net/mlx5: define explicit fields for Rx offloads
authorNelio Laranjeiro <nelio.laranjeiro@6wind.com>
Wed, 2 Nov 2016 10:39:38 +0000 (11:39 +0100)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Mon, 7 Nov 2016 17:42:22 +0000 (18:42 +0100)
This commit redefines the completion queue element structure as the
original lacks the required fields.

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
drivers/net/mlx5/mlx5_prm.h
drivers/net/mlx5/mlx5_rxtx.c

index 500f25a..7f31a2f 100644 (file)
@@ -158,7 +158,21 @@ struct mlx5_cqe {
 #if (RTE_CACHE_LINE_SIZE == 128)
        uint8_t padding[64];
 #endif
-       struct mlx5_cqe64 cqe64;
+       uint8_t pkt_info;
+       uint8_t rsvd0[11];
+       uint32_t rx_hash_res;
+       uint8_t rx_hash_type;
+       uint8_t rsvd1[11];
+       uint8_t hds_ip_ext;
+       uint8_t l4_hdr_type_etc;
+       uint16_t vlan_info;
+       uint8_t rsvd2[12];
+       uint32_t byte_cnt;
+       uint64_t timestamp;
+       uint8_t rsvd3[4];
+       uint16_t wqe_counter;
+       uint8_t rsvd4;
+       uint8_t op_own;
 };
 
 #endif /* RTE_PMD_MLX5_PRM_H_ */
index 3ad36a7..8d5f946 100644 (file)
  *   0 the first time.
  */
 static inline int
-check_cqe64_seen(volatile struct mlx5_cqe64 *cqe)
+check_cqe_seen(volatile struct mlx5_cqe *cqe)
 {
        static const uint8_t magic[] = "seen";
-       volatile uint8_t (*buf)[sizeof(cqe->rsvd40)] = &cqe->rsvd40;
+       volatile uint8_t (*buf)[sizeof(cqe->rsvd3)] = &cqe->rsvd3;
        int ret = 1;
        unsigned int i;
 
@@ -99,9 +99,9 @@ check_cqe64_seen(volatile struct mlx5_cqe64 *cqe)
 #endif /* NDEBUG */
 
 static inline int
-check_cqe64(volatile struct mlx5_cqe64 *cqe,
-           unsigned int cqes_n, const uint16_t ci)
-           __attribute__((always_inline));
+check_cqe(volatile struct mlx5_cqe *cqe,
+         unsigned int cqes_n, const uint16_t ci)
+         __attribute__((always_inline));
 
 /**
  * Check whether CQE is valid.
@@ -117,8 +117,8 @@ check_cqe64(volatile struct mlx5_cqe64 *cqe,
  *   0 on success, 1 on failure.
  */
 static inline int
-check_cqe64(volatile struct mlx5_cqe64 *cqe,
-               unsigned int cqes_n, const uint16_t ci)
+check_cqe(volatile struct mlx5_cqe *cqe,
+         unsigned int cqes_n, const uint16_t ci)
 {
        uint16_t idx = ci & cqes_n;
        uint8_t op_own = cqe->op_own;
@@ -136,14 +136,14 @@ check_cqe64(volatile struct mlx5_cqe64 *cqe,
                if ((syndrome == MLX5_CQE_SYNDROME_LOCAL_LENGTH_ERR) ||
                    (syndrome == MLX5_CQE_SYNDROME_REMOTE_ABORTED_ERR))
                        return 0;
-               if (!check_cqe64_seen(cqe))
+               if (!check_cqe_seen(cqe))
                        ERROR("unexpected CQE error %u (0x%02x)"
                              " syndrome 0x%02x",
                              op_code, op_code, syndrome);
                return 1;
        } else if ((op_code != MLX5_CQE_RESP_SEND) &&
                   (op_code != MLX5_CQE_REQ)) {
-               if (!check_cqe64_seen(cqe))
+               if (!check_cqe_seen(cqe))
                        ERROR("unexpected CQE opcode %u (0x%02x)",
                              op_code, op_code);
                return 1;
@@ -172,25 +172,25 @@ txq_complete(struct txq *txq)
        uint16_t elts_free = txq->elts_tail;
        uint16_t elts_tail;
        uint16_t cq_ci = txq->cq_ci;
-       volatile struct mlx5_cqe64 *cqe = NULL;
+       volatile struct mlx5_cqe *cqe = NULL;
        volatile struct mlx5_wqe *wqe;
 
        do {
-               volatile struct mlx5_cqe64 *tmp;
+               volatile struct mlx5_cqe *tmp;
 
-               tmp = &(*txq->cqes)[cq_ci & cqe_cnt].cqe64;
-               if (check_cqe64(tmp, cqe_n, cq_ci))
+               tmp = &(*txq->cqes)[cq_ci & cqe_cnt];
+               if (check_cqe(tmp, cqe_n, cq_ci))
                        break;
                cqe = tmp;
 #ifndef NDEBUG
                if (MLX5_CQE_FORMAT(cqe->op_own) == MLX5_COMPRESSED) {
-                       if (!check_cqe64_seen(cqe))
+                       if (!check_cqe_seen(cqe))
                                ERROR("unexpected compressed CQE, TX stopped");
                        return;
                }
                if ((MLX5_CQE_OPCODE(cqe->op_own) == MLX5_CQE_RESP_ERR) ||
                    (MLX5_CQE_OPCODE(cqe->op_own) == MLX5_CQE_REQ_ERR)) {
-                       if (!check_cqe64_seen(cqe))
+                       if (!check_cqe_seen(cqe))
                                ERROR("unexpected error CQE, TX stopped");
                        return;
                }
@@ -1088,13 +1088,12 @@ mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
  *   Packet type for struct rte_mbuf.
  */
 static inline uint32_t
-rxq_cq_to_pkt_type(volatile struct mlx5_cqe64 *cqe)
+rxq_cq_to_pkt_type(volatile struct mlx5_cqe *cqe)
 {
        uint32_t pkt_type;
        uint8_t flags = cqe->l4_hdr_type_etc;
-       uint8_t info = cqe->rsvd0[0];
 
-       if (info & MLX5_CQE_RX_TUNNEL_PACKET)
+       if (cqe->pkt_info & MLX5_CQE_RX_TUNNEL_PACKET)
                pkt_type =
                        TRANSPOSE(flags,
                                  MLX5_CQE_RX_OUTER_IPV4_PACKET,
@@ -1136,7 +1135,7 @@ rxq_cq_to_pkt_type(volatile struct mlx5_cqe64 *cqe)
  *   with error.
  */
 static inline int
-mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
+mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe *cqe,
                 uint16_t cqe_cnt, uint32_t *rss_hash)
 {
        struct rxq_zip *zip = &rxq->zip;
@@ -1147,7 +1146,7 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
        if (zip->ai) {
                volatile struct mlx5_mini_cqe8 (*mc)[8] =
                        (volatile struct mlx5_mini_cqe8 (*)[8])
-                       (uintptr_t)(&(*rxq->cqes)[zip->ca & cqe_cnt].cqe64);
+                       (uintptr_t)(&(*rxq->cqes)[zip->ca & cqe_cnt]);
 
                len = ntohl((*mc)[zip->ai & 7].byte_cnt);
                *rss_hash = ntohl((*mc)[zip->ai & 7].rx_hash_result);
@@ -1165,7 +1164,7 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
                        uint16_t end = zip->cq_ci;
 
                        while (idx != end) {
-                               (*rxq->cqes)[idx & cqe_cnt].cqe64.op_own =
+                               (*rxq->cqes)[idx & cqe_cnt].op_own =
                                        MLX5_CQE_INVALIDATE;
                                ++idx;
                        }
@@ -1177,7 +1176,7 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
                int ret;
                int8_t op_own;
 
-               ret = check_cqe64(cqe, cqe_n, rxq->cq_ci);
+               ret = check_cqe(cqe, cqe_n, rxq->cq_ci);
                if (unlikely(ret == 1))
                        return 0;
                ++rxq->cq_ci;
@@ -1186,7 +1185,7 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
                        volatile struct mlx5_mini_cqe8 (*mc)[8] =
                                (volatile struct mlx5_mini_cqe8 (*)[8])
                                (uintptr_t)(&(*rxq->cqes)[rxq->cq_ci &
-                                                         cqe_cnt].cqe64);
+                                                         cqe_cnt]);
 
                        /* Fix endianness. */
                        zip->cqe_cnt = ntohl(cqe->byte_cnt);
@@ -1230,12 +1229,11 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
  *   Offload flags (ol_flags) for struct rte_mbuf.
  */
 static inline uint32_t
-rxq_cq_to_ol_flags(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe)
+rxq_cq_to_ol_flags(struct rxq *rxq, volatile struct mlx5_cqe *cqe)
 {
        uint32_t ol_flags = 0;
        uint8_t l3_hdr = (cqe->l4_hdr_type_etc) & MLX5_CQE_L3_HDR_TYPE_MASK;
        uint8_t l4_hdr = (cqe->l4_hdr_type_etc) & MLX5_CQE_L4_HDR_TYPE_MASK;
-       uint8_t info = cqe->rsvd0[0];
 
        if ((l3_hdr == MLX5_CQE_L3_HDR_TYPE_IPV4) ||
            (l3_hdr == MLX5_CQE_L3_HDR_TYPE_IPV6))
@@ -1254,7 +1252,7 @@ rxq_cq_to_ol_flags(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe)
         * of PKT_RX_EIP_CKSUM_BAD because the latter is not functional
         * (its value is 0).
         */
-       if ((info & MLX5_CQE_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
+       if ((cqe->pkt_info & MLX5_CQE_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
                ol_flags |=
                        TRANSPOSE(~cqe->l4_hdr_type_etc,
                                  MLX5_CQE_RX_OUTER_IP_CSUM_OK,
@@ -1287,8 +1285,8 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
        const unsigned int sges_n = rxq->sges_n;
        struct rte_mbuf *pkt = NULL;
        struct rte_mbuf *seg = NULL;
-       volatile struct mlx5_cqe64 *cqe =
-               &(*rxq->cqes)[rxq->cq_ci & cqe_cnt].cqe64;
+       volatile struct mlx5_cqe *cqe =
+               &(*rxq->cqes)[rxq->cq_ci & cqe_cnt];
        unsigned int i = 0;
        unsigned int rq_ci = rxq->rq_ci << sges_n;
        int len; /* keep its value across iterations. */
@@ -1325,7 +1323,7 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
                        break;
                }
                if (!pkt) {
-                       cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_cnt].cqe64;
+                       cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_cnt];
                        len = mlx5_rx_poll_len(rxq, cqe, cqe_cnt,
                                               &rss_hash_res);
                        if (!len) {