From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail.droids-corp.org (zoll.droids-corp.org [94.23.50.67]) by dpdk.org (Postfix) with ESMTP id CA6E61B343; Thu, 18 Jan 2018 16:57:01 +0100 (CET) Received: from lfbn-lil-1-110-231.w90-45.abo.wanadoo.fr ([90.45.197.231] helo=droids-corp.org) by mail.droids-corp.org with esmtpsa (TLS1.0:RSA_AES_256_CBC_SHA1:256) (Exim 4.89) (envelope-from ) id 1ecCYn-0007wa-A6; Thu, 18 Jan 2018 16:57:02 +0100 Received: by droids-corp.org (sSMTP sendmail emulation); Thu, 18 Jan 2018 16:56:55 +0100 Date: Thu, 18 Jan 2018 16:56:55 +0100 From: Olivier Matz To: Tiwei Bie Cc: dev@dpdk.org, Yuanhan Liu , Maxime Coquelin , stable@dpdk.org Message-ID: <20180118155655.xhkikpq6d4ymuz4k@platinum> References: <20180118090733.12728-1-olivier.matz@6wind.com> <20180118090733.12728-3-olivier.matz@6wind.com> <20180118140549.tnqzaytw36grtaff@debian-xvivbkq.sh.intel.com> <20180118145553.4sttsn7l3dcdmvfw@platinum> <20180118154809.lkc7cjbagg3pg6k2@debian-xvivbkq.sh.intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20180118154809.lkc7cjbagg3pg6k2@debian-xvivbkq.sh.intel.com> User-Agent: NeoMutt/20170113 (1.7.2) Subject: Re: [dpdk-dev] [PATCH 2/3] net/virtio: rationalize queue flushing X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 18 Jan 2018 15:57:01 -0000 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 > > > > --- > > > > 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.