[Devel] [PATCH rh7 v2] mm/memcg: fix cache growth above cache.limit_in_bytes

[Denis V. Lunev]

On 7/30/20 6:57 PM, Evgenii Shatokhin wrote:
> On 30.07.2020 17:00, Denis V. Lunev wrote:
>> On 7/30/20 4:58 PM, Andrey Ryabinin wrote:
>>> Exceeding cache above cache.limit_in_bytes schedules high_work_func()
>>> which tries to reclaim 32 pages. If cache generated fast enough or
>>> it allows
>>> cgroup to steadily grow above cache.limit_in_bytes because we don't
>>> reclaim
>>> enough. Try to reclaim exceeded amount of cache instead.
>>>
>>> https://jira.sw.ru/browse/PSBM-106384
>>> Signed-off-by: Andrey Ryabinin <aryabinin at virtuozzo.com>
>>> ---
>>>
>>> Changes since v1: add bug link to changelog
>>>
>>>   mm/memcontrol.c | 10 +++++++---
>>>   1 file changed, 7 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
>>> index 3cf200f506c3..e5adb0e81cbb 100644
>>> --- a/mm/memcontrol.c
>>> +++ b/mm/memcontrol.c
>>> @@ -3080,12 +3080,16 @@ static void reclaim_high(struct mem_cgroup
>>> *memcg,
>>>   {
>>>         do {
>>> +        unsigned long cache_overused;
>>> +
>>>           if (page_counter_read(&memcg->memory) > memcg->high)
>>>               try_to_free_mem_cgroup_pages(memcg, nr_pages,
>>> gfp_mask, 0);
>>>   -        if (page_counter_read(&memcg->cache) > memcg->cache.limit)
>>> -            try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask,
>>> -                        MEM_CGROUP_RECLAIM_NOSWAP);
>>> +        cache_overused = page_counter_read(&memcg->cache) -
>>> +            memcg->cache.limit;
>>> +        if (cache_overused)
>>> +            try_to_free_mem_cgroup_pages(memcg, cache_overused,
>>> +                    gfp_mask, MEM_CGROUP_RECLAIM_NOSWAP);
>>>         } while ((memcg = parent_mem_cgroup(memcg)));
>>>   }
>> can we run some testing and after that create custom RK to make check
>> with
>> HostEurope on Monday?
>
> 1. Which kernel version(s)?

151.14

>
> 2. Would it be enough to prepare the live patch as .ko file and load
> it manually ('kpatch load') or RPM package is preferable?
>

sure

> If you plan to use the fix on more than one node, I think, an RPM
> package is easier to use. For a single node, *.ko file would be enough.
>
one node - for test purposes

Den