From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id B8888A0A02; Thu, 15 Apr 2021 04:02:57 +0200 (CEST) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 3CE56161EEE; Thu, 15 Apr 2021 04:02:57 +0200 (CEST) Received: from mga07.intel.com (mga07.intel.com [134.134.136.100]) by mails.dpdk.org (Postfix) with ESMTP id 03DE3161EE9 for ; Thu, 15 Apr 2021 04:02:54 +0200 (CEST) IronPort-SDR: 6crPmQV3rrdhZv+GFoEWa8l4YARBYoS2qcZfaSOaMe6JDvMq696d7YuxrUMvUifY/sYgG3Gdhx rftaDs2eV3zg== X-IronPort-AV: E=McAfee;i="6200,9189,9954"; a="258736870" X-IronPort-AV: E=Sophos;i="5.82,223,1613462400"; d="scan'208";a="258736870" Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by orsmga105.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 14 Apr 2021 19:02:53 -0700 IronPort-SDR: G5BJhAcu13m+lwImbUHS1sT4+DKbm+9fiqeyhPqTc8l0wwG1s+t4+XOYmOz6aaFac7bxisVWT+ q7kM621BWFwg== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.82,223,1613462400"; d="scan'208";a="452721494" Received: from fmsmsx606.amr.corp.intel.com ([10.18.126.86]) by fmsmga002.fm.intel.com with ESMTP; 14 Apr 2021 19:02:52 -0700 Received: from shsmsx606.ccr.corp.intel.com (10.109.6.216) by fmsmsx606.amr.corp.intel.com (10.18.126.86) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2106.2; Wed, 14 Apr 2021 19:02:51 -0700 Received: from shsmsx606.ccr.corp.intel.com (10.109.6.216) by SHSMSX606.ccr.corp.intel.com (10.109.6.216) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2106.2; Thu, 15 Apr 2021 10:02:49 +0800 Received: from shsmsx606.ccr.corp.intel.com ([10.109.6.216]) by SHSMSX606.ccr.corp.intel.com ([10.109.6.216]) with mapi id 15.01.2106.013; Thu, 15 Apr 2021 10:02:49 +0800 From: "Hu, Jiayu" To: "Jiang, Cheng1" , "maxime.coquelin@redhat.com" , "Xia, Chenbo" CC: "dev@dpdk.org" , "Yang, YvonneX" , "Wang, Yinan" , "Liu, Yong" Thread-Topic: [PATCH v7 2/4] vhost: add support for packed ring in async vhost Thread-Index: AQHXMPcwRVp2gR2nmUSLyBYlTWuWW6q0zOAQ Date: Thu, 15 Apr 2021 02:02:48 +0000 Message-ID: References: <20210317085426.10119-1-Cheng1.jiang@intel.com> <20210414061343.54919-1-Cheng1.jiang@intel.com> <20210414061343.54919-3-Cheng1.jiang@intel.com> In-Reply-To: <20210414061343.54919-3-Cheng1.jiang@intel.com> Accept-Language: en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: dlp-reaction: no-action dlp-version: 11.5.1.3 dlp-product: dlpe-windows x-originating-ip: [10.239.127.36] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Subject: Re: [dpdk-dev] [PATCH v7 2/4] vhost: add support for packed ring in async vhost X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 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 Cheng, > -----Original Message----- > From: Jiang, Cheng1 > Sent: Wednesday, April 14, 2021 2:14 PM > To: maxime.coquelin@redhat.com; Xia, Chenbo > Cc: dev@dpdk.org; Hu, Jiayu ; Yang, YvonneX > ; Wang, Yinan ; Liu, > Yong ; Jiang, Cheng1 > Subject: [PATCH v7 2/4] vhost: add support for packed ring in async vhost >=20 > For now async vhost data path only supports split ring. This patch > enables packed ring in async vhost data path to make async vhost > compatible with virtio 1.1 spec. >=20 > Signed-off-by: Cheng Jiang > --- > lib/librte_vhost/rte_vhost_async.h | 1 + > lib/librte_vhost/vhost.c | 49 ++-- > lib/librte_vhost/vhost.h | 15 +- > lib/librte_vhost/virtio_net.c | 432 +++++++++++++++++++++++++++-- > 4 files changed, 456 insertions(+), 41 deletions(-) >=20 > diff --git a/lib/librte_vhost/rte_vhost_async.h > b/lib/librte_vhost/rte_vhost_async.h > index c855ff875..6faa31f5a 100644 > --- a/lib/librte_vhost/rte_vhost_async.h > +++ b/lib/librte_vhost/rte_vhost_async.h > @@ -89,6 +89,7 @@ struct rte_vhost_async_channel_ops { > struct async_inflight_info { > struct rte_mbuf *mbuf; > uint16_t descs; /* num of descs inflight */ > + uint16_t nr_buffers; /* num of buffers inflight for packed ring */ > }; >=20 > /** > diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c > index a70fe01d8..f509186c6 100644 > --- a/lib/librte_vhost/vhost.c > +++ b/lib/librte_vhost/vhost.c > @@ -338,19 +338,22 @@ cleanup_device(struct virtio_net *dev, int destroy) > } >=20 > static void > -vhost_free_async_mem(struct vhost_virtqueue *vq) > +vhost_free_async_mem(struct virtio_net *dev, struct vhost_virtqueue *vq) > { > - if (vq->async_pkts_info) > - rte_free(vq->async_pkts_info); > - if (vq->async_descs_split) > + rte_free(vq->async_pkts_info); > + > + if (vq_is_packed(dev)) { > + rte_free(vq->async_buffers_packed); > + vq->async_buffers_packed =3D NULL; > + } else { > rte_free(vq->async_descs_split); > - if (vq->it_pool) > - rte_free(vq->it_pool); > - if (vq->vec_pool) > - rte_free(vq->vec_pool); > + vq->async_descs_split =3D NULL; > + } > + > + rte_free(vq->it_pool); > + rte_free(vq->vec_pool); >=20 > vq->async_pkts_info =3D NULL; > - vq->async_descs_split =3D NULL; > vq->it_pool =3D NULL; > vq->vec_pool =3D NULL; > } > @@ -360,10 +363,10 @@ free_vq(struct virtio_net *dev, struct > vhost_virtqueue *vq) > { > if (vq_is_packed(dev)) > rte_free(vq->shadow_used_packed); > - else { > + else > rte_free(vq->shadow_used_split); > - vhost_free_async_mem(vq); > - } > + > + vhost_free_async_mem(dev, vq); > rte_free(vq->batch_copy_elems); > if (vq->iotlb_pool) > rte_mempool_free(vq->iotlb_pool); > @@ -1626,10 +1629,9 @@ int rte_vhost_async_channel_register(int vid, > uint16_t queue_id, > if (unlikely(vq =3D=3D NULL || !dev->async_copy)) > return -1; >=20 > - /* packed queue is not supported */ > - if (unlikely(vq_is_packed(dev) || !f.async_inorder)) { > + if (unlikely(!f.async_inorder)) { > VHOST_LOG_CONFIG(ERR, > - "async copy is not supported on packed queue or > non-inorder mode " > + "async copy is not supported on non-inorder mode " > "(vid %d, qid: %d)\n", vid, queue_id); > return -1; > } > @@ -1667,12 +1669,19 @@ int rte_vhost_async_channel_register(int vid, > uint16_t queue_id, > vq->vec_pool =3D rte_malloc_socket(NULL, > VHOST_MAX_ASYNC_VEC * sizeof(struct iovec), > RTE_CACHE_LINE_SIZE, node); > - vq->async_descs_split =3D rte_malloc_socket(NULL, > + if (vq_is_packed(dev)) { > + vq->async_buffers_packed =3D rte_malloc_socket(NULL, > + vq->size * sizeof(struct vring_used_elem_packed), > + RTE_CACHE_LINE_SIZE, node); > + } else { > + vq->async_descs_split =3D rte_malloc_socket(NULL, > vq->size * sizeof(struct vring_used_elem), > RTE_CACHE_LINE_SIZE, node); > - if (!vq->async_descs_split || !vq->async_pkts_info || > - !vq->it_pool || !vq->vec_pool) { > - vhost_free_async_mem(vq); > + } > + > + if (!vq->async_buffers_packed || !vq->async_descs_split || async_buffers_packed and async_descs_split are two members of a union. Like the way processed in vhost_free_async_mem(), do you think it's better to check if they are NULL in if-else respectively? > + !vq->async_pkts_info || !vq->it_pool || !vq->vec_pool) { > + vhost_free_async_mem(dev, vq); > VHOST_LOG_CONFIG(ERR, > "async register failed: cannot allocate > memory for vq data " > "(vid %d, qid: %d)\n", vid, queue_id); > @@ -1728,7 +1737,7 @@ int rte_vhost_async_channel_unregister(int vid, > uint16_t queue_id) > goto out; > } >=20 > - vhost_free_async_mem(vq); > + vhost_free_async_mem(dev, vq); >=20 > vq->async_ops.transfer_data =3D NULL; > vq->async_ops.check_completed_copies =3D NULL; > diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h > index f628714c2..673335217 100644 > --- a/lib/librte_vhost/vhost.h > +++ b/lib/librte_vhost/vhost.h > @@ -201,9 +201,18 @@ struct vhost_virtqueue { > uint16_t async_pkts_idx; > uint16_t async_pkts_inflight_n; > uint16_t async_last_pkts_n; > - struct vring_used_elem *async_descs_split; > - uint16_t async_desc_idx; > - uint16_t last_async_desc_idx; > + union { > + struct vring_used_elem *async_descs_split; > + struct vring_used_elem_packed *async_buffers_packed; > + }; > + union { > + uint16_t async_desc_idx; > + uint16_t async_packed_buffer_idx; > + }; > + union { > + uint16_t last_async_desc_idx; > + uint16_t last_async_buffer_idx; > + }; >=20 > /* vq async features */ > bool async_inorder; > diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.= c > index 438bdafd1..54e11e3a5 100644 > --- a/lib/librte_vhost/virtio_net.c > +++ b/lib/librte_vhost/virtio_net.c > @@ -363,14 +363,14 @@ > vhost_shadow_dequeue_single_packed_inorder(struct vhost_virtqueue *vq, > } >=20 > static __rte_always_inline void > -vhost_shadow_enqueue_single_packed(struct virtio_net *dev, > - struct vhost_virtqueue *vq, > - uint32_t len[], > - uint16_t id[], > - uint16_t count[], > +vhost_shadow_enqueue_packed(struct vhost_virtqueue *vq, > + uint32_t *len, > + uint16_t *id, > + uint16_t *count, > uint16_t num_buffers) > { > uint16_t i; > + > for (i =3D 0; i < num_buffers; i++) { > /* enqueue shadow flush action aligned with batch num */ > if (!vq->shadow_used_idx) > @@ -382,6 +382,17 @@ vhost_shadow_enqueue_single_packed(struct > virtio_net *dev, > vq->shadow_aligned_idx +=3D count[i]; > vq->shadow_used_idx++; > } > +} > + > +static __rte_always_inline void > +vhost_shadow_enqueue_single_packed(struct virtio_net *dev, > + struct vhost_virtqueue *vq, > + uint32_t *len, > + uint16_t *id, > + uint16_t *count, > + uint16_t num_buffers) > +{ > + vhost_shadow_enqueue_packed(vq, len, id, count, num_buffers); >=20 > if (vq->shadow_aligned_idx >=3D PACKED_BATCH_SIZE) { > do_data_copy_enqueue(dev, vq); > @@ -1474,6 +1485,23 @@ store_dma_desc_info_split(struct > vring_used_elem *s_ring, struct vring_used_elem > } > } >=20 > +static __rte_always_inline void > +store_dma_desc_info_packed(struct vring_used_elem_packed *s_ring, > + struct vring_used_elem_packed *d_ring, > + uint16_t ring_size, uint16_t s_idx, uint16_t d_idx, uint16_t > count) > +{ > + uint16_t elem_size =3D sizeof(struct vring_used_elem_packed); > + > + if (d_idx + count <=3D ring_size) { > + rte_memcpy(d_ring + d_idx, s_ring + s_idx, count * > elem_size); > + } else { > + uint16_t size =3D ring_size - d_idx; > + > + rte_memcpy(d_ring + d_idx, s_ring + s_idx, size * elem_size); > + rte_memcpy(d_ring, s_ring + s_idx + size, (count - size) * > elem_size); > + } > +} > + > static __rte_noinline uint32_t > virtio_dev_rx_async_submit_split(struct virtio_net *dev, > struct vhost_virtqueue *vq, uint16_t queue_id, > @@ -1641,6 +1669,330 @@ virtio_dev_rx_async_submit_split(struct > virtio_net *dev, > return pkt_idx; > } >=20 > +static __rte_always_inline void > +vhost_update_used_packed(struct vhost_virtqueue *vq, > + struct vring_used_elem_packed *shadow_ring, > + uint16_t count) > +{ > + int i; > + uint16_t used_idx =3D vq->last_used_idx; > + uint16_t head_idx =3D vq->last_used_idx; > + uint16_t head_flags =3D 0; > + > + if (count =3D=3D 0) > + return; > + > + /* Split loop in two to save memory barriers */ > + for (i =3D 0; i < count; i++) { > + vq->desc_packed[used_idx].id =3D shadow_ring[i].id; > + vq->desc_packed[used_idx].len =3D shadow_ring[i].len; > + > + used_idx +=3D shadow_ring[i].count; > + if (used_idx >=3D vq->size) > + used_idx -=3D vq->size; > + } > + > + /* The ordering for storing desc flags needs to be enforced. */ > + rte_atomic_thread_fence(__ATOMIC_RELEASE); > + > + for (i =3D 0; i < count; i++) { > + uint16_t flags; > + > + if (vq->shadow_used_packed[i].len) > + flags =3D VRING_DESC_F_WRITE; > + else > + flags =3D 0; > + > + if (vq->used_wrap_counter) { > + flags |=3D VRING_DESC_F_USED; > + flags |=3D VRING_DESC_F_AVAIL; > + } else { > + flags &=3D ~VRING_DESC_F_USED; > + flags &=3D ~VRING_DESC_F_AVAIL; > + } > + > + if (i > 0) { > + vq->desc_packed[vq->last_used_idx].flags =3D flags; > + No need a blank line above. > + } else { > + head_idx =3D vq->last_used_idx; > + head_flags =3D flags; > + } > + > + vq_inc_last_used_packed(vq, shadow_ring[i].count); > + } > + > + vq->desc_packed[head_idx].flags =3D head_flags; > +} > + > +static __rte_always_inline int > +vhost_enqueue_async_single_packed(struct virtio_net *dev, > + struct vhost_virtqueue *vq, > + struct rte_mbuf *pkt, > + struct buf_vector *buf_vec, > + uint16_t *nr_descs, > + uint16_t *nr_buffers, > + struct vring_packed_desc *async_descs, > + struct iovec *src_iovec, struct iovec *dst_iovec, > + struct rte_vhost_iov_iter *src_it, > + struct rte_vhost_iov_iter *dst_it) > +{ > + uint16_t nr_vec =3D 0; > + uint16_t avail_idx =3D vq->last_avail_idx; > + uint16_t max_tries, tries =3D 0; > + uint16_t buf_id =3D 0; > + uint32_t len =3D 0; > + uint16_t desc_count =3D 0; > + uint32_t size =3D pkt->pkt_len + sizeof(struct > virtio_net_hdr_mrg_rxbuf); > + uint32_t buffer_len[vq->size]; > + uint16_t buffer_buf_id[vq->size]; > + uint16_t buffer_desc_count[vq->size]; > + *nr_buffers =3D 0; > + > + if (rxvq_is_mergeable(dev)) > + max_tries =3D vq->size - 1; > + else > + max_tries =3D 1; > + > + while (size > 0) { > + /* > + * if we tried all available ring items, and still > + * can't get enough buf, it means something abnormal > + * happened. > + */ > + if (unlikely(++tries > max_tries)) > + return -1; > + > + if (unlikely(fill_vec_buf_packed(dev, vq, avail_idx, > &desc_count, buf_vec, &nr_vec, > + &buf_id, &len, > VHOST_ACCESS_RW) < 0)) > + return -1; > + > + len =3D RTE_MIN(len, size); > + size -=3D len; > + > + buffer_len[*nr_buffers] =3D len; > + buffer_buf_id[*nr_buffers] =3D buf_id; > + buffer_desc_count[*nr_buffers] =3D desc_count; > + *nr_buffers +=3D 1; > + > + *nr_descs +=3D desc_count; > + avail_idx +=3D desc_count; > + if (avail_idx >=3D vq->size) > + avail_idx -=3D vq->size; > + } > + > + if (async_mbuf_to_desc(dev, vq, pkt, buf_vec, nr_vec, *nr_buffers, > src_iovec, dst_iovec, > + src_it, dst_it) < 0) > + return -1; > + /* store descriptors for DMA */ > + if (avail_idx >=3D *nr_descs) { > + rte_memcpy(async_descs, &vq->desc_packed[vq- > >last_avail_idx], > + *nr_descs * sizeof(struct vring_packed_desc)); > + } else { > + uint16_t nr_copy =3D vq->size - vq->last_avail_idx; > + rte_memcpy(async_descs, &vq->desc_packed[vq- > >last_avail_idx], > + nr_copy * sizeof(struct vring_packed_desc)); > + rte_memcpy(async_descs + nr_copy, vq->desc_packed, > + (*nr_descs - nr_copy) * sizeof(struct > vring_packed_desc)); > + } > + > + vhost_shadow_enqueue_packed(vq, buffer_len, buffer_buf_id, > buffer_desc_count, *nr_buffers); > + > + return 0; > +} > + > +static __rte_always_inline int16_t > +virtio_dev_rx_async_single_packed(struct virtio_net *dev, struct > vhost_virtqueue *vq, > + struct rte_mbuf *pkt, uint16_t *nr_descs, uint16_t > *nr_buffers, > + struct vring_packed_desc *async_descs, > + struct iovec *src_iovec, struct iovec *dst_iovec, > + struct rte_vhost_iov_iter *src_it, struct > rte_vhost_iov_iter *dst_it) > +{ > + struct buf_vector buf_vec[BUF_VECTOR_MAX]; > + *nr_descs =3D 0; > + *nr_buffers =3D 0; > + > + if (unlikely(vhost_enqueue_async_single_packed(dev, vq, pkt, > buf_vec, nr_descs, nr_buffers, > + async_descs, src_iovec, > dst_iovec, > + src_it, dst_it) < 0)) { > + VHOST_LOG_DATA(DEBUG, "(%d) failed to get enough desc > from vring\n", dev->vid); > + return -1; > + } > + > + VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end > index %d\n", > + dev->vid, vq->last_avail_idx, vq->last_avail_idx + > *nr_descs); > + > + return 0; > +} > + > +static __rte_always_inline void > +dma_error_handler_packed(struct vhost_virtqueue *vq, struct > vring_packed_desc *async_descs, > + uint16_t async_descs_idx, uint16_t slot_idx, uint32_t > nr_err, > + uint32_t *pkt_idx, uint32_t *num_async_pkts, > uint32_t *num_done_pkts) > +{ > + uint16_t descs_err =3D 0; > + uint16_t buffers_err =3D 0; > + struct async_inflight_info *pkts_info =3D vq->async_pkts_info; > + > + *num_async_pkts -=3D nr_err; > + *pkt_idx -=3D nr_err; > + /* calculate the sum of buffers and descs of DMA-error packets. */ > + while (nr_err-- > 0) { > + descs_err +=3D pkts_info[slot_idx % vq->size].descs; I notice there are several parts using "%" to wrap around index, but existed code uses "& (vq->size - 1)" instead. I think it's better to keep it consistent. Thanks, Jiayu