On Tue, Jul 04, 2023 at 07:37:17AM +1000, Dave Chinner wrote: > On Mon, Jul 03, 2023 at 03:13:15AM +0100, Matthew Wilcox wrote: > > On Sun, Jul 02, 2023 at 01:06:15PM -0700, Andrew Morton wrote: > > > On Wed, 28 Jun 2023 19:55:48 +0100 "Matthew Wilcox (Oracle)" <willy@xxxxxxxxxxxxx> wrote: > > > > > > > nr_to_write is a count of pages, so we need to decrease it by the number > > > > of pages in the folio we just wrote, not by 1. Most callers specify > > > > either LONG_MAX or 1, so are unaffected, but writeback_sb_inodes() > > > > might end up writing 512x as many pages as it asked for. > > > > > > 512 is a big number, Should we backport this? > > > > I'm really not sure. Maybe? I'm hoping one of the bots comes up with a > > meaningful performance change as a result of this patch and we find out. > > XFS is the only filesystem this would affect, right? AFAIA, nothing > else enables large folios and uses writeback through > write_cache_pages() at this point... Good point. Still, Intel's 0day has squawked about a loss of performance when large folios have _stopped_ being used, so they are at least testing with XFS.
