Re: [PATCH mm-unstable v4 5/5] mm/khugepaged: unify khugepaged and madv_collapse with collapse_single_pmd()

[Nico Pache]

On Tue, Mar 31, 2026 at 3:35 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Tue, 31 Mar 2026 22:00:58 +0200 "David Hildenbrand (Arm)" <david@xxxxxxxxxx> wrote:
>
> > > out_maybelock:
> > >     /* Caller expects us to hold mmap_lock on return */
> > >     if (mmap_unlocked) {
> > >             *lock_dropped = true;
> > >             mmap_read_lock(mm);
> > >     }
> > >
> >
> > Right.
> >
> > The original code used
> >
> >       bool mmap_locked = true;
> >
> > If the fix get squashed, maybe we should revert to that handling to
> > minimize the churn?
> >
> > (this is not in stable, right?)
>
> "mm/khugepaged: unify khugepaged and madv_collapse with
> collapse_single_pmd(): is presently in mm-stable.
>
> Doing a big rebase&rework isn't the done thing, I believe.  So queue it
> afterwards, add the Fixes: and tolerate the minor runtime bisection
> hole.  It ends up looking like a later hotfix.
>
>
> Below is what I added to mm-unstable.  I added a note-to-self to check
> in on Nico's question at
> https://lore.kernel.org/all/CAA1CXcCYLGQBy9nT-MSTPVK=XgWWtD6Af-QiA_6pXtM_2pybPg@xxxxxxxxxxxxxx/T/#u
>
>
> From: "Lorenzo Stoakes (Oracle)" <ljs@xxxxxxxxxx>
> Subject: mm/khugepaged: fix issue with tracking lock
> Date: Tue, 31 Mar 2026 13:11:18 +0100
>
> We are incorrectly treating lock_dropped to track both whether the lock is
> currently held and whether or not the lock was ever dropped.
>
> Update this change to account for this.
>
> Link: https://lkml.kernel.org/r/7760c811-e100-4d40-9217-0813c28314be@lucifer.local
> Fixes: 330f3758a3bc ("mm/khugepaged: unify khugepaged and madv_collapse with collapse_single_pmd()")
> Signed-off-by: Lorenzo Stoakes (Oracle) <ljs@xxxxxxxxxx>
> Reviewed-by: Lance Yang <lance.yang@xxxxxxxxx>
> Reviewed-by: Nico Pache <npache@xxxxxxxxxx>
> ...
> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> ---
>
>  mm/khugepaged.c |   12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
>
> --- a/mm/khugepaged.c~mm-khugepaged-fix-issue-with-tracking-lock
> +++ a/mm/khugepaged.c
> @@ -2828,6 +2828,7 @@ int madvise_collapse(struct vm_area_stru
>         unsigned long hstart, hend, addr;
>         enum scan_result last_fail = SCAN_FAIL;
>         int thps = 0;
> +       bool mmap_unlocked = false;
>
>         BUG_ON(vma->vm_start > start);
>         BUG_ON(vma->vm_end < end);
> @@ -2850,10 +2851,11 @@ int madvise_collapse(struct vm_area_stru
>         for (addr = hstart; addr < hend; addr += HPAGE_PMD_SIZE) {
>                 enum scan_result result = SCAN_FAIL;
>
> -               if (*lock_dropped) {
> +               if (mmap_unlocked) {
>                         cond_resched();
>                         mmap_read_lock(mm);
> -                       *lock_dropped = false;
> +                       mmap_unlocked = false;
> +                       *lock_dropped = true;
>                         result = hugepage_vma_revalidate(mm, addr, false, &vma,
>                                                          cc);
>                         if (result  != SCAN_SUCCEED) {
> @@ -2864,7 +2866,7 @@ int madvise_collapse(struct vm_area_stru
>                         hend = min(hend, vma->vm_end & HPAGE_PMD_MASK);
>                 }
>
> -               result = collapse_single_pmd(addr, vma, lock_dropped, cc);
> +               result = collapse_single_pmd(addr, vma, &mmap_unlocked, cc);
>
>                 switch (result) {
>                 case SCAN_SUCCEED:
> @@ -2893,8 +2895,10 @@ int madvise_collapse(struct vm_area_stru
>
>  out_maybelock:
>         /* Caller expects us to hold mmap_lock on return */
> -       if (*lock_dropped)
> +       if (mmap_unlocked) {
> +               *lock_dropped = true;

Hmm this was =false in David's example and = true here? Now i am also
confused hahaha. It seems Lorenzo's code is correct. I think i
responded to my own code changes, when I thought David was replying to
Lorenzo's.

I think we are all good with Lorenzo's changes.

Sorry for the confusion

-- Nico

>                 mmap_read_lock(mm);
> +       }
>  out_nolock:
>         mmap_assert_locked(mm);
>         mmdrop(mm);
> _
>