Although the oom_reaper is delayed and it gives the oom victim chance to clean up its address space this might take a while especially for processes with a large address space footprint. In those cases oom_reaper might start racing with the dying task and compete for shared resources - e.g. page table lock contention has been observed. Reduce those races by reaping the oom victim from the other end of the address space. It is also a significant improvement for process_mrelease(). When a process is killed, process_mrelease is used to reap the killed process and often runs concurrently with the dying task. The test data shows that after applying the patch, lock contention is greatly reduced during the procedure of reaping the killed process. The test is conducted on arm64. The following basic perf numbers show that applying this patch significantly reduces pte spin lock contention. Without the patch: |--99.57%-- oom_reaper | |--73.58%-- unmap_page_range | | |--8.67%-- [hit in function] | | |--41.59%-- __pte_offset_map_lock | | |--29.47%-- folio_remove_rmap_ptes | | |--16.11%-- tlb_flush_mmu | |--19.94%-- tlb_finish_mmu | |--3.21%-- folio_remove_rmap_ptes With the patch: |--99.53%-- oom_reaper | |--55.77%-- unmap_page_range | | |--20.49%-- [hit in function] | | |--58.30%-- folio_remove_rmap_ptes | | |--11.48%-- tlb_flush_mmu | | |--3.33%-- folio_mark_accessed | |--32.21%-- tlb_finish_mmu | |--6.93%-- folio_remove_rmap_ptes | |--0.69%-- __pte_offset_map_lock Detailed breakdowns for both scenarios are provided below. The cumulative time for oom_reaper plus exit_mmap(victim) in both cases is also summarized, making the performance improvements clear. +----------------------------------------------------------------+ | Category | Applying patch | Without patch | +-------------------------------+----------------+---------------+ | Total running time | 132.6 | 167.1 | | (exit_mmap + reaper work) | 72.4 + 60.2 | 90.7 + 76.4 | +-------------------------------+----------------+---------------+ | Time waiting for pte spinlock | 1.0 | 33.1 | | (exit_mmap + reaper work) | 0.4 + 0.6 | 10.0 + 23.1 | +-------------------------------+----------------+---------------+ | folio_remove_rmap_ptes time | 42.0 | 41.3 | | (exit_mmap + reaper work) | 18.4 + 23.6 | 22.4 + 18.9 | +----------------------------------------------------------------+ From this report, we can see that: 1. The reduction in total time comes mainly from the decrease in time spent on pte spinlock and other locks. 2. oom_reaper performs more work in some areas, but at the same time, exit_mmap also handles certain tasks more efficiently, such as folio_remove_rmap_ptes. Here is a more detailed perf report. [1] link: https://lore.kernel.org/all/20250915162619.5133-1-zhongjinji@honor.com/ Signed-off-by: zhongjinji Reviewed-by: Liam R. Howlett Reviewed-by: Lorenzo Stoakes Reviewed-by: Suren Baghdasaryan Acked-by: Shakeel Butt Acked-by: Michal Hocko --- mm/oom_kill.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) diff --git a/mm/oom_kill.c b/mm/oom_kill.c index 88356b66cc35..28fb36be332b 100644 --- a/mm/oom_kill.c +++ b/mm/oom_kill.c @@ -516,7 +516,7 @@ static bool __oom_reap_task_mm(struct mm_struct *mm) { struct vm_area_struct *vma; bool ret = true; - VMA_ITERATOR(vmi, mm, 0); + MA_STATE(mas, &mm->mm_mt, ULONG_MAX, ULONG_MAX); /* * Tell all users of get_user/copy_from_user etc... that the content @@ -526,7 +526,13 @@ static bool __oom_reap_task_mm(struct mm_struct *mm) */ set_bit(MMF_UNSTABLE, &mm->flags); - for_each_vma(vmi, vma) { + /* + * It might start racing with the dying task and compete for shared + * resources - e.g. page table lock contention has been observed. + * Reduce those races by reaping the oom victim from the other end + * of the address space. + */ + mas_for_each_rev(&mas, vma, 0) { if (vma->vm_flags & (VM_HUGETLB|VM_PFNMAP)) continue; -- 2.17.1