Add fast dequeue function just like enqueue function, fast dequeue
function will not support chained nor indirect descriptors, normal
function will handle that.
Signed-off-by: Marvin Liu <yong....@intel.com>
diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h
index f24026acd..329a7658b 100644
--- a/lib/librte_vhost/vhost.h
+++ b/lib/librte_vhost/vhost.h
@@ -41,6 +41,10 @@
/* Used in fast packed ring functions */
#define PACKED_DESC_PER_CACHELINE (RTE_CACHE_LINE_SIZE / sizeof(struct
vring_packed_desc))
+
+/* Indicated that normal path will handle */
+#define VIRTIO_DESC_NORMAL_FLAG (VRING_DESC_F_NEXT | VRING_DESC_F_INDIRECT)
+
/**
* Structure contains buffer address, length and descriptor index
* from vring to do scatter RX.
diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 410837122..a62e0feda 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -1613,6 +1613,158 @@ virtio_dev_tx_split(struct virtio_net *dev, struct
vhost_virtqueue *vq,
return i;
}
+static __rte_always_inline int
+vhost_dequeue_fast_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+ struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t
avail_idx,
+ uintptr_t *desc_addr, uint16_t *ids)
+{
+ bool wrap_counter = vq->avail_wrap_counter;
+ struct vring_packed_desc *descs = vq->desc_packed;
+ uint64_t len, len1, len2, len3;
+ uint32_t buf_offset = dev->vhost_hlen;
+
+ // check whether desc is cache aligned
+ if (unlikely(avail_idx & 0x3))
+ return -1;
+
+ // prefetch next cache line
+ if (unlikely(avail_idx < (vq->size - PACKED_DESC_PER_CACHELINE)))
+ rte_prefetch0((void *)(uintptr_t)&descs[avail_idx +
+ PACKED_DESC_PER_CACHELINE]);
+ else
+ rte_prefetch0((void *)(uintptr_t)&descs[0]);
+
+ if (unlikely(!desc_is_avail(&descs[avail_idx], wrap_counter)) |
+ unlikely(!desc_is_avail(&descs[avail_idx + 1], wrap_counter)) |
+ unlikely(!desc_is_avail(&descs[avail_idx + 2], wrap_counter)) |
+ unlikely(!desc_is_avail(&descs[avail_idx + 3], wrap_counter)))
+ return 1;
+
+ if (unlikely((descs[avail_idx].flags & VIRTIO_DESC_NORMAL_FLAG) |
+ (descs[avail_idx + 1].flags & VIRTIO_DESC_NORMAL_FLAG) |
+ (descs[avail_idx + 2].flags & VIRTIO_DESC_NORMAL_FLAG) |
+ (descs[avail_idx + 3].flags & VIRTIO_DESC_NORMAL_FLAG)))
+ return -1;
+
+ rte_smp_rmb();
+
+ len = descs[avail_idx].len;
+ len1 = descs[avail_idx + 1].len;
+ len2 = descs[avail_idx + 2].len;
+ len3 = descs[avail_idx + 3].len;
+
+ ids[0] = descs[avail_idx].id;
+ ids[1] = descs[avail_idx + 1].id;
+ ids[2] = descs[avail_idx + 2].id;
+ ids[3] = descs[avail_idx + 3].id;
+
+ desc_addr[0] = vhost_iova_to_vva(dev, vq,
+ descs[avail_idx].addr,
+ &len,
+ VHOST_ACCESS_RW);
+
+ desc_addr[1] = vhost_iova_to_vva(dev, vq,
+ descs[avail_idx + 1].addr,
+ &len1,
+ VHOST_ACCESS_RW);
+
+ desc_addr[2] = vhost_iova_to_vva(dev, vq,
+ descs[avail_idx + 2].addr,
+ &len2,
+ VHOST_ACCESS_RW);
+
+ desc_addr[3] = vhost_iova_to_vva(dev, vq,
+ descs[avail_idx + 3].addr,
+ &len3,
+ VHOST_ACCESS_RW);
+
+ if (unlikely((len != descs[avail_idx].len) |
+ (len1 != descs[avail_idx + 1].len) |
+ (len2 != descs[avail_idx + 2].len) |
+ (len3 != descs[avail_idx + 3].len))) {
+ return -1;
+ }
+
+ if (rte_pktmbuf_alloc_bulk(mbuf_pool, pkts ,4))
+ return -1;
+
+ if (unlikely(((uint64_t)(pkts[0]->buf_len - pkts[0]->data_off) <
+ (len + buf_offset)) |
+ ((uint64_t)(pkts[1]->buf_len - pkts[1]->data_off) <
+ (len1 + buf_offset)) |
+ ((uint64_t)(pkts[2]->buf_len - pkts[2]->data_off) <
+ (len2 + buf_offset)) |
+ ((uint64_t)(pkts[3]->buf_len - pkts[3]->data_off) <
+ (len3 + buf_offset)))) {
+ rte_pktmbuf_free(pkts[0]);
+ rte_pktmbuf_free(pkts[1]);
+ rte_pktmbuf_free(pkts[2]);
+ rte_pktmbuf_free(pkts[3]);
+ return -1;
+ }
+
+ pkts[0]->pkt_len = descs[avail_idx].len - buf_offset;
+ pkts[1]->pkt_len = descs[avail_idx + 1].len - buf_offset;
+ pkts[2]->pkt_len = descs[avail_idx + 2].len - buf_offset;
+ pkts[3]->pkt_len = descs[avail_idx + 3].len - buf_offset;
+
+ pkts[0]->data_len = pkts[0]->pkt_len;
+ pkts[1]->data_len = pkts[1]->pkt_len;
+ pkts[2]->data_len = pkts[2]->pkt_len;
+ pkts[3]->data_len = pkts[3]->pkt_len;
+
+ return 0;
+}
+
+static __rte_always_inline int
+virtio_dev_tx_fast_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
+ struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts)
+{
+ uint16_t avail_idx = vq->last_avail_idx;
+ uint32_t buf_offset = dev->vhost_hlen;
+ uintptr_t desc_addr[4];
+ uint16_t ids[4];
+ int ret;
+ struct virtio_net_hdr *hdr, *hdr1, *hdr2, *hdr3;
+
+ ret = vhost_dequeue_fast_packed(dev, vq, mbuf_pool, pkts, avail_idx,
+ desc_addr, ids);
+
+ if (ret)
+ return ret;
+
+ rte_memcpy(rte_pktmbuf_mtod_offset(pkts[0], void *, 0),
+ (void *)(uintptr_t)(desc_addr[0] + buf_offset),
+ pkts[0]->pkt_len);
+ rte_memcpy(rte_pktmbuf_mtod_offset(pkts[1], void *, 0),
+ (void *)(uintptr_t)(desc_addr[1] + buf_offset),
+ pkts[1]->pkt_len);
+ rte_memcpy(rte_pktmbuf_mtod_offset(pkts[2], void *, 0),
+ (void *)(uintptr_t)(desc_addr[2] + buf_offset),
+ pkts[2]->pkt_len);
+ rte_memcpy(rte_pktmbuf_mtod_offset(pkts[3], void *, 0),
+ (void *)(uintptr_t)(desc_addr[3] + buf_offset),
+ pkts[3]->pkt_len);
+
+ if (virtio_net_with_host_offload(dev)) {
+ hdr = (struct virtio_net_hdr *)((uintptr_t)desc_addr[0]);
+ hdr1 = (struct virtio_net_hdr *)((uintptr_t)desc_addr[1]);
+ hdr2 = (struct virtio_net_hdr *)((uintptr_t)desc_addr[2]);
+ hdr3 = (struct virtio_net_hdr *)((uintptr_t)desc_addr[3]);
+ vhost_dequeue_offload(hdr, pkts[0]);
+ vhost_dequeue_offload(hdr1, pkts[1]);
+ vhost_dequeue_offload(hdr2, pkts[2]);
+ vhost_dequeue_offload(hdr3, pkts[3]);
+ }
+
+ vq->last_avail_idx += PACKED_DESC_PER_CACHELINE;
+ if (vq->last_avail_idx >= vq->size) {
+ vq->last_avail_idx -= vq->size;
+ vq->avail_wrap_counter ^= 1;
+ }
+ return 0;
+}
+
static __rte_always_inline int
vhost_dequeue_normal_packed(struct virtio_net *dev,
struct vhost_virtqueue *vq,
--
2.17.1
