[dm-devel] [PATCH] dm: Avoid sleeping while holding the dm_bufio lock

Douglas Anderson dianders at chromium.org
Thu Nov 17 19:24:20 UTC 2016

We've seen in-field reports showing _lots_ (18 in one case, 41 in
another) of tasks all sitting there blocked on:


In the two cases analyzed, we see one task that looks like this:

  Workqueue: kverityd verity_prefetch_io


...and that looks to be the one holding the mutex.

The problem has been reproduced on fairly easily:
0. Be running Chrome OS w/ verity enabled on the root filesystem
1. Pick test patch: http://crosreview.com/412360
2. Install launchBalloons.sh and balloon.arm from
   ...that's just a memory stress test app.
3. On a 4GB rk3399 machine, run
     nice ./launchBalloons.sh 4 900 100000
   ...that tries to eat 4 * 900 MB of memory and keep accessing.
4. Login to the Chrome web browser and restore many tabs

With that, I've seen printouts like:
  DOUG: long bufio 90758 ms
...and stack trace always show's we're in dm_bufio_prefetch().

The problem is that we try to allocate memory with GFP_NOIO while
we're holding the dm_bufio lock.  Instead we should be using
GFP_NOWAIT.  Using GFP_NOIO can cause us to sleep while holding the
lock and that causes the above problems.

The current behavior explained by David Rientjes:

  It will still try reclaim initially because __GFP_WAIT (or
  __GFP_KSWAPD_RECLAIM) is set by GFP_NOIO.  This is the cause of
  contention on dm_bufio_lock() that the thread holds.  You want to
  pass GFP_NOWAIT instead of GFP_NOIO to alloc_buffer() when holding a
  mutex that can be contended by a concurrent slab shrinker (if
  count_objects didn't use a trylock, this pattern would trivially

Suggested-by: David Rientjes <rientjes at google.com>
Signed-off-by: Douglas Anderson <dianders at chromium.org>
Note that this change was developed and tested against the Chrome OS
4.4 kernel tree, not mainline.  Due to slight differences in verity
between mainline and Chrome OS it became too difficult to reproduce my
testing setup on mainline.  This patch still seems correct and
relevant to upstream, so I'm posting it.  If this is not acceptible to
you then please ignore this patch.

Also note that when I tested the Chrome OS 3.14 kernel tree I couldn't
reproduce the long delays described in the patch.  Presumably
something changed in either the kernel config or the memory management
code between the two kernel versions that made this crop up.  In a
similar vein, it is possible that problems described in this patch are
no longer reproducible upstream.  However, the arguments made in this
patch (that we don't want to block while holding the mutex) still
apply so I think the patch may still have merit.

 drivers/md/dm-bufio.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c
index b3ba142e59a4..3c767399cc59 100644
--- a/drivers/md/dm-bufio.c
+++ b/drivers/md/dm-bufio.c
@@ -827,7 +827,8 @@ static struct dm_buffer *__alloc_buffer_wait_no_callback(struct dm_bufio_client
 	 * dm-bufio is resistant to allocation failures (it just keeps
 	 * one buffer reserved in cases all the allocations fail).
 	 * So set flags to not try too hard:
-	 *	GFP_NOIO: don't recurse into the I/O layer
+	 *	GFP_NOWAIT: don't wait; if we need to sleep we'll release our
+	 *		    mutex and wait ourselves.
 	 *	__GFP_NORETRY: don't retry and rather return failure
 	 *	__GFP_NOMEMALLOC: don't use emergency reserves
 	 *	__GFP_NOWARN: don't print a warning in case of failure
@@ -837,7 +838,8 @@ static struct dm_buffer *__alloc_buffer_wait_no_callback(struct dm_bufio_client
 	while (1) {
 		if (dm_bufio_cache_size_latch != 1) {
-			b = alloc_buffer(c, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
+			b = alloc_buffer(c, GFP_NOWAIT | __GFP_NORETRY |
 			if (b)
 				return b;

More information about the dm-devel mailing list