Skip to contentSkip to conversation

net/virtio: fix refill order in packed ring datapath

PATCH Applied Cheng Jiang 3 participants 3 messages Jul 8, 2021 – Jul 20, 2021 Network drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net/virtio: fix refill order in packed ring datapath

The front-end should refill the descriptor with the mbuf indicated by the buff_id rather then the index of used descriptor. Back-end may return buffers out of order if async copy mode is enabled.

When initializing rxq, refill the descriptors in order as buff_id is not available at that time.

Fixes a76290c8f1cf ("net/virtio: implement Rx path for packed queues")
drivers/net/virtio/virtio_rxtx.c +51 -19
index 3ac847317f..d35875d9ce 100644
@@ -328,13 +328,35 @@ virtqueue_enqueue_recv_refill(struct virtqueue *vq, struct rte_mbuf **cookie,
 	return 0;
 }
 
+static inline void
+virtqueue_refill_single_packed(struct virtqueue *vq,
+			       struct vring_packed_desc *dp,
+			       struct rte_mbuf *cookie)
+{
+	uint16_t flags = vq->vq_packed.cached_flags;
+	struct virtio_hw *hw = vq->hw;
+
+	dp->addr = cookie->buf_iova +
+			RTE_PKTMBUF_HEADROOM - hw->vtnet_hdr_size;
+	dp->len = cookie->buf_len -
+		RTE_PKTMBUF_HEADROOM + hw->vtnet_hdr_size;
+
+	virtqueue_store_flags_packed(dp, flags,
+				     hw->weak_barriers);
+
+	if (++vq->vq_avail_idx >= vq->vq_nentries) {
+		vq->vq_avail_idx -= vq->vq_nentries;
+		vq->vq_packed.cached_flags ^=
+			VRING_PACKED_DESC_F_AVAIL_USED;
+		flags = vq->vq_packed.cached_flags;
+	}
+}
+
 static inline int
-virtqueue_enqueue_recv_refill_packed(struct virtqueue *vq,
+virtqueue_enqueue_recv_refill_packed_init(struct virtqueue *vq,
 				     struct rte_mbuf **cookie, uint16_t num)
 {
 	struct vring_packed_desc *start_dp = vq->vq_packed.ring.desc;
-	uint16_t flags = vq->vq_packed.cached_flags;
-	struct virtio_hw *hw = vq->hw;
 	struct vq_desc_extra *dxp;
 	uint16_t idx;
 	int i;
@@ -350,24 +372,34 @@ virtqueue_enqueue_recv_refill_packed(struct virtqueue *vq,
 		dxp->cookie = (void *)cookie[i];
 		dxp->ndescs = 1;
 
-		start_dp[idx].addr = cookie[i]->buf_iova +
-			RTE_PKTMBUF_HEADROOM - hw->vtnet_hdr_size;
-		start_dp[idx].len = cookie[i]->buf_len -
-			RTE_PKTMBUF_HEADROOM + hw->vtnet_hdr_size;
+		virtqueue_refill_single_packed(vq, &start_dp[idx], cookie[i]);
+	}
+	vq->vq_free_cnt = (uint16_t)(vq->vq_free_cnt - num);
+	return 0;
+}
 
-		vq->vq_desc_head_idx = dxp->next;
-		if (vq->vq_desc_head_idx == VQ_RING_DESC_CHAIN_END)
-			vq->vq_desc_tail_idx = vq->vq_desc_head_idx;
+static inline int
+virtqueue_enqueue_recv_refill_packed(struct virtqueue *vq,
+				     struct rte_mbuf **cookie, uint16_t num)
+{
+	struct vring_packed_desc *start_dp = vq->vq_packed.ring.desc;
+	struct vq_desc_extra *dxp;
+	uint16_t idx, did;
+	int i;
 
-		virtqueue_store_flags_packed(&start_dp[idx], flags,
-					     hw->weak_barriers);
+	if (unlikely(vq->vq_free_cnt == 0))
+		return -ENOSPC;
+	if (unlikely(vq->vq_free_cnt < num))
+		return -EMSGSIZE;
 
-		if (++vq->vq_avail_idx >= vq->vq_nentries) {
-			vq->vq_avail_idx -= vq->vq_nentries;
-			vq->vq_packed.cached_flags ^=
-				VRING_PACKED_DESC_F_AVAIL_USED;
-			flags = vq->vq_packed.cached_flags;
-		}
+	for (i = 0; i < num; i++) {
+		idx = vq->vq_avail_idx;
+		did = start_dp[idx].id;
+		dxp = &vq->vq_descx[did];
+		dxp->cookie = (void *)cookie[i];
+		dxp->ndescs = 1;
+
+		virtqueue_refill_single_packed(vq, &start_dp[idx], cookie[i]);
 	}
 	vq->vq_free_cnt = (uint16_t)(vq->vq_free_cnt - num);
 	return 0;
@@ -742,7 +774,7 @@ virtio_dev_rx_queue_setup_finish(struct rte_eth_dev *dev, uint16_t queue_idx)
 
 			/* Enqueue allocated buffers */
 			if (virtio_with_packed_queue(vq->hw))
-				error = virtqueue_enqueue_recv_refill_packed(vq,
+				error = virtqueue_enqueue_recv_refill_packed_init(vq,
 						&m, 1);
 			else
 				error = virtqueue_enqueue_recv_refill(vq,

--
2.17.1

Reviewed-by on v1 1/1
#1 lore raw
Quoted 4 lines · Cheng Jiang

The front-end should refill the descriptor with the mbuf indicated by the buff_id rather then the index of used descriptor. Back-end may return buffers out of order if async copy mode is enabled.

Quoted 9 lines

When initializing rxq, refill the descriptors in order as buff_id is not available at that time.

Fixes a76290c8f1cf ("net/virtio: implement Rx path for packed queues")

Thanks,
Maxime

Applied
#2 lore raw
In reply to Cheng Jiang
Quoted 16 lines

-----Original Message-----
From: Jiang, Cheng1 <[email protected]>
Sent: Thursday, July 8, 2021 5:58 PM
To: [email protected]; Xia, Chenbo <[email protected]>
Cc: [email protected]; Hu, Jiayu <[email protected]>; Liu, Yong <[email protected]>; Yang, YvonneX <[email protected]>; Jiang, Cheng1 <[email protected]>; [email protected]
Subject: [PATCH] net/virtio: fix refill order in packed ring datapath

The front-end should refill the descriptor with the mbuf indicated by the buff_id rather then the index of used descriptor. Back-end may return buffers out of order if async copy mode is enabled.

When initializing rxq, refill the descriptors in order as buff_id is not available at that time.

Quoted 7 lines
Fixes a76290c8f1cf ("net/virtio: implement Rx path for packed queues")

--
2.17.1

Applied to next-virtio/main, thanks.