linux-kernel - Re: [memcg] 45208c9105: aim7.jobs-per-min -14.0% regression

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Sun, 12 Sep 2021 21:29:14 +0800
From:   Feng Tang <feng.tang@...el.com>
To:     Hillf Danton <hdanton@...a.com>
Cc:     Shakeel Butt <shakeelb@...gle.com>,
        LKML <linux-kernel@...r.kernel.org>,
        Xing Zhengjun <zhengjun.xing@...ux.intel.com>,
        Linux MM <linux-mm@...ck.org>
Subject: Re: [memcg] 45208c9105: aim7.jobs-per-min -14.0% regression

On Sun, Sep 12, 2021 at 07:17:56PM +0800, Hillf Danton wrote:
[...]
> > +//	if (!(__this_cpu_inc_return(stats_flush_threshold) % MEMCG_CHARGE_BATCH))
> > +	if (!(__this_cpu_inc_return(stats_flush_threshold) % 128))
> >  		queue_work(system_unbound_wq, &stats_flush_work);
> >  }
> 
> Hi Feng,
> 
> Would you please check if it helps fix the regression to avoid queuing a
> queued work by adding and checking an atomic counter.
 
Hi Hillf,

I just tested your patch, and it didn't recover the regression, but
just reduced it from -14% to around -13%, similar to the patch
increasing the batch charge number.

Thanks,
Feng


> Hillf
> 
> --- x/mm/memcontrol.c
> +++ y/mm/memcontrol.c
> @@ -108,6 +108,7 @@ static void flush_memcg_stats_dwork(stru
>  static DECLARE_DEFERRABLE_WORK(stats_flush_dwork, flush_memcg_stats_dwork);
>  static void flush_memcg_stats_work(struct work_struct *w);
>  static DECLARE_WORK(stats_flush_work, flush_memcg_stats_work);
> +static atomic_t sfwork_queued;
>  static DEFINE_PER_CPU(unsigned int, stats_flush_threshold);
>  static DEFINE_SPINLOCK(stats_flush_lock);
>  
> @@ -660,8 +661,13 @@ void __mod_memcg_lruvec_state(struct lru
>  
>  	/* Update lruvec */
>  	__this_cpu_add(pn->lruvec_stats_percpu->state[idx], val);
> -	if (!(__this_cpu_inc_return(stats_flush_threshold) % MEMCG_CHARGE_BATCH))
> -		queue_work(system_unbound_wq, &stats_flush_work);
> +	if (!(__this_cpu_inc_return(stats_flush_threshold) %
> +	      MEMCG_CHARGE_BATCH)) {
> +		int queued = atomic_read(&sfwork_queued);
> +
> +		if (!queued && atomic_try_cmpxchg(&sfwork_queued, &queued, 1))
> +			queue_work(system_unbound_wq, &stats_flush_work);
> +	}
>  }
>  
>  /**
> @@ -5376,6 +5382,7 @@ static void flush_memcg_stats_dwork(stru
>  static void flush_memcg_stats_work(struct work_struct *w)
>  {
>  	mem_cgroup_flush_stats();
> +	atomic_dec(&sfwork_queued);
>  }
>  
>  static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu)