[PATCH] mm/vmscan: add pgstall to memcg

[PATCH] mm/vmscan: add pgstall to memcg

[zhouchuyi]

Add pgstall to indicate the times of entering direct reclaim of memcg, use
the item of ALLOCSTALL_NORMAL in vmevents to save the data.

Signed-off-by: zhouchuyi <zhouchuyi@bytedance.com>
---


Hi，

I notice memcg uses some items to indicate the situation of direct reclaim in memory.stat，such as pgsteal and pgscan. 
I think there is no harm in adding another item to show the times of entering direct reclaim.

Thanks,
zhouchuyi


 Documentation/admin-guide/cgroup-v2.rst | 7 +++++--
 mm/memcontrol.c                         | 2 ++
 mm/vmscan.c                             | 5 ++++-
 3 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst
index 176298f2f..44c5b15bd 100644
--- a/Documentation/admin-guide/cgroup-v2.rst
+++ b/Documentation/admin-guide/cgroup-v2.rst
@@ -1447,8 +1447,11 @@ PAGE_SIZE multiple when read back.
 
 	  pgsteal (npn)
 		Amount of reclaimed pages
-
-	  pgactivate (npn)
+          
+          pgstall (npn)
+	  	Number of entering direct reclaim
+	  
+          pgactivate (npn)
 		Amount of pages moved to the active LRU list
 
 	  pgdeactivate (npn)
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index abec50f31..f407084b2 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -1505,6 +1505,8 @@ static char *memory_stat_format(struct mem_cgroup *memcg)
 	seq_buf_printf(&s, "pgscan %lu\n",
 		       memcg_events(memcg, PGSCAN_KSWAPD) +
 		       memcg_events(memcg, PGSCAN_DIRECT));
+	seq_buf_printf(&s, "pgstall %lu\n", 
+			   memcg_events(memcg, ALLOCSTALL_NORMAL));
 	seq_buf_printf(&s, "pgsteal %lu\n",
 		       memcg_events(memcg, PGSTEAL_KSWAPD) +
 		       memcg_events(memcg, PGSTEAL_DIRECT));
diff --git a/mm/vmscan.c b/mm/vmscan.c
index f7d9a683e..107cec59a 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -3532,7 +3532,10 @@ static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
 
 	if (!cgroup_reclaim(sc))
 		__count_zid_vm_events(ALLOCSTALL, sc->reclaim_idx, 1);
-
+	else {
+		count_memcg_events(sc->target_mem_cgroup, 
+					ALLOCSTALL_NORMAL, 1);
+	}
 	do {
 		vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup,
 				sc->priority);
-- 
2.20.1

Re: [PATCH] mm/vmscan: add pgstall to memcg

[Andrew Morton]

(cc memcg developers)
On Thu, 23 Jun 2022 16:01:03 +0800 zhouchuyi <zhouchuyi@bytedance.com> wrote:

> Add pgstall to indicate the times of entering direct reclaim of memcg, use
> the item of ALLOCSTALL_NORMAL in vmevents to save the data.
> 
> Signed-off-by: zhouchuyi <zhouchuyi@bytedance.com>
> ---
> 
> 
> Hi，
> 
> I notice memcg uses some items to indicate the situation of direct reclaim in memory.stat，such as pgsteal and pgscan. 
> I think there is no harm in adding another item to show the times of entering direct reclaim.
> 

Why do you consider this useful?  Do you have a use case?  If so,
please describe it in detail.

>
> ...
>
> diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst
> index 176298f2f..44c5b15bd 100644
> --- a/Documentation/admin-guide/cgroup-v2.rst
> +++ b/Documentation/admin-guide/cgroup-v2.rst
> @@ -1447,8 +1447,11 @@ PAGE_SIZE multiple when read back.
>  
>  	  pgsteal (npn)
>  		Amount of reclaimed pages
> -
> -	  pgactivate (npn)
> +          
> +          pgstall (npn)
> +	  	Number of entering direct reclaim
> +	  
> +          pgactivate (npn)
>  		Amount of pages moved to the active LRU list
>  
>  	  pgdeactivate (npn)
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index abec50f31..f407084b2 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -1505,6 +1505,8 @@ static char *memory_stat_format(struct mem_cgroup *memcg)
>  	seq_buf_printf(&s, "pgscan %lu\n",
>  		       memcg_events(memcg, PGSCAN_KSWAPD) +
>  		       memcg_events(memcg, PGSCAN_DIRECT));
> +	seq_buf_printf(&s, "pgstall %lu\n", 
> +			   memcg_events(memcg, ALLOCSTALL_NORMAL));
>  	seq_buf_printf(&s, "pgsteal %lu\n",
>  		       memcg_events(memcg, PGSTEAL_KSWAPD) +
>  		       memcg_events(memcg, PGSTEAL_DIRECT));
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index f7d9a683e..107cec59a 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -3532,7 +3532,10 @@ static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
>  
>  	if (!cgroup_reclaim(sc))
>  		__count_zid_vm_events(ALLOCSTALL, sc->reclaim_idx, 1);
> -
> +	else {
> +		count_memcg_events(sc->target_mem_cgroup, 
> +					ALLOCSTALL_NORMAL, 1);
> +	}
>  	do {
>  		vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup,
>  				sc->priority);
> -- 
> 2.20.1

Re: [PATCH] mm/vmscan: add pgstall to memcg

[Chuyi Zhou]

在 2022/6/25 上午4:42, Andrew Morton 写道:
> (cc memcg developers)
> On Thu, 23 Jun 2022 16:01:03 +0800 zhouchuyi <zhouchuyi@bytedance.com> wrote:
> 
>> Add pgstall to indicate the times of entering direct reclaim of memcg, use
>> the item of ALLOCSTALL_NORMAL in vmevents to save the data.
>>
>> Signed-off-by: zhouchuyi <zhouchuyi@bytedance.com>
>> ---
>>
>>
>> Hi，
>>
>> I notice memcg uses some items to indicate the situation of direct reclaim in memory.stat，such as pgsteal and pgscan.
>> I think there is no harm in adding another item to show the times of entering direct reclaim.
>>
> 
> Why do you consider this useful?  Do you have a use case?  If so,
> please describe it in detail.
> 
Since there are allocstall in proc/vmstat which indicate the number of 
entering direct reclaim of !cgroup_reclaim, So I think it maybe useful. 
For example, users could adjust memory.max to prevent entering direct 
reclaim through this item.So what do you think about it ?
>>
>> ...
>>
>> diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst
>> index 176298f2f..44c5b15bd 100644
>> --- a/Documentation/admin-guide/cgroup-v2.rst
>> +++ b/Documentation/admin-guide/cgroup-v2.rst
>> @@ -1447,8 +1447,11 @@ PAGE_SIZE multiple when read back.
>>   
>>   	  pgsteal (npn)
>>   		Amount of reclaimed pages
>> -
>> -	  pgactivate (npn)
>> +
>> +          pgstall (npn)
>> +	  	Number of entering direct reclaim
>> +	
>> +          pgactivate (npn)
>>   		Amount of pages moved to the active LRU list
>>   
>>   	  pgdeactivate (npn)
>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
>> index abec50f31..f407084b2 100644
>> --- a/mm/memcontrol.c
>> +++ b/mm/memcontrol.c
>> @@ -1505,6 +1505,8 @@ static char *memory_stat_format(struct mem_cgroup *memcg)
>>   	seq_buf_printf(&s, "pgscan %lu\n",
>>   		       memcg_events(memcg, PGSCAN_KSWAPD) +
>>   		       memcg_events(memcg, PGSCAN_DIRECT));
>> +	seq_buf_printf(&s, "pgstall %lu\n",
>> +			   memcg_events(memcg, ALLOCSTALL_NORMAL));
>>   	seq_buf_printf(&s, "pgsteal %lu\n",
>>   		       memcg_events(memcg, PGSTEAL_KSWAPD) +
>>   		       memcg_events(memcg, PGSTEAL_DIRECT));
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index f7d9a683e..107cec59a 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -3532,7 +3532,10 @@ static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
>>   
>>   	if (!cgroup_reclaim(sc))
>>   		__count_zid_vm_events(ALLOCSTALL, sc->reclaim_idx, 1);
>> -
>> +	else {
>> +		count_memcg_events(sc->target_mem_cgroup,
>> +					ALLOCSTALL_NORMAL, 1);
>> +	}
>>   	do {
>>   		vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup,
>>   				sc->priority);
>> -- 
>> 2.20.1

Re: [PATCH] mm/vmscan: add pgstall to memcg

[Michal Hocko]

On Tue 12-07-22 17:28:50, Chuyi Zhou wrote:
> 
> 
> 在 2022/6/25 上午4:42, Andrew Morton 写道:
> > (cc memcg developers)
> > On Thu, 23 Jun 2022 16:01:03 +0800 zhouchuyi <zhouchuyi@bytedance.com> wrote:
> > 
> > > Add pgstall to indicate the times of entering direct reclaim of memcg, use
> > > the item of ALLOCSTALL_NORMAL in vmevents to save the data.
> > > 
> > > Signed-off-by: zhouchuyi <zhouchuyi@bytedance.com>
> > > ---
> > > 
> > > 
> > > Hi，
> > > 
> > > I notice memcg uses some items to indicate the situation of direct reclaim in memory.stat，such as pgsteal and pgscan.
> > > I think there is no harm in adding another item to show the times of entering direct reclaim.
> > > 
> > 
> > Why do you consider this useful?  Do you have a use case?  If so,
> > please describe it in detail.
> > 
> Since there are allocstall in proc/vmstat which indicate the number of
> entering direct reclaim of !cgroup_reclaim, So I think it maybe useful. For
> example, users could adjust memory.max to prevent entering direct reclaim
> through this item.So what do you think about it ?

We already do have events for that - have a look MEMCG_MAX. Can you work
with that?
-- 
Michal Hocko
SUSE Labs