From patchwork Fri Dec 8 05:31:20 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Srujana Challa X-Patchwork-Id: 134945 X-Patchwork-Delegate: maxime.coquelin@redhat.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 74A89436A4; Fri, 8 Dec 2023 06:31:29 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id BFAAB42FAD; Fri, 8 Dec 2023 06:31:28 +0100 (CET) Received: from mx0b-0016f401.pphosted.com (mx0b-0016f401.pphosted.com [67.231.156.173]) by mails.dpdk.org (Postfix) with ESMTP id 10DDB402CD for ; Fri, 8 Dec 2023 06:31:27 +0100 (CET) Received: from pps.filterd (m0045851.ppops.net [127.0.0.1]) by mx0b-0016f401.pphosted.com (8.17.1.19/8.17.1.19) with ESMTP id 3B81aM0s028741; Thu, 7 Dec 2023 21:31:27 -0800 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=marvell.com; h=from : to : cc : subject : date : message-id : mime-version : content-transfer-encoding : content-type; s=pfpt0220; bh=d5E5FkqXLBtD25s2+JCsnAhhcKeCEboJNHc+5PSvwUQ=; b=er3jEchfl1ecUvW1JF9ZsXal+gSaVt9ZSqpZYMoina+i4wUQr41+xuS10KOtw603q5GX Ufc75D0+cN8ouhXCugoZ0Xh+syMRflmNtsSU1y15QfBOFfHZqvWrkHPshfHRlnvtuF7g h9MzB0QH/PGDe2Fa7BEhGTyrgF13OGm7eE3fSIT6/XkU6WdMf9SWmLyAC6rk+VBkndGW aeN/KUW8KObeg9Ajljb+tr30+o1Gr05HL3T81YZKa3+WC73QtZzSubRl2isWKXLGx/gW TKQmLa5JobEUqmTGGxt1ymocgk0hrDHyTGjPM6jcW6nSAfcVCLMLFN0gxMbp87ul8JnJ 8w== Received: from dc5-exch02.marvell.com ([199.233.59.182]) by mx0b-0016f401.pphosted.com (PPS) with ESMTPS id 3uubddbscx-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-SHA384 bits=256 verify=NOT); Thu, 07 Dec 2023 21:31:26 -0800 Received: from DC5-EXCH01.marvell.com (10.69.176.38) by DC5-EXCH02.marvell.com (10.69.176.39) with Microsoft SMTP Server (TLS) id 15.0.1497.48; Thu, 7 Dec 2023 21:31:24 -0800 Received: from maili.marvell.com (10.69.176.80) by DC5-EXCH01.marvell.com (10.69.176.38) with Microsoft SMTP Server id 15.0.1497.48 via Frontend Transport; Thu, 7 Dec 2023 21:31:24 -0800 Received: from localhost.localdomain (unknown [10.28.36.175]) by maili.marvell.com (Postfix) with ESMTP id 7119F3F7050; Thu, 7 Dec 2023 21:31:22 -0800 (PST) From: Srujana Challa To: , , CC: , Subject: [PATCH 1/2] net/virtio-user: improve kick performance with notification area mapping Date: Fri, 8 Dec 2023 11:01:20 +0530 Message-ID: <20231208053121.152929-1-schalla@marvell.com> X-Mailer: git-send-email 2.25.1 MIME-Version: 1.0 X-Proofpoint-ORIG-GUID: OuUBJHd3fjIIGOpJmY21rE7A0tGkzyWp X-Proofpoint-GUID: OuUBJHd3fjIIGOpJmY21rE7A0tGkzyWp X-Proofpoint-Virus-Version: vendor=baseguard engine=ICAP:2.0.272,Aquarius:18.0.997,Hydra:6.0.619,FMLib:17.11.176.26 definitions=2023-12-08_02,2023-12-07_01,2023-05-22_02 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 This patch introduces new virtio-user callback to map the vq notification area and implements it for the vhost-vDPA backend. This is simply done by using mmap()/munmap() for the vhost-vDPA fd. This patch also adds a parameter for configuring feature bit VIRTIO_NET_F_NOTIFICATION_DATA. If feature is disabled, also update corresponding unsupported feature bit. And also adds code to write to queue notify address in notify callback. This will help in increasing the kick performance. Signed-off-by: Srujana Challa --- doc/guides/nics/virtio.rst | 5 ++ drivers/net/virtio/virtio_user/vhost.h | 1 + drivers/net/virtio/virtio_user/vhost_vdpa.c | 56 ++++++++++++++++++ .../net/virtio/virtio_user/virtio_user_dev.c | 52 +++++++++++++++-- .../net/virtio/virtio_user/virtio_user_dev.h | 5 +- drivers/net/virtio/virtio_user_ethdev.c | 57 ++++++++++++++++--- 6 files changed, 162 insertions(+), 14 deletions(-) diff --git a/doc/guides/nics/virtio.rst b/doc/guides/nics/virtio.rst index c22ce56a02..11dd6359e5 100644 --- a/doc/guides/nics/virtio.rst +++ b/doc/guides/nics/virtio.rst @@ -349,6 +349,11 @@ Below devargs are supported by the virtio-user vdev: election. (Default: 0 (disabled)) +#. ``notification_data``: + + It is used to enable virtio device notification data feature. + (Default: 1 (enabled)) + Virtio paths Selection and Usage -------------------------------- diff --git a/drivers/net/virtio/virtio_user/vhost.h b/drivers/net/virtio/virtio_user/vhost.h index f817cab77a..1bce00c7ac 100644 --- a/drivers/net/virtio/virtio_user/vhost.h +++ b/drivers/net/virtio/virtio_user/vhost.h @@ -90,6 +90,7 @@ struct virtio_user_backend_ops { int (*server_disconnect)(struct virtio_user_dev *dev); int (*server_reconnect)(struct virtio_user_dev *dev); int (*get_intr_fd)(struct virtio_user_dev *dev); + int (*map_notification_area)(struct virtio_user_dev *dev, bool map); }; extern struct virtio_user_backend_ops virtio_ops_user; diff --git a/drivers/net/virtio/virtio_user/vhost_vdpa.c b/drivers/net/virtio/virtio_user/vhost_vdpa.c index 2c36b26224..1eb0f9ec48 100644 --- a/drivers/net/virtio/virtio_user/vhost_vdpa.c +++ b/drivers/net/virtio/virtio_user/vhost_vdpa.c @@ -5,6 +5,7 @@ #include #include #include +#include #include #include #include @@ -622,6 +623,60 @@ vhost_vdpa_get_intr_fd(struct virtio_user_dev *dev __rte_unused) return -1; } +static int +unmap_notification_area(struct virtio_user_dev *dev, int nr_vrings) +{ + int i; + + for (i = 0; i < nr_vrings; i++) { + if (dev->notify_area[i]) + munmap(dev->notify_area[i], getpagesize()); + } + free(dev->notify_area); + + return 0; +} + +static int +vhost_vdpa_map_notification_area(struct virtio_user_dev *dev, bool map) +{ + struct vhost_vdpa_data *data = dev->backend_data; + int nr_vrings, i, page_size = getpagesize(); + uint16_t **notify_area; + + nr_vrings = dev->max_queue_pairs * 2; + if (dev->device_features & (1ull << VIRTIO_NET_F_CTRL_VQ)) + nr_vrings++; + + if (!map) + return unmap_notification_area(dev, nr_vrings); + + notify_area = malloc(nr_vrings * sizeof(*notify_area)); + if (!notify_area) { + PMD_DRV_LOG(ERR, "(%s) Failed to allocate notify area array", dev->path); + return -1; + } + for (i = 0; i < nr_vrings; i++) { + notify_area[i] = mmap(NULL, page_size, PROT_WRITE, MAP_SHARED | MAP_FILE, + data->vhostfd, i * page_size); + if (notify_area[i] == MAP_FAILED) { + PMD_DRV_LOG(ERR, "(%s) Map failed for notify address of queue %d\n", + dev->path, i); + goto map_err; + } + } + dev->notify_area = notify_area; + + return 0; + +map_err: + i--; + for (; i >= 0; i--) + munmap(notify_area[i], page_size); + free(notify_area); + return -1; +} + struct virtio_user_backend_ops virtio_ops_vdpa = { .setup = vhost_vdpa_setup, .destroy = vhost_vdpa_destroy, @@ -646,4 +701,5 @@ struct virtio_user_backend_ops virtio_ops_vdpa = { .dma_unmap = vhost_vdpa_dma_unmap_batch, .update_link_state = vhost_vdpa_update_link_state, .get_intr_fd = vhost_vdpa_get_intr_fd, + .map_notification_area = vhost_vdpa_map_notification_area, }; diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.c b/drivers/net/virtio/virtio_user/virtio_user_dev.c index af1f8c8237..578877d089 100644 --- a/drivers/net/virtio/virtio_user/virtio_user_dev.c +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.c @@ -18,6 +18,7 @@ #include #include #include +#include #include "vhost.h" #include "virtio_user_dev.h" @@ -413,6 +414,12 @@ virtio_user_dev_init_notify(struct virtio_user_dev *dev) dev->callfds[i] = callfd; dev->kickfds[i] = kickfd; } + dev->notify_area_mapped = false; + if (dev->ops->map_notification_area) { + if (dev->ops->map_notification_area(dev, true)) + goto err; + dev->notify_area_mapped = true; + } return 0; err: @@ -445,6 +452,11 @@ virtio_user_dev_uninit_notify(struct virtio_user_dev *dev) dev->callfds[i] = -1; } } + if (dev->ops->map_notification_area && dev->notify_area_mapped) { + /* Unmap notification area */ + dev->ops->map_notification_area(dev, false); + dev->notify_area_mapped = false; + } } static int @@ -674,12 +686,14 @@ virtio_user_free_vrings(struct virtio_user_dev *dev) 1ULL << VIRTIO_NET_F_GUEST_TSO6 | \ 1ULL << VIRTIO_F_IN_ORDER | \ 1ULL << VIRTIO_F_VERSION_1 | \ - 1ULL << VIRTIO_F_RING_PACKED) + 1ULL << VIRTIO_F_RING_PACKED | \ + 1ULL << VIRTIO_F_NOTIFICATION_DATA) int virtio_user_dev_init(struct virtio_user_dev *dev, char *path, uint16_t queues, int cq, int queue_size, const char *mac, char **ifname, int server, int mrg_rxbuf, int in_order, int packed_vq, + int notification_data, enum virtio_user_backend_type backend_type) { uint64_t backend_features; @@ -737,6 +751,9 @@ virtio_user_dev_init(struct virtio_user_dev *dev, char *path, uint16_t queues, if (!packed_vq) dev->unsupported_features |= (1ull << VIRTIO_F_RING_PACKED); + if (!notification_data) + dev->unsupported_features |= (1ull << VIRTIO_F_NOTIFICATION_DATA); + if (dev->mac_specified) dev->frontend_features |= (1ull << VIRTIO_NET_F_MAC); else @@ -1039,11 +1056,35 @@ static void virtio_user_control_queue_notify(struct virtqueue *vq, void *cookie) { struct virtio_user_dev *dev = cookie; - uint64_t buf = 1; + uint64_t notify_data = 1; + + if (!dev->notify_area_mapped) { + if (write(dev->kickfds[vq->vq_queue_index], ¬ify_data, sizeof(notify_data)) < 0) + PMD_DRV_LOG(ERR, "failed to kick backend: %s", + strerror(errno)); + return; + } else if (!virtio_with_feature(&dev->hw, VIRTIO_F_NOTIFICATION_DATA)) { + rte_write16(vq->vq_queue_index, vq->notify_addr); + return; + } - if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0) - PMD_DRV_LOG(ERR, "failed to kick backend: %s", - strerror(errno)); + if (virtio_with_packed_queue(&dev->hw)) { + /* Bit[0:15]: vq queue index + * Bit[16:30]: avail index + * Bit[31]: avail wrap counter + */ + notify_data = ((uint32_t)(!!(vq->vq_packed.cached_flags & + VRING_PACKED_DESC_F_AVAIL)) << 31) | + ((uint32_t)vq->vq_avail_idx << 16) | + vq->vq_queue_index; + } else { + /* Bit[0:15]: vq queue index + * Bit[16:31]: avail index + */ + notify_data = ((uint32_t)vq->vq_avail_idx << 16) | + vq->vq_queue_index; + } + rte_write32(notify_data, vq->notify_addr); } int @@ -1062,6 +1103,7 @@ virtio_user_dev_create_shadow_cvq(struct virtio_user_dev *dev, struct virtqueue scvq->cq.notify_queue = &virtio_user_control_queue_notify; scvq->cq.notify_cookie = dev; + scvq->notify_addr = vq->notify_addr; dev->scvq = scvq; return 0; diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.h b/drivers/net/virtio/virtio_user/virtio_user_dev.h index 7323d88302..29ec386da5 100644 --- a/drivers/net/virtio/virtio_user/virtio_user_dev.h +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.h @@ -64,6 +64,9 @@ struct virtio_user_dev { struct virtqueue *scvq; void *backend_data; + + bool notify_area_mapped; + uint16_t **notify_area; }; int virtio_user_dev_set_features(struct virtio_user_dev *dev); @@ -72,7 +75,7 @@ int virtio_user_stop_device(struct virtio_user_dev *dev); int virtio_user_dev_init(struct virtio_user_dev *dev, char *path, uint16_t queues, int cq, int queue_size, const char *mac, char **ifname, int server, int mrg_rxbuf, int in_order, - int packed_vq, + int packed_vq, int notification_data, enum virtio_user_backend_type backend_type); void virtio_user_dev_uninit(struct virtio_user_dev *dev); void virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx); diff --git a/drivers/net/virtio/virtio_user_ethdev.c b/drivers/net/virtio/virtio_user_ethdev.c index 3a31642899..241465ecdd 100644 --- a/drivers/net/virtio/virtio_user_ethdev.c +++ b/drivers/net/virtio/virtio_user_ethdev.c @@ -18,6 +18,7 @@ #include #include #include +#include #include "virtio_ethdev.h" #include "virtio_logs.h" @@ -232,6 +233,7 @@ virtio_user_setup_queue(struct virtio_hw *hw, struct virtqueue *vq) else virtio_user_setup_queue_split(vq, dev); + vq->notify_addr = dev->notify_area[vq->vq_queue_index]; if (dev->hw_cvq && hw->cvq && (virtnet_cq_to_vq(hw->cvq) == vq)) return virtio_user_dev_create_shadow_cvq(dev, vq); @@ -262,8 +264,8 @@ virtio_user_del_queue(struct virtio_hw *hw, struct virtqueue *vq) static void virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq) { - uint64_t buf = 1; struct virtio_user_dev *dev = virtio_user_get_dev(hw); + uint64_t notify_data = 1; if (hw->cvq && (virtnet_cq_to_vq(hw->cvq) == vq)) { virtio_user_handle_cq(dev, vq->vq_queue_index); @@ -271,9 +273,34 @@ virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq) return; } - if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0) - PMD_DRV_LOG(ERR, "failed to kick backend: %s", - strerror(errno)); + if (!dev->notify_area_mapped) { + if (write(dev->kickfds[vq->vq_queue_index], ¬ify_data, + sizeof(notify_data)) < 0) + PMD_DRV_LOG(ERR, "failed to kick backend: %s", + strerror(errno)); + return; + } else if (!virtio_with_feature(hw, VIRTIO_F_NOTIFICATION_DATA)) { + rte_write16(vq->vq_queue_index, vq->notify_addr); + return; + } + + if (virtio_with_packed_queue(hw)) { + /* Bit[0:15]: vq queue index + * Bit[16:30]: avail index + * Bit[31]: avail wrap counter + */ + notify_data = ((uint32_t)(!!(vq->vq_packed.cached_flags & + VRING_PACKED_DESC_F_AVAIL)) << 31) | + ((uint32_t)vq->vq_avail_idx << 16) | + vq->vq_queue_index; + } else { + /* Bit[0:15]: vq queue index + * Bit[16:31]: avail index + */ + notify_data = ((uint32_t)vq->vq_avail_idx << 16) | + vq->vq_queue_index; + } + rte_write32(notify_data, vq->notify_addr); } static int @@ -329,6 +356,8 @@ static const char *valid_args[] = { VIRTIO_USER_ARG_SPEED, #define VIRTIO_USER_ARG_VECTORIZED "vectorized" VIRTIO_USER_ARG_VECTORIZED, +#define VIRTIO_USER_ARG_NOTIFICATION_DATA "notification_data" + VIRTIO_USER_ARG_NOTIFICATION_DATA, NULL }; @@ -480,6 +509,7 @@ virtio_user_pmd_probe(struct rte_vdev_device *vdev) uint64_t in_order = 1; uint64_t packed_vq = 0; uint64_t vectorized = 0; + uint64_t notification_data = 1; char *path = NULL; char *ifname = NULL; char *mac_addr = NULL; @@ -637,6 +667,15 @@ virtio_user_pmd_probe(struct rte_vdev_device *vdev) } } + if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_NOTIFICATION_DATA) == 1) { + if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_NOTIFICATION_DATA, + &get_integer_arg, ¬ification_data) < 0) { + PMD_INIT_LOG(ERR, "error to parse %s", + VIRTIO_USER_ARG_NOTIFICATION_DATA); + goto end; + } + } + eth_dev = virtio_user_eth_dev_alloc(vdev); if (!eth_dev) { PMD_INIT_LOG(ERR, "virtio_user fails to alloc device"); @@ -645,9 +684,10 @@ virtio_user_pmd_probe(struct rte_vdev_device *vdev) dev = eth_dev->data->dev_private; hw = &dev->hw; - if (virtio_user_dev_init(dev, path, (uint16_t)queues, cq, - queue_size, mac_addr, &ifname, server_mode, - mrg_rxbuf, in_order, packed_vq, backend_type) < 0) { + if (virtio_user_dev_init(dev, path, (uint16_t)queues, cq, queue_size, + mac_addr, &ifname, server_mode, mrg_rxbuf, + in_order, packed_vq, notification_data, + backend_type) < 0) { PMD_INIT_LOG(ERR, "virtio_user_dev_init fails"); virtio_user_eth_dev_free(eth_dev); goto end; @@ -784,4 +824,5 @@ RTE_PMD_REGISTER_PARAM_STRING(net_virtio_user, "in_order=<0|1> " "packed_vq=<0|1> " "speed= " - "vectorized=<0|1>"); + "vectorized=<0|1> " + "notification_data=<0|1> ");