Re: [PATCH] mm, oom: avoid printk() iteration under RCU

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 17 Jul 2019 19:55:01 +0900 Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> wrote:

> Currently dump_tasks() might call printk() for many thousands times under
> RCU, which might take many minutes for slow consoles. Therefore, split
> dump_tasks() into three stages; take a snapshot of possible OOM victim
> candidates under RCU, dump the snapshot from reschedulable context, and
> destroy the snapshot.
> 
> In a future patch, the first stage would be moved to select_bad_process()
> and the third stage would be moved to after oom_kill_process(), and will
> simplify refcount handling.

Look straightforward enough.

>
> ...
>
>  static void dump_tasks(struct oom_control *oc)
>  {
> -	pr_info("Tasks state (memory values in pages):\n");
> -	pr_info("[  pid  ]   uid  tgid total_vm      rss pgtables_bytes swapents oom_score_adj name\n");
> +	static LIST_HEAD(list);

I don't think this needs to be static?

> +	struct task_struct *p;
> +	struct task_struct *t;
>  
>  	if (is_memcg_oom(oc))
> -		mem_cgroup_scan_tasks(oc->memcg, dump_task, oc);
> +		mem_cgroup_scan_tasks(oc->memcg, add_candidate_task, &list);
>  	else {
> -		struct task_struct *p;
> -
>  		rcu_read_lock();
>  		for_each_process(p)
> -			dump_task(p, oc);
> +			add_candidate_task(p, &list);
>  		rcu_read_unlock();
>  	}
> +	pr_info("Tasks state (memory values in pages):\n");
> +	pr_info("[  pid  ]   uid  tgid total_vm      rss pgtables_bytes swapents oom_score_adj name\n");
> +	list_for_each_entry(p, &list, oom_victim_list) {
> +		cond_resched();
> +		/* p may not have freeable memory in nodemask */
> +		if (!is_memcg_oom(oc) && !oom_cpuset_eligible(p, oc))
> +			continue;
> +		/* All of p's threads might have already detached their mm's. */
> +		t = find_lock_task_mm(p);
> +		if (!t)
> +			continue;
> +		pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu         %5hd %s\n",
> +			t->pid, from_kuid(&init_user_ns, task_uid(t)),
> +			t->tgid, t->mm->total_vm, get_mm_rss(t->mm),
> +			mm_pgtables_bytes(t->mm),
> +			get_mm_counter(t->mm, MM_SWAPENTS),
> +			t->signal->oom_score_adj, t->comm);
> +		task_unlock(t);
> +	}
> +	list_for_each_entry_safe(p, t, &list, oom_victim_list) {
> +		list_del(&p->oom_victim_list);
> +		put_task_struct(p);
> +	}
>  }