| Message ID | 554f3ce85edca54d14cc1e1b22c4207a3e8f36a7.1687296675.git.osandov@osandov.com (mailing list archive) |
|---|---|
| State | Superseded, archived |
| Headers | show |
| Series | xfs: CPU usage optimizations for realtime allocator | expand |
On Tue, Jun 20, 2023 at 02:32:16PM -0700, Omar Sandoval wrote: > From: Omar Sandoval <osandov@fb.com> > > As explained in the previous commit, xfs_rtallocate_extent_near() looks > for the end of a free extent when searching backwards from the target > bitmap block. Since the previous commit, it searches from the last > bitmap block it checked to the bitmap block containing the start of the > extent. > > This may still be more than necessary, since the free extent may not be > that long. We know the maximum size of the free extent from the realtime > summary. Use that to compute how many bitmap blocks we actually need to > check. > > Signed-off-by: Omar Sandoval <osandov@fb.com> > --- > fs/xfs/xfs_rtalloc.c | 25 +++++++++++++++++++++---- > 1 file changed, 21 insertions(+), 4 deletions(-) > > diff --git a/fs/xfs/xfs_rtalloc.c b/fs/xfs/xfs_rtalloc.c > index 4d9d0be2e616..2e2eb7c4a648 100644 > --- a/fs/xfs/xfs_rtalloc.c > +++ b/fs/xfs/xfs_rtalloc.c > @@ -517,12 +517,29 @@ xfs_rtallocate_extent_near( > * On the negative side of the starting location. > */ > else { /* i < 0 */ > + int maxblocks; > + > /* > - * Loop backwards through the bitmap blocks from > - * where we last checked up to where we are now. > - * There should be an extent which ends in this > - * bitmap block and is long enough. > + * Loop backwards to find the end of the extent > + * we found in the realtime summary. > + * > + * maxblocks is the maximum possible number of > + * bitmap blocks from the start of the extent to > + * the end of the extent. > */ > + if (maxlog == 0) > + maxblocks = 0; > + else if (maxlog < mp->m_blkbit_log) > + maxblocks = 1; > + else > + maxblocks = 2 << (maxlog - mp->m_blkbit_log); > + /* > + * We need to check bbno + i + maxblocks down to > + * bbno + i. We already checked bbno down to > + * bbno + j + 1, so we don't need to check those > + * again. > + */ > + j = min(i + maxblocks, j); Makes sense now with a fresher head... Reviewed-by: Darrick J. Wong <djwong@kernel.org> What does the xfsprogs version of this patchset look like? --D > for (; j >= i; j--) { > error = xfs_rtallocate_extent_block(mp, > tp, bbno + j, minlen, maxavail, > -- > 2.41.0 >
diff --git a/fs/xfs/xfs_rtalloc.c b/fs/xfs/xfs_rtalloc.c index 4d9d0be2e616..2e2eb7c4a648 100644 --- a/fs/xfs/xfs_rtalloc.c +++ b/fs/xfs/xfs_rtalloc.c @@ -517,12 +517,29 @@ xfs_rtallocate_extent_near( * On the negative side of the starting location. */ else { /* i < 0 */ + int maxblocks; + /* - * Loop backwards through the bitmap blocks from - * where we last checked up to where we are now. - * There should be an extent which ends in this - * bitmap block and is long enough. + * Loop backwards to find the end of the extent + * we found in the realtime summary. + * + * maxblocks is the maximum possible number of + * bitmap blocks from the start of the extent to + * the end of the extent. */ + if (maxlog == 0) + maxblocks = 0; + else if (maxlog < mp->m_blkbit_log) + maxblocks = 1; + else + maxblocks = 2 << (maxlog - mp->m_blkbit_log); + /* + * We need to check bbno + i + maxblocks down to + * bbno + i. We already checked bbno down to + * bbno + j + 1, so we don't need to check those + * again. + */ + j = min(i + maxblocks, j); for (; j >= i; j--) { error = xfs_rtallocate_extent_block(mp, tp, bbno + j, minlen, maxavail,