[pve-devel] applied: [PATCH v2 qemu] add patches to work around stuck guest IO with iothread and VirtIO block/SCSI

Fiona Ebner f.ebner at proxmox.com
Mon Feb 5 12:39:46 CET 2024

Am 02.02.24 um 19:42 schrieb Thomas Lamprecht:
> basically applied this, thanks.
> Basically, as I went for the v2 [0], as the series file needed some adaption
>  due to recent v8.1.5 patch I just made a new commit but kept your message and
> added an Originally-by Trailer, hope that's all right with you.

Yes, sure. Wanted to send a new version anyways :)

While the diff to v1 is pretty small, I did a quick test again and found
no obvious issue, while it still fixes my reproducer.

> [0]: https://lore.kernel.org/qemu-devel/20240202153158.788922-1-hreitz@redhat.com/
>      But still only the first two patches, the clean-up did not applied at 8.1.5 
>      and I did not bother checking that out closely.

Yes, there were some changes in the meantime and while a backport might
be removing the event_notifier_set() call from
virtio_blk_data_plane_start(), with the current code, there also is a
smp_wmb() and aio_context_acquire(s->ctx) in between
event_notifier_set() and virtio_queue_aio_attach_host_notifier(), so not
sure if that's okay in all scenarios. I don't think the backport is
necessary, because setting the event notifier twice will just kick the
virt queue twice.

More information about the pve-devel mailing list