Re: [PATCH 3/2] memcg: Facilitate termination of memcg OOM victims.

[Michal Hocko]

On Tue 08-01-19 23:21:23, Tetsuo Handa wrote:
> From: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
> 
> If memcg OOM events in different domains are pending, already OOM-killed
> threads needlessly wait for pending memcg OOM events in different domains.
> An out_of_memory() call is slow because it involves printk(). With slow
> serial consoles, out_of_memory() might take more than a second. Therefore,
> allowing killed processes to quickly call mmput() from exit_mm() from
> do_exit() will help calling __mmput() (which can reclaim more memory than
> the OOM reaper can reclaim) quickly.

Can you post it separately out of this thread please? It is really a
separate topic and I do not want to end with back and forth without
making a further progress.
 
> Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
> ---
>  mm/memcontrol.c | 17 +++++++++++------
>  1 file changed, 11 insertions(+), 6 deletions(-)
> 
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index 90eb2e2..a7d3ba9 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -1389,14 +1389,19 @@ static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
>  	};
>  	bool ret = true;
>  
> -	mutex_lock(&oom_lock);
> -
>  	/*
> -	 * multi-threaded tasks might race with oom_reaper and gain
> -	 * MMF_OOM_SKIP before reaching out_of_memory which can lead
> -	 * to out_of_memory failure if the task is the last one in
> -	 * memcg which would be a false possitive failure reported
> +	 * Multi-threaded tasks might race with oom_reaper() and gain
> +	 * MMF_OOM_SKIP before reaching out_of_memory(). But if current
> +	 * thread was already killed or is ready to terminate, there is
> +	 * no need to call out_of_memory() nor wait for oom_reaoer() to
> +	 * set MMF_OOM_SKIP. These three checks minimize possibility of
> +	 * needlessly calling out_of_memory() and try to call exit_mm()
> +	 * as soon as possible.
>  	 */
> +	if (mutex_lock_killable(&oom_lock))
> +		return true;
> +	if (fatal_signal_pending(current))
> +		goto unlock;
>  	if (tsk_is_oom_victim(current))
>  		goto unlock;
>  
> -- 
> 1.8.3.1
> 

-- 
Michal Hocko
SUSE Labs