From mboxrd@z Thu Jan  1 00:00:00 1970
From: Olivier Matz <olivier.matz@6wind.com>
Subject: Re: [PATCH v2 1/4] net/virtio: fix queue flushing with
 vector Rx enabled
Date: Mon, 22 Jan 2018 11:38:58 +0100
Message-ID: <20180122103858.n76gdigdmqeiwlik@platinum>
References: <20180118090733.12728-1-olivier.matz@6wind.com>
 <20180119155556.32597-1-olivier.matz@6wind.com>
 <20180119155556.32597-2-olivier.matz@6wind.com>
 <20180122025641.t4vgozvbjbms3edn@debian-xvivbkq.sh.intel.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: dev@dpdk.org, Yuanhan Liu <yliu@fridaylinux.org>,
 Maxime Coquelin <maxime.coquelin@redhat.com>, stable@dpdk.org
To: Tiwei Bie <tiwei.bie@intel.com>
Return-path: <dev-bounces@dpdk.org>
Content-Disposition: inline
In-Reply-To: <20180122025641.t4vgozvbjbms3edn@debian-xvivbkq.sh.intel.com>
List-Id: DPDK patches and discussions <dev.dpdk.org>
List-Unsubscribe: <https://dpdk.org/ml/options/dev>,
 <mailto:dev-request@dpdk.org?subject=unsubscribe>
List-Archive: <http://dpdk.org/ml/archives/dev/>
List-Post: <mailto:dev@dpdk.org>
List-Help: <mailto:dev-request@dpdk.org?subject=help>
List-Subscribe: <https://dpdk.org/ml/listinfo/dev>,
 <mailto:dev-request@dpdk.org?subject=subscribe>
Errors-To: dev-bounces@dpdk.org
Sender: "dev" <dev-bounces@dpdk.org>

On Mon, Jan 22, 2018 at 10:56:41AM +0800, Tiwei Bie wrote:
> On Fri, Jan 19, 2018 at 04:55:53PM +0100, Olivier Matz wrote:
> > When using vector Rx mode (use_simple_rx = 1), vq->vq_descx[] is not
> > kept up to date. To properly detach the mbufs in this case, browse
> > sw_ring[] instead, as it's done in virtqueue_rxvq_flush().
> > 
> > Since we need virtio_get_queue_type(), also move this function in
> > virtqueue.h as a static inline.
> > 
> > Fixes: fc3d66212fed ("virtio: add vector Rx")
> > Cc: stable@dpdk.org
> > 
> > Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
> > ---
> > 
> > Tiwei,
> > 
> > While it passes my test plan, please carefully check that what I did in
> > virtqueue_detatch_unused() is correct. I tried to reproduce what is done
> > in virtqueue_rxvq_flush(), but I may be mistaking due to the different
> > ring layout assumption and mbuf management between standard and vector.
> > 
> > Thanks
> > Olivier
> > 
> > 
> >  drivers/net/virtio/virtio_ethdev.c | 11 -----------
> >  drivers/net/virtio/virtqueue.c     | 17 +++++++++++++++--
> >  drivers/net/virtio/virtqueue.h     | 11 +++++++++++
> >  3 files changed, 26 insertions(+), 13 deletions(-)
> > 
> [...]
> >  struct rte_mbuf *
> >  virtqueue_detatch_unused(struct virtqueue *vq)
> >  {
> > +	struct virtio_hw *hw = vq->hw;
> >  	struct rte_mbuf *cookie;
> >  	int idx;
> > +	int type = virtio_get_queue_type(hw, vq->vq_queue_index);
> > +
> > +	if (vq == NULL)
> > +		return NULL;
> >  
> > -	if (vq != NULL)
> > -		for (idx = 0; idx < vq->vq_nentries; idx++) {
> > +	for (idx = 0; idx < vq->vq_nentries; idx++) {
> > +		if (hw->use_simple_rx && type == VTNET_RQ) {
> > +			cookie = vq->sw_ring[idx];
> > +			if (cookie != NULL) {
> > +				vq->sw_ring[idx] = NULL;
> 
> Thanks for working on this!
> 
> The vq->sw_ring[idx] isn't zeroed during Rx. So besides the
> check of (cookie != NULL), some other check is also needed
> to avoid freeing the mbufs already delivered to application.
> 
> The mbufs in below interval belong to application:
> 
> start: sw_ring[vq->vq_avail_idx & (vq->vq_nentries - 1)] (included)
> end: sw_ring[(vq->vq_avail_idx + vq->vq_free_cnt) & (vq->vq_nentries - 1)] (excluded)
> 
> PS. (vq->vq_avail_idx & (vq->vq_nentries - 1)) can be greater than
>     (vq->vq_avail_idx + vq->vq_free_cnt) & (vq->vq_nentries - 1).

Thank you for the review. What about something like this?

struct rte_mbuf *
virtqueue_detach_unused(struct virtqueue *vq)
{
	struct rte_mbuf *cookie;
	struct virtio_hw *hw;
	uint16_t start, end;
	int type, idx;

	if (vq == NULL)
		return NULL;

	hw = vq->hw;
	type = virtio_get_queue_type(hw, vq->vq_queue_index);
	start = vq->vq_avail_idx & (vq->vq_nentries - 1);
	end = (vq->vq_avail_idx + vq->vq_free_cnt) & (vq->vq_nentries - 1);
	end = (vq->vq_avail_idx + vq->vq_free_cnt) & (vq->vq_nentries - 1);

	for (idx = 0; idx < vq->vq_nentries; idx++) {
		if (hw->use_simple_rx && type == VTNET_RQ) {
			if (start <= end && idx >= start && idx < end)
				continue;
			if (start > end && (idx >= start || idx < end))
				continue;
			cookie = vq->sw_ring[idx];
			if (cookie == NULL)
				continue;
			vq->sw_ring[idx] = NULL;
			return cookie;
		} else {
			cookie = vq->vq_descx[idx].cookie;
			if (cookie != NULL) {
				vq->vq_descx[idx].cookie = NULL;
				return cookie;
			}
		}
	}

	return NULL;
}


Regards,
Olivier