Re: [PATCH v3] staging: android/lowmemorykiller: Don't grab tasklist_lock

[Oleg Nesterov <oleg@redhat.com>]

On 02/03, Anton Vorontsov wrote:
>
> @@ -132,7 +133,7 @@ static int lowmem_shrink(struct shrinker *s, struct shrink_control *sc)
>  	}
>  	selected_oom_adj = min_adj;
>
> -	read_lock(&tasklist_lock);
> +	rcu_read_lock();
>  	for_each_process(p) {
>  		struct mm_struct *mm;
>  		struct signal_struct *sig;
> @@ -180,12 +181,12 @@ static int lowmem_shrink(struct shrinker *s, struct shrink_control *sc)
>  		lowmem_deathpending = selected;
>  		task_handoff_register(&task_nb);
>  #endif
> -		force_sig(SIGKILL, selected);
> +		send_sig(SIGKILL, selected, 0);
>  		rem -= selected_tasksize;
>  	}
>  	lowmem_print(4, "lowmem_shrink %lu, %x, return %d\n",
>  		     sc->nr_to_scan, sc->gfp_mask, rem);
> -	read_unlock(&tasklist_lock);
> +	rcu_read_unlock();

I think this is correct. As for ->mm check please look at
find_lock_task_mm().

You can also remove the !sig check.

And, forgot to mention. There is another reason why mm != NULL
check is wrong (send_sig_all too). A kernel thread can do use_mm().
You should also check PF_KTHREAD.

Oleg.