c/pve-qemu-qoup
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

add patch to fix deadlock with VirtIO block and iothread during QMP stop

Browse Source 
Backported from commit bfa36802d1 ("virtio-blk: avoid using ioeventfd
state in irqfd conditional") because the rework/rename dataplane ->
ioeventfd didn't happen yet.

Reported in the community forum [0] and reproduced doing a backup loop
to PBS with suspend mode with fio doing heavy IO in the guest and
using an RBD storage (with krbd).

[0]: https://forum.proxmox.com/threads/141320

Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
This commit is contained in:
Fiona Ebner 2024-02-21 14:01:52 +01:00 committed by Thomas Lamprecht
parent 86460aef76
commit ed159bc32a
2 changed files with 62 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

61
debian/patches/extra/0013-virtio-blk-avoid-using-ioeventfd-state-in-irqfd-cond.patch vendored Normal file
View File

@ -0,0 +1,61 @@
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Stefan Hajnoczi <stefanha@redhat.com>
Date: Mon, 22 Jan 2024 12:26:25 -0500
Subject: [PATCH] virtio-blk: avoid using ioeventfd state in irqfd conditional
Requests that complete in an IOThread use irqfd to notify the guest
while requests that complete in the main loop thread use the traditional
qdev irq code path. The reason for this conditional is that the irq code
path requires the BQL:
if (s->ioeventfd_started && !s->ioeventfd_disabled) {
virtio_notify_irqfd(vdev, req->vq);
} else {
virtio_notify(vdev, req->vq);
}
There is a corner case where the conditional invokes the irq code path
instead of the irqfd code path:
static void virtio_blk_stop_ioeventfd(VirtIODevice *vdev)
{
...
/*
* Set ->ioeventfd_started to false before draining so that host notifiers
* are not detached/attached anymore.
*/
s->ioeventfd_started = false;
/* Wait for virtio_blk_dma_restart_bh() and in flight I/O to complete */
blk_drain(s->conf.conf.blk);
During blk_drain() the conditional produces the wrong result because
ioeventfd_started is false.
Use qemu_in_iothread() instead of checking the ioeventfd state.
Cc: qemu-stable@nongnu.org
Buglink: https://issues.redhat.com/browse/RHEL-15394
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20240122172625.415386-1-stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
[FE: backport: dataplane -> ioeventfd rework didn't happen yet]
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
---
hw/block/virtio-blk.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index 39e7f23fab..61bd1f6859 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -64,7 +64,7 @@ static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
iov_discard_undo(&req->inhdr_undo);
iov_discard_undo(&req->outhdr_undo);
virtqueue_push(req->vq, &req->elem, req->in_len);
- if (s->dataplane_started && !s->dataplane_disabled) {
+ if (qemu_in_iothread()) {
virtio_blk_data_plane_notify(s->dataplane, req->vq);
} else {
virtio_notify(vdev, req->vq);

1
debian/patches/series vendored
View File

@ -10,6 +10,7 @@ extra/0009-ui-clipboard-mark-type-as-not-available-when-there-i.patch
extra/0010-virtio-scsi-Attach-event-vq-notifier-with-no_poll.patch
extra/0011-virtio-Re-enable-notifications-after-drain.patch
extra/0012-qemu_init-increase-NOFILE-soft-limit-on-POSIX.patch
extra/0013-virtio-blk-avoid-using-ioeventfd-state-in-irqfd-cond.patch
bitmap-mirror/0001-drive-mirror-add-support-for-sync-bitmap-mode-never.patch
bitmap-mirror/0002-drive-mirror-add-support-for-conditional-and-always-.patch
bitmap-mirror/0003-mirror-add-check-for-bitmap-mode-without-bitmap.patch