[dpdk-dev] [PATCH] ENIC PMD receive path performance improvements.

johndale johndale at cisco.com
Tue Feb 16 06:36:49 CET 2016


- Simplify and reduce code path length of receive function.
- Put all receive stuff in enic_rx.c
- Reduce the number of posted_index updates (pay attention to rx_free_thresh)
- Remove the container structure around the rq mbuf ring
- Prefetch next Mbuf and descriptors while processing the current one
- Lookup table for converting CQ flags to mbuf flags.

Signed-off-by: johndale <johndale at cisco.com>
---
 drivers/net/enic/Makefile       |   1 +
 drivers/net/enic/base/vnic_rq.c |  99 ++--------
 drivers/net/enic/base/vnic_rq.h | 147 +-------------
 drivers/net/enic/enic.h         |  16 +-
 drivers/net/enic/enic_ethdev.c  |  21 +-
 drivers/net/enic/enic_main.c    | 319 +++++++++----------------------
 drivers/net/enic/enic_res.h     |  16 +-
 drivers/net/enic/enic_rx.c      | 413 ++++++++++++++++++++++++++++++++++++++++
 8 files changed, 555 insertions(+), 477 deletions(-)
 create mode 100644 drivers/net/enic/enic_rx.c

diff --git a/drivers/net/enic/Makefile b/drivers/net/enic/Makefile
index f0ee093..f316274 100644
--- a/drivers/net/enic/Makefile
+++ b/drivers/net/enic/Makefile
@@ -53,6 +53,7 @@ VPATH += $(SRCDIR)/src
 #
 SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_main.c
+SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_rx.c
 SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_clsf.c
 SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_res.c
 SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += base/vnic_cq.c
diff --git a/drivers/net/enic/base/vnic_rq.c b/drivers/net/enic/base/vnic_rq.c
index 1441604..cb62c5e 100644
--- a/drivers/net/enic/base/vnic_rq.c
+++ b/drivers/net/enic/base/vnic_rq.c
@@ -35,77 +35,21 @@
 #include "vnic_dev.h"
 #include "vnic_rq.h"
 
-static int vnic_rq_alloc_bufs(struct vnic_rq *rq)
-{
-	struct vnic_rq_buf *buf;
-	unsigned int i, j, count = rq->ring.desc_count;
-	unsigned int blks = VNIC_RQ_BUF_BLKS_NEEDED(count);
-
-	for (i = 0; i < blks; i++) {
-		rq->bufs[i] = kzalloc(VNIC_RQ_BUF_BLK_SZ(count), GFP_ATOMIC);
-		if (!rq->bufs[i])
-			return -ENOMEM;
-	}
-
-	for (i = 0; i < blks; i++) {
-		buf = rq->bufs[i];
-		for (j = 0; j < VNIC_RQ_BUF_BLK_ENTRIES(count); j++) {
-			buf->index = i * VNIC_RQ_BUF_BLK_ENTRIES(count) + j;
-			buf->desc = (u8 *)rq->ring.descs +
-				rq->ring.desc_size * buf->index;
-			if (buf->index + 1 == count) {
-				buf->next = rq->bufs[0];
-				break;
-			} else if (j + 1 == VNIC_RQ_BUF_BLK_ENTRIES(count)) {
-				buf->next = rq->bufs[i + 1];
-			} else {
-				buf->next = buf + 1;
-				buf++;
-			}
-		}
-	}
-
-	rq->to_use = rq->to_clean = rq->bufs[0];
-
-	return 0;
-}
-
-int vnic_rq_mem_size(struct vnic_rq *rq, unsigned int desc_count,
-	unsigned int desc_size)
-{
-	int mem_size = 0;
-
-	mem_size += vnic_dev_desc_ring_size(&rq->ring, desc_count, desc_size);
-
-	mem_size += VNIC_RQ_BUF_BLKS_NEEDED(rq->ring.desc_count) *
-		VNIC_RQ_BUF_BLK_SZ(rq->ring.desc_count);
-
-	return mem_size;
-}
-
 void vnic_rq_free(struct vnic_rq *rq)
 {
 	struct vnic_dev *vdev;
-	unsigned int i;
 
 	vdev = rq->vdev;
 
 	vnic_dev_free_desc_ring(vdev, &rq->ring);
 
-	for (i = 0; i < VNIC_RQ_BUF_BLKS_MAX; i++) {
-		if (rq->bufs[i]) {
-			kfree(rq->bufs[i]);
-			rq->bufs[i] = NULL;
-		}
-	}
-
 	rq->ctrl = NULL;
 }
 
 int vnic_rq_alloc(struct vnic_dev *vdev, struct vnic_rq *rq, unsigned int index,
 	unsigned int desc_count, unsigned int desc_size)
 {
-	int err;
+	int rc;
 	char res_name[NAME_MAX];
 	static int instance;
 
@@ -121,18 +65,9 @@ int vnic_rq_alloc(struct vnic_dev *vdev, struct vnic_rq *rq, unsigned int index,
 	vnic_rq_disable(rq);
 
 	snprintf(res_name, sizeof(res_name), "%d-rq-%d", instance++, index);
-	err = vnic_dev_alloc_desc_ring(vdev, &rq->ring, desc_count, desc_size,
+	rc = vnic_dev_alloc_desc_ring(vdev, &rq->ring, desc_count, desc_size,
 		rq->socket_id, res_name);
-	if (err)
-		return err;
-
-	err = vnic_rq_alloc_bufs(rq);
-	if (err) {
-		vnic_rq_free(rq);
-		return err;
-	}
-
-	return 0;
+	return rc;
 }
 
 void vnic_rq_init_start(struct vnic_rq *rq, unsigned int cq_index,
@@ -154,9 +89,6 @@ void vnic_rq_init_start(struct vnic_rq *rq, unsigned int cq_index,
 	iowrite32(fetch_index, &rq->ctrl->fetch_index);
 	iowrite32(posted_index, &rq->ctrl->posted_index);
 
-	rq->to_use = rq->to_clean =
-		&rq->bufs[fetch_index / VNIC_RQ_BUF_BLK_ENTRIES(count)]
-			[fetch_index % VNIC_RQ_BUF_BLK_ENTRIES(count)];
 }
 
 void vnic_rq_init(struct vnic_rq *rq, unsigned int cq_index,
@@ -176,6 +108,8 @@ void vnic_rq_init(struct vnic_rq *rq, unsigned int cq_index,
 		fetch_index, fetch_index,
 		error_interrupt_enable,
 		error_interrupt_offset);
+	rq->rxst_idx = 0;
+	rq->tot_pkts = 0;
 }
 
 void vnic_rq_error_out(struct vnic_rq *rq, unsigned int error)
@@ -212,21 +146,20 @@ int vnic_rq_disable(struct vnic_rq *rq)
 }
 
 void vnic_rq_clean(struct vnic_rq *rq,
-	void (*buf_clean)(struct vnic_rq *rq, struct vnic_rq_buf *buf))
+	void (*buf_clean)(struct rte_mbuf **buf))
 {
-	struct vnic_rq_buf *buf;
-	u32 fetch_index;
+	struct rte_mbuf **buf;
+	u32 fetch_index, i;
 	unsigned int count = rq->ring.desc_count;
 
-	buf = rq->to_clean;
-
-	while (vnic_rq_desc_used(rq) > 0) {
+	buf = &rq->mbuf_ring[0];
 
-		(*buf_clean)(rq, buf);
-
-		buf = rq->to_clean = buf->next;
-		rq->ring.desc_avail++;
+	for (i = 0; i < count; i++) {
+		(*buf_clean)(buf);
+		buf++;
 	}
+	rq->ring.desc_avail = count - 1;
+	rq->rx_nb_hold = 0;
 
 	/* Use current fetch_index as the ring starting point */
 	fetch_index = ioread32(&rq->ctrl->fetch_index);
@@ -235,9 +168,7 @@ void vnic_rq_clean(struct vnic_rq *rq,
 		/* Hardware surprise removal: reset fetch_index */
 		fetch_index = 0;
 	}
-	rq->to_use = rq->to_clean =
-		&rq->bufs[fetch_index / VNIC_RQ_BUF_BLK_ENTRIES(count)]
-			[fetch_index % VNIC_RQ_BUF_BLK_ENTRIES(count)];
+
 	iowrite32(fetch_index, &rq->ctrl->posted_index);
 
 	vnic_dev_clear_desc_ring(&rq->ring);
diff --git a/drivers/net/enic/base/vnic_rq.h b/drivers/net/enic/base/vnic_rq.h
index 0f5c3c1..e083ccc 100644
--- a/drivers/net/enic/base/vnic_rq.h
+++ b/drivers/net/enic/base/vnic_rq.h
@@ -66,42 +66,22 @@ struct vnic_rq_ctrl {
 	u32 pad10;
 };
 
-/* Break the vnic_rq_buf allocations into blocks of 32/64 entries */
-#define VNIC_RQ_BUF_MIN_BLK_ENTRIES 32
-#define VNIC_RQ_BUF_DFLT_BLK_ENTRIES 64
-#define VNIC_RQ_BUF_BLK_ENTRIES(entries) \
-	((unsigned int)((entries < VNIC_RQ_BUF_DFLT_BLK_ENTRIES) ? \
-	VNIC_RQ_BUF_MIN_BLK_ENTRIES : VNIC_RQ_BUF_DFLT_BLK_ENTRIES))
-#define VNIC_RQ_BUF_BLK_SZ(entries) \
-	(VNIC_RQ_BUF_BLK_ENTRIES(entries) * sizeof(struct vnic_rq_buf))
-#define VNIC_RQ_BUF_BLKS_NEEDED(entries) \
-	DIV_ROUND_UP(entries, VNIC_RQ_BUF_BLK_ENTRIES(entries))
-#define VNIC_RQ_BUF_BLKS_MAX VNIC_RQ_BUF_BLKS_NEEDED(4096)
-
-struct vnic_rq_buf {
-	struct vnic_rq_buf *next;
-	dma_addr_t dma_addr;
-	void *os_buf;
-	unsigned int os_buf_index;
-	unsigned int len;
-	unsigned int index;
-	void *desc;
-	uint64_t wr_id;
-};
-
 struct vnic_rq {
 	unsigned int index;
+	unsigned int posted_index;
 	struct vnic_dev *vdev;
-	struct vnic_rq_ctrl __iomem *ctrl;              /* memory-mapped */
+	struct vnic_rq_ctrl __iomem *ctrl;	/* memory-mapped */
 	struct vnic_dev_ring ring;
-	struct vnic_rq_buf *bufs[VNIC_RQ_BUF_BLKS_MAX];
-	struct vnic_rq_buf *to_use;
-	struct vnic_rq_buf *to_clean;
+	struct rte_mbuf **mbuf_ring;		/* array of allocated mbufs */
+	unsigned int mbuf_next_idx;		/* next mb to consume */
 	void *os_buf_head;
 	unsigned int pkts_outstanding;
-
+	uint16_t rx_nb_hold;
+	uint16_t rx_free_thresh;
 	unsigned int socket_id;
 	struct rte_mempool *mp;
+	uint16_t rxst_idx;
+	uint32_t tot_pkts;
 };
 
 static inline unsigned int vnic_rq_desc_avail(struct vnic_rq *rq)
@@ -116,119 +96,13 @@ static inline unsigned int vnic_rq_desc_used(struct vnic_rq *rq)
 	return rq->ring.desc_count - rq->ring.desc_avail - 1;
 }
 
-static inline void *vnic_rq_next_desc(struct vnic_rq *rq)
-{
-	return rq->to_use->desc;
-}
-
-static inline unsigned int vnic_rq_next_index(struct vnic_rq *rq)
-{
-	return rq->to_use->index;
-}
-
-static inline void vnic_rq_post(struct vnic_rq *rq,
-	void *os_buf, unsigned int os_buf_index,
-	dma_addr_t dma_addr, unsigned int len,
-	uint64_t wrid)
-{
-	struct vnic_rq_buf *buf = rq->to_use;
-
-	buf->os_buf = os_buf;
-	buf->os_buf_index = os_buf_index;
-	buf->dma_addr = dma_addr;
-	buf->len = len;
-	buf->wr_id = wrid;
-
-	buf = buf->next;
-	rq->to_use = buf;
-	rq->ring.desc_avail--;
-
-	/* Move the posted_index every nth descriptor
-	 */
-
-#ifndef VNIC_RQ_RETURN_RATE
-#define VNIC_RQ_RETURN_RATE		0xf	/* keep 2^n - 1 */
-#endif
-
-	if ((buf->index & VNIC_RQ_RETURN_RATE) == 0) {
-		/* Adding write memory barrier prevents compiler and/or CPU
-		 * reordering, thus avoiding descriptor posting before
-		 * descriptor is initialized. Otherwise, hardware can read
-		 * stale descriptor fields.
-		 */
-		wmb();
-		iowrite32(buf->index, &rq->ctrl->posted_index);
-	}
-}
-
-static inline void vnic_rq_post_commit(struct vnic_rq *rq,
-	void *os_buf, unsigned int os_buf_index,
-	dma_addr_t dma_addr, unsigned int len)
-{
-	struct vnic_rq_buf *buf = rq->to_use;
-
-	buf->os_buf = os_buf;
-	buf->os_buf_index = os_buf_index;
-	buf->dma_addr = dma_addr;
-	buf->len = len;
-
-	buf = buf->next;
-	rq->to_use = buf;
-	rq->ring.desc_avail--;
-
-	/* Move the posted_index every descriptor
-	 */
-
-	/* Adding write memory barrier prevents compiler and/or CPU
-	 * reordering, thus avoiding descriptor posting before
-	 * descriptor is initialized. Otherwise, hardware can read
-	 * stale descriptor fields.
-	 */
-	wmb();
-	iowrite32(buf->index, &rq->ctrl->posted_index);
-}
 
-static inline void vnic_rq_return_descs(struct vnic_rq *rq, unsigned int count)
-{
-	rq->ring.desc_avail += count;
-}
 
 enum desc_return_options {
 	VNIC_RQ_RETURN_DESC,
 	VNIC_RQ_DEFER_RETURN_DESC,
 };
 
-static inline int vnic_rq_service(struct vnic_rq *rq,
-	struct cq_desc *cq_desc, u16 completed_index,
-	int desc_return, int (*buf_service)(struct vnic_rq *rq,
-	struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
-	int skipped, void *opaque), void *opaque)
-{
-	struct vnic_rq_buf *buf;
-	int skipped;
-	int eop = 0;
-
-	buf = rq->to_clean;
-	while (1) {
-
-		skipped = (buf->index != completed_index);
-
-		if ((*buf_service)(rq, cq_desc, buf, skipped, opaque))
-			eop++;
-
-		if (desc_return == VNIC_RQ_RETURN_DESC)
-			rq->ring.desc_avail++;
-
-		rq->to_clean = buf->next;
-
-		if (!skipped)
-			break;
-
-		buf = rq->to_clean;
-	}
-	return eop;
-}
-
 static inline int vnic_rq_fill(struct vnic_rq *rq,
 	int (*buf_fill)(struct vnic_rq *rq))
 {
@@ -274,8 +148,5 @@ unsigned int vnic_rq_error_status(struct vnic_rq *rq);
 void vnic_rq_enable(struct vnic_rq *rq);
 int vnic_rq_disable(struct vnic_rq *rq);
 void vnic_rq_clean(struct vnic_rq *rq,
-	void (*buf_clean)(struct vnic_rq *rq, struct vnic_rq_buf *buf));
-int vnic_rq_mem_size(struct vnic_rq *rq, unsigned int desc_count,
-	unsigned int desc_size);
-
+	void (*buf_clean)(struct rte_mbuf **buf));
 #endif /* _VNIC_RQ_H_ */
diff --git a/drivers/net/enic/enic.h b/drivers/net/enic/enic.h
index 9e78305..8c914f5 100644
--- a/drivers/net/enic/enic.h
+++ b/drivers/net/enic/enic.h
@@ -45,6 +45,7 @@
 #include "vnic_nic.h"
 #include "vnic_rss.h"
 #include "enic_res.h"
+#include "cq_enet_desc.h"
 
 #define DRV_NAME		"enic_pmd"
 #define DRV_DESCRIPTION		"Cisco VIC Ethernet NIC Poll-mode Driver"
@@ -154,6 +155,16 @@ static inline struct enic *pmd_priv(struct rte_eth_dev *eth_dev)
 	return (struct enic *)eth_dev->data->dev_private;
 }
 
+#define RTE_LIBRTE_ENIC_ASSERT_ENABLE
+#ifdef RTE_LIBRTE_ENIC_ASSERT_ENABLE
+#define ASSERT(x) do {			\
+	if (!(x))			\
+		rte_panic("ENIC: x");	\
+} while (0)
+#else
+#define ASSERT(x)
+#endif
+
 extern void enic_fdir_stats_get(struct enic *enic,
 	struct rte_eth_fdir_stats *stats);
 extern int enic_fdir_add_fltr(struct enic *enic,
@@ -193,9 +204,10 @@ extern void enic_send_pkt(struct enic *enic, struct vnic_wq *wq,
 			  uint16_t ol_flags, uint16_t vlan_tag);
 
 extern void enic_post_wq_index(struct vnic_wq *wq);
-extern int enic_poll(struct vnic_rq *rq, struct rte_mbuf **rx_pkts,
-	unsigned int budget, unsigned int *work_done);
 extern int enic_probe(struct enic *enic);
 extern int enic_clsf_init(struct enic *enic);
 extern void enic_clsf_destroy(struct enic *enic);
+uint16_t enic_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
+			uint16_t nb_pkts);
+
 #endif /* _ENIC_H_ */
diff --git a/drivers/net/enic/enic_ethdev.c b/drivers/net/enic/enic_ethdev.c
index 2a88043..e54919b 100644
--- a/drivers/net/enic/enic_ethdev.c
+++ b/drivers/net/enic/enic_ethdev.c
@@ -270,6 +270,10 @@ static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
 		return ret;
 	}
 
+	enic->rq[queue_idx].rx_free_thresh = rx_conf->rx_free_thresh;
+	dev_debug(enic, "Set queue_id:%u free thresh:%u\n", queue_idx,
+			enic->rq[queue_idx].rx_free_thresh);
+
 	return enicpmd_dev_setup_intr(enic);
 }
 
@@ -429,6 +433,9 @@ static void enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
 		DEV_TX_OFFLOAD_IPV4_CKSUM  |
 		DEV_TX_OFFLOAD_UDP_CKSUM   |
 		DEV_TX_OFFLOAD_TCP_CKSUM;
+	device_info->default_rxconf = (struct rte_eth_rxconf) {
+		.rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
+	};
 }
 
 static void enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
@@ -538,18 +545,6 @@ static uint16_t enicpmd_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 	return index;
 }
 
-static uint16_t enicpmd_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
-	uint16_t nb_pkts)
-{
-	struct vnic_rq *rq = (struct vnic_rq *)rx_queue;
-	unsigned int work_done;
-
-	if (enic_poll(rq, rx_pkts, (unsigned int)nb_pkts, &work_done))
-		dev_err(enic, "error in enicpmd poll\n");
-
-	return work_done;
-}
-
 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
 	.dev_configure        = enicpmd_dev_configure,
 	.dev_start            = enicpmd_dev_start,
@@ -606,7 +601,7 @@ static int eth_enicpmd_dev_init(struct rte_eth_dev *eth_dev)
 	enic->port_id = eth_dev->data->port_id;
 	enic->rte_dev = eth_dev;
 	eth_dev->dev_ops = &enicpmd_eth_dev_ops;
-	eth_dev->rx_pkt_burst = &enicpmd_recv_pkts;
+	eth_dev->rx_pkt_burst = &enic_recv_pkts;
 	eth_dev->tx_pkt_burst = &enicpmd_xmit_pkts;
 
 	pdev = eth_dev->pci_dev;
diff --git a/drivers/net/enic/enic_main.c b/drivers/net/enic/enic_main.c
index 07a9810..d897fcf 100644
--- a/drivers/net/enic/enic_main.c
+++ b/drivers/net/enic/enic_main.c
@@ -60,6 +60,17 @@
 #include "vnic_nic.h"
 #include "enic_vnic_wq.h"
 
+static inline struct rte_mbuf *
+rte_rxmbuf_alloc(struct rte_mempool *mp)
+{
+	struct rte_mbuf *m;
+
+	m = __rte_mbuf_raw_alloc(mp);
+	__rte_mbuf_sanity_check_raw(m, 0);
+	return m;
+}
+
+
 static inline int enic_is_sriov_vf(struct enic *enic)
 {
 	return enic->pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
@@ -90,6 +101,25 @@ enic_rxmbuf_alloc(struct rte_mempool *mp)
 	return m;
 }
 
+static void
+enic_rxmbuf_queue_release(struct enic *enic, struct vnic_rq *rq)
+{
+	uint16_t i;
+
+	if (!rq || !rq->mbuf_ring) {
+		dev_debug(enic, "Pointer to rq or mbuf_ring is NULL");
+		return;
+	}
+
+	for (i = 0; i < enic->config.rq_desc_count; i++) {
+		if (rq->mbuf_ring[i]) {
+			rte_pktmbuf_free_seg(rq->mbuf_ring[i]);
+			rq->mbuf_ring[i] = NULL;
+		}
+	}
+}
+
+
 void enic_set_hdr_split_size(struct enic *enic, u16 split_hdr_size)
 {
 	vnic_set_hdr_split_size(enic->vdev, split_hdr_size);
@@ -262,13 +292,13 @@ void enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
 }
 
 static void
-enic_free_rq_buf(__rte_unused struct vnic_rq *rq, struct vnic_rq_buf *buf)
+enic_free_rq_buf(struct rte_mbuf **mbuf)
 {
-	if (!buf->os_buf)
+	if (*mbuf == NULL)
 		return;
 
-	rte_pktmbuf_free((struct rte_mbuf *)buf->os_buf);
-	buf->os_buf = NULL;
+	rte_pktmbuf_free(*mbuf);
+	mbuf = NULL;
 }
 
 void enic_init_vnic_resources(struct enic *enic)
@@ -314,221 +344,47 @@ void enic_init_vnic_resources(struct enic *enic)
 }
 
 
-static int enic_rq_alloc_buf(struct vnic_rq *rq)
+static int
+enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
 {
-	struct enic *enic = vnic_dev_priv(rq->vdev);
+	struct rte_mbuf *mb;
+	struct rq_enet_desc *rqd = rq->ring.descs;
+	unsigned i;
 	dma_addr_t dma_addr;
-	struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
-	uint8_t type = RQ_ENET_TYPE_ONLY_SOP;
-	u16 split_hdr_size = vnic_get_hdr_split_size(enic->vdev);
-	struct rte_mbuf *mbuf = enic_rxmbuf_alloc(rq->mp);
-	struct rte_mbuf *hdr_mbuf = NULL;
-
-	if (!mbuf) {
-		dev_err(enic, "mbuf alloc in enic_rq_alloc_buf failed\n");
-		return -1;
-	}
-
-	if (unlikely(split_hdr_size)) {
-		if (vnic_rq_desc_avail(rq) < 2) {
-			rte_mempool_put(mbuf->pool, mbuf);
-			return -1;
-		}
-		hdr_mbuf = enic_rxmbuf_alloc(rq->mp);
-		if (!hdr_mbuf) {
-			rte_mempool_put(mbuf->pool, mbuf);
-			dev_err(enic,
-				"hdr_mbuf alloc in enic_rq_alloc_buf failed\n");
-			return -1;
-		}
-
-		hdr_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
-
-		hdr_mbuf->nb_segs = 2;
-		hdr_mbuf->port = enic->port_id;
-		hdr_mbuf->next = mbuf;
-
-		dma_addr = (dma_addr_t)
-		    (hdr_mbuf->buf_physaddr + hdr_mbuf->data_off);
-
-		rq_enet_desc_enc(desc, dma_addr, type, split_hdr_size);
-
-		vnic_rq_post(rq, (void *)hdr_mbuf, 0 /*os_buf_index*/, dma_addr,
-			(unsigned int)split_hdr_size, 0 /*wrid*/);
-
-		desc = vnic_rq_next_desc(rq);
-		type = RQ_ENET_TYPE_NOT_SOP;
-	} else {
-		mbuf->nb_segs = 1;
-		mbuf->port = enic->port_id;
-	}
-
-	mbuf->data_off = RTE_PKTMBUF_HEADROOM;
-	mbuf->next = NULL;
-
-	dma_addr = (dma_addr_t)
-	    (mbuf->buf_physaddr + mbuf->data_off);
-
-	rq_enet_desc_enc(desc, dma_addr, type, mbuf->buf_len);
-
-	vnic_rq_post(rq, (void *)mbuf, 0 /*os_buf_index*/, dma_addr,
-		(unsigned int)mbuf->buf_len, 0 /*wrid*/);
-
-	return 0;
-}
-
-static int enic_rq_indicate_buf(struct vnic_rq *rq,
-	struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
-	int skipped, void *opaque)
-{
-	struct enic *enic = vnic_dev_priv(rq->vdev);
-	struct rte_mbuf **rx_pkt_bucket = (struct rte_mbuf **)opaque;
-	struct rte_mbuf *rx_pkt = NULL;
-	struct rte_mbuf *hdr_rx_pkt = NULL;
-
-	u8 type, color, eop, sop, ingress_port, vlan_stripped;
-	u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
-	u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
-	u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
-	u8 packet_error;
-	u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
-	u32 rss_hash;
-
-	cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
-		&type, &color, &q_number, &completed_index,
-		&ingress_port, &fcoe, &eop, &sop, &rss_type,
-		&csum_not_calc, &rss_hash, &bytes_written,
-		&packet_error, &vlan_stripped, &vlan_tci, &checksum,
-		&fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
-		&fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
-		&ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
-		&fcs_ok);
-
-	rx_pkt = (struct rte_mbuf *)buf->os_buf;
-	buf->os_buf = NULL;
 
-	if (unlikely(packet_error)) {
-		dev_err(enic, "packet error\n");
-		rx_pkt->data_len = 0;
-		return 0;
-	}
+	dev_debug(enic, "queue %u, allocating %u rx queue mbufs", rq->index,
+		  rq->ring.desc_count);
 
-	if (unlikely(skipped)) {
-		rx_pkt->data_len = 0;
-		return 0;
-	}
-
-	if (likely(!vnic_get_hdr_split_size(enic->vdev))) {
-		/* No header split configured */
-		*rx_pkt_bucket = rx_pkt;
-		rx_pkt->pkt_len = bytes_written;
-
-		if (ipv4) {
-			rx_pkt->packet_type = RTE_PTYPE_L3_IPV4;
-			if (!csum_not_calc) {
-				if (unlikely(!ipv4_csum_ok))
-					rx_pkt->ol_flags |= PKT_RX_IP_CKSUM_BAD;
-
-				if ((tcp || udp) && (!tcp_udp_csum_ok))
-					rx_pkt->ol_flags |= PKT_RX_L4_CKSUM_BAD;
-			}
-		} else if (ipv6)
-			rx_pkt->packet_type = RTE_PTYPE_L3_IPV6;
-	} else {
-		/* Header split */
-		if (sop && !eop) {
-			/* This piece is header */
-			*rx_pkt_bucket = rx_pkt;
-			rx_pkt->pkt_len = bytes_written;
-		} else {
-			if (sop && eop) {
-				/* The packet is smaller than split_hdr_size */
-				*rx_pkt_bucket = rx_pkt;
-				rx_pkt->pkt_len = bytes_written;
-				if (ipv4) {
-					rx_pkt->packet_type = RTE_PTYPE_L3_IPV4;
-					if (!csum_not_calc) {
-						if (unlikely(!ipv4_csum_ok))
-							rx_pkt->ol_flags |=
-							    PKT_RX_IP_CKSUM_BAD;
-
-						if ((tcp || udp) &&
-						    (!tcp_udp_csum_ok))
-							rx_pkt->ol_flags |=
-							    PKT_RX_L4_CKSUM_BAD;
-					}
-				} else if (ipv6)
-					rx_pkt->packet_type = RTE_PTYPE_L3_IPV6;
-			} else {
-				/* Payload */
-				hdr_rx_pkt = *rx_pkt_bucket;
-				hdr_rx_pkt->pkt_len += bytes_written;
-				if (ipv4) {
-					hdr_rx_pkt->packet_type =
-						RTE_PTYPE_L3_IPV4;
-					if (!csum_not_calc) {
-						if (unlikely(!ipv4_csum_ok))
-							hdr_rx_pkt->ol_flags |=
-							    PKT_RX_IP_CKSUM_BAD;
-
-						if ((tcp || udp) &&
-						    (!tcp_udp_csum_ok))
-							hdr_rx_pkt->ol_flags |=
-							    PKT_RX_L4_CKSUM_BAD;
-					}
-				} else if (ipv6)
-					hdr_rx_pkt->packet_type =
-						RTE_PTYPE_L3_IPV6;
-			}
+	for (i = 0; i < rq->ring.desc_count; i++, rqd++) {
+		mb = rte_rxmbuf_alloc(rq->mp);
+		if (mb == NULL) {
+			dev_err(enic, "RX mbuf alloc failed queue_id=%u",
+			(unsigned)rq->index);
+			return -ENOMEM;
 		}
-	}
 
-	rx_pkt->data_len = bytes_written;
+		dma_addr = (dma_addr_t)(mb->buf_physaddr + mb->data_off);
 
-	if (rss_hash) {
-		rx_pkt->ol_flags |= PKT_RX_RSS_HASH;
-		rx_pkt->hash.rss = rss_hash;
+		rq_enet_desc_enc(rqd, dma_addr, RQ_ENET_TYPE_ONLY_SOP,
+				 mb->buf_len);
+		rq->mbuf_ring[i] = mb;
 	}
 
-	if (vlan_tci) {
-		rx_pkt->ol_flags |= PKT_RX_VLAN_PKT;
-		rx_pkt->vlan_tci = vlan_tci;
-	}
+	/* make sure all prior writes are complete before doing the PIO write */
+	rte_rmb();
 
-	return eop;
-}
+	/* Post all but the last 2 cache lines' worth of descriptors */
+	rq->posted_index = rq->ring.desc_count - (2 * RTE_CACHE_LINE_SIZE
+			/ sizeof(struct rq_enet_desc));
+	rq->rx_nb_hold = 0;
 
-static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
-	__rte_unused u8 type, u16 q_number, u16 completed_index, void *opaque)
-{
-	struct enic *enic = vnic_dev_priv(vdev);
-
-	return vnic_rq_service(&enic->rq[q_number], cq_desc,
-		completed_index, VNIC_RQ_RETURN_DESC,
-		enic_rq_indicate_buf, opaque);
-
-}
+	dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
+		enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
+	iowrite32(rq->posted_index, &rq->ctrl->posted_index);
+	rte_rmb();
 
-int enic_poll(struct vnic_rq *rq, struct rte_mbuf **rx_pkts,
-	unsigned int budget, unsigned int *work_done)
-{
-	struct enic *enic = vnic_dev_priv(rq->vdev);
-	unsigned int cq = enic_cq_rq(enic, rq->index);
-	int err = 0;
-
-	*work_done = vnic_cq_service(&enic->cq[cq],
-		budget, enic_rq_service, (void *)rx_pkts);
-
-	if (*work_done) {
-		vnic_rq_fill(rq, enic_rq_alloc_buf);
+	return 0;
 
-		/* Need at least one buffer on ring to get going */
-		if (vnic_rq_desc_used(rq) == 0) {
-			dev_err(enic, "Unable to alloc receive buffers\n");
-			err = -1;
-		}
-	}
-	return err;
 }
 
 static void *
@@ -576,6 +432,7 @@ enic_intr_handler(__rte_unused struct rte_intr_handle *handle,
 int enic_enable(struct enic *enic)
 {
 	unsigned int index;
+	int err;
 	struct rte_eth_dev *eth_dev = enic->rte_dev;
 
 	eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
@@ -586,15 +443,11 @@ int enic_enable(struct enic *enic)
 		dev_warning(enic, "Init of hash table for clsf failed."\
 			"Flow director feature will not work\n");
 
-	/* Fill RQ bufs */
 	for (index = 0; index < enic->rq_count; index++) {
-		vnic_rq_fill(&enic->rq[index], enic_rq_alloc_buf);
-
-		/* Need at least one buffer on ring to get going
-		*/
-		if (vnic_rq_desc_used(&enic->rq[index]) == 0) {
-			dev_err(enic, "Unable to alloc receive buffers\n");
-			return -1;
+		err = enic_alloc_rx_queue_mbufs(enic, &enic->rq[index]);
+		if (err) {
+			dev_err(enic, "Failed to alloc RX queue mbufs\n");
+			return err;
 		}
 	}
 
@@ -636,6 +489,9 @@ void enic_free_rq(void *rxq)
 	struct vnic_rq *rq = (struct vnic_rq *)rxq;
 	struct enic *enic = vnic_dev_priv(rq->vdev);
 
+	enic_rxmbuf_queue_release(enic, rq);
+	rte_free(rq->mbuf_ring);
+	rq->mbuf_ring = NULL;
 	vnic_rq_free(rq);
 	vnic_cq_free(&enic->cq[rq->index]);
 }
@@ -664,7 +520,7 @@ int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
 	unsigned int socket_id, struct rte_mempool *mp,
 	uint16_t nb_desc)
 {
-	int err;
+	int rc;
 	struct vnic_rq *rq = &enic->rq[queue_idx];
 
 	rq->socket_id = socket_id;
@@ -687,23 +543,35 @@ int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
 	}
 
 	/* Allocate queue resources */
-	err = vnic_rq_alloc(enic->vdev, &enic->rq[queue_idx], queue_idx,
-		enic->config.rq_desc_count,
-		sizeof(struct rq_enet_desc));
-	if (err) {
+	rc = vnic_rq_alloc(enic->vdev, rq, queue_idx,
+		enic->config.rq_desc_count, sizeof(struct rq_enet_desc));
+	if (rc) {
 		dev_err(enic, "error in allocation of rq\n");
-		return err;
+		goto err_exit;
 	}
 
-	err = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
+	rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
 		socket_id, enic->config.rq_desc_count,
 		sizeof(struct cq_enet_rq_desc));
-	if (err) {
-		vnic_rq_free(rq);
+	if (rc) {
 		dev_err(enic, "error in allocation of cq for rq\n");
+		goto err_free_rq_exit;
 	}
 
-	return err;
+	/* Allocate the mbuf ring */
+	rq->mbuf_ring = (struct rte_mbuf **)rte_zmalloc_socket("rq->mbuf_ring",
+			sizeof(struct rte_mbuf *) * enic->config.rq_desc_count,
+			RTE_CACHE_LINE_SIZE, rq->socket_id);
+
+	if (rq->mbuf_ring != NULL)
+		return 0;
+
+	/* cleanup on error */
+	vnic_cq_free(&enic->cq[queue_idx]);
+err_free_rq_exit:
+	vnic_rq_free(rq);
+err_exit:
+	return -ENOMEM;
 }
 
 void enic_free_wq(void *txq)
@@ -790,6 +658,7 @@ int enic_disable(struct enic *enic)
 
 	for (i = 0; i < enic->wq_count; i++)
 		vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
+
 	for (i = 0; i < enic->rq_count; i++)
 		vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
 	for (i = 0; i < enic->cq_count; i++)
@@ -1074,7 +943,7 @@ int enic_probe(struct enic *enic)
 
 	/* Set ingress vlan rewrite mode before vnic initialization */
 	err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
-		IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN);
+		IG_VLAN_REWRITE_MODE_PASS_THRU);
 	if (err) {
 		dev_err(enic,
 			"Failed to set ingress vlan rewrite mode, aborting.\n");
diff --git a/drivers/net/enic/enic_res.h b/drivers/net/enic/enic_res.h
index 49f7e22..33f2e84 100644
--- a/drivers/net/enic/enic_res.h
+++ b/drivers/net/enic/enic_res.h
@@ -52,6 +52,7 @@
 #define ENIC_UNICAST_PERFECT_FILTERS	32
 
 #define ENIC_NON_TSO_MAX_DESC		16
+#define ENIC_DEFAULT_RX_FREE_THRESH	32
 
 #define ENIC_SETTING(enic, f) ((enic->config.flags & VENETF_##f) ? 1 : 0)
 
@@ -133,21 +134,6 @@ static inline void enic_queue_wq_desc_tso(struct vnic_wq *wq,
 		WQ_ENET_OFFLOAD_MODE_TSO,
 		eop, 1 /* SOP */, eop, loopback);
 }
-static inline void enic_queue_rq_desc(struct vnic_rq *rq,
-	void *os_buf, unsigned int os_buf_index,
-	dma_addr_t dma_addr, unsigned int len)
-{
-	struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
-	u64 wrid = 0;
-	u8 type = os_buf_index ?
-		RQ_ENET_TYPE_NOT_SOP : RQ_ENET_TYPE_ONLY_SOP;
-
-	rq_enet_desc_enc(desc,
-		(u64)dma_addr | VNIC_PADDR_TARGET,
-		type, (u16)len);
-
-	vnic_rq_post(rq, os_buf, os_buf_index, dma_addr, len, wrid);
-}
 
 struct enic;
 
diff --git a/drivers/net/enic/enic_rx.c b/drivers/net/enic/enic_rx.c
new file mode 100644
index 0000000..45c2752
--- /dev/null
+++ b/drivers/net/enic/enic_rx.c
@@ -0,0 +1,413 @@
+/*
+ * Copyright 2008-2014 Cisco Systems, Inc.  All rights reserved.
+ * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
+ *
+ * Copyright (c) 2014, Cisco Systems, Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+ * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+#include <rte_mbuf.h>
+#include <rte_ethdev.h>
+#include <rte_prefetch.h>
+
+#include "enic_compat.h"
+#include "rq_enet_desc.h"
+#include "enic.h"
+
+#define RTE_PMD_USE_PREFETCH
+
+#ifdef RTE_PMD_USE_PREFETCH
+/*
+ * Prefetch a cache line into all cache levels.
+ */
+#define rte_enic_prefetch(p) rte_prefetch0(p)
+#else
+#define rte_enic_prefetch(p) do {} while (0)
+#endif
+
+#ifdef RTE_PMD_PACKET_PREFETCH
+#define rte_packet_prefetch(p) rte_prefetch1(p)
+#else
+#define rte_packet_prefetch(p) do {} while (0)
+#endif
+
+static inline struct rte_mbuf *
+rte_rxmbuf_alloc(struct rte_mempool *mp)
+{
+	struct rte_mbuf *m;
+
+	m = __rte_mbuf_raw_alloc(mp);
+	__rte_mbuf_sanity_check_raw(m, 0);
+	return m;
+}
+
+static inline uint16_t
+enic_cq_rx_desc_ciflags(struct cq_enet_rq_desc *crd)
+{
+	return le16_to_cpu(crd->completed_index_flags) & ~CQ_DESC_COMP_NDX_MASK;
+}
+
+static inline uint16_t
+enic_cq_rx_desc_bwflags(struct cq_enet_rq_desc *crd)
+{
+	return(le16_to_cpu(crd->bytes_written_flags) &
+		~CQ_ENET_RQ_DESC_BYTES_WRITTEN_MASK);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_packet_error(uint16_t bwflags)
+{
+	return((bwflags & CQ_ENET_RQ_DESC_FLAGS_TRUNCATED) ==
+		CQ_ENET_RQ_DESC_FLAGS_TRUNCATED);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_eop(uint16_t ciflags)
+{
+	return (ciflags & CQ_ENET_RQ_DESC_FLAGS_EOP)
+		== CQ_ENET_RQ_DESC_FLAGS_EOP;
+}
+
+static inline uint8_t
+enic_cq_rx_desc_csum_not_calc(struct cq_enet_rq_desc *cqrd)
+{
+	return ((le16_to_cpu(cqrd->q_number_rss_type_flags) &
+		CQ_ENET_RQ_DESC_FLAGS_CSUM_NOT_CALC) ==
+		CQ_ENET_RQ_DESC_FLAGS_CSUM_NOT_CALC);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_ipv4_csum_ok(struct cq_enet_rq_desc *cqrd)
+{
+	return ((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_IPV4_CSUM_OK) ==
+		CQ_ENET_RQ_DESC_FLAGS_IPV4_CSUM_OK);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_tcp_udp_csum_ok(struct cq_enet_rq_desc *cqrd)
+{
+	return((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_TCP_UDP_CSUM_OK) ==
+		CQ_ENET_RQ_DESC_FLAGS_TCP_UDP_CSUM_OK);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_rss_type(struct cq_enet_rq_desc *cqrd)
+{
+	return (uint8_t)((le16_to_cpu(cqrd->q_number_rss_type_flags) >>
+		CQ_DESC_Q_NUM_BITS) & CQ_ENET_RQ_DESC_RSS_TYPE_MASK);
+}
+
+static inline uint32_t
+enic_cq_rx_desc_rss_hash(struct cq_enet_rq_desc *cqrd)
+{
+	return le32_to_cpu(cqrd->rss_hash);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_ipv4(struct cq_enet_rq_desc *cqrd)
+{
+	return ((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_IPV4) ==
+		CQ_ENET_RQ_DESC_FLAGS_IPV4);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_ipv6(struct cq_enet_rq_desc *cqrd)
+{
+	return ((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_IPV6) ==
+		CQ_ENET_RQ_DESC_FLAGS_IPV6);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_udp(struct cq_enet_rq_desc *cqrd)
+{
+	return ((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_UDP)
+		== CQ_ENET_RQ_DESC_FLAGS_UDP);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_tcp(struct cq_enet_rq_desc *cqrd)
+{
+	return ((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_TCP)
+		== CQ_ENET_RQ_DESC_FLAGS_TCP);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_ipv4_frag(struct cq_enet_rq_desc *cqrd)
+{
+	return ((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_IPV4_FRAGMENT)
+		== CQ_ENET_RQ_DESC_FLAGS_IPV4_FRAGMENT);
+}
+
+static inline uint8_t
+enic_cq_rx_desc_fcs_ok(struct cq_enet_rq_desc *cqrd)
+{
+	return ((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_FCS_OK) ==
+		CQ_ENET_RQ_DESC_FLAGS_FCS_OK);
+}
+
+static inline uint16_t
+enic_cq_rx_desc_vlan(struct cq_enet_rq_desc *cqrd)
+{
+	return le16_to_cpu(cqrd->vlan);
+}
+
+static inline uint16_t
+enic_cq_rx_desc_n_bytes(struct cq_desc *cqd)
+{
+	struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
+	return le16_to_cpu(cqrd->bytes_written_flags) &
+		CQ_ENET_RQ_DESC_BYTES_WRITTEN_MASK;
+}
+
+static inline uint64_t
+enic_cq_rx_to_pkt_err_flags(struct cq_desc *cqd)
+{
+	struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
+	uint16_t bwflags;
+	uint64_t pkt_err_flags = 0;
+
+	bwflags = enic_cq_rx_desc_bwflags(cqrd);
+
+	/* Check for packet error. Can't be more specific than MAC error */
+	if (enic_cq_rx_desc_packet_error(bwflags)) {
+#ifdef RTE_NEXT_ABI
+		pkt_err_flags |= PKT_RX_MAC_ERR;
+#else
+		pkt_err_flags |= PKT_EXT_RX_PKT_ERROR;
+#endif
+	}
+
+	/* Check for bad FCS. MAC error isn't quite, but no other choice */
+	if (!enic_cq_rx_desc_fcs_ok(cqrd)) {
+#ifdef RTE_NEXT_ABI
+		pkt_err_flags |= PKT_RX_MAC_ERR;
+#else
+		pkt_err_flags |= PKT_EXT_RX_BAD_FCS;
+#endif
+	}
+	return pkt_err_flags;
+}
+
+/*
+ * Lookup table to translate RX CQ flags to mbuf flags.
+ */
+static inline uint32_t
+enic_cq_rx_flags_to_pkt_type(struct cq_desc *cqd)
+{
+	struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
+	uint8_t cqrd_flags = cqrd->flags;
+	static const uint32_t cq_type_table[128] __rte_cache_aligned = {
+		[32] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4,
+		[34] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4
+			| RTE_PTYPE_L4_UDP,
+		[36] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4
+			| RTE_PTYPE_L4_TCP,
+		[96] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4
+			| RTE_PTYPE_L4_FRAG,
+		[16] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6,
+		[18] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6
+			| RTE_PTYPE_L4_UDP,
+		[20] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6
+			| RTE_PTYPE_L4_TCP,
+		[80] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6
+			| RTE_PTYPE_L4_FRAG,
+		/* All others reserved */
+	};
+	cqrd_flags &= CQ_ENET_RQ_DESC_FLAGS_IPV4_FRAGMENT
+		| CQ_ENET_RQ_DESC_FLAGS_IPV4 | CQ_ENET_RQ_DESC_FLAGS_IPV6
+		| CQ_ENET_RQ_DESC_FLAGS_TCP | CQ_ENET_RQ_DESC_FLAGS_UDP;
+	return cq_type_table[cqrd_flags];
+}
+
+static inline void
+enic_cq_rx_to_pkt_flags(struct cq_desc *cqd, struct rte_mbuf *mbuf)
+{
+	struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
+	uint16_t ciflags, bwflags, pkt_flags = 0;
+	ciflags = enic_cq_rx_desc_ciflags(cqrd);
+	bwflags = enic_cq_rx_desc_bwflags(cqrd);
+
+	ASSERT(mbuf->ol_flags == 0);
+
+	/* flags are meaningless if !EOP */
+	if (unlikely(!enic_cq_rx_desc_eop(ciflags)))
+		goto mbuf_flags_done;
+
+	/* VLAN stripping */
+	if (bwflags & CQ_ENET_RQ_DESC_FLAGS_VLAN_STRIPPED) {
+		pkt_flags |= PKT_RX_VLAN_PKT;
+		mbuf->vlan_tci = enic_cq_rx_desc_vlan(cqrd);
+	} else {
+		mbuf->vlan_tci = 0;
+	}
+
+	/* RSS flag */
+	if (enic_cq_rx_desc_rss_type(cqrd)) {
+		pkt_flags |= PKT_RX_RSS_HASH;
+		mbuf->hash.rss = enic_cq_rx_desc_rss_hash(cqrd);
+	}
+
+	/* checksum flags */
+	if (!enic_cq_rx_desc_csum_not_calc(cqrd) &&
+		(mbuf->packet_type & RTE_PTYPE_L3_IPV4)) {
+		if (unlikely(!enic_cq_rx_desc_ipv4_csum_ok(cqrd)))
+			pkt_flags |= PKT_RX_IP_CKSUM_BAD;
+		if (mbuf->packet_type & (RTE_PTYPE_L4_UDP | RTE_PTYPE_L4_TCP)) {
+			if (unlikely(!enic_cq_rx_desc_tcp_udp_csum_ok(cqrd)))
+				pkt_flags |= PKT_RX_L4_CKSUM_BAD;
+		}
+	}
+
+ mbuf_flags_done:
+	mbuf->ol_flags = pkt_flags;
+}
+
+static inline uint32_t
+enic_ring_add(uint32_t n_descriptors, uint32_t i0, uint32_t i1)
+{
+	uint32_t d = i0 + i1;
+	ASSERT(i0 < n_descriptors);
+	ASSERT(i1 < n_descriptors);
+	d -= (d >= n_descriptors) ? n_descriptors : 0;
+	return d;
+}
+
+
+uint16_t
+enic_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
+	       uint16_t nb_pkts)
+{
+	struct vnic_rq *rq = rx_queue;
+	struct enic *enic = vnic_dev_priv(rq->vdev);
+	unsigned int rx_id;
+	struct rte_mbuf *nmb, *rxmb;
+	uint16_t nb_rx = 0;
+	uint16_t nb_hold;
+	struct vnic_cq *cq;
+	volatile struct cq_desc *cqd_ptr;
+	uint8_t color;
+
+	ASSERT(nb_pkts <= rq->ring.desc_count);
+
+	cq = &enic->cq[enic_cq_rq(enic, rq->index)];
+	rx_id = cq->to_clean;		/* index of cqd, rqd, mbuf_table */
+	cqd_ptr = (struct cq_desc *)(cq->ring.descs) + rx_id;
+
+	nb_hold = rq->rx_nb_hold;	/* mbufs held by software */
+
+	while (nb_rx < nb_pkts) {
+		uint16_t rx_pkt_len;
+		volatile struct rq_enet_desc *rqd_ptr;
+		dma_addr_t dma_addr;
+		struct cq_desc cqd;
+		uint64_t ol_err_flags;
+
+
+		color = (cqd_ptr->type_color >> CQ_DESC_COLOR_SHIFT)
+			& CQ_DESC_COLOR_MASK;
+		if (color == cq->last_color)
+			break;
+
+		/* get the cq descriptor and rq pointer */
+		cqd = *cqd_ptr;
+		rqd_ptr = (struct rq_enet_desc *)(rq->ring.descs) + rx_id;
+
+		/* allocate a new mbuf */
+		nmb = rte_rxmbuf_alloc(rq->mp);
+		if (nmb == NULL) {
+			dev_err(enic, "RX mbuf alloc failed port=%u qid=%u",
+			enic->port_id, (unsigned)rq->index);
+			rte_eth_devices[enic->port_id].
+					data->rx_mbuf_alloc_failed++;
+			break;
+		}
+
+		/* Check for FCS or packet errors */
+		ol_err_flags = enic_cq_rx_to_pkt_err_flags(&cqd);
+		if (ol_err_flags == 0)
+			rx_pkt_len = enic_cq_rx_desc_n_bytes(&cqd);
+		else
+			rx_pkt_len = 0;
+
+		/* Increment cqd, rqd, mbuf_table index */
+		rx_id++;
+		if (unlikely(rx_id == rq->ring.desc_count)) {
+			rx_id = 0;
+			cq->last_color = cq->last_color ? 0 : 1;
+		}
+
+		/* Prefetch next mbuf & desc while processing current one */
+		cqd_ptr = (struct cq_desc *)(cq->ring.descs) + rx_id;
+		rte_enic_prefetch(cqd_ptr);
+		rte_enic_prefetch(rq->mbuf_ring[rx_id]);
+		rte_enic_prefetch((struct rq_enet_desc *)(rq->ring.descs)
+				 + rx_id);
+
+		/* Get the mbuf to return and replace with one just allocated */
+		rxmb = rq->mbuf_ring[rx_id];
+		rq->mbuf_ring[rx_id] = nmb;
+		dma_addr = (dma_addr_t)(nmb->buf_physaddr + nmb->data_off);
+		rqd_ptr->address = rte_cpu_to_le_64(dma_addr);
+		rqd_ptr->length_type = cpu_to_le16(nmb->buf_len);
+
+		/* Fill in the rest of the mbuf */
+		rxmb->data_off = RTE_PKTMBUF_HEADROOM;
+		rxmb->nb_segs = 1;
+		rxmb->next = NULL;
+		rxmb->pkt_len = rx_pkt_len;
+		rxmb->data_len = rx_pkt_len;
+		rxmb->port = enic->port_id;
+		rxmb->ol_flags = ol_err_flags;
+		if (!ol_err_flags)
+			enic_cq_rx_to_pkt_flags(&cqd, rxmb);
+		rxmb->packet_type = enic_cq_rx_flags_to_pkt_type(&cqd);
+
+		/* prefetch mbuf data for caller */
+		rte_packet_prefetch(RTE_PTR_ADD(rxmb->buf_addr,
+				    RTE_PKTMBUF_HEADROOM));
+
+		/* store the mbuf address into the next entry of the array */
+		rx_pkts[nb_rx++] = rxmb;
+	}
+
+	nb_hold += nb_rx;
+	cq->to_clean = rx_id;
+
+	if (nb_hold > rq->rx_free_thresh) {
+		rq->posted_index = enic_ring_add(rq->ring.desc_count,
+				rq->posted_index, nb_hold);
+		nb_hold = 0;
+		rte_mb();
+		iowrite32(rq->posted_index, &rq->ctrl->posted_index);
+	}
+
+	rq->rx_nb_hold = nb_hold;
+
+	return nb_rx;
+}
-- 
2.7.0



More information about the dev mailing list