Re: [PATCH] mm: page_alloc: unreserve highatomic page blocks before oom

[Michal Hocko <mhocko@suse.com>]

On Tue 31-10-23 18:43:55, Charan Teja Kalla wrote:
> Thanks Michal/Pavan!!
> 
> On 10/31/2023 1:44 PM, Michal Hocko wrote:
> > On Mon 30-10-23 18:09:50, Charan Teja Kalla wrote:
> >> __alloc_pages_direct_reclaim() is called from slowpath allocation where
> >> high atomic reserves can be unreserved after there is a progress in
> >> reclaim and yet no suitable page is found. Later should_reclaim_retry()
> >> gets called from slow path allocation to decide if the reclaim needs to
> >> be retried before OOM kill path is taken.
> >>
> >> should_reclaim_retry() checks the available(reclaimable + free pages)
> >> memory against the min wmark levels of a zone and returns:
> >> a)  true, if it is above the min wmark so that slow path allocation will
> >> do the reclaim retries.
> >> b) false, thus slowpath allocation takes oom kill path.
> >>
> >> should_reclaim_retry() can also unreserves the high atomic reserves
> >> **but only after all the reclaim retries are exhausted.**
> >>
> >> In a case where there are almost none reclaimable memory and free pages
> >> contains mostly the high atomic reserves but allocation context can't
> >> use these high atomic reserves, makes the available memory below min
> >> wmark levels hence false is returned from should_reclaim_retry() leading
> >> the allocation request to take OOM kill path. This is an early oom kill
> >> because high atomic reserves are holding lot of free memory and 
> >> unreserving of them is not attempted.
> > 
> > OK, I see. So we do not release those reserved pages because OOM hits
> > too early. 
> > 
> >> (early)OOM is encountered on a machine in the below state(excerpt from
> >> the oom kill logs):
> >> [  295.998653] Normal free:7728kB boost:0kB min:804kB low:1004kB
> >> high:1204kB reserved_highatomic:8192KB active_anon:4kB inactive_anon:0kB
> >> active_file:24kB inactive_file:24kB unevictable:1220kB writepending:0kB
> >> present:70732kB managed:49224kB mlocked:0kB bounce:0kB free_pcp:688kB
> >> local_pcp:492kB free_cma:0kB
> >> [  295.998656] lowmem_reserve[]: 0 32
> >> [  295.998659] Normal: 508*4kB (UMEH) 241*8kB (UMEH) 143*16kB (UMEH)
> >> 33*32kB (UH) 7*64kB (UH) 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB
> >> 0*4096kB = 7752kB
> > 
> > OK, this is quite interesting as well. The system is really tiny and 8MB
> > of reserved memory is indeed really high. How come those reservations
> > have grown that high?
> 
> Actually it is a VM running on the Linux kernel.
> 
> Regarding the reservations, I think it is because of the 'max_managed '
> calculations in the below:
> static void reserve_highatomic_pageblock(struct page *page, ....) {
>     ....
>   /*
>    * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
>    * Check is race-prone but harmless.
>    */
>     max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
> 
>     if (zone->nr_reserved_highatomic >= max_managed)
>             goto out;
> 
>     zone->nr_reserved_highatomic += pageblock_nr_pages;
>     set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
>     move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
> out:
> }
> 
> Since we are always appending the 1% of zone managed pages count to
> pageblock_nr_pages, the minimum it is turning into 2 pageblocks as the
> 'nr_reserved_highatomic' is incremented/decremented in pageblock size
> granules.
> 
> And for my case the 8M out of ~50M is turned out to be 16%, which is high.
> 
> If the below looks fine to you, I can raise this as a separate change:

Yes, please. Having a full page block (4MB) sounds still too much for
such a tiny system. Maybe there shouldn't be any reservation. But
definitely worth a separate patch.

> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 2a2536d..41441ced 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -1886,7 +1886,9 @@ static void reserve_highatomic_pageblock(struct
> page *page, struct zone *zone)
>          * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
>          * Check is race-prone but harmless.
>          */
> -       max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
> +       max_managed = max_t(unsigned long,
> +                       ALIGN(zone_managed_pages(zone) / 100,
> pageblock_nr_pages),
> +                       pageblock_nr_pages);
>         if (zone->nr_reserved_highatomic >= max_managed)
>                 return;
> 
> >>
> >> Per above log, the free memory of ~7MB exist in the high atomic
> >> reserves is not freed up before falling back to oom kill path.
> >>
> >> This fix includes unreserving these atomic reserves in the OOM path
> >> before going for a kill. The side effect of unreserving in oom kill path
> >> is that these free pages are checked against the high wmark. If
> >> unreserved from should_reclaim_retry()/__alloc_pages_direct_reclaim(),
> >> they are checked against the min wmark levels.
> > 
> > I do not like the fix much TBH. I think the logic should live in
> 
> yeah, This code looks way too cleaner to me. Let me know If I can raise
> V2 with the below, suggested-by you.

Sure, go ahead.
 
> I think another thing system is missing here is draining the pcp lists.
> min:804kB low:1004kB high:1204kB free_pcp:688kB

Yes, but this seems like negligible even under a small system like that.
Does it actually help to keep system in balance? I would expect that the
OOM is just imminent no matter the draining. Anyway if this makes any
difference then just make it a separate patch please.
-- 
Michal Hocko
SUSE Labs