Re: DIO hangs in 6.4.0-rc2

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, May 16, 2023 at 01:28:00PM +0100, Matthew Wilcox wrote:
> Plain 6.4.0-rc2 with a relatively minor change to the futex code that
> I cannot believe was in any way responsible for this.
> 
> kworkers blocked all over the place.  Some on XFS_ILOCK_EXCL.  Some on
> xfs_buf_lock.  One in xfs_btree_split() calling wait_for_completion.
> 
> This was an overnight test run that is now dead, so I can't get any
> more info from the locked up kernel.  I have the vmlinux if some
> decoding of offsets is useful.

This is likely the same AGF try-lock bug that was discovered in this
thread:

https://lore.kernel.org/linux-xfs/202305090905.aff4e0e6-oliver.sang@xxxxxxxxx/

The fact that the try-lock was ignored means that out of order AGF
locking can be attempted, and the try-lock prevents deadlocks from
occurring.

Can you try the patch below - I was going to send it for review
anyway this morning so it can't hurt to see if it also fixes this
issue.

-Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx

xfs: restore allocation trylock iteration

From: Dave Chinner <dchinner@xxxxxxxxxx>

It was accidentally dropped when refactoring the allocation code,
resulting in the AG iteration always doing blocking AG iteration.
This results in a small performance regression for a specific fsmark
test that runs more user data writer threads than there are AGs.

Reported-by: kernel test robot <oliver.sang@xxxxxxxxx>
Fixes: 2edf06a50f5b ("xfs: factor xfs_alloc_vextent_this_ag() for _iterate_ags()")
Signed-off-by: Dave Chinner <dchinner@xxxxxxxxxx>
---
 fs/xfs/libxfs/xfs_alloc.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

diff --git a/fs/xfs/libxfs/xfs_alloc.c b/fs/xfs/libxfs/xfs_alloc.c
index fdfa08cbf4db..61eb65be17f3 100644
--- a/fs/xfs/libxfs/xfs_alloc.c
+++ b/fs/xfs/libxfs/xfs_alloc.c
@@ -3187,7 +3187,8 @@ xfs_alloc_vextent_check_args(
  */
 static int
 xfs_alloc_vextent_prepare_ag(
-	struct xfs_alloc_arg	*args)
+	struct xfs_alloc_arg	*args,
+	uint32_t		flags)
 {
 	bool			need_pag = !args->pag;
 	int			error;
@@ -3196,7 +3197,7 @@ xfs_alloc_vextent_prepare_ag(
 		args->pag = xfs_perag_get(args->mp, args->agno);
 
 	args->agbp = NULL;
-	error = xfs_alloc_fix_freelist(args, 0);
+	error = xfs_alloc_fix_freelist(args, flags);
 	if (error) {
 		trace_xfs_alloc_vextent_nofix(args);
 		if (need_pag)
@@ -3336,7 +3337,7 @@ xfs_alloc_vextent_this_ag(
 		return error;
 	}
 
-	error = xfs_alloc_vextent_prepare_ag(args);
+	error = xfs_alloc_vextent_prepare_ag(args, 0);
 	if (!error && args->agbp)
 		error = xfs_alloc_ag_vextent_size(args);
 
@@ -3380,7 +3381,7 @@ xfs_alloc_vextent_iterate_ags(
 	for_each_perag_wrap_range(mp, start_agno, restart_agno,
 			mp->m_sb.sb_agcount, agno, args->pag) {
 		args->agno = agno;
-		error = xfs_alloc_vextent_prepare_ag(args);
+		error = xfs_alloc_vextent_prepare_ag(args, flags);
 		if (error)
 			break;
 		if (!args->agbp) {
@@ -3546,7 +3547,7 @@ xfs_alloc_vextent_exact_bno(
 		return error;
 	}
 
-	error = xfs_alloc_vextent_prepare_ag(args);
+	error = xfs_alloc_vextent_prepare_ag(args, 0);
 	if (!error && args->agbp)
 		error = xfs_alloc_ag_vextent_exact(args);
 
@@ -3587,7 +3588,7 @@ xfs_alloc_vextent_near_bno(
 	if (needs_perag)
 		args->pag = xfs_perag_grab(mp, args->agno);
 
-	error = xfs_alloc_vextent_prepare_ag(args);
+	error = xfs_alloc_vextent_prepare_ag(args, 0);
 	if (!error && args->agbp)
 		error = xfs_alloc_ag_vextent_near(args);
 



[Index of Archives]     [XFS Filesystem Development (older mail)]     [Linux Filesystem Development]     [Linux Audio Users]     [Yosemite Trails]     [Linux Kernel]     [Linux RAID]     [Linux SCSI]


  Powered by Linux