Re: [PATCH v9 3/6] mm: add folio_mark_dropbehind() to mark a folio dropbehind and account it

[Mike Snitzer <snitzer@kernel.org>]

On Sun, Oct 25, 2026 at 10:12:09PM +0000, sashiko-bot@kernel.org wrote:
> Can WB_DONTCACHE_DIRTY still be double-counted here if a stale
> PAGECACHE_TAG_DIRTY coincides with an unlocked dirtier that has just set
> PG_dirty?

Yes, in that window.  The folio must be clean with a stale dirty tag.
On XFS that happens when iomap_writeback_handle_eof() skips a folio that
a truncate put beyond EOF.  Then an unlocked dirtier sets PG_dirty, and
folio_mark_dropbehind() or filemap_clear_dropbehind() runs on that folio
before the dirtier takes the i_pages lock.  In that gap, a counted folio
and an uncounted one look the same: PG_dirty and the dirty tag are both
set.  Closing it needs a new page flag, or a tag clear in the paths that
uncount a folio.  Both are core mm changes.

Only the counter changes.  WB_DONTCACHE_DIRTY tells
wb_check_start_dontcache() how many pages the DONTCACHE writeback kick
writes.  A count that is too high makes the kick write more dirty pages.
A count below zero makes it write none.  Data and page state do not
change.  Before 1/6, the clear in __filemap_get_folio_mpol() decided by
PG_dirty alone, so this window and a wider one were already there.

I do not plan to change this in this series.  If linux-mm wants it
closed, I can follow up with whichever of the two changes linux-mm
prefers.