On Mon, Jun 30, 2025 at 07:45:42AM +0200, Christoph Hellwig wrote: > On Fri, Jun 27, 2025 at 12:38:52PM -0400, Brian Foster wrote: > > > Move unlocking the folio out of iomap_writeback_folio into the caller. > > > This means the end writeback machinery is now run with the folio locked > > > when no writeback happend, or writeback completed extremely fast. > > > > > > > I notice that folio_end_dropbehind_write() (via folio_end_writeback()) > > wants to trylock the folio in order to do its thing. Is this going to > > cause issues with that (i.e. prevent invalidations)? > > Good point. It renders the filemap_end_dropbehind_write call (the > function got renamed in 6.16-rc) essentially useless. OTOH this is > the case where no writeback happened due to a race, so it isn't needed > to start with. But it might be worth documenting that fact. > I'll have to read through it again on the next round, but yeah if it's just a calling context caveat then please do leave a note somewhere that it's known and why it's ok. Thanks. Brian
