On Thu, Apr 28, 2016 at 10:22:18AM +0100, Ian Campbell wrote:
> On 26 April 2016 at 14:47, Stefan Hajnoczi <> wrote:
> > I need to look seriously into the locking so we can fix it. The fact
> > that rx code calls directly into tx code to send control packets is
> > causing trouble.
>
> You mean that you intend to fix this in the frontend implementation
> rather than by changing the protocol? Sounds great if its possible!
Yes, I think virtio_transport.c needs to be more careful about tx during
rx.
Interestingly the vhost "backend" uses a workqueue and a send packet
list so it doesn't block. I'll take a look at using the same approach
in virtio_transport.c.
Stefan
