On Wed, Aug 29, 2018 at 2:24 PM Roman Gushchin [off-list ref] wrote:

On Tue, Aug 21, 2018 at 03:10:52PM -0700, Shakeel Butt wrote:

quoted

On Tue, Aug 21, 2018 at 2:36 PM Roman Gushchin [off-list ref] wrote:

quoted

If CONFIG_VMAP_STACK is set, kernel stacks are allocated
using __vmalloc_node_range() with __GFP_ACCOUNT. So kernel
stack pages are charged against corresponding memory cgroups
on allocation and uncharged on releasing them.

The problem is that we do cache kernel stacks in small
per-cpu caches and do reuse them for new tasks, which can
belong to different memory cgroups.

Each stack page still holds a reference to the original cgroup,
so the cgroup can't be released until the vmap area is released.

To make this happen we need more than two subsequent exits
without forks in between on the current cpu, which makes it
very unlikely to happen. As a result, I saw a significant number
of dying cgroups (in theory, up to 2 * number_of_cpu +
number_of_tasks), which can't be released even by significant
memory pressure.

As a cgroup structure can take a significant amount of memory
(first of all, per-cpu data like memcg statistics), it leads
to a noticeable waste of memory.

Signed-off-by: Roman Gushchin <redacted>

Reviewed-by: Shakeel Butt <redacted>

BTW this makes a very good use-case for optimizing kmem uncharging
similar to what you did for skmem uncharging.

The only thing I'm slightly worried here is that it can make
reclaiming of memory cgroups harder. Probably, it's still ok,
but let me first finish the work I'm doing on optimizing the
whole memcg reclaim process, and then return to this case.

Yes, maybe we can disable that optimization for offlined memcgs.
Anyways, we can discuss this later as you have suggested.

Shakeel