as cgroup iterators
(css_next_descendant_pre) didn't visit root themselves.

Nevertheless cgroup iterators have been reworked to visit root by commit
bd8815a6d802 ("cgroup: make css_for_each_descendant() and friends
include the origin css in the iteration") when the root bypass have been
dropped in __mem_cgroup_iter_next.  This means that css_put is not
called for root and so css along with mem_cgroup and other cgroup
internal object tied by css lifetime are never freed.

Fix the issue by reintroducing root check in __mem_cgroup_iter_next and
do not take css reference for it.

This reference counting magic protects us also from another issue, an
endless loop reported by Hugh Dickins when reclaim races with root
removal and css_tryget called by iterator internally would fail.  There
would be no other nodes to visit so __mem_cgroup_iter_next would return
NULL and mem_cgroup_iter would interpret it as "start looping from root
again" and so mem_cgroup_iter would loop forever internally.

Signed-off-by: Michal Hocko 
Reported-by: Hugh Dickins 
Tested-by: Hugh Dickins 
Cc: Johannes Weiner 
Cc: Greg Thelen 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

memcg: fix endless loop caused by mem_cgroup_iter

2014-02-13T21:50:11+00:00

commit ecc736fc3c71c411a9d201d8588c9e7e049e5d8c upstream.

Hugh has reported an endless loop when the hardlimit reclaim sees the
same group all the time.  This might happen when the reclaim races with
the memcg removal.

shrink_zone
                                                [rmdir root]
  mem_cgroup_iter(root, NULL, reclaim)
    // prev = NULL
    rcu_read_lock()
    mem_cgroup_iter_load
      last_visited = iter->last_visited   // gets root || NULL
      css_tryget(last_visited)            // failed
      last_visited = NULL                 [1]
    memcg = root = __mem_cgroup_iter_next(root, NULL)
    mem_cgroup_iter_update
      iter->last_visited = root;
    reclaim->generation = iter->generation

 mem_cgroup_iter(root, root, reclaim)
   // prev = root
   rcu_read_lock
    mem_cgroup_iter_load
      last_visited = iter->last_visited   // gets root
      css_tryget(last_visited)            // failed
    [1]

The issue seemed to be introduced by commit 5f5781619718 ("memcg: relax
memcg iter caching") which has replaced unconditional css_get/css_put by
css_tryget/css_put for the cached iterator.

This patch fixes the issue by skipping css_tryget on the root of the
tree walk in mem_cgroup_iter_load and symmetrically doesn't release it
in mem_cgroup_iter_update.

Signed-off-by: Michal Hocko 
Reported-by: Hugh Dickins 
Tested-by: Hugh Dickins 
Cc: Johannes Weiner 
Cc: Greg Thelen 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm/page-writeback.c: do not count anon pages as dirtyable memory

2014-02-13T21:50:11+00:00

commit a1c3bfb2f67ef766de03f1f56bdfff9c8595ab14 upstream.

The VM is currently heavily tuned to avoid swapping.  Whether that is
good or bad is a separate discussion, but as long as the VM won't swap
to make room for dirty cache, we can not consider anonymous pages when
calculating the amount of dirtyable memory, the baseline to which
dirty_background_ratio and dirty_ratio are applied.

A simple workload that occupies a significant size (40+%, depending on
memory layout, storage speeds etc.) of memory with anon/tmpfs pages and
uses the remainder for a streaming writer demonstrates this problem.  In
that case, the actual cache pages are a small fraction of what is
considered dirtyable overall, which results in an relatively large
portion of the cache pages to be dirtied.  As kswapd starts rotating
these, random tasks enter direct reclaim and stall on IO.

Only consider free pages and file pages dirtyable.

Signed-off-by: Johannes Weiner 
Reported-by: Tejun Heo 
Tested-by: Tejun Heo 
Reviewed-by: Rik van Riel 
Cc: Mel Gorman 
Cc: Wu Fengguang 
Reviewed-by: Michal Hocko 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman