Re: [PATCH mm-unstable v14 03/16] introduce collapse_single_pmd to unify khugepaged and madvise_collapse

[Lance Yang <lance.yang@linux.dev>]

On 2026/1/24 07:26, Nico Pache wrote:
> On Thu, Jan 22, 2026 at 10:08 PM Lance Yang <lance.yang@linux.dev> wrote:
>>
>>
>>
>> On 2026/1/23 03:28, Nico Pache wrote:
>>> The khugepaged daemon and madvise_collapse have two different
>>> implementations that do almost the same thing.
>>>
>>> Create collapse_single_pmd to increase code reuse and create an entry
>>> point to these two users.
>>>
>>> Refactor madvise_collapse and collapse_scan_mm_slot to use the new
>>> collapse_single_pmd function. This introduces a minor behavioral change
>>> that is most likely an undiscovered bug. The current implementation of
>>> khugepaged tests collapse_test_exit_or_disable before calling
>>> collapse_pte_mapped_thp, but we weren't doing it in the madvise_collapse
>>> case. By unifying these two callers madvise_collapse now also performs
>>> this check. We also modify the return value to be SCAN_ANY_PROCESS which
>>> properly indicates that this process is no longer valid to operate on.
>>>
>>> We also guard the khugepaged_pages_collapsed variable to ensure its only
>>> incremented for khugepaged.
>>>
>>> Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
>>> Reviewed-by: Lance Yang <lance.yang@linux.dev>
>>> Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
>>> Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
>>> Reviewed-by: Zi Yan <ziy@nvidia.com>
>>> Acked-by: David Hildenbrand <david@redhat.com>
>>> Signed-off-by: Nico Pache <npache@redhat.com>
>>> ---
>>
>> I think this patch introduces some functional changes compared to previous
>> version[1] ...
>>
>> Maybe we should drop the r-b tags and let folks take another look?
>>
>> There might be an issue with the vma access in madvise_collapse(). See
>> below:
>>
>> [1]
>> https://lore.kernel.org/linux-mm/20251201174627.23295-3-npache@redhat.com/
>>
>>>    mm/khugepaged.c | 106 +++++++++++++++++++++++++++---------------------
>>>    1 file changed, 60 insertions(+), 46 deletions(-)
>>>
>>> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
>>> index fefcbdca4510..59e5a5588d85 100644
>>> --- a/mm/khugepaged.c
>>> +++ b/mm/khugepaged.c
>>> @@ -2394,6 +2394,54 @@ static enum scan_result collapse_scan_file(struct mm_struct *mm, unsigned long a
>>>        return result;
>>>    }
>>>
>>> +/*
>>> + * Try to collapse a single PMD starting at a PMD aligned addr, and return
>>> + * the results.
>>> + */
>>> +static enum scan_result collapse_single_pmd(unsigned long addr,
>>> +             struct vm_area_struct *vma, bool *mmap_locked,
>>> +             struct collapse_control *cc)
>>> +{
>>> +     struct mm_struct *mm = vma->vm_mm;
>>> +     enum scan_result result;
>>> +     struct file *file;
>>> +     pgoff_t pgoff;
>>> +
>>> +     if (vma_is_anonymous(vma)) {
>>> +             result = collapse_scan_pmd(mm, vma, addr, mmap_locked, cc);
>>> +             goto end;
>>> +     }
>>> +
>>> +     file = get_file(vma->vm_file);
>>> +     pgoff = linear_page_index(vma, addr);
>>> +
>>> +     mmap_read_unlock(mm);
>>> +     *mmap_locked = false;
>>> +     result = collapse_scan_file(mm, addr, file, pgoff, cc);
>>> +     fput(file);
>>> +
>>> +     if (result != SCAN_PTE_MAPPED_HUGEPAGE)
>>> +             goto end;
>>> +
>>> +     mmap_read_lock(mm);
>>> +     *mmap_locked = true;
>>> +     if (collapse_test_exit_or_disable(mm)) {
>>> +             mmap_read_unlock(mm);
>>> +             *mmap_locked = false;
>>> +             return SCAN_ANY_PROCESS;
>>> +     }
>>> +     result = try_collapse_pte_mapped_thp(mm, addr, !cc->is_khugepaged);
>>> +     if (result == SCAN_PMD_MAPPED)
>>> +             result = SCAN_SUCCEED;
>>> +     mmap_read_unlock(mm);
>>> +     *mmap_locked = false;
>>> +
>>> +end:
>>> +     if (cc->is_khugepaged && result == SCAN_SUCCEED)
>>> +             ++khugepaged_pages_collapsed;
>>> +     return result;
>>> +}
>>> +
>>>    static unsigned int collapse_scan_mm_slot(unsigned int pages, enum scan_result *result,
>>>                                            struct collapse_control *cc)
>>>        __releases(&khugepaged_mm_lock)
>>> @@ -2466,34 +2514,9 @@ static unsigned int collapse_scan_mm_slot(unsigned int pages, enum scan_result *
>>>                        VM_BUG_ON(khugepaged_scan.address < hstart ||
>>>                                  khugepaged_scan.address + HPAGE_PMD_SIZE >
>>>                                  hend);
>>> -                     if (!vma_is_anonymous(vma)) {
>>> -                             struct file *file = get_file(vma->vm_file);
>>> -                             pgoff_t pgoff = linear_page_index(vma,
>>> -                                             khugepaged_scan.address);
>>> -
>>> -                             mmap_read_unlock(mm);
>>> -                             mmap_locked = false;
>>> -                             *result = collapse_scan_file(mm,
>>> -                                     khugepaged_scan.address, file, pgoff, cc);
>>> -                             fput(file);
>>> -                             if (*result == SCAN_PTE_MAPPED_HUGEPAGE) {
>>> -                                     mmap_read_lock(mm);
>>> -                                     if (collapse_test_exit_or_disable(mm))
>>> -                                             goto breakouterloop;
>>> -                                     *result = try_collapse_pte_mapped_thp(mm,
>>> -                                             khugepaged_scan.address, false);
>>> -                                     if (*result == SCAN_PMD_MAPPED)
>>> -                                             *result = SCAN_SUCCEED;
>>> -                                     mmap_read_unlock(mm);
>>> -                             }
>>> -                     } else {
>>> -                             *result = collapse_scan_pmd(mm, vma,
>>> -                                     khugepaged_scan.address, &mmap_locked, cc);
>>> -                     }
>>> -
>>> -                     if (*result == SCAN_SUCCEED)
>>> -                             ++khugepaged_pages_collapsed;
>>>
>>> +                     *result = collapse_single_pmd(khugepaged_scan.address,
>>> +                                                   vma, &mmap_locked, cc);
>>>                        /* move to next address */
>>>                        khugepaged_scan.address += HPAGE_PMD_SIZE;
>>>                        progress += HPAGE_PMD_NR;
>>> @@ -2799,6 +2822,7 @@ int madvise_collapse(struct vm_area_struct *vma, unsigned long start,
>>>                        cond_resched();
>>>                        mmap_read_lock(mm);
>>>                        mmap_locked = true;
>>> +                     *lock_dropped = true;
>>>                        result = hugepage_vma_revalidate(mm, addr, false, &vma,
>>>                                                         cc);
>>>                        if (result  != SCAN_SUCCEED) {
>>> @@ -2809,17 +2833,17 @@ int madvise_collapse(struct vm_area_struct *vma, unsigned long start,
>>>                        hend = min(hend, vma->vm_end & HPAGE_PMD_MASK);
>>>                }
>>>                mmap_assert_locked(mm);
>>> -             if (!vma_is_anonymous(vma)) {
>>> -                     struct file *file = get_file(vma->vm_file);
>>> -                     pgoff_t pgoff = linear_page_index(vma, addr);
>>>
>>> -                     mmap_read_unlock(mm);
>>> -                     mmap_locked = false;
>>> +             result = collapse_single_pmd(addr, vma, &mmap_locked, cc);
>>> +
>>> +             if (!mmap_locked)
>>>                        *lock_dropped = true;
>>> -                     result = collapse_scan_file(mm, addr, file, pgoff, cc);
>>>
>>> -                     if (result == SCAN_PAGE_DIRTY_OR_WRITEBACK && !triggered_wb &&
>>> -                         mapping_can_writeback(file->f_mapping)) {
>>> +             if (result == SCAN_PAGE_DIRTY_OR_WRITEBACK && !triggered_wb) {
>>> +                     struct file *file = get_file(vma->vm_file);
>>> +                     pgoff_t pgoff = linear_page_index(vma, addr);
>>
>>
>> After collapse_single_pmd() returns, mmap_lock might have been released.
>> Between
>> that unlock and here, another thread could unmap/remap the VMA, making
>> the vma
>> pointer stale when we access vma->vm_file?
> 
> + Shivank, I thought they were on the CC list.
> 
> Hey! I thought of this case, but then figured it was no different than
> what is currently implemented for the writeback-retry logic, since the
> mmap lock is dropped and not revalidated. BUT I failed to consider
> that the file reference is held throughout that time.
> 
> I thought of moving the functionality into collapse_single_pmd(), but
> figured I'd keep it in madvise_collapse() as it's the sole user of
> that functionality. Given the potential file ref issue, that may be
> the best solution, and I dont think it should be too difficult. I'll
> queue that up, and also drop the r-b tags as you suggested.
> 
> Ok, here's my solution, does this look like the right approach?:

Hey! Thanks for the quick fix!

> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index 59e5a5588d85..dda9fdc35767 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -2418,6 +2418,14 @@ static enum scan_result
> collapse_single_pmd(unsigned long addr,
>          mmap_read_unlock(mm);
>          *mmap_locked = false;
>          result = collapse_scan_file(mm, addr, file, pgoff, cc);
> +
> +       if (!cc->is_khugepaged && result == SCAN_PAGE_DIRTY_OR_WRITEBACK &&
> +           mapping_can_writeback(file->f_mapping)) {
> +               loff_t lstart = (loff_t)pgoff << PAGE_SHIFT;
> +               loff_t lend = lstart + HPAGE_PMD_SIZE - 1;
> +
> +               filemap_write_and_wait_range(file->f_mapping, lstart, lend);
> +       }
>          fput(file);
> 
>          if (result != SCAN_PTE_MAPPED_HUGEPAGE)
> @@ -2840,19 +2848,8 @@ int madvise_collapse(struct vm_area_struct
> *vma, unsigned long start,
>                          *lock_dropped = true;
> 
>                  if (result == SCAN_PAGE_DIRTY_OR_WRITEBACK && !triggered_wb) {
> -                       struct file *file = get_file(vma->vm_file);
> -                       pgoff_t pgoff = linear_page_index(vma, addr);
> -
> -                       if (mapping_can_writeback(file->f_mapping)) {
> -                               loff_t lstart = (loff_t)pgoff << PAGE_SHIFT;
> -                               loff_t lend = lstart + HPAGE_PMD_SIZE - 1;
> -
> -
> filemap_write_and_wait_range(file->f_mapping, lstart, lend);
> -                               triggered_wb = true;
> -                               fput(file);
> -                               goto retry;
> -                       }
> -                       fput(file);
> +                       triggered_wb = true;
> +                       goto retry;
>                  }
> 
>                  switch (result) {
> 
> 
> 
> -- Nico

 From a quick glimpse, that looks good to me ;)

Only madvise needs writeback and then retry once, and khugepaged just
skips dirty pages and moves on.

Now, we grab the file reference before dropping mmap_lock, then only
use the file pointer during writeback - no vma access after unlock.

So even if the VMA gets unmapped, we're safe, IIUC.

[...]