Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 28 Mar 2023 14:35:41 -0400
From:   Johannes Weiner <hannes@cmpxchg.org>
To:     Yosry Ahmed <yosryahmed@google.com>
Cc:     Tejun Heo <tj@kernel.org>, Josef Bacik <josef@toxicpanda.com>,
        Jens Axboe <axboe@kernel.dk>,
        Zefan Li <lizefan.x@bytedance.com>,
        Michal Hocko <mhocko@kernel.org>,
        Roman Gushchin <roman.gushchin@linux.dev>,
        Shakeel Butt <shakeelb@google.com>,
        Muchun Song <muchun.song@linux.dev>,
        Andrew Morton <akpm@linux-foundation.org>,
        Michal =?iso-8859-1?Q?Koutn=FD?= <mkoutny@suse.com>,
        Vasily Averin <vasily.averin@linux.dev>,
        cgroups@vger.kernel.org, linux-block@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        bpf@vger.kernel.org
Subject: Re: [PATCH v1 6/9] memcg: sleep during flushing stats in safe
 contexts
Message-ID: <ZCMzfQuo9IhWVzRA@cmpxchg.org>
References: <20230328061638.203420-1-yosryahmed@google.com>
 <20230328061638.203420-7-yosryahmed@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230328061638.203420-7-yosryahmed@google.com>
Precedence: bulk

On Tue, Mar 28, 2023 at 06:16:35AM +0000, Yosry Ahmed wrote:
> @@ -642,24 +642,57 @@ static void __mem_cgroup_flush_stats(void)
>  	 * from memcg flushers (e.g. reclaim, refault, etc).
>  	 */
>  	if (atomic_xchg(&stats_flush_ongoing, 1))
> -		return;
> +		return false;
>  
>  	WRITE_ONCE(flush_next_time, jiffies_64 + 2*FLUSH_TIME);
> -	cgroup_rstat_flush_atomic(root_mem_cgroup->css.cgroup);
> +	return true;
> +}
> +
> +static void mem_cgroup_post_stats_flush(void)
> +{
>  	atomic_set(&stats_flush_threshold, 0);
>  	atomic_set(&stats_flush_ongoing, 0);
>  }
>  
> -void mem_cgroup_flush_stats(void)
> +static bool mem_cgroup_should_flush_stats(void)
>  {
> -	if (atomic_read(&stats_flush_threshold) > num_online_cpus())
> -		__mem_cgroup_flush_stats();
> +	return atomic_read(&stats_flush_threshold) > num_online_cpus();
> +}
> +
> +/* atomic functions, safe to call from any context */
> +static void __mem_cgroup_flush_stats_atomic(void)
> +{
> +	if (mem_cgroup_pre_stats_flush()) {
> +		cgroup_rstat_flush_atomic(root_mem_cgroup->css.cgroup);
> +		mem_cgroup_post_stats_flush();
> +	}
> +}

I'm afraid I wasn't very nuanced with my complaint about the bool
parameter in the previous version. In this case, when you can do a
common helper for a couple of API functions defined right below it,
and the callers don't spread throughout the codebase, using bools
makes things simpler while still being easily understandable:

static void do_flush_stats(bool may_sleep)
{
	if (atomic_xchg(&stats_flush_ongoing, 1))
		return;

	WRITE_ONCE(flush_next_time, jiffies_64 + 2*FLUSH_TIME);
	atomic_set(&stats_flush_threshold, 0);

	if (!may_sleep)
		cgroup_rstat_flush_atomic(root_mem_cgroup->css.cgroup);
	else
		cgroup_rstat_flush(root_mem_cgroup->css.cgroup);

	atomic_set(&stats_flush_ongoing, 0);
}

void mem_cgroup_flush_stats(void)
{
	if (atomic_read(&stats_flush_threshold) > num_online_cpus())
		do_flush_stats(true);
}

void mem_cgroup_flush_stats_atomic(void)
{
	if (atomic_read(&stats_flush_threshold) > num_online_cpus())
		do_flush_stats(false);
}

>  void mem_cgroup_flush_stats_ratelimited(void)
>  {
>  	if (time_after64(jiffies_64, READ_ONCE(flush_next_time)))
> -		mem_cgroup_flush_stats();
> +		mem_cgroup_flush_stats_atomic();
> +}

This should probably be mem_cgroup_flush_stats_atomic_ratelimited().

(Whee, kinda long, but that's alright. Very specialized caller...)

Btw, can you guys think of a reason against moving the threshold check
into the common function? It would then apply to the time-limited
flushes as well, but that shouldn't hurt anything. This would make the
code even simpler:

static void do_flush_stats(bool may_sleep)
{
	if (atomic_read(&stats_flush_threshold) <= num_online_cpus())
		return;

	if (atomic_xchg(&stats_flush_ongoing, 1))
		return;

	WRITE_ONCE(flush_next_time, jiffies_64 + 2*FLUSH_TIME);
	atomic_set(&stats_flush_threshold, 0);

	if (!may_sleep)
		cgroup_rstat_flush_atomic(root_mem_cgroup->css.cgroup);
	else
		cgroup_rstat_flush(root_mem_cgroup->css.cgroup);

	atomic_set(&stats_flush_ongoing, 0);
}

void mem_cgroup_flush_stats(void)
{
	do_flush_stats(true);
}

void mem_cgroup_flush_stats_atomic(void)
{
	do_flush_stats(false);
}

void mem_cgroup_flush_stats_atomic_ratelimited(void)
{
	if (time_after64(jiffies_64, READ_ONCE(flush_next_time)))
		do_flush_stats(false);
}

> @@ -2845,7 +2845,7 @@ static void prepare_scan_count(pg_data_t *pgdat, struct scan_control *sc)
>  	 * Flush the memory cgroup stats, so that we read accurate per-memcg
>  	 * lruvec stats for heuristics.
>  	 */
> -	mem_cgroup_flush_stats();
> +	mem_cgroup_flush_stats_atomic();

I'm thinking this one could be non-atomic as well. It's called fairly
high up in reclaim without any locks held.