vhost: optimize packed ring enqueue
authorMarvin Liu <yong.liu@intel.com>
Thu, 24 Oct 2019 16:08:29 +0000 (00:08 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Fri, 25 Oct 2019 17:20:47 +0000 (19:20 +0200)
Optimize vhost device packed ring enqueue function by splitting batch
and single functions. Packets can be filled into one desc will be
handled by batch and others will be handled by single as before.

Signed-off-by: Marvin Liu <yong.liu@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
lib/librte_vhost/virtio_net.c

index b09e03f..1c63262 100644 (file)
@@ -778,64 +778,6 @@ fill_vec_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
        return 0;
 }
 
-/*
- * Returns -1 on fail, 0 on success
- */
-static inline int
-reserve_avail_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
-                               uint32_t size, struct buf_vector *buf_vec,
-                               uint16_t *nr_vec, uint16_t *num_buffers,
-                               uint16_t *nr_descs)
-{
-       uint16_t avail_idx;
-       uint16_t vec_idx = 0;
-       uint16_t max_tries, tries = 0;
-
-       uint16_t buf_id = 0;
-       uint32_t len = 0;
-       uint16_t desc_count;
-
-       *num_buffers = 0;
-       avail_idx = vq->last_avail_idx;
-
-       if (rxvq_is_mergeable(dev))
-               max_tries = vq->size - 1;
-       else
-               max_tries = 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, &vec_idx,
-                                               &buf_id, &len,
-                                               VHOST_ACCESS_RW) < 0))
-                       return -1;
-
-               len = RTE_MIN(len, size);
-               update_shadow_used_ring_packed(vq, buf_id, len, desc_count);
-               size -= len;
-
-               avail_idx += desc_count;
-               if (avail_idx >= vq->size)
-                       avail_idx -= vq->size;
-
-               *nr_descs += desc_count;
-               *num_buffers += 1;
-       }
-
-       *nr_vec = vec_idx;
-
-       return 0;
-}
-
 static __rte_noinline void
 copy_vnet_hdr_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
                struct buf_vector *buf_vec,
@@ -1118,7 +1060,7 @@ virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
        return pkt_idx;
 }
 
-static __rte_unused int
+static __rte_always_inline int
 virtio_dev_rx_batch_packed(struct virtio_net *dev,
                           struct vhost_virtqueue *vq,
                           struct rte_mbuf **pkts)
@@ -1194,7 +1136,7 @@ virtio_dev_rx_batch_packed(struct virtio_net *dev,
        return 0;
 }
 
-static __rte_unused int16_t
+static __rte_always_inline int16_t
 virtio_dev_rx_single_packed(struct virtio_net *dev,
                            struct vhost_virtqueue *vq,
                            struct rte_mbuf *pkt)
@@ -1221,49 +1163,40 @@ virtio_dev_rx_single_packed(struct virtio_net *dev,
 }
 
 static __rte_noinline uint32_t
-virtio_dev_rx_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
-       struct rte_mbuf **pkts, uint32_t count)
+virtio_dev_rx_packed(struct virtio_net *dev,
+                    struct vhost_virtqueue *vq,
+                    struct rte_mbuf **pkts,
+                    uint32_t count)
 {
        uint32_t pkt_idx = 0;
-       uint16_t num_buffers;
-       struct buf_vector buf_vec[BUF_VECTOR_MAX];
+       uint32_t remained = count;
 
-       for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
-               uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
-               uint16_t nr_vec = 0;
-               uint16_t nr_descs = 0;
+       do {
+               rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
 
-               if (unlikely(reserve_avail_buf_packed(dev, vq,
-                                               pkt_len, buf_vec, &nr_vec,
-                                               &num_buffers, &nr_descs) < 0)) {
-                       VHOST_LOG_DEBUG(VHOST_DATA,
-                               "(%d) failed to get enough desc from vring\n",
-                               dev->vid);
-                       vq->shadow_used_idx -= num_buffers;
-                       break;
+               if (remained >= PACKED_BATCH_SIZE) {
+                       if (!virtio_dev_rx_batch_packed(dev, vq, pkts)) {
+                               pkt_idx += PACKED_BATCH_SIZE;
+                               remained -= PACKED_BATCH_SIZE;
+                               continue;
+                       }
                }
 
-               VHOST_LOG_DEBUG(VHOST_DATA, "(%d) current index %d | end index %d\n",
-                       dev->vid, vq->last_avail_idx,
-                       vq->last_avail_idx + num_buffers);
-
-               if (copy_mbuf_to_desc(dev, vq, pkts[pkt_idx],
-                                               buf_vec, nr_vec,
-                                               num_buffers) < 0) {
-                       vq->shadow_used_idx -= num_buffers;
+               if (virtio_dev_rx_single_packed(dev, vq, pkts[pkt_idx]))
                        break;
-               }
+               pkt_idx++;
+               remained--;
 
-               vq_inc_last_avail_packed(vq, nr_descs);
-       }
-
-       do_data_copy_enqueue(dev, vq);
+       } while (pkt_idx < count);
 
-       if (likely(vq->shadow_used_idx)) {
+       if (vq->shadow_used_idx) {
+               do_data_copy_enqueue(dev, vq);
                vhost_flush_enqueue_shadow_packed(dev, vq);
-               vhost_vring_call_packed(dev, vq);
        }
 
+       if (pkt_idx)
+               vhost_vring_call_packed(dev, vq);
+
        return pkt_idx;
 }