lkml.org 
[lkml]   [2011]   [May]   [5]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 1/3] writeback: introduce wbc.tagged_sync for the WB_SYNC_NONE sync stage
On Thu 05-05-11 20:14:02, Wu Fengguang wrote:
> On Thu, May 05, 2011 at 05:00:59AM +0800, Jan Kara wrote:
> > On Mon 02-05-11 11:17:51, Wu Fengguang wrote:
> > > sync(2) is performed in two stages: the WB_SYNC_NONE sync and the
> > > WB_SYNC_ALL sync. Tag the first stage with wbc.tagged_sync and do
> > > livelock prevention for it, too.
> > >
> > > Note that writeback_inodes_sb() is called by not only sync(), they are
> > > treated the same because the other callers need also need livelock
> > > prevention.
> > I was thinking about this and could not find any - which other callers
> > of writeback_inodes_sb() need the livelock prevention?
>
> For example, the writeback_inodes_sb_if_idle() call from ext4.
> In general anyone that pass get_nr_dirty_pages() as work->nr_pages
> may be highly over-estimating the work set.
OK, I see what you mean. I agree using tagging in these cases probably
makes sense.

> It won't be directly livelocked since ext4 won't wait for completion,
> however there is possibility the works queued behind are delayed and
> livelocked.
Actually it will, writeback_inodes_sb() waits for completion (because of
s_umount locking). ext4 should use writeback_inodes_sb_if_idle_nr() (with
some relatively small number) anyway to avoid the pauses.

> Ideally simple ->nr_pages works should be given lower priority and
> even may be merged with each other, and that would be future work.
Merging works wanting to do the same thing would be nice. I though about
it some time ago for a while but getting all the combinations right and
making the merging code resonably simple was hard so I postponed it for
later because it was not urgent.

Honza
--
Jan Kara <jack@suse.cz>
SUSE Labs, CR


\
 
 \ /
  Last update: 2011-05-05 15:57    [W:0.044 / U:0.244 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site