Re: [PATCH 0/11] Per-bdi writeback flusher threads v8

[Jens Axboe <jens.axboe@oracle.com>]

On Wed, May 27 2009, Theodore Tso wrote:
> On Wed, May 27, 2009 at 01:53:53PM -0400, Theodore Tso wrote:
> > On Wed, May 27, 2009 at 10:47:54AM -0400, Theodore Tso wrote:
> > > 
> > > I'll retry the test with your stock writeback-v8 git branch w/o any
> > > ext4 patches planned the next mere window mainline to see if I get the
> > > same soft lockup, but I thought I should give you an early heads up.
> > 
> > Confirmed.  I had to run fsstress twice, but I was able to trigger a
> > soft hangup with just the per-bdi v8 patches using ext4.
> 
> As you requested, here's the .config file which I used.  This was on a
> Lenovo S10 (N270 Atom dual-core CPU, 1.5 gigs of memory, 5400 rpm hdd).

If you have time, can you rerun with this little patch? It moves the
super sync to a separate thread. Thanks!

diff --git a/mm/backing-dev.c b/mm/backing-dev.c
index f71588c..8b30f29 100644
--- a/mm/backing-dev.c
+++ b/mm/backing-dev.c
@@ -218,10 +218,14 @@ static __init int bdi_class_init(void)
 }
 postcore_initcall(bdi_class_init);
 
+static int bdi_sync_supers(void *unused);
+
 static int __init default_bdi_init(void)
 {
 	int err;
 
+	kthread_run(bdi_sync_supers, NULL, "bdi-super");
+
 	err = bdi_init(&default_backing_dev_info);
 	if (!err)
 		bdi_register(&default_backing_dev_info, NULL, "default");
@@ -412,6 +416,20 @@ static void bdi_flush_io(struct backing_dev_info *bdi)
 	generic_sync_bdi_inodes(NULL, &wbc);
 }
 
+static int bdi_sync_supers(void *unused)
+{
+	while (!kthread_should_stop()) {
+		schedule_timeout(dirty_expire_interval * 10);
+
+		/*
+		 * Do this periodically, like kupdated() did before.
+		 */
+		sync_supers();
+	}
+
+	return 0;
+}
+
 static int bdi_forker_task(void *ptr)
 {
 	struct bdi_writeback *me = ptr;
@@ -424,11 +442,6 @@ static int bdi_forker_task(void *ptr)
 		struct bdi_writeback *wb;
 
 		/*
-		 * Do this periodically, like kupdated() did before.
-		 */
-		sync_supers();
-
-		/*
 		 * Temporary measure, we want to make sure we don't see
 		 * dirty data on the default backing_dev_info
 		 */

-- 
Jens Axboe