> -----Original Message-----
> From: Maxime Coquelin <maxime.coque...@redhat.com>
> Sent: Tuesday, February 7, 2023 11:18 PM
> To: dev@dpdk.org; Xia, Chenbo <chenbo....@intel.com>;
> david.march...@redhat.com; epere...@redhat.com; step...@networkplumber.org
> Cc: Maxime Coquelin <maxime.coque...@redhat.com>
> Subject: [PATCH v2 10/21] net/virtio: alloc Rx SW ring only if vectorized
> path
>
> This patch only allocates the SW ring when vectorized
> datapath is used. It also moves the SW ring and fake mbuf
> in the virtnet_rx struct since this is Rx-only.
>
> Signed-off-by: Maxime Coquelin <maxime.coque...@redhat.com>
> ---
> drivers/net/virtio/virtio_ethdev.c | 88 ++++++++++++-------
> drivers/net/virtio/virtio_rxtx.c | 8 +-
> drivers/net/virtio/virtio_rxtx.h | 4 +-
> drivers/net/virtio/virtio_rxtx_simple.h | 2 +-
> .../net/virtio/virtio_rxtx_simple_altivec.c | 4 +-
> drivers/net/virtio/virtio_rxtx_simple_neon.c | 4 +-
> drivers/net/virtio/virtio_rxtx_simple_sse.c | 4 +-
> drivers/net/virtio/virtqueue.c | 6 +-
> drivers/net/virtio/virtqueue.h | 1 -
> 9 files changed, 72 insertions(+), 49 deletions(-)
>
> diff --git a/drivers/net/virtio/virtio_ethdev.c
> b/drivers/net/virtio/virtio_ethdev.c
> index f839a24d12..14c5dc9059 100644
> --- a/drivers/net/virtio/virtio_ethdev.c
> +++ b/drivers/net/virtio/virtio_ethdev.c
> @@ -339,6 +339,47 @@ virtio_free_queue_headers(struct virtqueue *vq)
> *hdr_mem = 0;
> }
>
> +static int
> +virtio_rxq_sw_ring_alloc(struct virtqueue *vq, int numa_node)
> +{
> + void *sw_ring;
> + struct rte_mbuf *mbuf;
> + size_t size;
> +
> + /* SW ring is only used with vectorized datapath */
> + if (!vq->hw->use_vec_rx)
> + return 0;
> +
> + size = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq->vq_nentries) * sizeof(vq-
> >rxq.sw_ring[0]);
> +
> + sw_ring = rte_zmalloc_socket("sw_ring", size, RTE_CACHE_LINE_SIZE,
> numa_node);
> + if (!sw_ring) {
> + PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
> + return -ENOMEM;
> + }
> +
> + mbuf = rte_zmalloc_socket("sw_ring", sizeof(*mbuf),
> RTE_CACHE_LINE_SIZE, numa_node);
> + if (!mbuf) {
> + PMD_INIT_LOG(ERR, "can not allocate fake mbuf");
> + rte_free(sw_ring);
> + return -ENOMEM;
> + }
> +
> + vq->rxq.sw_ring = sw_ring;
> + vq->rxq.fake_mbuf = mbuf;
> +
> + return 0;
> +}
> +
> +static void
> +virtio_rxq_sw_ring_free(struct virtqueue *vq)
> +{
> + rte_free(vq->rxq.fake_mbuf);
> + vq->rxq.fake_mbuf = NULL;
> + rte_free(vq->rxq.sw_ring);
> + vq->rxq.sw_ring = NULL;
> +}
> +
> static int
> virtio_init_queue(struct rte_eth_dev *dev, uint16_t queue_idx)
> {
> @@ -346,14 +387,11 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t
> queue_idx)
> const struct rte_memzone *mz = NULL;
> unsigned int vq_size, size;
> struct virtio_hw *hw = dev->data->dev_private;
> - struct virtnet_rx *rxvq = NULL;
> struct virtnet_ctl *cvq = NULL;
> struct virtqueue *vq;
> - void *sw_ring = NULL;
> int queue_type = virtio_get_queue_type(hw, queue_idx);
> int ret;
> int numa_node = dev->device->numa_node;
> - struct rte_mbuf *fake_mbuf = NULL;
>
> PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
> queue_idx, numa_node);
> @@ -441,28 +479,9 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t
> queue_idx)
> }
>
> if (queue_type == VTNET_RQ) {
> - size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
> - sizeof(vq->sw_ring[0]);
> -
> - sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
> - RTE_CACHE_LINE_SIZE, numa_node);
> - if (!sw_ring) {
> - PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
> - ret = -ENOMEM;
> + ret = virtio_rxq_sw_ring_alloc(vq, numa_node);
> + if (ret)
> goto free_hdr_mz;
> - }
> -
> - fake_mbuf = rte_zmalloc_socket("sw_ring", sizeof(*fake_mbuf),
> - RTE_CACHE_LINE_SIZE, numa_node);
> - if (!fake_mbuf) {
> - PMD_INIT_LOG(ERR, "can not allocate fake mbuf");
> - ret = -ENOMEM;
> - goto free_sw_ring;
> - }
> -
> - vq->sw_ring = sw_ring;
> - rxvq = &vq->rxq;
> - rxvq->fake_mbuf = fake_mbuf;
> } else if (queue_type == VTNET_TQ) {
> virtqueue_txq_indirect_headers_init(vq);
> } else if (queue_type == VTNET_CQ) {
> @@ -486,9 +505,8 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t
> queue_idx)
>
> clean_vq:
> hw->cvq = NULL;
> - rte_free(fake_mbuf);
> -free_sw_ring:
> - rte_free(sw_ring);
> + if (queue_type == VTNET_RQ)
> + virtio_rxq_sw_ring_free(vq);
> free_hdr_mz:
> virtio_free_queue_headers(vq);
> free_mz:
> @@ -519,7 +537,7 @@ virtio_free_queues(struct virtio_hw *hw)
> queue_type = virtio_get_queue_type(hw, i);
> if (queue_type == VTNET_RQ) {
> rte_free(vq->rxq.fake_mbuf);
> - rte_free(vq->sw_ring);
> + rte_free(vq->rxq.sw_ring);
> }
>
> virtio_free_queue_headers(vq);
> @@ -2214,6 +2232,11 @@ eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
>
> rte_spinlock_init(&hw->state_lock);
>
> + if (vectorized) {
> + hw->use_vec_rx = 1;
> + hw->use_vec_tx = 1;
> + }
> +
> /* reset device and negotiate default features */
> ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
> if (ret < 0)
> @@ -2221,12 +2244,11 @@ eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
>
> if (vectorized) {
> if (!virtio_with_packed_queue(hw)) {
> - hw->use_vec_rx = 1;
> + hw->use_vec_tx = 0;
> } else {
> -#if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
> - hw->use_vec_rx = 1;
> - hw->use_vec_tx = 1;
> -#else
> +#if !defined(CC_AVX512_SUPPORT) && !defined(RTE_ARCH_ARM)
> + hw->use_vec_rx = 0;
> + hw->use_vec_tx = 0;
> PMD_DRV_LOG(INFO,
> "building environment do not support packed ring
> vectorized");
> #endif
> diff --git a/drivers/net/virtio/virtio_rxtx.c
> b/drivers/net/virtio/virtio_rxtx.c
> index 4f69b97f41..2d0afd3302 100644
> --- a/drivers/net/virtio/virtio_rxtx.c
> +++ b/drivers/net/virtio/virtio_rxtx.c
> @@ -737,9 +737,11 @@ virtio_dev_rx_queue_setup_finish(struct rte_eth_dev
> *dev, uint16_t queue_idx)
> virtio_rxq_vec_setup(rxvq);
> }
>
> - memset(rxvq->fake_mbuf, 0, sizeof(*rxvq->fake_mbuf));
> - for (desc_idx = 0; desc_idx < RTE_PMD_VIRTIO_RX_MAX_BURST;
> desc_idx++)
> - vq->sw_ring[vq->vq_nentries + desc_idx] = rxvq->fake_mbuf;
> + if (hw->use_vec_rx) {
> + memset(rxvq->fake_mbuf, 0, sizeof(*rxvq->fake_mbuf));
> + for (desc_idx = 0; desc_idx < RTE_PMD_VIRTIO_RX_MAX_BURST;
> desc_idx++)
> + vq->rxq.sw_ring[vq->vq_nentries + desc_idx] = rxvq-
> >fake_mbuf;
> + }
>
> if (hw->use_vec_rx && !virtio_with_packed_queue(hw)) {
> while (vq->vq_free_cnt >= RTE_VIRTIO_VPMD_RX_REARM_THRESH) {
> diff --git a/drivers/net/virtio/virtio_rxtx.h
> b/drivers/net/virtio/virtio_rxtx.h
> index 57af630110..afc4b74534 100644
> --- a/drivers/net/virtio/virtio_rxtx.h
> +++ b/drivers/net/virtio/virtio_rxtx.h
> @@ -18,8 +18,8 @@ struct virtnet_stats {
> };
>
> struct virtnet_rx {
> - /* dummy mbuf, for wraparound when processing RX ring. */
> - struct rte_mbuf *fake_mbuf;
> + struct rte_mbuf **sw_ring; /**< RX software ring. */
> + struct rte_mbuf *fake_mbuf; /**< dummy mbuf, for wraparound when
> processing RX ring. */
> uint64_t mbuf_initializer; /**< value to init mbufs. */
> struct rte_mempool *mpool; /**< mempool for mbuf allocation */
>
> diff --git a/drivers/net/virtio/virtio_rxtx_simple.h
> b/drivers/net/virtio/virtio_rxtx_simple.h
> index 8e235f4dbc..79196ed86e 100644
> --- a/drivers/net/virtio/virtio_rxtx_simple.h
> +++ b/drivers/net/virtio/virtio_rxtx_simple.h
> @@ -26,7 +26,7 @@ virtio_rxq_rearm_vec(struct virtnet_rx *rxvq)
> struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
>
> desc_idx = vq->vq_avail_idx & (vq->vq_nentries - 1);
> - sw_ring = &vq->sw_ring[desc_idx];
> + sw_ring = &vq->rxq.sw_ring[desc_idx];
> start_dp = &vq->vq_split.ring.desc[desc_idx];
>
> ret = rte_mempool_get_bulk(rxvq->mpool, (void **)sw_ring,
> diff --git a/drivers/net/virtio/virtio_rxtx_simple_altivec.c
> b/drivers/net/virtio/virtio_rxtx_simple_altivec.c
> index e7f0ed6068..542ec3d952 100644
> --- a/drivers/net/virtio/virtio_rxtx_simple_altivec.c
> +++ b/drivers/net/virtio/virtio_rxtx_simple_altivec.c
> @@ -103,8 +103,8 @@ virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf
> **rx_pkts,
>
> desc_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
> rused = &vq->vq_split.ring.used->ring[desc_idx];
> - sw_ring = &vq->sw_ring[desc_idx];
> - sw_ring_end = &vq->sw_ring[vq->vq_nentries];
> + sw_ring = &vq->rxq.sw_ring[desc_idx];
> + sw_ring_end = &vq->rxq.sw_ring[vq->vq_nentries];
>
> rte_prefetch0(rused);
>
> diff --git a/drivers/net/virtio/virtio_rxtx_simple_neon.c
> b/drivers/net/virtio/virtio_rxtx_simple_neon.c
> index 7fd92d1b0c..7139b31d78 100644
> --- a/drivers/net/virtio/virtio_rxtx_simple_neon.c
> +++ b/drivers/net/virtio/virtio_rxtx_simple_neon.c
> @@ -101,8 +101,8 @@ virtio_recv_pkts_vec(void *rx_queue,
>
> desc_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
> rused = &vq->vq_split.ring.used->ring[desc_idx];
> - sw_ring = &vq->sw_ring[desc_idx];
> - sw_ring_end = &vq->sw_ring[vq->vq_nentries];
> + sw_ring = &vq->rxq.sw_ring[desc_idx];
> + sw_ring_end = &vq->rxq.sw_ring[vq->vq_nentries];
>
> rte_prefetch_non_temporal(rused);
>
> diff --git a/drivers/net/virtio/virtio_rxtx_simple_sse.c
> b/drivers/net/virtio/virtio_rxtx_simple_sse.c
> index 7577f5e86d..6a18741b6d 100644
> --- a/drivers/net/virtio/virtio_rxtx_simple_sse.c
> +++ b/drivers/net/virtio/virtio_rxtx_simple_sse.c
> @@ -101,8 +101,8 @@ virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf
> **rx_pkts,
>
> desc_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
> rused = &vq->vq_split.ring.used->ring[desc_idx];
> - sw_ring = &vq->sw_ring[desc_idx];
> - sw_ring_end = &vq->sw_ring[vq->vq_nentries];
> + sw_ring = &vq->rxq.sw_ring[desc_idx];
> + sw_ring_end = &vq->rxq.sw_ring[vq->vq_nentries];
>
> rte_prefetch0(rused);
>
> diff --git a/drivers/net/virtio/virtqueue.c
> b/drivers/net/virtio/virtqueue.c
> index fb651a4ca3..7a84796513 100644
> --- a/drivers/net/virtio/virtqueue.c
> +++ b/drivers/net/virtio/virtqueue.c
> @@ -38,9 +38,9 @@ virtqueue_detach_unused(struct virtqueue *vq)
> continue;
> if (start > end && (idx >= start || idx < end))
> continue;
> - cookie = vq->sw_ring[idx];
> + cookie = vq->rxq.sw_ring[idx];
> if (cookie != NULL) {
> - vq->sw_ring[idx] = NULL;
> + vq->rxq.sw_ring[idx] = NULL;
> return cookie;
> }
> } else {
> @@ -100,7 +100,7 @@ virtqueue_rxvq_flush_split(struct virtqueue *vq)
> uep = &vq->vq_split.ring.used->ring[used_idx];
> if (hw->use_vec_rx) {
> desc_idx = used_idx;
> - rte_pktmbuf_free(vq->sw_ring[desc_idx]);
> + rte_pktmbuf_free(vq->rxq.sw_ring[desc_idx]);
> vq->vq_free_cnt++;
> } else if (hw->use_inorder_rx) {
> desc_idx = (uint16_t)uep->id;
> diff --git a/drivers/net/virtio/virtqueue.h
> b/drivers/net/virtio/virtqueue.h
> index d453c3ec26..d7f8ee79bb 100644
> --- a/drivers/net/virtio/virtqueue.h
> +++ b/drivers/net/virtio/virtqueue.h
> @@ -206,7 +206,6 @@ struct virtqueue {
> * or virtual address for virtio_user. */
>
> uint16_t *notify_addr;
> - struct rte_mbuf **sw_ring; /**< RX software ring. */
> struct vq_desc_extra vq_descx[];
> };
>
> --
> 2.39.1
Reviewed-by: Chenbo Xia <chenbo....@intel.com>
