]> git.baikalelectronics.ru Git - kernel.git/commitdiff
mm, memcg: do not high throttle allocators based on wraparound
authorJakub Kicinski <kuba@kernel.org>
Fri, 10 Apr 2020 21:32:19 +0000 (14:32 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 10 Apr 2020 22:36:20 +0000 (15:36 -0700)
If a cgroup violates its memory.high constraints, we may end up unduly
penalising it.  For example, for the following hierarchy:

  A:   max high, 20 usage
  A/B: 9 high, 10 usage
  A/C: max high, 10 usage

We would end up doing the following calculation below when calculating
high delay for A/B:

  A/B: 10 - 9 = 1...
  A:   20 - PAGE_COUNTER_MAX = 21, so set max_overage to 21.

This gets worse with higher disparities in usage in the parent.

I have no idea how this disappeared from the final version of the patch,
but it is certainly Not Good(tm).  This wasn't obvious in testing because,
for a simple cgroup hierarchy with only one child, the result is usually
roughly the same.  It's only in more complex hierarchies that things go
really awry (although still, the effects are limited to a maximum of 2
seconds in schedule_timeout_killable at a maximum).

[chris@chrisdown.name: changelog]
Fixes: e26733e0d0ec ("mm, memcg: throttle allocators based on ancestral memory.high")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@vger.kernel.org> [5.4.x]
Link: http://lkml.kernel.org/r/20200331152424.GA1019937@chrisdown.name
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c

index 05b4ec2c6499daff7f658a95b6d5ab5f9d5605c7..5beea03dd58ad872d679376ee2b9f82cbe68f0fb 100644 (file)
@@ -2336,6 +2336,9 @@ static unsigned long calculate_high_delay(struct mem_cgroup *memcg,
                usage = page_counter_read(&memcg->memory);
                high = READ_ONCE(memcg->high);
 
+               if (usage <= high)
+                       continue;
+
                /*
                 * Prevent division by 0 in overage calculation by acting as if
                 * it was a threshold of 1 page