On Tue 26-04-11 13:37:06, Wu Fengguang wrote:
> On Sat, Apr 23, 2011 at 05:12:55AM +0800, Jan Kara wrote:
> > On Fri 22-04-11 10:24:59, Wu Fengguang wrote:
> > > > 2) The intention of both bdi_flush_io() and balance_dirty_pages() is to
> > > > write .nr_to_write pages. So they should either do queue_io()
> > > > unconditionally (I kind of like that for simplicity) or they should requeue
> > > > once if they have not written enough - otherwise it could happen that they
> > > > are called just at the moment when b_io contains a single inode with a few
> > > > dirty pages and they end up doing almost nothing.
> > >
> > > It makes much more sense to keep the policy consistent. When the
> > > flusher and the throttled tasks are both actively manipulating the
> > > shared lists but in different ways, how are we going to analyze the
> > > resulted mixture behavior?
> > >
> > > Note that bdi_flush_io() and balance_dirty_pages() both have outer
> > > loops to retry writeout, so smallish b_io is not a problem at all.
> > Well, it changes how balance_dirty_pages() behaves in some corner cases
> > (I'm not that much concerned about bdi_flush_io() because that is a last
> > resort thing anyway). But I see your point in consistency as well.
> >
> > > > 3) I guess your patch does not compile because queue_io() is static ;).
> > >
> > > Yeah, good spot~ :) Here is the updated patch. I feel like moving
> > > bdi_flush_io() to fs-writeback.c rather than exporting the low level
> > > queue_io() (and enable others to conveniently change the queue policy!).
> > >
> > > balance_dirty_pages() cannot be moved.. so I plan to submit it after
> > > any IO-less merges. It's a cleanup patch after all.
> > Can't we just have a wrapper in fs/fs-writeback.c that will do:
> > spin_lock(&bdi->wb.list_lock);
> > if (list_empty(&bdi->wb.b_io))
> > queue_io(&bdi->wb, &wbc);
> > writeback_inodes_wb(&bdi->wb, &wbc);
> > spin_unlock(&bdi->wb.list_lock);
> >
> > And call it wherever we need? We can then also unexport
> > writeback_inodes_wb() which is not really a function someone would want to
> > call externally after your changes.
>
> OK, this avoids the need to move bdi_flush_io(). Here is the updated
> patch, do you see any more problems?
Yes, with this patch I think your change to the queueing logic is OK.
Thanks.
Honza
>
> Thanks,
> Fengguang
> ---
> Subject: writeback: elevate queue_io() into wb_writeback()
> Date: Thu Apr 21 12:06:32 CST 2011
>
> Code refactor for more logical code layout.
> No behavior change.
>
> - remove the mis-named __writeback_inodes_sb()
>
> - wb_writeback()/writeback_inodes_wb() will decide when to queue_io()
> before calling __writeback_inodes_wb()
>
> Signed-off-by: Wu Fengguang <fengguang.wu@xxxxxxxxx>
> ---
> fs/fs-writeback.c | 27 ++++++++++++---------------
> 1 file changed, 12 insertions(+), 15 deletions(-)
>
> --- linux-next.orig/fs/fs-writeback.c 2011-04-26 13:20:17.000000000 +0800
> +++ linux-next/fs/fs-writeback.c 2011-04-26 13:30:19.000000000 +0800
> @@ -570,17 +570,13 @@ static int writeback_sb_inodes(struct su
> return 1;
> }
>
> -void writeback_inodes_wb(struct bdi_writeback *wb,
> - struct writeback_control *wbc)
> +static void __writeback_inodes_wb(struct bdi_writeback *wb,
> + struct writeback_control *wbc)
> {
> int ret = 0;
>
> if (!wbc->wb_start)
> wbc->wb_start = jiffies; /* livelock avoidance */
> - spin_lock(&wb->list_lock);
> -
> - if (list_empty(&wb->b_io))
> - queue_io(wb, wbc);
>
> while (!list_empty(&wb->b_io)) {
> struct inode *inode = wb_inode(wb->b_io.prev);
> @@ -596,19 +592,16 @@ void writeback_inodes_wb(struct bdi_writ
> if (ret)
> break;
> }
> - spin_unlock(&wb->list_lock);
> /* Leave any unwritten inodes on b_io */
> }
>
> -static void __writeback_inodes_sb(struct super_block *sb,
> - struct bdi_writeback *wb, struct writeback_control *wbc)
> +void writeback_inodes_wb(struct bdi_writeback *wb,
> + struct writeback_control *wbc)
> {
> - WARN_ON(!rwsem_is_locked(&sb->s_umount));
> -
> spin_lock(&wb->list_lock);
> if (list_empty(&wb->b_io))
> queue_io(wb, wbc);
> - writeback_sb_inodes(sb, wb, wbc, true);
> + __writeback_inodes_wb(wb, wbc);
> spin_unlock(&wb->list_lock);
> }
>
> @@ -674,7 +667,7 @@ static long wb_writeback(struct bdi_writ
> * The intended call sequence for WB_SYNC_ALL writeback is:
> *
> * wb_writeback()
> - * __writeback_inodes_sb() <== called only once
> + * writeback_sb_inodes() <== called only once
> * write_cache_pages() <== called once for each inode
> * (quickly) tag currently dirty pages
> * (maybe slowly) sync all tagged pages
> @@ -722,10 +715,14 @@ static long wb_writeback(struct bdi_writ
>
> retry:
> trace_wbc_writeback_start(&wbc, wb->bdi);
> + spin_lock(&wb->list_lock);
> + if (list_empty(&wb->b_io))
> + queue_io(wb, &wbc);
> if (work->sb)
> - __writeback_inodes_sb(work->sb, wb, &wbc);
> + writeback_sb_inodes(work->sb, wb, &wbc, true);
> else
> - writeback_inodes_wb(wb, &wbc);
> + __writeback_inodes_wb(wb, &wbc);
> + spin_unlock(&wb->list_lock);
> trace_wbc_writeback_written(&wbc, wb->bdi);
>
> work->nr_pages -= write_chunk - wbc.nr_to_write;
--
Jan Kara <jack@xxxxxxx>
SUSE Labs, CR
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>