* Re: [PATCH] virtio-scsi: set dataplane_started to false upon failure
2026-08-25 15:48 [PATCH] virtio-scsi: set dataplane_started to false upon failure Stefan Hajnoczi
@ 2026-08-26 8:33 ` Philippe Mathieu-Daudé
2026-09-08 19:37 ` Stefan Hajnoczi
2026-09-12 9:13 ` Michael Tokarev
2 siblings, 0 replies; 5+ messages in thread
From: Philippe Mathieu-Daudé @ 2026-08-26 8:33 UTC (permalink / raw)
To: Stefan Hajnoczi, qemu-devel; +Cc: Fam Zheng, Paolo Bonzini, Michael S. Tsirkin
On 25/8/26 17:48, Stefan Hajnoczi wrote:
> Commit e24a47c5b73e ("virtio-scsi: do not use vring in dataplane") set
> dataplane_started to true when virtio_scsi_dataplane_start() fails. This
> is likely a typo since dataplane is not functional after failure.
>
> The non-ioeventfd fallback is broken because dataplane_started is true
> after virtio_scsi_dataplane_start() failure. As a result, the following
> assertion failure occurs when a virtio-scsi device exceeds the open file
> descriptor ulimit when setting up notifiers (eventfds):
>
> $ ulimit -n 104 # exact number can vary based on QEMU version
> $ qemu-system-x86_64 -M q35,accel=kvm -m 1G -cpu host -smp 14 \
> --blockdev file,node-name=drive0,filename=test.img,cache.direct=on,aio=native \
> --blockdev file,node-name=drive1,filename=data01.img,cache.direct=on,aio=native \
> --device virtio-scsi-pci,id=virtio-scsi-pci0 \
> --device scsi-hd,drive=drive0,bus=virtio-scsi-pci0.0 \
> --device virtio-scsi-pci,id=virtio-scsi-pci1 \
> --device scsi-hd,drive=drive1,bus=virtio-scsi-pci1.0
> qemu: virtio-scsi: Failed to set guest notifiers (-24), ensure -accel kvm is set.
> qemu: virtio_bus_start_ioeventfd: failed. Fallback to userspace (slower).
> qemu-system-x86_64: ../hw/scsi/virtio-scsi.c:1017: virtio_scsi_reset: Assertion `!s->dataplane_started' failed.
> Aborted (core dumped)
>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Fixes: e24a47c5b73e ("virtio-scsi: do not use vring in dataplane")
> Buglink: https://redhat.atlassian.net/browse/RHEL-193328
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> ---
> hw/scsi/virtio-scsi-dataplane.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
Reviewed-by: Philippe Mathieu-Daudé <philmd@oss.qualcomm.com>
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH] virtio-scsi: set dataplane_started to false upon failure
2026-08-25 15:48 [PATCH] virtio-scsi: set dataplane_started to false upon failure Stefan Hajnoczi
2026-08-26 8:33 ` Philippe Mathieu-Daudé
@ 2026-09-08 19:37 ` Stefan Hajnoczi
2026-09-12 9:13 ` Michael Tokarev
2 siblings, 0 replies; 5+ messages in thread
From: Stefan Hajnoczi @ 2026-09-08 19:37 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: Fam Zheng, Paolo Bonzini, Michael S. Tsirkin, qemu-devel
[-- Attachment #1: Type: text/plain, Size: 2380 bytes --]
On Tue, Aug 25, 2026 at 11:48:54AM -0400, Stefan Hajnoczi wrote:
> Commit e24a47c5b73e ("virtio-scsi: do not use vring in dataplane") set
> dataplane_started to true when virtio_scsi_dataplane_start() fails. This
> is likely a typo since dataplane is not functional after failure.
>
> The non-ioeventfd fallback is broken because dataplane_started is true
> after virtio_scsi_dataplane_start() failure. As a result, the following
> assertion failure occurs when a virtio-scsi device exceeds the open file
> descriptor ulimit when setting up notifiers (eventfds):
>
> $ ulimit -n 104 # exact number can vary based on QEMU version
> $ qemu-system-x86_64 -M q35,accel=kvm -m 1G -cpu host -smp 14 \
> --blockdev file,node-name=drive0,filename=test.img,cache.direct=on,aio=native \
> --blockdev file,node-name=drive1,filename=data01.img,cache.direct=on,aio=native \
> --device virtio-scsi-pci,id=virtio-scsi-pci0 \
> --device scsi-hd,drive=drive0,bus=virtio-scsi-pci0.0 \
> --device virtio-scsi-pci,id=virtio-scsi-pci1 \
> --device scsi-hd,drive=drive1,bus=virtio-scsi-pci1.0
> qemu: virtio-scsi: Failed to set guest notifiers (-24), ensure -accel kvm is set.
> qemu: virtio_bus_start_ioeventfd: failed. Fallback to userspace (slower).
> qemu-system-x86_64: ../hw/scsi/virtio-scsi.c:1017: virtio_scsi_reset: Assertion `!s->dataplane_started' failed.
> Aborted (core dumped)
>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
Ping
Paolo: Can you take this through your SCSI tree? Thanks!
Stefan
> Fixes: e24a47c5b73e ("virtio-scsi: do not use vring in dataplane")
> Buglink: https://redhat.atlassian.net/browse/RHEL-193328
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> ---
> hw/scsi/virtio-scsi-dataplane.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/hw/scsi/virtio-scsi-dataplane.c b/hw/scsi/virtio-scsi-dataplane.c
> index 95f13fb7c28..bcb6e441335 100644
> --- a/hw/scsi/virtio-scsi-dataplane.c
> +++ b/hw/scsi/virtio-scsi-dataplane.c
> @@ -230,7 +230,7 @@ fail_host_notifiers:
> fail_guest_notifiers:
> s->dataplane_fenced = true;
> s->dataplane_starting = false;
> - s->dataplane_started = true;
> + s->dataplane_started = false;
> return -ENOSYS;
> }
>
> --
> 2.55.0
>
>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH] virtio-scsi: set dataplane_started to false upon failure
2026-08-25 15:48 [PATCH] virtio-scsi: set dataplane_started to false upon failure Stefan Hajnoczi
2026-08-26 8:33 ` Philippe Mathieu-Daudé
2026-09-08 19:37 ` Stefan Hajnoczi
@ 2026-09-12 9:13 ` Michael Tokarev
2026-09-12 15:09 ` Stefan Hajnoczi
2 siblings, 1 reply; 5+ messages in thread
From: Michael Tokarev @ 2026-09-12 9:13 UTC (permalink / raw)
To: Stefan Hajnoczi, qemu-devel
Cc: Fam Zheng, Paolo Bonzini, Michael S. Tsirkin, QEMU Stable
On 8/25/26 18:48, Stefan Hajnoczi wrote:
> Commit e24a47c5b73e ("virtio-scsi: do not use vring in dataplane") set
> dataplane_started to true when virtio_scsi_dataplane_start() fails. This
> is likely a typo since dataplane is not functional after failure.
>
> The non-ioeventfd fallback is broken because dataplane_started is true
> after virtio_scsi_dataplane_start() failure. As a result, the following
> assertion failure occurs when a virtio-scsi device exceeds the open file
> descriptor ulimit when setting up notifiers (eventfds):
>
> $ ulimit -n 104 # exact number can vary based on QEMU version
> $ qemu-system-x86_64 -M q35,accel=kvm -m 1G -cpu host -smp 14 \
> --blockdev file,node-name=drive0,filename=test.img,cache.direct=on,aio=native \
> --blockdev file,node-name=drive1,filename=data01.img,cache.direct=on,aio=native \
> --device virtio-scsi-pci,id=virtio-scsi-pci0 \
> --device scsi-hd,drive=drive0,bus=virtio-scsi-pci0.0 \
> --device virtio-scsi-pci,id=virtio-scsi-pci1 \
> --device scsi-hd,drive=drive1,bus=virtio-scsi-pci1.0
> qemu: virtio-scsi: Failed to set guest notifiers (-24), ensure -accel kvm is set.
> qemu: virtio_bus_start_ioeventfd: failed. Fallback to userspace (slower).
> qemu-system-x86_64: ../hw/scsi/virtio-scsi.c:1017: virtio_scsi_reset: Assertion `!s->dataplane_started' failed.
> Aborted (core dumped)
>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Fixes: e24a47c5b73e ("virtio-scsi: do not use vring in dataplane")
> Buglink: https://redhat.atlassian.net/browse/RHEL-193328
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
I'm picking this one for currently active qemu stable series.
Please let me know if I shouldn't. Maybe it's a too specific
bug but it's not good to throw assertion failures to the user.
Thanks,
/mjt
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH] virtio-scsi: set dataplane_started to false upon failure
2026-09-12 9:13 ` Michael Tokarev
@ 2026-09-12 15:09 ` Stefan Hajnoczi
0 siblings, 0 replies; 5+ messages in thread
From: Stefan Hajnoczi @ 2026-09-12 15:09 UTC (permalink / raw)
To: Michael Tokarev
Cc: Stefan Hajnoczi, qemu-devel, Fam Zheng, Paolo Bonzini,
Michael S. Tsirkin, QEMU Stable
On Sat, Sep 12, 2026 at 5:14 AM Michael Tokarev <mjt@tls.msk.ru> wrote:
>
> On 8/25/26 18:48, Stefan Hajnoczi wrote:
> > Commit e24a47c5b73e ("virtio-scsi: do not use vring in dataplane") set
> > dataplane_started to true when virtio_scsi_dataplane_start() fails. This
> > is likely a typo since dataplane is not functional after failure.
> >
> > The non-ioeventfd fallback is broken because dataplane_started is true
> > after virtio_scsi_dataplane_start() failure. As a result, the following
> > assertion failure occurs when a virtio-scsi device exceeds the open file
> > descriptor ulimit when setting up notifiers (eventfds):
> >
> > $ ulimit -n 104 # exact number can vary based on QEMU version
> > $ qemu-system-x86_64 -M q35,accel=kvm -m 1G -cpu host -smp 14 \
> > --blockdev file,node-name=drive0,filename=test.img,cache.direct=on,aio=native \
> > --blockdev file,node-name=drive1,filename=data01.img,cache.direct=on,aio=native \
> > --device virtio-scsi-pci,id=virtio-scsi-pci0 \
> > --device scsi-hd,drive=drive0,bus=virtio-scsi-pci0.0 \
> > --device virtio-scsi-pci,id=virtio-scsi-pci1 \
> > --device scsi-hd,drive=drive1,bus=virtio-scsi-pci1.0
> > qemu: virtio-scsi: Failed to set guest notifiers (-24), ensure -accel kvm is set.
> > qemu: virtio_bus_start_ioeventfd: failed. Fallback to userspace (slower).
> > qemu-system-x86_64: ../hw/scsi/virtio-scsi.c:1017: virtio_scsi_reset: Assertion `!s->dataplane_started' failed.
> > Aborted (core dumped)
> >
> > Cc: Paolo Bonzini <pbonzini@redhat.com>
> > Fixes: e24a47c5b73e ("virtio-scsi: do not use vring in dataplane")
> > Buglink: https://redhat.atlassian.net/browse/RHEL-193328
> > Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> I'm picking this one for currently active qemu stable series.
> Please let me know if I shouldn't. Maybe it's a too specific
> bug but it's not good to throw assertion failures to the user.
Sounds good.
Stefan
^ permalink raw reply [flat|nested] 5+ messages in thread