From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 7EF35A2EFC for ; Wed, 18 Sep 2019 07:20:51 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id CD9DD1BF7B; Wed, 18 Sep 2019 07:20:50 +0200 (CEST) Received: from mga04.intel.com (mga04.intel.com [192.55.52.120]) by dpdk.org (Postfix) with ESMTP id AFD331BF2E for ; Wed, 18 Sep 2019 07:20:48 +0200 (CEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from orsmga006.jf.intel.com ([10.7.209.51]) by fmsmga104.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 17 Sep 2019 22:20:47 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.64,519,1559545200"; d="scan'208";a="191611539" Received: from fmsmsx105.amr.corp.intel.com ([10.18.124.203]) by orsmga006.jf.intel.com with ESMTP; 17 Sep 2019 22:20:47 -0700 Received: from fmsmsx126.amr.corp.intel.com (10.18.125.43) by FMSMSX105.amr.corp.intel.com (10.18.124.203) with Microsoft SMTP Server (TLS) id 14.3.439.0; Tue, 17 Sep 2019 22:20:46 -0700 Received: from shsmsx106.ccr.corp.intel.com (10.239.4.159) by FMSMSX126.amr.corp.intel.com (10.18.125.43) with Microsoft SMTP Server (TLS) id 14.3.439.0; Tue, 17 Sep 2019 22:20:46 -0700 Received: from shsmsx104.ccr.corp.intel.com ([169.254.5.32]) by SHSMSX106.ccr.corp.intel.com ([169.254.10.86]) with mapi id 14.03.0439.000; Wed, 18 Sep 2019 13:20:44 +0800 From: "Wang, Yinan" To: Joyce Kong , "dev@dpdk.org" CC: "nd@arm.com" , "maxime.coquelin@redhat.com" , "Richardson, Bruce" , "Bie, Tiwei" , "Wang, Zhihong" , "amorenoz@redhat.com" , "Wang, Xiao W" , "Liu, Yong" , "jfreimann@redhat.com" , "honnappa.nagarahalli@arm.com" , "gavin.hu@arm.com" Thread-Topic: [PATCH v4 2/2] virtio: one way barrier for packed vring desc used flags Thread-Index: AQHVbRjcLB4pBw6rI0OWy+JSYu6nqqcw5oUg Date: Wed, 18 Sep 2019 05:20:44 +0000 Message-ID: References: <1566893979-3290-1-git-send-email-joyce.kong@arm.com> <1568698107-36668-3-git-send-email-joyce.kong@arm.com> In-Reply-To: <1568698107-36668-3-git-send-email-joyce.kong@arm.com> Accept-Language: en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-titus-metadata-40: eyJDYXRlZ29yeUxhYmVscyI6IiIsIk1ldGFkYXRhIjp7Im5zIjoiaHR0cDpcL1wvd3d3LnRpdHVzLmNvbVwvbnNcL0ludGVsMyIsImlkIjoiYzQzMGY4ZjQtODAzNi00ODkwLTk1MjktYTFiNzE0N2U5MmIxIiwicHJvcHMiOlt7Im4iOiJDVFBDbGFzc2lmaWNhdGlvbiIsInZhbHMiOlt7InZhbHVlIjoiQ1RQX05UIn1dfV19LCJTdWJqZWN0TGFiZWxzIjpbXSwiVE1DVmVyc2lvbiI6IjE3LjEwLjE4MDQuNDkiLCJUcnVzdGVkTGFiZWxIYXNoIjoiM1lzeDdvdjI2UGN6dkR0cFpXeWFTOXhnR1RmVFlwYmk4dEU0cENFSFRJNXZja3RlUThtMjc3U3ROSUREeHpiMCJ9 x-ctpclassification: CTP_NT dlp-product: dlpe-windows dlp-version: 11.2.0.6 dlp-reaction: no-action x-originating-ip: [10.239.127.40] Content-Type: text/plain; charset="iso-2022-jp" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Subject: Re: [dpdk-dev] [PATCH v4 2/2] virtio: one way barrier for packed vring desc used flags 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: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Hi Joyce, I test performance impact of your patch set with code base commit id: d03d8= 622db48918d14bfe805641b1766ecc40088, after applying your v4 patch set , pac= ked ring shows small performance drop as below: PVP vhost/virtio 1c1q test commit:d03d8622db48918 apply v4 patch set pvp_virtio11_mergeable 7.218 7.147 pvp_virtio11_normal 7.217 7.182 Thanks, Yinan > -----Original Message----- > From: Joyce Kong [mailto:joyce.kong@arm.com] > Sent: 2019=1B$BG/=1B(B9=1B$B7n=1B(B17=1B$BF|=1B(B 13:28 > To: dev@dpdk.org > Cc: nd@arm.com; maxime.coquelin@redhat.com; Wang, Yinan > ; Richardson, Bruce ; B= ie, > Tiwei ; Wang, Zhihong ; > amorenoz@redhat.com; Wang, Xiao W ; Liu, Yong > ; jfreimann@redhat.com; honnappa.nagarahalli@arm.com; > gavin.hu@arm.com > Subject: [PATCH v4 2/2] virtio: one way barrier for packed vring desc use= d flags >=20 > In case VIRTIO_F_ORDER_PLATFORM(36) is not negotiated, then the frontend > and backend are assumed to be implemented in software, that is they can r= un on > identical CPUs in an SMP configuration. > Thus a weak form of memory barriers like rte_smp_r/wmb, other than > rte_cio_r/wmb, is sufficient for this case(vq->hw->weak_barriers =3D=3D 1= ) and > yields better performance. > For the above case, this patch helps yielding even better performance by > replacing the two-way barriers with C11 one-way barriers for used flags i= n packed > ring. >=20 > Signed-off-by: Joyce Kong > Reviewed-by: Gavin Hu > Reviewed-by: Phil Yang > Reviewed-by: Maxime Coquelin > --- > drivers/net/virtio/virtio_rxtx.c | 12 +++++++--- > drivers/net/virtio/virtio_user/virtio_user_dev.c | 4 ++-- > drivers/net/virtio/virtqueue.h | 28 > +++++++++++++++++++++++- > lib/librte_vhost/virtio_net.c | 5 ++--- > 4 files changed, 40 insertions(+), 9 deletions(-) >=20 > diff --git a/drivers/net/virtio/virtio_rxtx.c b/drivers/net/virtio/virtio= _rxtx.c > index a87ffe1..2f0879c 100644 > --- a/drivers/net/virtio/virtio_rxtx.c > +++ b/drivers/net/virtio/virtio_rxtx.c > @@ -122,9 +122,11 @@ virtqueue_dequeue_burst_rx_packed(struct virtqueue > *vq, >=20 > for (i =3D 0; i < num; i++) { > used_idx =3D vq->vq_used_cons_idx; > + /* desc_is_used has a load-acquire or rte_cio_rmb inside > + * and wait for used desc in virtqueue. > + */ > if (!desc_is_used(&desc[used_idx], vq)) > return i; > - virtio_rmb(vq->hw->weak_barriers); > len[i] =3D desc[used_idx].len; > id =3D desc[used_idx].id; > cookie =3D (struct rte_mbuf *)vq->vq_descx[id].cookie; @@ -233,8 > +235,10 @@ virtio_xmit_cleanup_inorder_packed(struct virtqueue *vq, int n= um) > struct vq_desc_extra *dxp; >=20 > used_idx =3D vq->vq_used_cons_idx; > + /* desc_is_used has a load-acquire or rte_cio_rmb inside > + * and wait for used desc in virtqueue. > + */ > while (num > 0 && desc_is_used(&desc[used_idx], vq)) { > - virtio_rmb(vq->hw->weak_barriers); > id =3D desc[used_idx].id; > do { > curr_id =3D used_idx; > @@ -265,8 +269,10 @@ virtio_xmit_cleanup_normal_packed(struct virtqueue > *vq, int num) > struct vq_desc_extra *dxp; >=20 > used_idx =3D vq->vq_used_cons_idx; > + /* desc_is_used has a load-acquire or rte_cio_rmb inside > + * and wait for used desc in virtqueue. > + */ > while (num-- && desc_is_used(&desc[used_idx], vq)) { > - virtio_rmb(vq->hw->weak_barriers); > id =3D desc[used_idx].id; > dxp =3D &vq->vq_descx[id]; > vq->vq_used_cons_idx +=3D dxp->ndescs; > diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.c > b/drivers/net/virtio/virtio_user/virtio_user_dev.c > index 7911c39..1c575d0 100644 > --- a/drivers/net/virtio/virtio_user/virtio_user_dev.c > +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.c > @@ -698,8 +698,8 @@ virtio_user_handle_cq_packed(struct virtio_user_dev > *dev, uint16_t queue_idx) > if (vq->used_wrap_counter) > flags |=3D VRING_PACKED_DESC_F_AVAIL_USED; >=20 > - rte_smp_wmb(); > - vring->desc[vq->used_idx].flags =3D flags; > + __atomic_store_n(&vring->desc[vq->used_idx].flags, flags, > + __ATOMIC_RELEASE); >=20 > vq->used_idx +=3D n_descs; > if (vq->used_idx >=3D dev->queue_size) { diff --git > a/drivers/net/virtio/virtqueue.h b/drivers/net/virtio/virtqueue.h index > b728ff8..8d7f197 100644 > --- a/drivers/net/virtio/virtqueue.h > +++ b/drivers/net/virtio/virtqueue.h > @@ -54,6 +54,32 @@ virtio_wmb(uint8_t weak_barriers) > rte_cio_wmb(); > } >=20 > +static inline uint16_t > +virtqueue_fetch_flags_packed(struct vring_packed_desc *dp, > + uint8_t weak_barriers) > +{ > + uint16_t flags; > + > + if (weak_barriers) { > +/* x86 prefers to using rte_smp_rmb over __atomic_load_n as it reports > + * a better perf(~1.5%), which comes from the saved branch by the compil= er. > + * The if and else branch are identical with the smp and cio barriers > +both > + * defined as compiler barriers on x86. > + */ > +#ifdef RTE_ARCH_X86_64 > + flags =3D dp->flags; > + rte_smp_rmb(); > +#else > + flags =3D __atomic_load_n(&dp->flags, __ATOMIC_ACQUIRE); #endif > + } else { > + flags =3D dp->flags; > + rte_cio_rmb(); > + } > + > + return flags; > +} > + > static inline void > virtqueue_store_flags_packed(struct vring_packed_desc *dp, > uint16_t flags, uint8_t weak_barriers) @@ -307,7 +333,7 > @@ desc_is_used(struct vring_packed_desc *desc, struct virtqueue *vq) { > uint16_t used, avail, flags; >=20 > - flags =3D desc->flags; > + flags =3D virtqueue_fetch_flags_packed(desc, vq->hw->weak_barriers); > used =3D !!(flags & VRING_PACKED_DESC_F_USED); > avail =3D !!(flags & VRING_PACKED_DESC_F_AVAIL); >=20 > diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.= c index > e7463ff..241d467 100644 > --- a/lib/librte_vhost/virtio_net.c > +++ b/lib/librte_vhost/virtio_net.c > @@ -110,8 +110,6 @@ flush_shadow_used_ring_packed(struct virtio_net *dev, > used_idx -=3D vq->size; > } >=20 > - rte_smp_wmb(); > - > for (i =3D 0; i < vq->shadow_used_idx; i++) { > uint16_t flags; >=20 > @@ -147,7 +145,8 @@ flush_shadow_used_ring_packed(struct virtio_net *dev, > } > } >=20 > - vq->desc_packed[head_idx].flags =3D head_flags; > + __atomic_store_n(&vq->desc_packed[head_idx].flags, head_flags, > + __ATOMIC_RELEASE); >=20 > vhost_log_cache_used_vring(dev, vq, > head_idx * > -- > 2.7.4