Re: [Qemu-devel] [PATCH V2 RFT] vhost_net: don't set backend for the uninitialized virtqueue

[Michael S. Tsirkin]

On Tue, Mar 26, 2019 at 02:59:19PM +0800, Jason Wang wrote:
> 
> On 2019/3/25 下午8:32, Michael S. Tsirkin wrote:
> > On Mon, Mar 25, 2019 at 11:56:13AM +0800, Jason Wang wrote:
> > > We used to set backend unconditionally, this won't work for some
> > > guests (e.g windows driver) who may not initialize all virtqueues. For
> > > kernel backend, this will fail since it may try to validate the rings
> > > during setting backend.
> > > 
> > > Fixing this by simply skipping the backend set when we find desc is
> > > not ready.
> > > 
> > > Signed-off-by: Jason Wang <address@hidden>
> > > ---
> > >   hw/net/vhost_net.c | 15 +++++++++++++++
> > >   1 file changed, 15 insertions(+)
> > > 
> > > diff --git a/hw/net/vhost_net.c b/hw/net/vhost_net.c
> > > index be3cc88370..04fd924d15 100644
> > > --- a/hw/net/vhost_net.c
> > > +++ b/hw/net/vhost_net.c
> > > @@ -221,6 +221,7 @@ static int vhost_net_start_one(struct vhost_net *net,
> > >                                  VirtIODevice *dev)
> > >   {
> > >       struct vhost_vring_file file = { };
> > > +    hwaddr a;
> > >       int r;
> > >       net->dev.nvqs = 2;
> > > @@ -244,6 +245,13 @@ static int vhost_net_start_one(struct vhost_net *net,
> > >           qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
> > >           file.fd = net->backend;
> > >           for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
> > > +            a = virtio_queue_get_desc_addr(dev,
> > > +                                           net->dev.vq_index +
> > > +                                           file.index);
> > > +            if (a == 0) {
> > > +                /* Queue might not be ready for start */
> > > +                continue;
> > > +            }
> > >               r = vhost_net_set_backend(&net->dev, &file);
> > >               if (r < 0) {
> > >                   r = -errno;
> > > @@ -256,6 +264,13 @@ fail:
> > >       file.fd = -1;
> > >       if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
> > >           while (file.index-- > 0) {
> > > +            a = virtio_queue_get_desc_addr(dev,
> > > +                                           net->dev.vq_index +
> > > +                                           file.index);
> > > +            if (a == 0) {
> > > +                /* Queue might not be ready for start */
> > > +                continue;
> > > +            }
> > >               int r = vhost_net_set_backend(&net->dev, &file);
> > >               assert(r >= 0);
> > >           }
> > 
> > I think we want an API that explicitly says "queue is enabled".
> > For 0.X it will return !!addr. For 1.X it will return enabled.
> 
> 
> For 1.x, desc.addr won't be set until queue_enabled is set through
> virtio_queue_set_vrings(). And it looks to me ccw did something similar of 
> CMD_SET_VQ.
> 
> So we're ok actually?
> 
> Thanks
> 

OK maybe but why can't we use an explicit API?
0.X can use addr != 0 trick since there's no queue_enabled.

> > 
> > 
> > At the moment enabled seems to be saved but ignored at least
> > in case of virtio-pci and I think that's the real bug.
> > 
> > 
> > > -- 
> > > 2.19.1