Re: [dpdk-dev] [PATCH] net/virtio-user: fix multiple queues fail in server mode

[Tiwei Bie <tiwei.bie@intel.com>]

On Thu, May 10, 2018 at 11:04:59AM +0800, Yang, Zhiyong wrote:
> Hi  Tiwei,
> 
> > -----Original Message-----
> > From: Bie, Tiwei
> > Sent: Thursday, May 10, 2018 10:47 AM
> > To: Yang, Zhiyong <zhiyong.yang@intel.com>
> > Cc: dev@dpdk.org; maxime.coquelin@redhat.com; Yao, Lei A
> > <lei.a.yao@intel.com>; stable@dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH] net/virtio-user: fix multiple queues fail in
> > server mode
> > 
> > On Wed, May 09, 2018 at 05:49:36PM +0800, zhiyong.yang@intel.com wrote:
> > > This patch fixes multiple queues failure when virtio-user works in
> > > server mode.
> > >
> > > Fixes: bd8f50a45d0f ("net/virtio-user: support server mode")
> > > Cc: stable@dpdk.org
> > > Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com>
> > > ---
> > >  drivers/net/virtio/virtio_user/vhost_user.c      | 3 +++
> > >  drivers/net/virtio/virtio_user/virtio_user_dev.c | 9 +++++++++
> > >  2 files changed, 12 insertions(+)
> > >
> > > diff --git a/drivers/net/virtio/virtio_user/vhost_user.c
> > > b/drivers/net/virtio/virtio_user/vhost_user.c
> > > index a6df97a00..a9e53d7b5 100644
> > > --- a/drivers/net/virtio/virtio_user/vhost_user.c
> > > +++ b/drivers/net/virtio/virtio_user/vhost_user.c
> > > @@ -263,6 +263,9 @@ vhost_user_sock(struct virtio_user_dev *dev,
> > >
> > >  	PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
> > >
> > > +	if (vhostfd < 0)
> > > +		return -1;
> > 
> > I think this is just a workaround to avoid printing error messages in server
> > mode.
> > 
> > Ideally, vhost_user_sock() shouldn't be called with vhostfd < 0.
> > 
> > If we want this workaround, we should only allow this in server mode. I.e. do
> > the check like this:
> > 
> > 	if (dev->is_server && vhostfd < 0)
> > 		return -1;
>  
> Ok. 
> 
> > > +
> > >  	msg.request = req;
> > >  	msg.flags = VHOST_USER_VERSION;
> > >  	msg.size = 0;
> > > diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.c
> > > b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> > > index 38b8bc90d..e988dc3f4 100644
> > > --- a/drivers/net/virtio/virtio_user/virtio_user_dev.c
> > > +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> > > @@ -126,6 +126,10 @@ virtio_user_start_device(struct virtio_user_dev *dev)
> > >  	features &= ~(1ull << VIRTIO_NET_F_MAC);
> > >  	/* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */
> > >  	features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
> > > +	/* Also disable features which depend on VIRTIO_NET_F_CTRL_VQ */
> > > +	features &= ~(1ull << VIRTIO_NET_F_CTRL_RX);
> > > +	features &= ~(1ull << VIRTIO_NET_F_CTRL_VLAN);
> > > +	features &= ~(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
> > 
> > Is this really related to this fix?
> > 
> 
> Yes.  if vhost user side receives these feature bits and  
> Vhost user doesn’t  support these features and will close the socket connection.
> If we have disable ctrl_VQ,  also should disable them.
> 

If cq is disabled, all the related features (including
VIRTIO_NET_F_MQ) should have been disabled.

The features in this function should already be a subset
of the features supported by the vhost backend. The thing
that's not right is that, in server mode, virtio-user
doesn't do GET_FEATURES. And we need to do it.

Thanks

> 
> > >  	features &= ~(1ull << VIRTIO_NET_F_STATUS);
> > >  	ret = dev->ops->send_request(dev, VHOST_USER_SET_FEATURES,
> > &features);
> > >  	if (ret < 0)
> > > @@ -488,6 +492,11 @@ virtio_user_handle_ctrl_msg(struct
> > virtio_user_dev *dev, struct vring *vring,
> > >  		queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
> > >  		status = virtio_user_handle_mq(dev, queues);
> > >  	}
> > > +	/* Server mode can't enable queue pairs if vhostfd is not connected,
> > > +	 * we suppose that status always returns 0 in this case.
> > > +	 * /
> > 
> > As Ferruh pointed, a typo here.
> Ok. fix it.
> 
> > 
> > > +	if (dev->is_server && dev->vhostfd < 0)
> > > +		status = 0;
> > 
> > When the connection to the backend isn't established in server mode, what
> > virtio_user_handle_mq() can't do is to enable the queue pairs. But other
> > checks in that function are still valid. So I would suggest moving above code
> > to virtio_user_handle_mq().
> Ok. 
> 
> Thanks
> Zhiyong