On Sun, Oct 27, 2024 at 2:36 PM Vlastimil Babka <vbabka@xxxxxxx> wrote: > > On 10/27/24 21:17, Yu Zhao wrote: > > On Sun, Oct 27, 2024 at 1:53 PM Vlastimil Babka <vbabka@xxxxxxx> wrote: > >> > >> On 10/26/24 05:36, Yu Zhao wrote: > >> > OOM kills due to vastly overestimated free highatomic reserves were > >> > observed: > >> > > >> > ... invoked oom-killer: gfp_mask=0x100cca(GFP_HIGHUSER_MOVABLE), order=0 ... > >> > Node 0 Normal free:1482936kB boost:0kB min:410416kB low:739404kB high:1068392kB reserved_highatomic:1073152KB ... > >> > Node 0 Normal: 1292*4kB (ME) 1920*8kB (E) 383*16kB (UE) 220*32kB (ME) 340*64kB (E) 2155*128kB (UE) 3243*256kB (UE) 615*512kB (U) 1*1024kB (M) 0*2048kB 0*4096kB = 1477408kB > >> > > >> > The second line above shows that the OOM kill was due to the following > >> > condition: > >> > > >> > free (1482936kB) - reserved_highatomic (1073152kB) = 409784KB < min (410416kB) > >> > > >> > And the third line shows there were no free pages in any > >> > MIGRATE_HIGHATOMIC pageblocks, which otherwise would show up as type > >> > 'H'. Therefore __zone_watermark_unusable_free() underestimated the > >> > usable free memory by over 1GB, which resulted in the unnecessary OOM > >> > kill above. > >> > > >> > The comments in __zone_watermark_unusable_free() warns about the > >> > potential risk, i.e., > >> > > >> > If the caller does not have rights to reserves below the min > >> > watermark then subtract the high-atomic reserves. This will > >> > over-estimate the size of the atomic reserve but it avoids a search. > >> > > >> > However, it is possible to keep track of free pages in reserved > >> > highatomic pageblocks with a new per-zone counter nr_free_highatomic > >> > protected by the zone lock, to avoid a search when calculating the > >> > >> It's only possible to track this reliably since the "mm: page_alloc: > >> freelist migratetype hygiene" patchset was merged, which explains why > >> nr_reserved_highatomic was used until now, even if it's imprecise. > > > > I just refreshed my memory by quickly going through the discussion > > around that series and didn't find anything that helps me understand > > the above. More pointers please? > > For example: > > - a page is on pcplist in MIGRATE_MOVABLE list > - we reserve its pageblock as highatomic, which does nothing to the page on > the pcplist > - page above is flushed from pcplist to zone freelist, but it remembers it > was MIGRATE_MOVABLE, merges with another buddy/buddies from the > now-highatomic list, the resulting order-X page ends up on the movable > freelist despite being in highatomic pageblock. The counter of free > highatomic is now wrong wrt the freelist reality This is the part I don't follow: how is it wrong w.r.t. the freelist reality? The new nr_free_highatomic should reflect how many pages are exactly on free_list[MIGRATE_HIGHATOMIC], because it's updated accordingly. (My current understanding is that, in this case, the reservation itself is messed up, i.e., under-reserved.) > The series has addressed various scenarios like that, where page can end up > on the wrong freelist.
