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 based on arm64.

Without the patch:
|--99.57%-- oom_reaper
|    |--0.28%-- [hit in function]
|    |--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
|    |    |--1.66%-- folio_mark_accessed
|    |    |--0.74%-- free_swap_and_cache_nr
|    |    |--0.69%-- __tlb_remove_folio_pages
|    |--19.94%-- tlb_finish_mmu
|    |--3.21%-- folio_remove_rmap_ptes
|    |--1.16%-- __tlb_remove_folio_pages
|    |--1.16%-- folio_mark_accessed
|    |--0.36%-- __pte_offset_map_lock

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
|    |    |--2.65%-- __tlb_remove_folio_pages
|    |    |--1.37%-- _raw_spin_lock
|    |    |--0.68%-- __mod_lruvec_page_state
|    |    |--0.51%-- __pte_offset_map_lock
|    |--32.21%-- tlb_finish_mmu
|    |--6.93%-- folio_remove_rmap_ptes
|    |--1.90%-- __tlb_remove_folio_pages
|    |--1.55%-- folio_mark_accessed
|    |--0.69%-- __pte_offset_map_lock

Signed-off-by: zhongjinji <zhongjinji@honor.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Acked-by: Shakeel Butt <shakeel.butt@linux.dev>
Acked-by: Michal Hocko <mhocko@suse.com>
---
 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

On Wed 10-09-25 22:37:26, zhongjinji wrote:
> 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 based on arm64.
> 
> Without the patch:
> |--99.57%-- oom_reaper
> |    |--0.28%-- [hit in function]
> |    |--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
> |    |    |--1.66%-- folio_mark_accessed
> |    |    |--0.74%-- free_swap_and_cache_nr
> |    |    |--0.69%-- __tlb_remove_folio_pages
> |    |--19.94%-- tlb_finish_mmu
> |    |--3.21%-- folio_remove_rmap_ptes
> |    |--1.16%-- __tlb_remove_folio_pages
> |    |--1.16%-- folio_mark_accessed
> |    |--0.36%-- __pte_offset_map_lock
> 
> 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
> |    |    |--2.65%-- __tlb_remove_folio_pages
> |    |    |--1.37%-- _raw_spin_lock
> |    |    |--0.68%-- __mod_lruvec_page_state
> |    |    |--0.51%-- __pte_offset_map_lock
> |    |--32.21%-- tlb_finish_mmu
> |    |--6.93%-- folio_remove_rmap_ptes
> |    |--1.90%-- __tlb_remove_folio_pages
> |    |--1.55%-- folio_mark_accessed
> |    |--0.69%-- __pte_offset_map_lock

I do not object to the patch but this profile is not telling much really
as already pointed out in prior versions as we do not know the base
those percentages are from. It would be really much more helpful to
measure the elapse time for the oom_repaer and exit_mmap to see those
gains.
 
> Signed-off-by: zhongjinji <zhongjinji@honor.com>
> Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
> Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
> Reviewed-by: Suren Baghdasaryan <surenb@google.com>
> Acked-by: Shakeel Butt <shakeel.butt@linux.dev>
> Acked-by: Michal Hocko <mhocko@suse.com>
> ---
>  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

-- 
Michal Hocko
SUSE Labs