From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga11.intel.com (mga11.intel.com [192.55.52.93]) by dpdk.org (Postfix) with ESMTP id 70F59591F for ; Thu, 6 Oct 2016 16:37:32 +0200 (CEST) Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by fmsmga102.fm.intel.com with ESMTP; 06 Oct 2016 07:37:32 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.31,454,1473145200"; d="scan'208";a="1066970631" Received: from fmsmsx104.amr.corp.intel.com ([10.18.124.202]) by fmsmga002.fm.intel.com with ESMTP; 06 Oct 2016 07:37:32 -0700 Received: from fmsmsx121.amr.corp.intel.com (10.18.125.36) by fmsmsx104.amr.corp.intel.com (10.18.124.202) with Microsoft SMTP Server (TLS) id 14.3.248.2; Thu, 6 Oct 2016 07:37:31 -0700 Received: from shsmsx152.ccr.corp.intel.com (10.239.6.52) by fmsmsx121.amr.corp.intel.com (10.18.125.36) with Microsoft SMTP Server (TLS) id 14.3.248.2; Thu, 6 Oct 2016 07:37:31 -0700 Received: from shsmsx102.ccr.corp.intel.com ([169.254.2.15]) by SHSMSX152.ccr.corp.intel.com ([169.254.6.95]) with mapi id 14.03.0248.002; Thu, 6 Oct 2016 22:37:29 +0800 From: "Xu, Qian Q" To: Yuanhan Liu , "dev@dpdk.org" CC: Maxime Coquelin Thread-Topic: [dpdk-dev] [PATCH v2 4/7] vhost: add dequeue zero copy Thread-Index: AQHSFVDqzmgwdw6bVEGvN7mzPH4lkqCbkgeg Date: Thu, 6 Oct 2016 14:37:27 +0000 Message-ID: <82F45D86ADE5454A95A89742C8D1410E391FCDA5@shsmsx102.ccr.corp.intel.com> References: <1471939839-29778-1-git-send-email-yuanhan.liu@linux.intel.com> <1474604007-5221-1-git-send-email-yuanhan.liu@linux.intel.com> <1474604007-5221-5-git-send-email-yuanhan.liu@linux.intel.com> In-Reply-To: <1474604007-5221-5-git-send-email-yuanhan.liu@linux.intel.com> Accept-Language: en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [10.239.127.40] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Subject: Re: [dpdk-dev] [PATCH v2 4/7] vhost: add dequeue zero copy X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 06 Oct 2016 14:37:33 -0000 See the bottom.=20 -----Original Message----- From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Yuanhan Liu Sent: Friday, September 23, 2016 5:13 AM To: dev@dpdk.org Cc: Maxime Coquelin ; Yuanhan Liu Subject: [dpdk-dev] [PATCH v2 4/7] vhost: add dequeue zero copy The basic idea of dequeue zero copy is, instead of copying data from the de= sc buf, here we let the mbuf reference the desc buf addr directly. Doing so, however, has one major issue: we can't update the used ring at th= e end of rte_vhost_dequeue_burst. Because we don't do the copy here, an upd= ate of the used ring would let the driver to reclaim the desc buf. As a res= ult, DPDK might reference a stale memory region. To update the used ring properly, this patch does several tricks: - when mbuf references a desc buf, refcnt is added by 1. This is to pin lock the mbuf, so that a mbuf free from the DPDK won't actually free it, instead, refcnt is subtracted by 1. - We chain all those mbuf together (by tailq) And we check it every time on the rte_vhost_dequeue_burst entrance, to see if the mbuf is freed (when refcnt equals to 1). If that happens, it means we are the last user of this mbuf and we are safe to update the used ring. - "struct zcopy_mbuf" is introduced, to associate an mbuf with the right desc idx. Dequeue zero copy is introduced for performance reason, and some rough test= s show about 50% perfomance boost for packet size 1500B. For small packets,= (e.g. 64B), it actually slows a bit down (well, it could up to 15%). That = is expected because this patch introduces some extra works, and it outweigh= s the benefit from saving few bytes copy. Signed-off-by: Yuanhan Liu --- v2: - use unlikely/likely for dequeue_zero_copy check, as it's not enabled by default, as well as it has some limitations in vm2nic case. - handle the case that a desc buf might across 2 host phys pages - reset nr_zmbuf to 0 at set_vring_num - set the zmbuf_size to vq->size, but not the double of it. --- lib/librte_vhost/vhost.c | 2 + lib/librte_vhost/vhost.h | 22 +++++- lib/librte_vhost/vhost_user.c | 42 +++++++++- lib/librte_vhost/virtio_ne= t.c | 173 +++++++++++++++++++++++++++++++++++++----- 4 files changed, 219 insertions(+), 20 deletions(-) diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c index 4609= 5c3..ab25649 100644 --- a/lib/librte_vhost/vhost.c +++ b/lib/librte_vhost/vhost.c @@ -141,6 +141,8 @@ init_vring_queue(struct vhost_virtqueue *vq, int qp_idx= ) /* always set the default vq pair to enabled */ if (qp_idx =3D=3D 0) vq->enabled =3D 1; + + TAILQ_INIT(&vq->zmbuf_list); } =20 static void diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h index 8565= fa1..be8a398 100644 --- a/lib/librte_vhost/vhost.h +++ b/lib/librte_vhost/vhost.h @@ -36,6 +36,7 @@ #include #include #include +#include #include #include =20 @@ -61,6 +62,19 @@ struct buf_vector { uint32_t desc_idx; }; =20 +/* + * A structure to hold some fields needed in zero copy code path, + * mainly for associating an mbuf with the right desc_idx. + */ +struct zcopy_mbuf { + struct rte_mbuf *mbuf; + uint32_t desc_idx; + uint16_t in_use; + + TAILQ_ENTRY(zcopy_mbuf) next; +}; +TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf); + /** * Structure contains variables relevant to RX/TX virtqueues. */ @@ -85,6 +99,12 @@ struct vhost_virtqueue { =20 /* Physical address of used ring, for logging */ uint64_t log_guest_addr; + + uint16_t nr_zmbuf; + uint16_t zmbuf_size; + uint16_t last_zmbuf_idx; + struct zcopy_mbuf *zmbufs; + struct zcopy_mbuf_list zmbuf_list; } __rte_cache_aligned; =20 /* Old kernels have no such macro defined */ @@ -135,6 +155,7 @@ struct vi= rtio_net { /* to tell if we need broadcast rarp packet */ rte_atomic16_t broadcast_rarp; uint32_t virt_qp_nb; + int dequeue_zero_copy; struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2]; #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) char ifname[IF_NAME_SZ]; @@ -146,7 +167,6 @@ struct virtio_net { uint32_t nr_guest_pages; uint32_t max_guest_pages; struct guest_page *guest_pages; - } __rte_cache_aligned; =20 /** diff --git a/lib/librte_vhost/vhost_user.c b/lib/librte_vhost/vhost_user.c = index a92377a..ac40408 100644 --- a/lib/librte_vhost/vhost_user.c +++ b/lib/librte_vhost/vhost_user.c @@ -180,7 +180,23 @@ static int vhost_user_set_vring_num(struct virtio_net *dev, struct vhost_vring_state *state) { - dev->virtqueue[state->index]->size =3D state->num; + struct vhost_virtqueue *vq =3D dev->virtqueue[state->index]; + + vq->size =3D state->num; + + if (dev->dequeue_zero_copy) { + vq->nr_zmbuf =3D 0; + vq->last_zmbuf_idx =3D 0; + vq->zmbuf_size =3D vq->size; + vq->zmbufs =3D rte_zmalloc(NULL, vq->zmbuf_size * + sizeof(struct zcopy_mbuf), 0); + if (vq->zmbufs =3D=3D NULL) { + RTE_LOG(WARNING, VHOST_CONFIG, + "failed to allocate mem for zero copy; " + "zero copy is force disabled\n"); + dev->dequeue_zero_copy =3D 0; + } + } =20 return 0; } @@ -662,11 +678,32 @@ vhost_user_set_vring_kick(struct virtio_net *dev, str= uct VhostUserMsg *pmsg) vq->kickfd =3D file.fd; =20 if (virtio_is_ready(dev) && !(dev->flags & VIRTIO_DEV_RUNNING)) { + if (dev->dequeue_zero_copy) { + RTE_LOG(INFO, VHOST_CONFIG, + "Tx zero copy is enabled\n"); + } + if (notify_ops->new_device(dev->vid) =3D=3D 0) dev->flags |=3D VIRTIO_DEV_RUNNING; } } =20 +static void +free_zmbufs(struct vhost_virtqueue *vq) { + struct zcopy_mbuf *zmbuf, *next; + + for (zmbuf =3D TAILQ_FIRST(&vq->zmbuf_list); + zmbuf !=3D NULL; zmbuf =3D next) { + next =3D TAILQ_NEXT(zmbuf, next); + + rte_pktmbuf_free(zmbuf->mbuf); + TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next); + } + + rte_free(vq->zmbufs); +} + /* * when virtio is stopped, qemu will send us the GET_VRING_BASE message. */ @@ -695,6 +732,9 @@ vhost_user_get_vring_base(struct virtio_net *dev, =20 dev->virtqueue[state->index]->kickfd =3D VIRTIO_UNINITIALIZED_EVENTFD; =20 + if (dev->dequeue_zero_copy) + free_zmbufs(dev->virtqueue[state->index]); + return 0; } =20 diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c = index 1c2ee47..215542c 100644 --- a/lib/librte_vhost/virtio_net.c +++ b/lib/librte_vhost/virtio_net.c @@ -678,6 +678,43 @@ make_rarp_packet(struct rte_mbuf *rarp_mbuf, const str= uct ether_addr *mac) return 0; } =20 +static inline struct zcopy_mbuf *__attribute__((always_inline))=20 +get_zmbuf(struct vhost_virtqueue *vq) { + uint16_t i; + uint16_t last; + int tries =3D 0; + + /* search [last_zmbuf_idx, zmbuf_size) */ + i =3D vq->last_zmbuf_idx; + last =3D vq->zmbuf_size; + +again: + for (; i < last; i++) { + if (vq->zmbufs[i].in_use =3D=3D 0) { + vq->last_zmbuf_idx =3D i + 1; + vq->zmbufs[i].in_use =3D 1; + return &vq->zmbufs[i]; + } + } + + tries++; + if (tries =3D=3D 1) { + /* search [0, last_zmbuf_idx) */ + i =3D 0; + last =3D vq->last_zmbuf_idx; + goto again; + } + + return NULL; +} + +static inline void __attribute__((always_inline)) put_zmbuf(struct=20 +zcopy_mbuf *zmbuf) { + zmbuf->in_use =3D 0; +} + static inline int __attribute__((always_inline)) copy_desc_to_mbuf(struct= virtio_net *dev, struct vhost_virtqueue *vq, struct rte_mbuf *m, uint16_t desc_idx, @@ -701,6 +738,27 @@ copy_desc_= to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq, if (unlikely(!desc_addr)) return -1; =20 + if (unlikely(dev->dequeue_zero_copy)) { + struct zcopy_mbuf *zmbuf; + + zmbuf =3D get_zmbuf(vq); + if (!zmbuf) + return -1; + zmbuf->mbuf =3D m; + zmbuf->desc_idx =3D desc_idx; + + /* + * Pin lock the mbuf; we will check later to see whether + * the mbuf is freed (when we are the last user) or not. + * If that's the case, we then could update the used ring + * safely. + */ + rte_mbuf_refcnt_update(m, 1); + + vq->nr_zmbuf +=3D 1; + TAILQ_INSERT_TAIL(&vq->zmbuf_list, zmbuf, next); + } + hdr =3D (struct virtio_net_hdr *)((uintptr_t)desc_addr); rte_prefetch0(hdr); =20 this function copy_desc_to_mbuf has changed on the dpdk-next-virtio repo. B= ased on current dpdk-next-virtio repo, the commit ID is as below:=20 commit b4f7b43cd9d3b6413f41221051d03a23bc5f5fbe Author: Zhiyong Yang Date: Thu Sep 29 20:35:49 2016 +0800 Then you will find the parameter "struct vhost_virtqueue *vq" is removed, s= o if apply your patch on that commit ID, the build will fail, since no vq d= efinition but we used it in the function.=20 Could you check? Thx.=20 =3D=3D Build lib/librte_table /home/qxu10/dpdk-zero/lib/librte_vhost/virtio_net.c: In function 'copy_desc= _to_mbuf': /home/qxu10/dpdk-zero/lib/librte_vhost/virtio_net.c:745:21: error: 'vq' und= eclared (first use in this function) zmbuf =3D get_zmbuf(vq); ^ /home/qxu10/dpdk-zero/lib/librte_vhost/virtio_net.c:745:21: note: each unde= clared identifier is reported only once for each function it appears in /home/qxu10/dpdk-zero/mk/internal/rte.compile-pre.mk:138: recipe for target= 'virtio_net.o' failed make[5]: *** [virtio_net.o] Error 1 /home/qxu10/dpdk-zero/mk/rte.subdir.mk:61: recipe for target 'librte_vhost'= failed make[4]: *** [librte_vhost] Error 2 make[4]: *** Waiting for unfinished jobs....