From: Chuck Lever <chuck.lever@xxxxxxxxxx>
Even after commit 9f87fc4d72f5 ("block: queue data commands from the
flush state machine at the head"), I'm still seeing hangs on NFS
exports that reside on SATA devices.
Bisected to commit 615939a2ae73 ("blk-mq: defer to the normal
submission path for post-flush requests"). I've tested reverting
that commit on several kernels from 0aa69d53ac7c ("Merge tag
'for-6.5/io_uring-2023-06-23' of git://git.kernel.dk/linux") to
v6.5-rc5, and it seems to address the problem.
I also confirmed that commit 9f87fc4d72f5 ("block: queue data
commands from the flush state machine at the head") is still
necessary.
There is still no root cause.
Fixes: 615939a2ae73 ("blk-mq: defer to the normal submission path for post-flush requests")
Signed-off-by: Chuck Lever <chuck.lever@xxxxxxxxxx>
---
block/blk-flush.c | 11 -----------
1 file changed, 11 deletions(-)
diff --git a/block/blk-flush.c b/block/blk-flush.c
index 8220517c2d67..e4d12f2c344c 100644
--- a/block/blk-flush.c
+++ b/block/blk-flush.c
@@ -437,17 +437,6 @@ bool blk_insert_flush(struct request *rq)
* Queue for normal execution.
*/
return false;
- case REQ_FSEQ_DATA | REQ_FSEQ_POSTFLUSH:
- /*
- * Initialize the flush fields and completion handler to trigger
- * the post flush, and then just pass the command on.
- */
- blk_rq_init_flush(rq);
- rq->flush.seq |= REQ_FSEQ_POSTFLUSH;
- spin_lock_irq(&fq->mq_flush_lock);
- list_move_tail(&rq->flush.list, &fq->flush_data_in_flight);
- spin_unlock_irq(&fq->mq_flush_lock);
- return false;
default:
/*
* Mark the request as part of a flush sequence and submit it
