Re: [dpdk-stable] [PATCH 2/3] net/virtio: rationalize queue flushing

[Olivier Matz <olivier.matz@6wind.com>]

On Thu, Jan 18, 2018 at 11:48:09PM +0800, Tiwei Bie wrote:
> On Thu, Jan 18, 2018 at 03:55:53PM +0100, Olivier Matz wrote:
> > On Thu, Jan 18, 2018 at 10:05:49PM +0800, Tiwei Bie wrote:
> > > On Thu, Jan 18, 2018 at 10:07:32AM +0100, Olivier Matz wrote:
> > > > Rationalize the function virtio_dev_free_mbufs():
> > > > 
> > > > - skip NULL vqs instead of crashing: this is required for the
> > > >   next commit
> > > > - use the same kind of loop than in virtio_free_queues()
> > > > - also flush mbufs from the control queue (this is useless yet)
> > > > - factorize common code between rxq, txq, cq
> > > > 
> > > > Cc: stable@dpdk.org
> > > > 
> > > > Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
> > > > ---
> > > >  drivers/net/virtio/virtio_ethdev.c | 55 ++++++++++++++++++--------------------
> > > >  1 file changed, 26 insertions(+), 29 deletions(-)
> > > > 
> > > > diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
> > > > index c7426951c..d8b3b8ed7 100644
> > > > --- a/drivers/net/virtio/virtio_ethdev.c
> > > > +++ b/drivers/net/virtio/virtio_ethdev.c
> > > > @@ -1868,47 +1868,44 @@ virtio_dev_start(struct rte_eth_dev *dev)
> > > >  
> > > >  static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
> > > >  {
> > > [...]
> > > >  
> > > > -		mbuf_num = 0;
> > > > -		while ((buf = virtqueue_detach_unused(txvq->vq)) != NULL) {
> > > > +		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
> > > 
> > > Thanks for working on this! The virtqueue_detach_unused() can't
> > > handle the vector Rx case correctly. Because vq->vq_descx[] is
> > > initialized for vector Rx, but isn't updated by the vector Rx.
> > > So together with the next commit, it may cause problems during
> > > dev_stop/dev_configure/dev_start if vector Rx is used.
> > 
> > What would be the appropriate way to fix it?
> > 
> > Either we fix the vector functions to set vq->vq_descx, or we find
> > another method to flush the mbufs in case of vector rx. Can we use
> > vq->sw_ring[] to get the mbufs instead?
> 
> The first one may hurt the performance. If not, it would be the
> best choice IMO. For the second one and your question, there is
> some related code in virtqueue_rxvq_flush() you could refer to:
> 
> https://github.com/DPDK/dpdk/blob/e00093f381a1/drivers/net/virtio/virtqueue.c#L51

Yes, I've just seen it too :)
I'll give it a try.