]> git.ipfire.org Git - thirdparty/kernel/stable.git/commit
virtio-blk: don't keep queue frozen during system suspend
authorMing Lei <ming.lei@redhat.com>
Tue, 12 Nov 2024 12:58:21 +0000 (20:58 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 2 Jan 2025 09:34:20 +0000 (10:34 +0100)
commit92d5139b91147ab372a17daf5dc27a5b9278e516
treecb7baf75890be025a2601d6a6e2591ad1ab028ac
parentce55818b2d3a999f886af91679589e4644ff1dc8
virtio-blk: don't keep queue frozen during system suspend

[ Upstream commit 7678abee0867e6b7fb89aa40f6e9f575f755fb37 ]

Commit 4ce6e2db00de ("virtio-blk: Ensure no requests in virtqueues before
deleting vqs.") replaces queue quiesce with queue freeze in virtio-blk's
PM callbacks. And the motivation is to drain inflight IOs before suspending.

block layer's queue freeze looks very handy, but it is also easy to cause
deadlock, such as, any attempt to call into bio_queue_enter() may run into
deadlock if the queue is frozen in current context. There are all kinds
of ->suspend() called in suspend context, so keeping queue frozen in the
whole suspend context isn't one good idea. And Marek reported lockdep
warning[1] caused by virtio-blk's freeze queue in virtblk_freeze().

[1] https://lore.kernel.org/linux-block/ca16370e-d646-4eee-b9cc-87277c89c43c@samsung.com/

Given the motivation is to drain in-flight IOs, it can be done by calling
freeze & unfreeze, meantime restore to previous behavior by keeping queue
quiesced during suspend.

Cc: Yi Sun <yi.sun@unisoc.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: virtualization@lists.linux.dev
Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Link: https://lore.kernel.org/r/20241112125821.1475793-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/block/virtio_blk.c