net/mlx5: reduce memory overhead of Rx/Tx descriptors

PMD uses only power of two number of descriptors, storing the number of
elements in log2 helps to reduce the size of the container to store it.

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>

diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 205ce9c..bf4232a 100644 (file)
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -808,7 +808,7 @@ recover:
                if (rehash)
                        ret = rxq_rehash(dev, rxq_ctrl);
                else
-                       ret = rxq_ctrl_setup(dev, rxq_ctrl, rxq->elts_n,
+                       ret = rxq_ctrl_setup(dev, rxq_ctrl, 1 << rxq->elts_n,
                                             rxq_ctrl->socket, NULL, rxq->mp);
                if (!ret)
                        continue;
@@ -1314,7 +1314,7 @@ mlx5_secondary_data_setup(struct priv *priv)
                if (txq_ctrl != NULL) {
                        if (txq_ctrl_setup(priv->dev,
                                           primary_txq_ctrl,
-                                          primary_txq->elts_n,
+                                          1 << primary_txq->elts_n,
                                           primary_txq_ctrl->socket,
                                           NULL) == 0) {
                                txq_ctrl->txq.stats.idx =

diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 8e02440..a784c8e 100644 (file)
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -723,7 +723,7 @@ rxq_free_elts(struct rxq_ctrl *rxq_ctrl)
        if (rxq_ctrl->rxq.elts == NULL)
                return;
 
-       for (i = 0; (i != rxq_ctrl->rxq.elts_n); ++i) {
+       for (i = 0; (i != (1u << rxq_ctrl->rxq.elts_n)); ++i) {
                if ((*rxq_ctrl->rxq.elts)[i] != NULL)
                        rte_pktmbuf_free_seg((*rxq_ctrl->rxq.elts)[i]);
                (*rxq_ctrl->rxq.elts)[i] = NULL;
@@ -807,7 +807,7 @@ rxq_cleanup(struct rxq_ctrl *rxq_ctrl)
 int
 rxq_rehash(struct rte_eth_dev *dev, struct rxq_ctrl *rxq_ctrl)
 {
-       unsigned int elts_n = rxq_ctrl->rxq.elts_n;
+       unsigned int elts_n = 1 << rxq_ctrl->rxq.elts_n;
        unsigned int i;
        struct ibv_exp_wq_attr mod;
        int err;
@@ -870,7 +870,7 @@ rxq_setup(struct rxq_ctrl *tmpl)
        struct ibv_cq *ibcq = tmpl->cq;
        struct mlx5_cq *cq = to_mxxx(cq, cq);
        struct mlx5_rwq *rwq = container_of(tmpl->wq, struct mlx5_rwq, wq);
-       struct rte_mbuf *(*elts)[tmpl->rxq.elts_n] =
+       struct rte_mbuf *(*elts)[1 << tmpl->rxq.elts_n] =
                rte_calloc_socket("RXQ", 1, sizeof(*elts), 0, tmpl->socket);
 
        if (cq->cqe_sz != RTE_CACHE_LINE_SIZE) {
@@ -924,7 +924,7 @@ rxq_ctrl_setup(struct rte_eth_dev *dev, struct rxq_ctrl *rxq_ctrl,
                .priv = priv,
                .socket = socket,
                .rxq = {
-                       .elts_n = desc,
+                       .elts_n = log2above(desc),
                        .mp = mp,
                },
        };
@@ -1148,7 +1148,7 @@ rxq_ctrl_setup(struct rte_eth_dev *dev, struct rxq_ctrl *rxq_ctrl,
        }
        /* Reuse buffers from original queue if possible. */
        if (rxq_ctrl->rxq.elts_n) {
-               assert(rxq_ctrl->rxq.elts_n == desc);
+               assert(1 << rxq_ctrl->rxq.elts_n == desc);
                assert(rxq_ctrl->rxq.elts != tmpl.rxq.elts);
                ret = rxq_alloc_elts(&tmpl, desc, rxq_ctrl->rxq.elts);
        } else

diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 7efa76a..ebc094e 100644 (file)
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -166,8 +166,8 @@ txq_complete(struct txq *txq) __attribute__((always_inline));
 static inline void
 txq_complete(struct txq *txq)
 {
-       const unsigned int elts_n = txq->elts_n;
        const unsigned int cqe_n = txq->cqe_n;
+       const unsigned int elts_n = 1 << txq->elts_n;
        const unsigned int cqe_cnt = cqe_n - 1;
        uint16_t elts_free = txq->elts_tail;
        uint16_t elts_tail;
@@ -468,7 +468,7 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 {
        struct txq *txq = (struct txq *)dpdk_txq;
        uint16_t elts_head = txq->elts_head;
-       const unsigned int elts_n = txq->elts_n;
+       const unsigned int elts_n = 1 << txq->elts_n;
        unsigned int i = 0;
        unsigned int j = 0;
        unsigned int max;
@@ -680,7 +680,7 @@ mlx5_tx_burst_mpw(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 {
        struct txq *txq = (struct txq *)dpdk_txq;
        uint16_t elts_head = txq->elts_head;
-       const unsigned int elts_n = txq->elts_n;
+       const unsigned int elts_n = 1 << txq->elts_n;
        unsigned int i = 0;
        unsigned int j = 0;
        unsigned int max;
@@ -884,7 +884,7 @@ mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
 {
        struct txq *txq = (struct txq *)dpdk_txq;
        uint16_t elts_head = txq->elts_head;
-       const unsigned int elts_n = txq->elts_n;
+       const unsigned int elts_n = 1 << txq->elts_n;
        unsigned int i = 0;
        unsigned int j = 0;
        unsigned int max;
@@ -1272,8 +1272,8 @@ uint16_t
 mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 {
        struct rxq *rxq = dpdk_rxq;
-       const unsigned int wqe_cnt = rxq->elts_n - 1;
        const unsigned int cqe_cnt = rxq->cqe_n - 1;
+       const unsigned int wqe_cnt = (1 << rxq->elts_n) - 1;
        const unsigned int sges_n = rxq->sges_n;
        struct rte_mbuf *pkt = NULL;
        struct rte_mbuf *seg = NULL;

diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index b3cc8ee..71a7d58 100644 (file)
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -54,6 +54,7 @@
 #endif
 #include <rte_mbuf.h>
 #include <rte_mempool.h>
+#include <rte_common.h>
 #ifdef PEDANTIC
 #pragma GCC diagnostic error "-Wpedantic"
 #endif
@@ -109,16 +110,16 @@ struct rxq {
        unsigned int vlan_strip:1; /* Enable VLAN stripping. */
        unsigned int crc_present:1; /* CRC must be subtracted. */
        unsigned int sges_n:2; /* Log 2 of SGEs (max buffers per packet). */
+       unsigned int elts_n:4; /* Log 2 of Mbufs. */
+       unsigned int port_id:8;
+       volatile uint32_t *rq_db;
+       volatile uint32_t *cq_db;
        uint16_t rq_ci;
        uint16_t cq_ci;
-       uint16_t elts_n;
        uint16_t cqe_n; /* Number of CQ elements. */
-       uint16_t port_id;
        volatile struct mlx5_wqe_data_seg(*wqes)[];
        volatile struct mlx5_cqe(*cqes)[];
        struct rxq_zip zip; /* Compressed context. */
-       volatile uint32_t *rq_db;
-       volatile uint32_t *cq_db;
        struct rte_mbuf *(*elts)[];
        struct rte_mempool *mp;
        struct mlx5_rxq_stats stats;
@@ -238,15 +239,16 @@ struct hash_rxq {
 };
 
 /* TX queue descriptor. */
+RTE_STD_C11
 struct txq {
        uint16_t elts_head; /* Current index in (*elts)[]. */
        uint16_t elts_tail; /* First element awaiting completion. */
        uint16_t elts_comp; /* Counter since last completion request. */
-       uint16_t elts_n; /* (*elts)[] length. */
        uint16_t cq_ci; /* Consumer index for completion queue. */
        uint16_t cqe_n; /* Number of CQ elements. */
        uint16_t wqe_ci; /* Consumer index for work queue. */
        uint16_t wqe_n; /* Number of WQ elements. */
+       uint16_t elts_n:4; /* (*elts)[] length (in log2). */
        uint16_t bf_offset; /* Blueflame offset. */
        uint16_t bf_buf_size; /* Blueflame size. */
        uint16_t max_inline; /* Multiple of RTE_CACHE_LINE_SIZE to inline. */

diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c
index 269ffe6..d9e61ed 100644 (file)
--- a/drivers/net/mlx5/mlx5_txq.c
+++ b/drivers/net/mlx5/mlx5_txq.c
@@ -101,7 +101,7 @@ txq_alloc_elts(struct txq_ctrl *txq_ctrl, unsigned int elts_n)
 static void
 txq_free_elts(struct txq_ctrl *txq_ctrl)
 {
-       unsigned int elts_n = txq_ctrl->txq.elts_n;
+       unsigned int elts_n = 1 << txq_ctrl->txq.elts_n;
        unsigned int elts_head = txq_ctrl->txq.elts_head;
        unsigned int elts_tail = txq_ctrl->txq.elts_tail;
        struct rte_mbuf *(*elts)[elts_n] = txq_ctrl->txq.elts;
@@ -227,7 +227,7 @@ txq_setup(struct txq_ctrl *tmpl, struct txq_ctrl *txq_ctrl)
                (volatile struct mlx5_cqe (*)[])
                (uintptr_t)cq->active_buf->buf;
        tmpl->txq.elts =
-               (struct rte_mbuf *(*)[tmpl->txq.elts_n])
+               (struct rte_mbuf *(*)[1 << tmpl->txq.elts_n])
                ((uintptr_t)txq_ctrl + sizeof(*txq_ctrl));
        return 0;
 }
@@ -277,7 +277,7 @@ txq_ctrl_setup(struct rte_eth_dev *dev, struct txq_ctrl *txq_ctrl,
        }
        (void)conf; /* Thresholds configuration (ignored). */
        assert(desc > MLX5_TX_COMP_THRESH);
-       tmpl.txq.elts_n = desc;
+       tmpl.txq.elts_n = log2above(desc);
        /* MRs will be registered in mp2mr[] later. */
        attr.rd = (struct ibv_exp_res_domain_init_attr){
                .comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |