Re: [Qemu-devel] [RFT 0/4] Don't start virtqueues that are not enabled for vhost

[Jason Wang]

On 2019/3/24 下午6:53, Yuri Benditovich wrote:
> Hi Jason,
>
> This series does not do the job. Test case: tap, 4 queues, 2 CPU (so
> only 2 queues are enabled)
> For Q0 and Q1 vhost_net_start_one succeeds,
> For Q2:
> vhost_net_start_one calls vhost_dev_start (the call succeeds, does not
> start queue that is not enabled),
> then vhost_net_start_one calls vhost_net_set_backend, the call fails
>
> Thanks,
> Yuri


Thanks a lot for the testing. I miss the case of 
vhost_net_set_backend(). Let me post a simplified V2.


> On Fri, Mar 22, 2019 at 11:28 AM Jason Wang <address@hidden> wrote:
> > Hi:
> >
> > This series try to avoid starting virtqueue that is not enabled. This
> > is done through querying it through a bus specific way and skip the
> > virtqueues if not enabled when starting vhost virtqueues.
> >
> > Only PCI is implemented, maybe it's better to move the enable flag to
> > virito genenic virtqueue structure.
> >
> > Yuri, Could you please to test this series to see if it solves the
> > issues when using windows driver?
> >
> > Thanks
> >
> > Jason Wang (4):
> >    virtio-bus: introduce a new method for querying the queue status
> >    virtio-pci: set enabled for legacy device
> >    virtio-pci: implement queue_enabled
> >    vhost_net: don't start vhost for the virtqueue that is not enabled
> >
> >   hw/virtio/vhost.c              | 11 +++++++++++
> >   hw/virtio/virtio-pci.c         | 12 +++++++++++-
> >   include/hw/virtio/virtio-bus.h |  4 ++++
> >   3 files changed, 26 insertions(+), 1 deletion(-)
> >
> > --
> > 2.19.1