From: MengEn Sun <mengensun@tencent.com>

when a cpu stops tick, quiet_vmstat may flush all the per cpu
statistics counter.

while, the shepherd is needed those counters to kick the
vmstat_work.

when a cpu in nohz with a lot of pcp pages, and do not do page
allocating and freeing. the pcp pages of the cpu may not be hold
for a long time

we make shepherd keep a eye on the pcp high_min and high_max

Reviewed-by: Jinliang Zheng <alexjlzheng@tencent.com>
Signed-off-by: MengEn Sun <mengensun@tencent.com>
---
 mm/vmstat.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/mm/vmstat.c b/mm/vmstat.c
index 1917c034c045..32768fb8f769 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -2024,8 +2024,13 @@ static bool need_update(int cpu)
 
 	for_each_populated_zone(zone) {
 		struct per_cpu_zonestat *pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
+		struct per_cpu_pages *pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
 		struct per_cpu_nodestat *n;
 
+		/* in onhz or nohz full make pcp decay work */
+		if (pcp->high_max > pcp->high_min)
+			return true;
+
 		/*
 		 * The fast way of checking if there are any vmstat diffs.
 		 */
-- 
2.43.5

On Fri, 11 Oct 2024 18:36:10 +0800 alexjlzheng@gmail.com wrote:

> Subject: [PATCH] mm: make pcp decay work with onhz

"NOHZ".

> Date: Fri, 11 Oct 2024 18:36:10 +0800
> X-Mailer: git-send-email 2.39.3
> 
> From: MengEn Sun <mengensun@tencent.com>
> 
> when a cpu stops tick, quiet_vmstat may flush all the per cpu
> statistics counter.
> 
> while, the shepherd is needed those counters to kick the
> vmstat_work.
> 
> when a cpu in nohz with a lot of pcp pages, and do not do page
> allocating and freeing. the pcp pages of the cpu may not be hold
> for a long time
> 
> we make shepherd keep a eye on the pcp high_min and high_max

I can see what you're saying here, but it's hard to understand.  Please
spend a little time clarifying the text?  And please start sentences
with a capital letter!

> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -2024,8 +2024,13 @@ static bool need_update(int cpu)
>  
>  	for_each_populated_zone(zone) {
>  		struct per_cpu_zonestat *pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
> +		struct per_cpu_pages *pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
>  		struct per_cpu_nodestat *n;
>  
> +		/* in onhz or nohz full make pcp decay work */

"NOHZ".

> +		if (pcp->high_max > pcp->high_min)
> +			return true;
> +
>  		/*
>  		 * The fast way of checking if there are any vmstat diffs.
>  		 */
> --

On Sat, 12 Oct 2024 15:43:28 -0700 akpm@linux-foundation.org wrote:
> On Fri, 11 Oct 2024 18:36:10 +0800 alexjlzheng@gmail.com wrote:
> 
> > Subject: [PATCH] mm: make pcp decay work with onhz
> 
> "NOHZ".
> 
> > Date: Fri, 11 Oct 2024 18:36:10 +0800
> > X-Mailer: git-send-email 2.39.3
> > 
> > From: MengEn Sun <mengensun@tencent.com>
> > 
> > when a cpu stops tick, quiet_vmstat may flush all the per cpu
> > statistics counter.
> > 
> > while, the shepherd is needed those counters to kick the
> > vmstat_work.
> > 
> > when a cpu in nohz with a lot of pcp pages, and do not do page
> > allocating and freeing. the pcp pages of the cpu may not be hold
> > for a long time
> > 
> > we make shepherd keep a eye on the pcp high_min and high_max
> 
> I can see what you're saying here, but it's hard to understand.  Please
> spend a little time clarifying the text?  And please start sentences
> with a capital letter!

Thank you for your suggestion
I will create a v2 version, and make text clarifyed

> 
> > --- a/mm/vmstat.c
> > +++ b/mm/vmstat.c
> > @@ -2024,8 +2024,13 @@ static bool need_update(int cpu)
> >  
> >  	for_each_populated_zone(zone) {
> >  		struct per_cpu_zonestat *pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
> > +		struct per_cpu_pages *pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
> >  		struct per_cpu_nodestat *n;
> >  
> > +		/* in onhz or nohz full make pcp decay work */
> 
> "NOHZ".
> 
> > +		if (pcp->high_max > pcp->high_min)
> > +			return true;
> > +
> >  		/*
> >  		 * The fast way of checking if there are any vmstat diffs.
> >  		 */
> > --