diff mbox series

[v1,06/14] vhost: flush vhost enqueue shadow ring by burst

Message ID 20190905161421.55981-7-yong.liu@intel.com (mailing list archive)
State Superseded, archived
Delegated to: Maxime Coquelin
Headers
Series vhost packed ring performance optimization |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK

Commit Message

Marvin Liu Sept. 5, 2019, 4:14 p.m. UTC 
  Buffer vhost enqueue shadow ring update, flush shadow ring until
buffered descriptors number exceed one burst. Thus virtio can receive
packets at a faster frequency.

Signed-off-by: Marvin Liu <yong.liu@intel.com>
diff mbox series

Patch

diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index b33f29ba0..86552cbeb 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -143,6 +143,7 @@  struct vhost_virtqueue {
 		struct vring_used_elem_packed *shadow_used_packed;
 	};
 	uint16_t                shadow_used_idx;
+	uint16_t                enqueue_shadow_count;
 	struct vhost_vring_addr ring_addrs;
 
 	struct batch_copy_elem	*batch_copy_elems;
diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 7116c389d..dffd466d5 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -157,6 +157,24 @@  flush_shadow_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
 	vhost_log_cache_sync(dev, vq);
 }
 
+static __rte_always_inline void
+update_enqueue_shadow_packed(struct vhost_virtqueue *vq, uint16_t desc_idx,
+	uint32_t len, uint16_t count)
+{
+	/* enqueue shadow flush action aligned with burst num */
+	if (!vq->shadow_used_idx)
+		vq->enqueue_shadow_count = vq->last_used_idx &
+						PACKED_BURST_MASK;
+
+	uint16_t i = vq->shadow_used_idx++;
+
+	vq->shadow_used_packed[i].id  = desc_idx;
+	vq->shadow_used_packed[i].len = len;
+	vq->shadow_used_packed[i].count = count;
+
+	vq->enqueue_shadow_count += count;
+}
+
 static __rte_always_inline void
 update_shadow_packed(struct vhost_virtqueue *vq,
 			 uint16_t desc_idx, uint32_t len, uint16_t count)
@@ -197,6 +215,22 @@  do_data_copy_dequeue(struct vhost_virtqueue *vq)
 	vq->batch_copy_nb_elems = 0;
 }
 
+static __rte_always_inline void
+flush_enqueue_packed(struct virtio_net *dev,
+	struct vhost_virtqueue *vq, uint32_t len[], uint16_t id[],
+	uint16_t count[], uint16_t num_buffers)
+{
+	int i;
+	for (i = 0; i < num_buffers; i++) {
+		update_enqueue_shadow_packed(vq, id[i], len[i], count[i]);
+
+		if (vq->enqueue_shadow_count >= PACKED_DESCS_BURST) {
+			do_data_copy_enqueue(dev, vq);
+			flush_shadow_packed(dev, vq);
+		}
+	}
+}
+
 /* avoid write operation when necessary, to lessen cache issues */
 #define ASSIGN_UNLESS_EQUAL(var, val) do {	\
 	if ((var) != (val))			\
@@ -798,6 +832,9 @@  vhost_enqueue_single_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		max_tries = 1;
 
 	uint16_t num_buffers = 0;
+	uint32_t buffer_len[max_tries];
+	uint16_t buffer_buf_id[max_tries];
+	uint16_t buffer_desc_count[max_tries];
 
 	while (size > 0) {
 		/*
@@ -820,6 +857,10 @@  vhost_enqueue_single_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
 
 		size -= len;
 
+		buffer_len[num_buffers] = len;
+		buffer_buf_id[num_buffers] = buf_id;
+		buffer_desc_count[num_buffers] = desc_count;
+
 		avail_idx += desc_count;
 		if (avail_idx >= vq->size)
 			avail_idx -= vq->size;
@@ -834,6 +875,9 @@  vhost_enqueue_single_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
 		return 0;
 	}
 
+	flush_enqueue_packed(dev, vq, buffer_len, buffer_buf_id,
+			buffer_desc_count, num_buffers);
+
 	return 0;
 }