xfs: don't serialise direct IO reads on page cache checks
authorDave Chinner <dchinner@redhat.com>
Thu, 1 Dec 2011 23:27:39 +0000 (17:27 -0600)
committerGreg Kroah-Hartman <gregkh@suse.de>
Fri, 9 Dec 2011 16:54:29 +0000 (08:54 -0800)
commit686da49e5aa50117d8d824c579c3fd9e0318fbc6
treec489f0761a86b4f4f3b6741bbcaa56114d734317
parentf06b70aa96019273ed984f76c79ad786ac216f7e
xfs: don't serialise direct IO reads on page cache checks

commit 0c38a2512df272b14ef4238b476a2e4f70da1479 upstream.

There is no need to grab the i_mutex of the IO lock in exclusive
mode if we don't need to invalidate the page cache. Taking these
locks on every direct IO effective serialises them as taking the IO
lock in exclusive mode has to wait for all shared holders to drop
the lock. That only happens when IO is complete, so effective it
prevents dispatch of concurrent direct IO reads to the same inode.

Fix this by taking the IO lock shared to check the page cache state,
and only then drop it and take the IO lock exclusively if there is
work to be done. Hence for the normal direct IO case, no exclusive
locking will occur.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Tested-by: Joern Engel <joern@logfs.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
Cc: Ben Myers <bpm@sgi.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
fs/xfs/xfs_file.c