Message-ID: <877c10mju3.fsf@DESKTOP-5N7EMDA>
Date: Wed, 25 Jun 2025 14:11:32 +0800
From: "Huang, Ying" <ying.huang@...ux.alibaba.com>
To: Li Zhijian <lizhijian@...itsu.com>
Cc: linux-mm@...ck.org,  akpm@...ux-foundation.org,
  linux-kernel@...r.kernel.org,  y-goto@...itsu.com,  Ingo Molnar
 <mingo@...hat.com>,  Peter Zijlstra <peterz@...radead.org>,  Juri Lelli
 <juri.lelli@...hat.com>,  Vincent Guittot <vincent.guittot@...aro.org>,
  Dietmar Eggemann <dietmar.eggemann@....com>,  Steven Rostedt
 <rostedt@...dmis.org>,  Ben Segall <bsegall@...gle.com>,  Mel Gorman
 <mgorman@...e.de>,  Valentin Schneider <vschneid@...hat.com>,
  lkp@...el.com
Subject: Re: [PATCH RFC v2] mm: memory-tiering: Fix PGPROMOTE_CANDIDATE
 accounting

Li Zhijian <lizhijian@...itsu.com> writes:

> Goto-san reported confusing pgpromote statistics where
> the pgpromote_success count significantly exceeded pgpromote_candidate.
>
> On a system with three nodes (nodes 0-1: DRAM 4GB, node 2: NVDIMM 4GB):
>  # Enable demotion only
>  echo 1 > /sys/kernel/mm/numa/demotion_enabled
>  numactl -m 0-1 memhog -r200 3500M >/dev/null &
>  pid=$!
>  sleep 2
>  numactl memhog -r100 2500M >/dev/null &
>  sleep 10
>  kill -9 $pid # terminate the 1st memhog
>  # Enable promotion
>  echo 2 > /proc/sys/kernel/numa_balancing
>
> After a few seconds, we observeed `pgpromote_candidate < pgpromote_success`
> $ grep -e pgpromote /proc/vmstat
> pgpromote_success 2579
> pgpromote_candidate 0
>
> In this scenario, after terminating the first memhog, the conditions for
> pgdat_free_space_enough() are quickly met, triggering promotion.
> However, these migrated pages are only accounted for in PGPROMOTE_SUCCESS,
> not in PGPROMOTE_CANDIDATE.
>
> This update increments PGPROMOTE_CANDIDATE within the free space branch
> when a promotion decision is made, which may alter the mechanism of the
> rate limit. Consequently, it becomes easier to reach the rate limit than
> it was previously.
>
> For example:
> Rate Limit = 100 pages/sec
> Scenario:
>   T0: 90 free-space migrations
>   T0+100ms: 20-page migration request
>
> Before:
>   Rate limit is *not* reached: 0 + 20 = 20 < 100
>   PGPROMOTE_CANDIDATE: 20
> After:
>   Rate limit is reached: 90 + 20 = 110 > 100
>   PGPROMOTE_CANDIDATE: 110
>
> Due to the fact that the rate limit mechanism recalculates every second,
> theoretically, only within that one second can the transition from
> pgdat_free_space_enough() to !pgdat_free_space_enough() in top-tier
> remaining memory be affected.
>
> Moreover, previously, within this one-second span, promotions caused by
> pgdat_free_space_enough() are not restricted by rate limits.
> This theoretically makes it easier to cause application latency.
>
> The current modification can better control the rate limit in cases of
> transition from pgdat_free_space_enough() to !pgdat_free_space_enough()
> within one second.
>
> Cc: Huang Ying <ying.huang@...ux.alibaba.com>
> Cc: Ingo Molnar <mingo@...hat.com>
> Cc: Peter Zijlstra <peterz@...radead.org>
> Cc: Juri Lelli <juri.lelli@...hat.com>
> Cc: Vincent Guittot <vincent.guittot@...aro.org>
> Cc: Dietmar Eggemann <dietmar.eggemann@....com>
> Cc: Steven Rostedt <rostedt@...dmis.org>
> Cc: Ben Segall <bsegall@...gle.com>
> Cc: Mel Gorman <mgorman@...e.de>
> Cc: Valentin Schneider <vschneid@...hat.com>
> Reported-by: Yasunori Gotou (Fujitsu) <y-goto@...itsu.com>
> Signed-off-by: Li Zhijian <lizhijian@...itsu.com>
> ---
> V2:
> Fix compiling error # Reported by LKP
>
> As Ying suggested, we need to assess whether this change causes regression.
> However, considering the stringent conditions this patch involves,
> properly evaluating it may be challenging, as the outcomes depend on your
> perspective. Much like in a zero-sum game, if someone benefits, another
> might lose.
>
> If there are subsequent results, I will update them here.

I understand that it's hard to identify all possible regressions.
However, at least done some test to check some common use cases?

> Cc: lkp@...el.com
> Here, I hope to leverage the existing LKP benchmark to evaluate the
> potential impacts. The ideal evaluation conditions are:
> 1. Installed with DRAM + NVDIMM (which can be simulated).
> 2. NVDIMM is used as system RAM (configurable via daxctl).
> 3. Promotion is enabled (`echo 2 > /proc/sys/kernel/numa_balancing`).
>
> Alternative:
> We can indeed eliminate the potential impact within
> pgdat_free_space_enough(), so that the rate limit behavior remains as
> before.
>
> For instance, consider the following change:
>                 if (pgdat_free_space_enough(pgdat)) {
>                         /* workload changed, reset hot threshold */
>                         pgdat->nbp_threshold = 0;
> +                        pgdat->nbp_rl_nr_cand += nr;
>                         mod_node_page_state(pgdat, PGPROMOTE_CANDIDATE, nr);
>                         return true;
>                 }
>
> RFC:
> I am uncertain whether we originally intended for this discrepancy or if
> it was overlooked.
>
> However, the current situation where pgpromote_candidate < pgpromote_success
> is indeed confusing when interpreted literally.
> ---
>  kernel/sched/fair.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 7a14da5396fb..505b40f8897a 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -1940,11 +1940,13 @@ bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio,
>  		struct pglist_data *pgdat;
>  		unsigned long rate_limit;
>  		unsigned int latency, th, def_th;
> +		long nr = folio_nr_pages(folio);
>  
>  		pgdat = NODE_DATA(dst_nid);
>  		if (pgdat_free_space_enough(pgdat)) {
>  			/* workload changed, reset hot threshold */
>  			pgdat->nbp_threshold = 0;
> +			mod_node_page_state(pgdat, PGPROMOTE_CANDIDATE, nr);
>  			return true;
>  		}
>  
> @@ -1958,8 +1960,7 @@ bool should_numa_migrate_memory(struct task_struct *p, struct folio *folio,
>  		if (latency >= th)
>  			return false;
>  
> -		return !numa_promotion_rate_limit(pgdat, rate_limit,
> -						  folio_nr_pages(folio));
> +		return !numa_promotion_rate_limit(pgdat, rate_limit, nr);
>  	}
>  
>  	this_cpupid = cpu_pid_to_cpupid(dst_cpu, current->pid);

---
Best Regards,
Huang, Ying

Powered by blists - more mailing lists