On 08.10.2021 14:47, Vlastimil Babka wrote:
> On 10/8/21 11:24, Vasily Averin wrote:
>> __alloc_pages_bulk can call __count_zid_vm_events and zone_statistics
>> with nr_account = 0.
>
> But that's not a bug, right? Just an effective no-op that's not commonly
> happening, so is it worth the check?
Why not?
Yes, it's not a bug, it just makes the kernel a bit more efficient in a very unlikely case.
However, it looks strange and makes uninformed code reviewers like me worry about possible
problems inside the affected functions. No one else calls these functions from 0.
>> Fixes: 3e23060b2d0b ("mm/page_alloc: batch the accounting updates in the bulk allocator")
>> Signed-off-by: Vasily Averin <vvs@xxxxxxxxxxxxx>
>> ---
>> mm/page_alloc.c | 7 ++++---
>> 1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 602819a232e5..e67113452ee8 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -5364,9 +5364,10 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
>> }
>>
>> local_unlock_irqrestore(&pagesets.lock, flags);
>> -
>> - __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account);
>> - zone_statistics(ac.preferred_zoneref->zone, zone, nr_account);
>> + if (nr_account) {
>> + __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account);
>> + zone_statistics(ac.preferred_zoneref->zone, zone, nr_account);
>> + }
>> if (objcg)
>> memcg_bulk_post_charge_hook(objcg, nr_pre_charge - nr_account);
>>
>>
>
