writeback: use higher precision calculation in domain_dirty_limits()
authorTejun Heo <tj@kernel.org>
Fri, 27 May 2016 18:34:46 +0000 (14:34 -0400)
committerJens Axboe <axboe@fb.com>
Mon, 30 May 2016 14:54:40 +0000 (08:54 -0600)
commit62a584fe05eef1f80ed49a286a29328f1a224fb9
treecf34bae00e267a9e5fb94a70d7f24f893d325e6b
parentb02b1fbdd338f70e49efa9ca9899214134526701
writeback: use higher precision calculation in domain_dirty_limits()

As vm.dirty_[background_]bytes can't be applied verbatim to multiple
cgroup writeback domains, they get converted to percentages in
domain_dirty_limits() and applied the same way as
vm.dirty_[background]ratio.  However, if the specified bytes is lower
than 1% of available memory, the calculated ratios become zero and the
writeback domain gets throttled constantly.

Fix it by using per-PAGE_SIZE instead of percentage for ratio
calculations.  Also, the updated DIV_ROUND_UP() usages now should
yield 1/4096 (0.0244%) as the minimum ratio as long as the specified
bytes are above zero.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Miao Xie <miaoxie@huawei.com>
Link: http://lkml.kernel.org/g/57333E75.3080309@huawei.com
Cc: stable@vger.kernel.org # v4.2+
Fixes: 9fc3a43e1757 ("writeback: separate out domain_dirty_limits()")
Reviewed-by: Jan Kara <jack@suse.cz>
Adjusted comment based on Jan's suggestion.
Signed-off-by: Jens Axboe <axboe@fb.com>
mm/page-writeback.c