On Fri, Aug 15, 2025 at 11:38 AM Joanne Koong <joannelkoong@xxxxxxxxx> wrote:
>
> On Thu, Aug 14, 2025 at 9:38 AM Darrick J. Wong <djwong@xxxxxxxxxx> wrote:
> >
> > On Thu, Jul 31, 2025 at 05:21:31PM -0700, Joanne Koong wrote:
> > > Add granular dirty and writeback accounting for large folios. These
> > > stats are used by the mm layer for dirty balancing and throttling.
> > > Having granular dirty and writeback accounting helps prevent
> > > over-aggressive balancing and throttling.
> > >
> > > There are 4 places in iomap this commit affects:
> > > a) filemap dirtying, which now calls filemap_dirty_folio_pages()
> > > b) writeback_iter with setting the wbc->no_stats_accounting bit and
> > > calling clear_dirty_for_io_stats()
> > > c) starting writeback, which now calls __folio_start_writeback()
> > > d) ending writeback, which now calls folio_end_writeback_pages()
> > >
> > > This relies on using the ifs->state dirty bitmap to track dirty pages in
> > > the folio. As such, this can only be utilized on filesystems where the
> > > block size >= PAGE_SIZE.
> >
> > Apologies for my slow responses this month. :)
>
> No worries at all, thanks for looking at this.
> >
> > I wonder, does this cause an observable change in the writeback
> > accounting and throttling behavior for non-fuse filesystems like XFS
> > that use large folios? I *think* this does actually reduce throttling
> > for XFS, but it might not be so noticeable because the limits are much
> > more generous outside of fuse?
>
> I haven't run any benchmarks on non-fuse filesystems yet but that's
> what I would expect too. Will run some benchmarks to see!
I ran some benchmarks on xfs for the contrived test case I used for
fuse (eg writing 2 GB in 128 MB chunks and then doing 50k 50-byte
random writes) and I don't see any noticeable performance difference.
I re-tested it on fuse but this time with strictlimiting disabled and
didn't notice any difference on that either, probably because with
strictlimiting off we don't run into the upper limit in that test so
there's no extra throttling that needs to be mitigated.
It's unclear to me how often (if at all?) real workloads run up
against their dirty/writeback limits.
>
> >
> > > Signed-off-by: Joanne Koong <joannelkoong@xxxxxxxxx>
> > > ---
> > > fs/iomap/buffered-io.c | 136 ++++++++++++++++++++++++++++++++++++++---
> > > 1 file changed, 128 insertions(+), 8 deletions(-)
> > >
> > > diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
> > > index bcc6e0e5334e..626c3c8399cc 100644
> > > --- a/fs/iomap/buffered-io.c
> > > +++ b/fs/iomap/buffered-io.c
> > > @@ -20,6 +20,8 @@ struct iomap_folio_state {
> > > spinlock_t state_lock;
> > > unsigned int read_bytes_pending;
> > > atomic_t write_bytes_pending;
> > > + /* number of pages being currently written back */
> > > + unsigned nr_pages_writeback;
> > >
> > > /*
> > > * Each block has two bits in this bitmap:
> > > @@ -81,6 +83,25 @@ static inline bool ifs_block_is_dirty(struct folio *folio,
> > > return test_bit(block + blks_per_folio, ifs->state);
> > > }
> > >
> > > +static unsigned ifs_count_dirty_pages(struct folio *folio)
> > > +{
> > > + struct iomap_folio_state *ifs = folio->private;
> > > + struct inode *inode = folio->mapping->host;
> > > + unsigned block_size = 1 << inode->i_blkbits;
> > > + unsigned start_blk = 0;
> > > + unsigned end_blk = min((unsigned)(i_size_read(inode) >> inode->i_blkbits),
> > > + i_blocks_per_folio(inode, folio));
> > > + unsigned nblks = 0;
> > > +
> > > + while (start_blk < end_blk) {
> > > + if (ifs_block_is_dirty(folio, ifs, start_blk))
> > > + nblks++;
> > > + start_blk++;
> > > + }
> >
> > Hmm, isn't this bitmap_weight(ifs->state, blks_per_folio) ?
> >
> > Ohh wait no, the dirty bitmap doesn't start on a byte boundary because
> > the format of the bitmap is [uptodate bits][dirty bits].
> >
> > Maybe those two should be reversed, because I bet the dirty state gets
> > changed a lot more over the lifetime of a folio than the uptodate bits.
>
> I think there's the find_next_bit() helper (which Christoph also
> pointed out) that could probably be used here instead. Or at least
> that's how I see a lot of the driver code doing it for their bitmaps.
