X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fthunderx%2Fnicvf_rxtx.c;h=d0ea95f1d11cdaec83d8d5eb29c42a057a83f7d6;hb=1aacc3d388d3aa9b4fdefcaa111ef177f39ad57f;hp=aa278afee3c5c1ab7b74f1e5b1d1fce37ead09f9;hpb=42473d6782896bf6281d347f61cbcfaf8df17d0f;p=dpdk.git

diff --git a/drivers/net/thunderx/nicvf_rxtx.c b/drivers/net/thunderx/nicvf_rxtx.c
index aa278afee3..d0ea95f1d1 100644
--- a/drivers/net/thunderx/nicvf_rxtx.c
+++ b/drivers/net/thunderx/nicvf_rxtx.c
@@ -13,7 +13,7 @@
 #include <rte_common.h>
 #include <rte_cycles.h>
 #include <rte_errno.h>
-#include <rte_ethdev.h>
+#include <rte_ethdev_driver.h>
 #include <rte_ether.h>
 #include <rte_log.h>
 #include <rte_mbuf.h>
@@ -25,7 +25,7 @@
 #include "nicvf_rxtx.h"
 #include "nicvf_logs.h"
 
-static inline void __hot
+static inline void __rte_hot
 fill_sq_desc_header(union sq_entry_t *entry, struct rte_mbuf *pkt)
 {
 	/* Local variable sqe to avoid read from sq desc memory*/
@@ -61,7 +61,15 @@ fill_sq_desc_header(union sq_entry_t *entry, struct rte_mbuf *pkt)
 	entry->buff[0] = sqe.buff[0];
 }
 
-void __hot
+static inline void __rte_hot
+fill_sq_desc_header_zero_w1(union sq_entry_t *entry,
+				struct rte_mbuf *pkt)
+{
+	fill_sq_desc_header(entry, pkt);
+	entry->buff[1] = 0ULL;
+}
+
+void __rte_hot
 nicvf_single_pool_free_xmited_buffers(struct nicvf_txq *sq)
 {
 	int j = 0;
@@ -84,7 +92,7 @@ nicvf_single_pool_free_xmited_buffers(struct nicvf_txq *sq)
 	NICVF_TX_ASSERT(sq->xmit_bufs >= 0);
 }
 
-void __hot
+void __rte_hot
 nicvf_multi_pool_free_xmited_buffers(struct nicvf_txq *sq)
 {
 	uint32_t n = 0;
@@ -107,7 +115,7 @@ nicvf_multi_pool_free_xmited_buffers(struct nicvf_txq *sq)
 	NICVF_TX_ASSERT(sq->xmit_bufs >= 0);
 }
 
-static inline uint32_t __hot
+static inline uint32_t __rte_hot
 nicvf_free_tx_desc(struct nicvf_txq *sq)
 {
 	return ((sq->head - sq->tail - 1) & sq->qlen_mask);
@@ -116,7 +124,7 @@ nicvf_free_tx_desc(struct nicvf_txq *sq)
 /* Send Header + Packet */
 #define TX_DESC_PER_PKT 2
 
-static inline uint32_t __hot
+static inline uint32_t __rte_hot
 nicvf_free_xmitted_buffers(struct nicvf_txq *sq, struct rte_mbuf **tx_pkts,
 			    uint16_t nb_pkts)
 {
@@ -134,7 +142,7 @@ nicvf_free_xmitted_buffers(struct nicvf_txq *sq, struct rte_mbuf **tx_pkts,
 	return free_desc;
 }
 
-uint16_t __hot
+uint16_t __rte_hot
 nicvf_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 {
 	int i;
@@ -162,16 +170,18 @@ nicvf_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 		free_desc -= TX_DESC_PER_PKT;
 	}
 
-	sq->tail = tail;
-	sq->xmit_bufs += i;
-	rte_wmb();
+	if (likely(i)) {
+		sq->tail = tail;
+		sq->xmit_bufs += i;
+		rte_wmb();
 
-	/* Inform HW to xmit the packets */
-	nicvf_addr_write(sq->sq_door, i * TX_DESC_PER_PKT);
+		/* Inform HW to xmit the packets */
+		nicvf_addr_write(sq->sq_door, i * TX_DESC_PER_PKT);
+	}
 	return i;
 }
 
-uint16_t __hot
+uint16_t __rte_hot
 nicvf_xmit_pkts_multiseg(void *tx_queue, struct rte_mbuf **tx_pkts,
 			 uint16_t nb_pkts)
 {
@@ -202,7 +212,7 @@ nicvf_xmit_pkts_multiseg(void *tx_queue, struct rte_mbuf **tx_pkts,
 		used_bufs += nb_segs;
 
 		txbuffs[tail] = NULL;
-		fill_sq_desc_header(desc_ptr + tail, pkt);
+		fill_sq_desc_header_zero_w1(desc_ptr + tail, pkt);
 		tail = (tail + 1) & qlen_mask;
 
 		txbuffs[tail] = pkt;
@@ -218,12 +228,14 @@ nicvf_xmit_pkts_multiseg(void *tx_queue, struct rte_mbuf **tx_pkts,
 		}
 	}
 
-	sq->tail = tail;
-	sq->xmit_bufs += used_bufs;
-	rte_wmb();
+	if (likely(used_desc)) {
+		sq->tail = tail;
+		sq->xmit_bufs += used_bufs;
+		rte_wmb();
 
-	/* Inform HW to xmit the packets */
-	nicvf_addr_write(sq->sq_door, used_desc);
+		/* Inform HW to xmit the packets */
+		nicvf_addr_write(sq->sq_door, used_desc);
+	}
 	return i;
 }
 
@@ -321,13 +333,27 @@ static const uint32_t ptype_table[16][16] __rte_cache_aligned = {
 	[L3_OTHER][L4_NVGRE] = RTE_PTYPE_TUNNEL_NVGRE,
 };
 
-static inline uint32_t __hot
+static inline uint32_t __rte_hot
 nicvf_rx_classify_pkt(cqe_rx_word0_t cqe_rx_w0)
 {
 	return ptype_table[cqe_rx_w0.l3_type][cqe_rx_w0.l4_type];
 }
 
-static inline int __hot
+static inline uint64_t __rte_hot
+nicvf_set_olflags(const cqe_rx_word0_t cqe_rx_w0)
+{
+	static const uint64_t flag_table[3] __rte_cache_aligned = {
+		PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_GOOD,
+		PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_UNKNOWN,
+		PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_BAD,
+	};
+
+	const uint8_t idx = (cqe_rx_w0.err_opcode == CQE_RX_ERR_L4_CHK) << 1 |
+		(cqe_rx_w0.err_opcode == CQE_RX_ERR_IP_CHK);
+	return flag_table[idx];
+}
+
+static inline int __rte_hot
 nicvf_fill_rbdr(struct nicvf_rxq *rxq, int to_fill)
 {
 	int i;
@@ -359,15 +385,14 @@ nicvf_fill_rbdr(struct nicvf_rxq *rxq, int to_fill)
 		ltail++;
 	}
 
-	while (__atomic_load_n(&rbdr->tail, __ATOMIC_RELAXED) != next_tail)
-		rte_pause();
+	rte_wait_until_equal_32(&rbdr->tail, next_tail, __ATOMIC_RELAXED);
 
 	__atomic_store_n(&rbdr->tail, ltail, __ATOMIC_RELEASE);
 	nicvf_addr_write(door, to_fill);
 	return to_fill;
 }
 
-static inline int32_t __hot
+static inline int32_t __rte_hot
 nicvf_rx_pkts_to_process(struct nicvf_rxq *rxq, uint16_t nb_pkts,
 			 int32_t available_space)
 {
@@ -378,18 +403,20 @@ nicvf_rx_pkts_to_process(struct nicvf_rxq *rxq, uint16_t nb_pkts,
 	return RTE_MIN(nb_pkts, available_space);
 }
 
-static inline void __hot
+static inline void __rte_hot
 nicvf_rx_offload(cqe_rx_word0_t cqe_rx_w0, cqe_rx_word2_t cqe_rx_w2,
 		 struct rte_mbuf *pkt)
 {
 	if (likely(cqe_rx_w0.rss_alg)) {
 		pkt->hash.rss = cqe_rx_w2.rss_tag;
 		pkt->ol_flags |= PKT_RX_RSS_HASH;
+
 	}
 }
 
-uint16_t __hot
-nicvf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+static __rte_always_inline uint16_t
+nicvf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
+		const uint32_t flag)
 {
 	uint32_t i, to_process;
 	struct cqe_rx_t *cqe_rx;
@@ -420,7 +447,19 @@ nicvf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 		rb0_ptr = *((uint64_t *)cqe_rx + rbptr_offset);
 		pkt = (struct rte_mbuf *)nicvf_mbuff_phy2virt
 				(rb0_ptr - cqe_rx_w1.align_pad, mbuf_phys_off);
-		pkt->ol_flags = 0;
+
+		if (flag & NICVF_RX_OFFLOAD_NONE)
+			pkt->ol_flags = 0;
+		if (flag & NICVF_RX_OFFLOAD_CKSUM)
+			pkt->ol_flags = nicvf_set_olflags(cqe_rx_w0);
+		if (flag & NICVF_RX_OFFLOAD_VLAN_STRIP) {
+			if (unlikely(cqe_rx_w0.vlan_stripped)) {
+				pkt->ol_flags |= PKT_RX_VLAN
+							| PKT_RX_VLAN_STRIPPED;
+				pkt->vlan_tci =
+					rte_cpu_to_be_16(cqe_rx_w2.vlan_tci);
+			}
+		}
 		pkt->data_len = cqe_rx_w3.rb0_sz;
 		pkt->pkt_len = cqe_rx_w3.rb0_sz;
 		pkt->packet_type = nicvf_rx_classify_pkt(cqe_rx_w0);
@@ -445,11 +484,43 @@ nicvf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 	return to_process;
 }
 
-static inline uint16_t __hot
+uint16_t __rte_hot
+nicvf_recv_pkts_no_offload(void *rx_queue, struct rte_mbuf **rx_pkts,
+		uint16_t nb_pkts)
+{
+	return nicvf_recv_pkts(rx_queue, rx_pkts, nb_pkts,
+			NICVF_RX_OFFLOAD_NONE);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_cksum(void *rx_queue, struct rte_mbuf **rx_pkts,
+		uint16_t nb_pkts)
+{
+	return nicvf_recv_pkts(rx_queue, rx_pkts, nb_pkts,
+			NICVF_RX_OFFLOAD_CKSUM);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_vlan_strip(void *rx_queue, struct rte_mbuf **rx_pkts,
+		uint16_t nb_pkts)
+{
+	return nicvf_recv_pkts(rx_queue, rx_pkts, nb_pkts,
+			NICVF_RX_OFFLOAD_NONE | NICVF_RX_OFFLOAD_VLAN_STRIP);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_cksum_vlan_strip(void *rx_queue, struct rte_mbuf **rx_pkts,
+		uint16_t nb_pkts)
+{
+	return nicvf_recv_pkts(rx_queue, rx_pkts, nb_pkts,
+			NICVF_RX_OFFLOAD_CKSUM | NICVF_RX_OFFLOAD_VLAN_STRIP);
+}
+
+static __rte_always_inline uint16_t __rte_hot
 nicvf_process_cq_mseg_entry(struct cqe_rx_t *cqe_rx,
 			uint64_t mbuf_phys_off,
 			struct rte_mbuf **rx_pkt, uint8_t rbptr_offset,
-			uint64_t mbuf_init)
+			uint64_t mbuf_init, const uint32_t flag)
 {
 	struct rte_mbuf *pkt, *seg, *prev;
 	cqe_rx_word0_t cqe_rx_w0;
@@ -467,12 +538,22 @@ nicvf_process_cq_mseg_entry(struct cqe_rx_t *cqe_rx,
 	pkt = (struct rte_mbuf *)nicvf_mbuff_phy2virt
 			(rb_ptr[0] - cqe_rx_w1.align_pad, mbuf_phys_off);
 
-	pkt->ol_flags = 0;
 	pkt->pkt_len = cqe_rx_w1.pkt_len;
 	pkt->data_len = rb_sz[nicvf_frag_num(0)];
 	nicvf_mbuff_init_mseg_update(
 				pkt, mbuf_init, cqe_rx_w1.align_pad, nb_segs);
 	pkt->packet_type = nicvf_rx_classify_pkt(cqe_rx_w0);
+	if (flag & NICVF_RX_OFFLOAD_NONE)
+		pkt->ol_flags = 0;
+	if (flag & NICVF_RX_OFFLOAD_CKSUM)
+		pkt->ol_flags = nicvf_set_olflags(cqe_rx_w0);
+	if (flag & NICVF_RX_OFFLOAD_VLAN_STRIP) {
+		if (unlikely(cqe_rx_w0.vlan_stripped)) {
+			pkt->ol_flags |= PKT_RX_VLAN
+				| PKT_RX_VLAN_STRIPPED;
+			pkt->vlan_tci = rte_cpu_to_be_16(cqe_rx_w2.vlan_tci);
+		}
+	}
 	nicvf_rx_offload(cqe_rx_w0, cqe_rx_w2, pkt);
 
 	*rx_pkt = pkt;
@@ -491,9 +572,9 @@ nicvf_process_cq_mseg_entry(struct cqe_rx_t *cqe_rx,
 	return nb_segs;
 }
 
-uint16_t __hot
+static __rte_always_inline uint16_t __rte_hot
 nicvf_recv_pkts_multiseg(void *rx_queue, struct rte_mbuf **rx_pkts,
-			 uint16_t nb_pkts)
+			 uint16_t nb_pkts, const uint32_t flag)
 {
 	union cq_entry_t *cq_entry;
 	struct cqe_rx_t *cqe_rx;
@@ -515,7 +596,7 @@ nicvf_recv_pkts_multiseg(void *rx_queue, struct rte_mbuf **rx_pkts,
 		cq_entry = &desc[cqe_head];
 		cqe_rx = (struct cqe_rx_t *)cq_entry;
 		nb_segs = nicvf_process_cq_mseg_entry(cqe_rx, mbuf_phys_off,
-			rx_pkts + i, rbptr_offset, mbuf_init);
+			rx_pkts + i, rbptr_offset, mbuf_init, flag);
 		buffers_consumed += nb_segs;
 		cqe_head = (cqe_head + 1) & cqe_mask;
 		nicvf_prefetch_store_keep(rx_pkts[i]);
@@ -535,6 +616,38 @@ nicvf_recv_pkts_multiseg(void *rx_queue, struct rte_mbuf **rx_pkts,
 	return to_process;
 }
 
+uint16_t __rte_hot
+nicvf_recv_pkts_multiseg_no_offload(void *rx_queue, struct rte_mbuf **rx_pkts,
+		uint16_t nb_pkts)
+{
+	return nicvf_recv_pkts_multiseg(rx_queue, rx_pkts, nb_pkts,
+			NICVF_RX_OFFLOAD_NONE);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_multiseg_cksum(void *rx_queue, struct rte_mbuf **rx_pkts,
+		uint16_t nb_pkts)
+{
+	return nicvf_recv_pkts_multiseg(rx_queue, rx_pkts, nb_pkts,
+			NICVF_RX_OFFLOAD_CKSUM);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_multiseg_vlan_strip(void *rx_queue, struct rte_mbuf **rx_pkts,
+		uint16_t nb_pkts)
+{
+	return nicvf_recv_pkts_multiseg(rx_queue, rx_pkts, nb_pkts,
+			NICVF_RX_OFFLOAD_NONE | NICVF_RX_OFFLOAD_VLAN_STRIP);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_multiseg_cksum_vlan_strip(void *rx_queue,
+		struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+{
+	return nicvf_recv_pkts_multiseg(rx_queue, rx_pkts, nb_pkts,
+			NICVF_RX_OFFLOAD_CKSUM | NICVF_RX_OFFLOAD_VLAN_STRIP);
+}
+
 uint32_t
 nicvf_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t queue_idx)
 {