Re: [PATCH 5/7] shrink support for memcg kmem controller

[Glauber Costa <glommer-bzQdu9zFT3WakBO8gow8eQ@public.gmane.org>]

On 02/22/2012 03:35 AM, Suleiman Souhlal wrote:
> On Tue, Feb 21, 2012 at 3:34 AM, Glauber Costa<glommer-bzQdu9zFT3WakBO8gow8eQ@public.gmane.org>  wrote:
>
>> @@ -5055,8 +5117,21 @@ int memcg_kmem_newpage(struct mem_cgroup *memcg, struct page *page, unsigned lon
>>   {
>>         unsigned long size = pages<<  PAGE_SHIFT;
>>         struct res_counter *fail;
>> +       int ret;
>> +       bool do_softlimit;
>> +
>> +       ret = res_counter_charge(memcg_kmem(memcg), size,&fail);
>> +       if (unlikely(mem_cgroup_event_ratelimit(memcg,
>> +                                               MEM_CGROUP_TARGET_THRESH))) {
>> +
>> +               do_softlimit = mem_cgroup_event_ratelimit(memcg,
>> +                                               MEM_CGROUP_TARGET_SOFTLIMIT);
>> +               mem_cgroup_threshold(memcg);
>> +               if (unlikely(do_softlimit))
>> +                       mem_cgroup_update_tree(memcg, page);
>> +       }
>>
>> -       return res_counter_charge(memcg_kmem(memcg), size,&fail);
>> +       return ret;
>>   }
>
> It seems like this might cause a lot of kernel memory allocations to
> fail whenever we are at the limit, even if we have a lot of
> reclaimable memory, when we don't have independent accounting.
>
> Would it be better to use __mem_cgroup_try_charge() here, when we
> don't have independent accounting, in order to deal with this
> situation?
>

Yes, it would.
I'll work on that.