Re: [PATCH -next] mm: vmscan: correct nr_requested tracing in scan_folios

[Chen Ridong]

On 2025/12/4 19:54, David Hildenbrand (Red Hat) wrote:
> On 12/4/25 01:46, Chen Ridong wrote:
>>
>>
>> On 2025/12/3 19:33, David Hildenbrand (Red Hat) wrote:
>>> On 12/3/25 10:40, Chen Ridong wrote:
>>>> From: Chen Ridong <chenridong@xxxxxxxxxx>
>>>>
>>>> When enabling vmscan tracing, it is observed that nr_requested is always
>>>> 4096, which is confusing.
>>>>
>>>>           mm_vmscan_lru_isolate: classzone=3 order=0 nr_requested=4096 ...
>>>>           mm_vmscan_lru_isolate: classzone=3 order=0 nr_requested=4096 ...
>>>>           mm_vmscan_lru_isolate: classzone=3 order=0 nr_requested=4096 ...
>>>>           mm_vmscan_lru_isolate: classzone=3 order=0 nr_requested=4096 ...
>>>>           mm_vmscan_lru_isolate: classzone=3 order=0 nr_requested=4096 ...
>>>>           mm_vmscan_lru_isolate: classzone=3 order=0 nr_requested=4096 ...
>>>>           mm_vmscan_lru_isolate: classzone=3 order=0 nr_requested=4096 ...
>>>>
>>>> This is because it prints MAX_LRU_BATCH, which is meaningless as it's a
>>>> constant. To fix this, modify it to print nr_to_scan as isolate_lru_folios
>>>> does.
>>>>
>>>> Fixes: 8c2214fc9a47 ("mm: multi-gen LRU: reuse some legacy trace events")
>>>> Signed-off-by: Chen Ridong <chenridong@xxxxxxxxxx>
>>>> ---
>>>>    mm/vmscan.c | 2 +-
>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>>>> index fddd168a9737..8cfafd50a7a8 100644
>>>> --- a/mm/vmscan.c
>>>> +++ b/mm/vmscan.c
>>>> @@ -4601,7 +4601,7 @@ static int scan_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>>>>        count_memcg_events(memcg, item, isolated);
>>>>        count_memcg_events(memcg, PGREFILL, sorted);
>>>>        __count_vm_events(PGSCAN_ANON + type, isolated);
>>>> -    trace_mm_vmscan_lru_isolate(sc->reclaim_idx, sc->order, MAX_LRU_BATCH,
>>>> +    trace_mm_vmscan_lru_isolate(sc->reclaim_idx, sc->order, nr_to_scan,
>>>>                    scanned, skipped, isolated,
>>>
>>> We do that in isolate_lru_folios().
>>>
>>> Given that we do
>>>
>>>      int remaining = min(nr_to_scan, MAX_LRU_BATCH);
>>>
>>> and effectively cap it, I wonder if we would want to trace that capped valued instead of
>>> MAX_LRU_BATCH.
>>>
>>
>> I prefer tracing nr_to_scan, as it reflects the original target number of pages we intended to scan.
> 
> But it's misleading, because we're also tracing "scanned, skipped, isolated", and one might wonder
> how it relates to nr_to_scan?
> 
>> Even if nr_to_scan exceeds MAX_LRU_BATCH, we can still deduce that it was effectively capped by
>> examining the actual scanned, skipped, or isolated counts. However, if we trace min(nr_to_scan,
>> MAX_LRU_BATCH) instead, we would lose visibility into what the original nr_to_scan value was.
> 
> Is that really required for the purpose we are tracing here?
> 

Thank you David,

I've seen Lance's response and agree with your point. Using min(nr_to_scan, MAX_LRU_BATCH) would
indeed be more appropriate for the trace, as it reflects the actual capped value used during scanning.

I'll update the patch accordingly.

-- 
Best regards,
Ridong