Re: [PATCH v2] vmscan: add a vmscan event for reclaim_pages

[Vlastimil Babka <vbabka@suse.cz>]

On 10/16/24 15:24, 김재원 wrote:
>>On 10/11/24 14:49, Jaewon Kim wrote:
>>> The reclaim_folio_list uses a dummy reclaim_stat and is not being
>>> used. To know the memory stat, add a new trace event. This is useful how
>>> how many pages are not reclaimed or why.
>>> 
>>> This is an example.
>>> mm_vmscan_reclaim_pages: nid=0 nr_scanned=112 nr_reclaimed=112 nr_dirty=0 nr_writeback=0 nr_congested=0 nr_immediate=0 nr_activate_anon=0 nr_activate_file=0 nr_ref_keep=0 nr_unmap_fail=0
>>> 
>>> Currenlty reclaim_folio_list is only called by reclaim_pages, and
>>> reclaim_pages is used by damon and madvise. In the latest Android,
>>> reclaim_pages is also used by shmem to reclaim all pages in a
>>> address_space.
>>> 
>>> Signed-off-by: Jaewon Kim <jaewon31.kim@samsung.com>
>>> ---
>>> v2: remove reclaim_stat_add function and call the trace on each node
>>> v1: introduce a new trace event
>>> ---
>>>  include/trace/events/vmscan.h | 45 +++++++++++++++++++++++++++++++++++
>>>  mm/vmscan.c                   | 16 +++++++++----
>>>  2 files changed, 56 insertions(+), 5 deletions(-)
>>> 
>>> diff --git a/include/trace/events/vmscan.h b/include/trace/events/vmscan.h
>>> index 1a488c30afa5..490958fa10de 100644
>>> --- a/include/trace/events/vmscan.h
>>> +++ b/include/trace/events/vmscan.h
>>> @@ -346,6 +346,51 @@ TRACE_EVENT(mm_vmscan_write_folio,
>>>                  show_reclaim_flags(__entry->reclaim_flags))
>>>  );
>>>  
>>> +TRACE_EVENT(mm_vmscan_reclaim_pages,
>>> +
>>> +        TP_PROTO(int nid,
>>> +                unsigned long nr_scanned, unsigned long nr_reclaimed,
>>> +                struct reclaim_stat *stat),
>>> +
>>> +        TP_ARGS(nid, nr_scanned, nr_reclaimed, stat),
>>> +
>>> +        TP_STRUCT__entry(
>>> +                __field(int, nid)
>>> +                __field(unsigned long, nr_scanned)
>>> +                __field(unsigned long, nr_reclaimed)
>>> +                __field(unsigned long, nr_dirty)
>>> +                __field(unsigned long, nr_writeback)
>>> +                __field(unsigned long, nr_congested)
>>> +                __field(unsigned long, nr_immediate)
>>> +                __field(unsigned int, nr_activate0)
>>> +                __field(unsigned int, nr_activate1)
>>> +                __field(unsigned long, nr_ref_keep)
>>> +                __field(unsigned long, nr_unmap_fail)
>>> +        ),
>>> +
>>> +        TP_fast_assign(
>>> +                __entry->nid = nid;
>>> +                __entry->nr_scanned = nr_scanned;
>>> +                __entry->nr_reclaimed = nr_reclaimed;
>>> +                __entry->nr_dirty = stat->nr_dirty;
>>> +                __entry->nr_writeback = stat->nr_writeback;
>>> +                __entry->nr_congested = stat->nr_congested;
>>> +                __entry->nr_immediate = stat->nr_immediate;
>>> +                __entry->nr_activate0 = stat->nr_activate[0];
>>> +                __entry->nr_activate1 = stat->nr_activate[1];
>>> +                __entry->nr_ref_keep = stat->nr_ref_keep;
>>> +                __entry->nr_unmap_fail = stat->nr_unmap_fail;
>>> +        ),
>>> +
>>> +        TP_printk("nid=%d nr_scanned=%ld nr_reclaimed=%ld nr_dirty=%ld nr_writeback=%ld nr_congested=%ld nr_immediate=%ld nr_activate_anon=%d nr_activate_file=%d nr_ref_keep=%ld nr_unmap_fail=%ld",
>>> +                __entry->nid,
>>> +                __entry->nr_scanned, __entry->nr_reclaimed,
>>> +                __entry->nr_dirty, __entry->nr_writeback,
>>> +                __entry->nr_congested, __entry->nr_immediate,
>>> +                __entry->nr_activate0, __entry->nr_activate1,
>>> +                __entry->nr_ref_keep, __entry->nr_unmap_fail)
>>> +);
>>> +
>>>  TRACE_EVENT(mm_vmscan_lru_shrink_inactive,
>>>  
>>>          TP_PROTO(int nid,
>>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>>> index 749cdc110c74..0c2c36bf4c5a 100644
>>> --- a/mm/vmscan.c
>>> +++ b/mm/vmscan.c
>>> @@ -2126,9 +2126,10 @@ static void shrink_active_list(unsigned long nr_to_scan,
>>>  }
>>>  
>>>  static unsigned int reclaim_folio_list(struct list_head *folio_list,
>>> +                                      unsigned int nr_scanned,
>>>                                        struct pglist_data *pgdat)
>>>  {
>>> -        struct reclaim_stat dummy_stat;
>>> +        struct reclaim_stat stat;
>>>          unsigned int nr_reclaimed;
>>>          struct folio *folio;
>>>          struct scan_control sc = {
>>> @@ -2139,12 +2140,13 @@ static unsigned int reclaim_folio_list(struct list_head *folio_list,
>>>                  .no_demotion = 1,
>>>          };
>>>  
>>> -        nr_reclaimed = shrink_folio_list(folio_list, pgdat, &sc, &dummy_stat, true);
>>> +        nr_reclaimed = shrink_folio_list(folio_list, pgdat, &sc, &stat, true);
>>>          while (!list_empty(folio_list)) {
>>>                  folio = lru_to_folio(folio_list);
>>>                  list_del(&folio->lru);
>>>                  folio_putback_lru(folio);
>>>          }
>>> +        trace_mm_vmscan_reclaim_pages(pgdat->node_id, nr_scanned, nr_reclaimed, &stat);
>>
>>Why is the new calculation of nr_scanned needed? Could you just take a delta
>>of sc->nr_scanned, i.e. after - before calling shrink_folio_list() ?
> 
> Oh correct. 
> 
> Except the case of (!folio_trylock(folio)), shrink_folio_list would count on sc->nr_scanned.
> I don't understand why we do not count this lock case though.

Guess it means we couldn't really scan it. Should be probably rare anyway.