On Thu, Aug 04, 2022 at 02:56:59PM -0400, Stefan Hajnoczi wrote:
> On Wed, Aug 03, 2022 at 12:28:24PM -0400, Stefan Hajnoczi wrote:
> > As soon as virtio_scsi_data_plane_start() attaches host notifiers the
> > IOThread may start virtqueue processing. There is a race between
> > IOThread virtqueue processing and virtio_scsi_data_plane_start() because
> > it only assigns s->dataplane_started after attaching host notifiers.
> >
> > When a virtqueue handler function in the IOThread calls
> > virtio_scsi_defer_to_dataplane() it may see !s->dataplane_started and
> > attempt to start dataplane even though we're already in the IOThread:
> >
> > #0 0x7f67b360857c __pthread_kill_implementation (libc.so.6 + 0xa257c)
> > #1 0x7f67b35bbd56 raise (libc.so.6 + 0x55d56)
> > #2 0x7f67b358e833 abort (libc.so.6 + 0x28833)
> > #3 0x7f67b358e75b __assert_fail_base.cold (libc.so.6 + 0x2875b)
> > #4 0x7f67b35b4cd6 __assert_fail (libc.so.6 + 0x4ecd6)
> > #5 0x55ca87fd411b memory_region_transaction_commit (qemu-kvm +
> > 0x67511b)
> > #6 0x55ca87e17811 virtio_pci_ioeventfd_assign (qemu-kvm + 0x4b8811)
> > #7 0x55ca87e14836 virtio_bus_set_host_notifier (qemu-kvm + 0x4b5836)
> > #8 0x55ca87f8e14e virtio_scsi_set_host_notifier (qemu-kvm + 0x62f14e)
> > #9 0x55ca87f8dd62 virtio_scsi_dataplane_start (qemu-kvm + 0x62ed62)
> > #10 0x55ca87e14610 virtio_bus_start_ioeventfd (qemu-kvm + 0x4b5610)
> > #11 0x55ca87f8c29a virtio_scsi_handle_ctrl (qemu-kvm + 0x62d29a)
> > #12 0x55ca87fa5902 virtio_queue_host_notifier_read (qemu-kvm +
> > 0x646902)
> > #13 0x55ca882c099e aio_dispatch_handler (qemu-kvm + 0x96199e)
> > #14 0x55ca882c1761 aio_poll (qemu-kvm + 0x962761)
> > #15 0x55ca880e1052 iothread_run (qemu-kvm + 0x782052)
> > #16 0x55ca882c562a qemu_thread_start (qemu-kvm + 0x96662a)
> >
> > This patch assigns s->dataplane_started before attaching host notifiers
> > so that virtqueue handler functions that run in the IOThread before
> > virtio_scsi_data_plane_start() returns correctly identify that dataplane
> > does not need to be started.
> >
> > Note that s->dataplane_started does not need the AioContext lock because
> > it is set before attaching host notifiers and cleared after detaching
> > host notifiers. In other words, the IOThread always sees the value true
> > and the main loop thread does not modify it while the IOThread is
> > active.
> >
> > Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=2099541
> > Reported-by: Qing Wang
>
> Qing Wang has confirmed that this solves the assertion failures.
>
> Paolo/Michael: Can this still be merged for QEMU 7.1?
>
> Stefan
As a bugfix, for sure.
Can I have your ack?
--
MST
