Re: [PATCH stable] mm/khugepaged: write all dirty file folios when collapsing

[Lance Yang]

On Fri, Jul 03, 2026 at 10:55:42AM +0200, David Hildenbrand (Arm) wrote:
>On 7/2/26 18:54, Pedro Falcato wrote:
>> As-is, khugepaged and writable-file opening exclude each other. A file
>> cannot be open writeable and have THPs (because the filesystem is not aware
>> of them). khugepaged will never collapse file pages for files that are
>> opened writeable. On an open(O_RDWR/O_WRONLY), the page cache for that
>> particular file is dropped. This is fine because nothing could've been
>> dirtied.
>> 
>> However, there is an edge-case: collapse_file() might not be able to
>> coexist with concurrent writers, but it can coexist with dirty folios
>> (from previous writers). Therefore, the following can happen:
>> 
>> open(file, O_RDWR)
>> write(file)
>> close(file)
>
>Okay, folios are dirty.
>
>> madvise(file_mapping, MADV_COLLAPSE, some non-dirty range)
>
>collapse_file() has
>
>	if (!is_shmem && (folio_test_dirty(folio) ||
>			  folio_test_writeback(folio))) {
>		...
>		result = SCAN_PAGE_DIRTY_OR_WRITEBACK;
>		goto out_unlock;
>	}
>
>Making us abort collapse.
>
>What am I missing?

Hmm ... dirty folios can be outside the range being collapsed ...

For example:

  write/dirty:      [6M, 8M)
  MADV_COLLAPSE:    [0M, 2M)

collapse_file() only checks the folios in the collapse range, so the
dirty/writeback check passes for [0M, 2M). But after that, for the old
READ_ONLY_THP_FOR_FS case, nr_thps gets bumped for the mapping.

Then a later writable open can hit ...

  filemap_nr_thps(mapping)
    -> truncate_inode_pages(mapping, 0)

and that drops page cache for the whole mapping, including the dirty
folios at [6M, 8M) ...

Cheers, Lance