/* Verbs header. */
/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
#ifdef PEDANTIC
-#pragma GCC diagnostic ignored "-pedantic"
+#pragma GCC diagnostic ignored "-Wpedantic"
#endif
#include <infiniband/verbs.h>
+#include <infiniband/mlx5_hw.h>
+#include <infiniband/arch.h>
#ifdef PEDANTIC
-#pragma GCC diagnostic error "-pedantic"
+#pragma GCC diagnostic error "-Wpedantic"
#endif
/* DPDK headers don't like -pedantic. */
#ifdef PEDANTIC
-#pragma GCC diagnostic ignored "-pedantic"
+#pragma GCC diagnostic ignored "-Wpedantic"
#endif
#include <rte_mbuf.h>
#include <rte_mempool.h>
#include <rte_prefetch.h>
#include <rte_common.h>
#include <rte_branch_prediction.h>
+#include <rte_ether.h>
#ifdef PEDANTIC
-#pragma GCC diagnostic error "-pedantic"
+#pragma GCC diagnostic error "-Wpedantic"
#endif
#include "mlx5.h"
#include "mlx5_rxtx.h"
#include "mlx5_autoconf.h"
#include "mlx5_defs.h"
+#include "mlx5_prm.h"
+
+static inline int
+check_cqe(volatile struct mlx5_cqe *cqe,
+ unsigned int cqes_n, const uint16_t ci)
+ __attribute__((always_inline));
+
+static inline void
+txq_complete(struct txq *txq) __attribute__((always_inline));
+
+static inline uint32_t
+txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
+ __attribute__((always_inline));
+
+static inline void
+mlx5_tx_dbrec(struct txq *txq, volatile struct mlx5_wqe *wqe)
+ __attribute__((always_inline));
+
+static inline uint32_t
+rxq_cq_to_pkt_type(volatile struct mlx5_cqe *cqe)
+ __attribute__((always_inline));
+
+static inline int
+mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe *cqe,
+ uint16_t cqe_cnt, uint32_t *rss_hash)
+ __attribute__((always_inline));
+
+static inline uint32_t
+rxq_cq_to_ol_flags(struct rxq *rxq, volatile struct mlx5_cqe *cqe)
+ __attribute__((always_inline));
+
+#ifndef NDEBUG
/**
- * Manage TX completions.
+ * Verify or set magic value in CQE.
*
- * When sending a burst, mlx5_tx_burst() posts several WRs.
- * To improve performance, a completion event is only required once every
- * MLX5_PMD_TX_PER_COMP_REQ sends. Doing so discards completion information
- * for other WRs, but this information would not be used anyway.
+ * @param cqe
+ * Pointer to CQE.
+ *
+ * @return
+ * 0 the first time.
+ */
+static inline int
+check_cqe_seen(volatile struct mlx5_cqe *cqe)
+{
+ static const uint8_t magic[] = "seen";
+ volatile uint8_t (*buf)[sizeof(cqe->rsvd0)] = &cqe->rsvd0;
+ int ret = 1;
+ unsigned int i;
+
+ for (i = 0; i < sizeof(magic) && i < sizeof(*buf); ++i)
+ if (!ret || (*buf)[i] != magic[i]) {
+ ret = 0;
+ (*buf)[i] = magic[i];
+ }
+ return ret;
+}
+
+#endif /* NDEBUG */
+
+/**
+ * Check whether CQE is valid.
+ *
+ * @param cqe
+ * Pointer to CQE.
+ * @param cqes_n
+ * Size of completion queue.
+ * @param ci
+ * Consumer index.
+ *
+ * @return
+ * 0 on success, 1 on failure.
+ */
+static inline int
+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;
+ uint8_t op_owner = MLX5_CQE_OWNER(op_own);
+ uint8_t op_code = MLX5_CQE_OPCODE(op_own);
+
+ if (unlikely((op_owner != (!!(idx))) || (op_code == MLX5_CQE_INVALID)))
+ return 1; /* No CQE. */
+#ifndef NDEBUG
+ if ((op_code == MLX5_CQE_RESP_ERR) ||
+ (op_code == MLX5_CQE_REQ_ERR)) {
+ volatile struct mlx5_err_cqe *err_cqe = (volatile void *)cqe;
+ uint8_t syndrome = err_cqe->syndrome;
+
+ if ((syndrome == MLX5_CQE_SYNDROME_LOCAL_LENGTH_ERR) ||
+ (syndrome == MLX5_CQE_SYNDROME_REMOTE_ABORTED_ERR))
+ return 0;
+ 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_cqe_seen(cqe))
+ ERROR("unexpected CQE opcode %u (0x%02x)",
+ op_code, op_code);
+ return 1;
+ }
+#endif /* NDEBUG */
+ return 0;
+}
+
+/**
+ * Return the address of the WQE.
*
* @param txq
* Pointer to TX queue structure.
+ * @param wqe_ci
+ * WQE consumer index.
*
* @return
- * 0 on success, -1 on failure.
+ * WQE address.
*/
-static int
+static inline uintptr_t *
+tx_mlx5_wqe(struct txq *txq, uint16_t ci)
+{
+ ci &= ((1 << txq->wqe_n) - 1);
+ return (uintptr_t *)((uintptr_t)txq->wqes + ci * MLX5_WQE_SIZE);
+}
+
+/**
+ * Manage TX completions.
+ *
+ * When sending a burst, mlx5_tx_burst() posts several WRs.
+ *
+ * @param txq
+ * Pointer to TX queue structure.
+ */
+static inline void
txq_complete(struct txq *txq)
{
- unsigned int elts_comp = txq->elts_comp;
- unsigned int elts_tail = txq->elts_tail;
- const unsigned int elts_n = txq->elts_n;
- int wcs_n;
+ const unsigned int elts_n = 1 << txq->elts_n;
+ const unsigned int cqe_n = 1 << txq->cqe_n;
+ const unsigned int cqe_cnt = cqe_n - 1;
+ uint16_t elts_free = txq->elts_tail;
+ uint16_t elts_tail;
+ uint16_t cq_ci = txq->cq_ci;
+ volatile struct mlx5_cqe *cqe = NULL;
+ volatile struct mlx5_wqe_ctrl *ctrl;
- if (unlikely(elts_comp == 0))
- return 0;
-#ifdef DEBUG_SEND
- DEBUG("%p: processing %u work requests completions",
- (void *)txq, elts_comp);
+ do {
+ volatile struct mlx5_cqe *tmp;
+
+ 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_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_cqe_seen(cqe))
+ ERROR("unexpected error CQE, TX stopped");
+ return;
+ }
+#endif /* NDEBUG */
+ ++cq_ci;
+ } while (1);
+ if (unlikely(cqe == NULL))
+ return;
+ txq->wqe_pi = ntohs(cqe->wqe_counter);
+ ctrl = (volatile struct mlx5_wqe_ctrl *)
+ tx_mlx5_wqe(txq, txq->wqe_pi);
+ elts_tail = ctrl->ctrl3;
+ assert(elts_tail < (1 << txq->wqe_n));
+ /* Free buffers. */
+ while (elts_free != elts_tail) {
+ struct rte_mbuf *elt = (*txq->elts)[elts_free];
+ unsigned int elts_free_next =
+ (elts_free + 1) & (elts_n - 1);
+ struct rte_mbuf *elt_next = (*txq->elts)[elts_free_next];
+
+#ifndef NDEBUG
+ /* Poisoning. */
+ memset(&(*txq->elts)[elts_free],
+ 0x66,
+ sizeof((*txq->elts)[elts_free]));
#endif
- wcs_n = txq->if_cq->poll_cnt(txq->cq, elts_comp);
- if (unlikely(wcs_n == 0))
- return 0;
- if (unlikely(wcs_n < 0)) {
- DEBUG("%p: ibv_poll_cq() failed (wcs_n=%d)",
- (void *)txq, wcs_n);
- return -1;
+ RTE_MBUF_PREFETCH_TO_FREE(elt_next);
+ /* Only one segment needs to be freed. */
+ rte_pktmbuf_free_seg(elt);
+ elts_free = elts_free_next;
}
- elts_comp -= wcs_n;
- assert(elts_comp <= txq->elts_comp);
- /*
- * Assume WC status is successful as nothing can be done about it
- * anyway.
- */
- elts_tail += wcs_n * txq->elts_comp_cd_init;
- if (elts_tail >= elts_n)
- elts_tail -= elts_n;
+ txq->cq_ci = cq_ci;
txq->elts_tail = elts_tail;
- txq->elts_comp = elts_comp;
- return 0;
+ /* Update the consumer index. */
+ rte_wmb();
+ *txq->cq_db = htonl(cq_ci);
}
/**
* @return
* mr->lkey on success, (uint32_t)-1 on failure.
*/
-static uint32_t
-txq_mp2mr(struct txq *txq, const struct rte_mempool *mp)
+static inline uint32_t
+txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
{
unsigned int i;
- struct ibv_mr *mr;
+ uint32_t lkey = (uint32_t)-1;
for (i = 0; (i != RTE_DIM(txq->mp2mr)); ++i) {
if (unlikely(txq->mp2mr[i].mp == NULL)) {
}
if (txq->mp2mr[i].mp == mp) {
assert(txq->mp2mr[i].lkey != (uint32_t)-1);
- assert(txq->mp2mr[i].mr->lkey == txq->mp2mr[i].lkey);
- return txq->mp2mr[i].lkey;
+ assert(htonl(txq->mp2mr[i].mr->lkey) ==
+ txq->mp2mr[i].lkey);
+ lkey = txq->mp2mr[i].lkey;
+ break;
}
}
- /* Add a new entry, register MR first. */
- DEBUG("%p: discovered new memory pool \"%s\" (%p)",
- (void *)txq, mp->name, (const void *)mp);
- mr = ibv_reg_mr(txq->priv->pd,
- (void *)mp->elt_va_start,
- (mp->elt_va_end - mp->elt_va_start),
- (IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE));
- if (unlikely(mr == NULL)) {
- DEBUG("%p: unable to configure MR, ibv_reg_mr() failed.",
- (void *)txq);
- return (uint32_t)-1;
- }
- if (unlikely(i == RTE_DIM(txq->mp2mr))) {
- /* Table is full, remove oldest entry. */
- DEBUG("%p: MR <-> MP table full, dropping oldest entry.",
- (void *)txq);
- --i;
- claim_zero(ibv_dereg_mr(txq->mp2mr[0].mr));
- memmove(&txq->mp2mr[0], &txq->mp2mr[1],
- (sizeof(txq->mp2mr) - sizeof(txq->mp2mr[0])));
- }
- /* Store the new entry. */
- txq->mp2mr[i].mp = mp;
- txq->mp2mr[i].mr = mr;
- txq->mp2mr[i].lkey = mr->lkey;
- DEBUG("%p: new MR lkey for MP \"%s\" (%p): 0x%08" PRIu32,
- (void *)txq, mp->name, (const void *)mp, txq->mp2mr[i].lkey);
- return txq->mp2mr[i].lkey;
+ if (unlikely(lkey == (uint32_t)-1))
+ lkey = txq_mp2mr_reg(txq, mp, i);
+ return lkey;
}
-struct txq_mp2mr_mbuf_check_data {
- const struct rte_mempool *mp;
- int ret;
-};
-
/**
- * Callback function for rte_mempool_obj_iter() to check whether a given
- * mempool object looks like a mbuf.
+ * Ring TX queue doorbell.
*
- * @param[in, out] arg
- * Context data (struct txq_mp2mr_mbuf_check_data). Contains mempool pointer
- * and return value.
- * @param[in] start
- * Object start address.
- * @param[in] end
- * Object end address.
- * @param index
- * Unused.
- *
- * @return
- * Nonzero value when object is not a mbuf.
- */
-static void
-txq_mp2mr_mbuf_check(void *arg, void *start, void *end,
- uint32_t index __rte_unused)
-{
- struct txq_mp2mr_mbuf_check_data *data = arg;
- struct rte_mbuf *buf =
- (void *)((uintptr_t)start + data->mp->header_size);
-
- (void)index;
- /* Check whether mbuf structure fits element size and whether mempool
- * pointer is valid. */
- if (((uintptr_t)end >= (uintptr_t)(buf + 1)) &&
- (buf->pool == data->mp))
- data->ret = 0;
- else
- data->ret = -1;
-}
-
-/**
- * Iterator function for rte_mempool_walk() to register existing mempools and
- * fill the MP to MR cache of a TX queue.
- *
- * @param[in] mp
- * Memory Pool to register.
- * @param *arg
+ * @param txq
* Pointer to TX queue structure.
+ * @param wqe
+ * Pointer to the last WQE posted in the NIC.
*/
-void
-txq_mp2mr_iter(const struct rte_mempool *mp, void *arg)
+static inline void
+mlx5_tx_dbrec(struct txq *txq, volatile struct mlx5_wqe *wqe)
{
- struct txq *txq = arg;
- struct txq_mp2mr_mbuf_check_data data = {
- .mp = mp,
- .ret = -1,
- };
+ uint64_t *dst = (uint64_t *)((uintptr_t)txq->bf_reg);
+ volatile uint64_t *src = ((volatile uint64_t *)wqe);
- /* Discard empty mempools. */
- if (mp->size == 0)
- return;
- /* Register mempool only if the first element looks like a mbuf. */
- rte_mempool_obj_iter((void *)mp->elt_va_start,
- 1,
- mp->header_size + mp->elt_size + mp->trailer_size,
- 1,
- mp->elt_pa,
- mp->pg_num,
- mp->pg_shift,
- txq_mp2mr_mbuf_check,
- &data);
- if (data.ret)
- return;
- txq_mp2mr(txq, mp);
+ rte_wmb();
+ *txq->qp_db = htonl(txq->wqe_ci);
+ /* Ensure ordering between DB record and BF copy. */
+ rte_wmb();
+ *dst = *src;
}
-#if MLX5_PMD_SGE_WR_N > 1
-
/**
- * Copy scattered mbuf contents to a single linear buffer.
+ * DPDK callback to check the status of a tx descriptor.
*
- * @param[out] linear
- * Linear output buffer.
- * @param[in] buf
- * Scattered input buffer.
+ * @param tx_queue
+ * The tx queue.
+ * @param[in] offset
+ * The index of the descriptor in the ring.
*
* @return
- * Number of bytes copied to the output buffer or 0 if not large enough.
+ * The status of the tx descriptor.
*/
-static unsigned int
-linearize_mbuf(linear_t *linear, struct rte_mbuf *buf)
+int
+mlx5_tx_descriptor_status(void *tx_queue, uint16_t offset)
{
- unsigned int size = 0;
- unsigned int offset;
-
- do {
- unsigned int len = DATA_LEN(buf);
+ struct txq *txq = tx_queue;
+ const unsigned int elts_n = 1 << txq->elts_n;
+ const unsigned int elts_cnt = elts_n - 1;
+ unsigned int used;
- offset = size;
- size += len;
- if (unlikely(size > sizeof(*linear)))
- return 0;
- memcpy(&(*linear)[offset],
- rte_pktmbuf_mtod(buf, uint8_t *),
- len);
- buf = NEXT(buf);
- } while (buf != NULL);
- return size;
+ txq_complete(txq);
+ used = (txq->elts_head - txq->elts_tail) & elts_cnt;
+ if (offset < used)
+ return RTE_ETH_TX_DESC_FULL;
+ return RTE_ETH_TX_DESC_DONE;
}
/**
- * Handle scattered buffers for mlx5_tx_burst().
+ * DPDK callback to check the status of a rx descriptor.
*
- * @param txq
- * TX queue structure.
- * @param segs
- * Number of segments in buf.
- * @param elt
- * TX queue element to fill.
- * @param[in] buf
- * Buffer to process.
- * @param elts_head
- * Index of the linear buffer to use if necessary (normally txq->elts_head).
- * @param[out] sges
- * Array filled with SGEs on success.
+ * @param rx_queue
+ * The rx queue.
+ * @param[in] offset
+ * The index of the descriptor in the ring.
*
* @return
- * A structure containing the processed packet size in bytes and the
- * number of SGEs. Both fields are set to (unsigned int)-1 in case of
- * failure.
+ * The status of the tx descriptor.
*/
-static struct tx_burst_sg_ret {
- unsigned int length;
- unsigned int num;
-}
-tx_burst_sg(struct txq *txq, unsigned int segs, struct txq_elt *elt,
- struct rte_mbuf *buf, unsigned int elts_head,
- struct ibv_sge (*sges)[MLX5_PMD_SGE_WR_N])
+int
+mlx5_rx_descriptor_status(void *rx_queue, uint16_t offset)
{
- unsigned int sent_size = 0;
- unsigned int j;
- int linearize = 0;
-
- /* When there are too many segments, extra segments are
- * linearized in the last SGE. */
- if (unlikely(segs > RTE_DIM(*sges))) {
- segs = (RTE_DIM(*sges) - 1);
- linearize = 1;
- }
- /* Update element. */
- elt->buf = buf;
- /* Register segments as SGEs. */
- for (j = 0; (j != segs); ++j) {
- struct ibv_sge *sge = &(*sges)[j];
- uint32_t lkey;
-
- /* Retrieve Memory Region key for this memory pool. */
- lkey = txq_mp2mr(txq, txq_mb2mp(buf));
- if (unlikely(lkey == (uint32_t)-1)) {
- /* MR does not exist. */
- DEBUG("%p: unable to get MP <-> MR association",
- (void *)txq);
- /* Clean up TX element. */
- elt->buf = NULL;
- goto stop;
- }
- /* Update SGE. */
- sge->addr = rte_pktmbuf_mtod(buf, uintptr_t);
- if (txq->priv->vf)
- rte_prefetch0((volatile void *)
- (uintptr_t)sge->addr);
- sge->length = DATA_LEN(buf);
- sge->lkey = lkey;
- sent_size += sge->length;
- buf = NEXT(buf);
+ struct rxq *rxq = rx_queue;
+ struct rxq_zip *zip = &rxq->zip;
+ volatile struct mlx5_cqe *cqe;
+ const unsigned int cqe_n = (1 << rxq->cqe_n);
+ const unsigned int cqe_cnt = cqe_n - 1;
+ unsigned int cq_ci;
+ unsigned int used;
+
+ /* if we are processing a compressed cqe */
+ if (zip->ai) {
+ used = zip->cqe_cnt - zip->ca;
+ cq_ci = zip->cq_ci;
+ } else {
+ used = 0;
+ cq_ci = rxq->cq_ci;
}
- /* If buf is not NULL here and is not going to be linearized,
- * nb_segs is not valid. */
- assert(j == segs);
- assert((buf == NULL) || (linearize));
- /* Linearize extra segments. */
- if (linearize) {
- struct ibv_sge *sge = &(*sges)[segs];
- linear_t *linear = &(*txq->elts_linear)[elts_head];
- unsigned int size = linearize_mbuf(linear, buf);
-
- assert(segs == (RTE_DIM(*sges) - 1));
- if (size == 0) {
- /* Invalid packet. */
- DEBUG("%p: packet too large to be linearized.",
- (void *)txq);
- /* Clean up TX element. */
- elt->buf = NULL;
- goto stop;
- }
- /* If MLX5_PMD_SGE_WR_N is 1, free mbuf immediately. */
- if (RTE_DIM(*sges) == 1) {
- do {
- struct rte_mbuf *next = NEXT(buf);
+ cqe = &(*rxq->cqes)[cq_ci & cqe_cnt];
+ while (check_cqe(cqe, cqe_n, cq_ci) == 0) {
+ int8_t op_own;
+ unsigned int n;
- rte_pktmbuf_free_seg(buf);
- buf = next;
- } while (buf != NULL);
- elt->buf = NULL;
- }
- /* Update SGE. */
- sge->addr = (uintptr_t)&(*linear)[0];
- sge->length = size;
- sge->lkey = txq->mr_linear->lkey;
- sent_size += size;
- /* Include last segment. */
- segs++;
+ op_own = cqe->op_own;
+ if (MLX5_CQE_FORMAT(op_own) == MLX5_COMPRESSED)
+ n = ntohl(cqe->byte_cnt);
+ else
+ n = 1;
+ cq_ci += n;
+ used += n;
+ cqe = &(*rxq->cqes)[cq_ci & cqe_cnt];
}
- return (struct tx_burst_sg_ret){
- .length = sent_size,
- .num = segs,
- };
-stop:
- return (struct tx_burst_sg_ret){
- .length = -1,
- .num = -1,
- };
+ used = RTE_MIN(used, (1U << rxq->elts_n) - 1);
+ if (offset < used)
+ return RTE_ETH_RX_DESC_DONE;
+ return RTE_ETH_RX_DESC_AVAIL;
}
-#endif /* MLX5_PMD_SGE_WR_N > 1 */
-
/**
* DPDK callback for TX.
*
mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
{
struct txq *txq = (struct txq *)dpdk_txq;
- unsigned int elts_head = txq->elts_head;
- const unsigned int elts_n = txq->elts_n;
- unsigned int elts_comp_cd = txq->elts_comp_cd;
- unsigned int elts_comp = 0;
- unsigned int i;
+ uint16_t elts_head = txq->elts_head;
+ const unsigned int elts_n = 1 << txq->elts_n;
+ unsigned int i = 0;
+ unsigned int j = 0;
+ unsigned int k = 0;
unsigned int max;
- int err;
+ uint16_t max_wqe;
+ unsigned int comp;
+ volatile struct mlx5_wqe_v *wqe = NULL;
+ unsigned int segs_n = 0;
+ struct rte_mbuf *buf = NULL;
+ uint8_t *raw;
- assert(elts_comp_cd != 0);
+ if (unlikely(!pkts_n))
+ return 0;
+ /* Prefetch first packet cacheline. */
+ rte_prefetch0(*pkts);
+ /* Start processing. */
txq_complete(txq);
max = (elts_n - (elts_head - txq->elts_tail));
if (max > elts_n)
max -= elts_n;
- assert(max >= 1);
- assert(max <= elts_n);
- /* Always leave one free entry in the ring. */
- --max;
- if (max == 0)
+ max_wqe = (1u << txq->wqe_n) - (txq->wqe_ci - txq->wqe_pi);
+ if (unlikely(!max_wqe))
return 0;
- if (max > pkts_n)
- max = pkts_n;
- for (i = 0; (i != max); ++i) {
- struct rte_mbuf *buf = pkts[i];
- unsigned int elts_head_next =
- (((elts_head + 1) == elts_n) ? 0 : elts_head + 1);
- struct txq_elt *elt_next = &(*txq->elts)[elts_head_next];
- struct txq_elt *elt = &(*txq->elts)[elts_head];
- unsigned int segs = NB_SEGS(buf);
+ do {
+ volatile rte_v128u32_t *dseg = NULL;
+ uint32_t length;
+ unsigned int ds = 0;
+ uintptr_t addr;
+ uint64_t naddr;
+ uint16_t pkt_inline_sz = MLX5_WQE_DWORD_SIZE + 2;
+ uint16_t tso_header_sz = 0;
+ uint16_t ehdr;
+ uint8_t cs_flags = 0;
+ uint64_t tso = 0;
#ifdef MLX5_PMD_SOFT_COUNTERS
- unsigned int sent_size = 0;
+ uint32_t total_length = 0;
#endif
- uint32_t send_flags = 0;
-
- /* Clean up old buffer. */
- if (likely(elt->buf != NULL)) {
- struct rte_mbuf *tmp = elt->buf;
- /* Faster than rte_pktmbuf_free(). */
- do {
- struct rte_mbuf *next = NEXT(tmp);
+ /* first_seg */
+ buf = *(pkts++);
+ segs_n = buf->nb_segs;
+ /*
+ * Make sure there is enough room to store this packet and
+ * that one ring entry remains unused.
+ */
+ assert(segs_n);
+ if (max < segs_n + 1)
+ break;
+ max -= segs_n;
+ --segs_n;
+ if (!segs_n)
+ --pkts_n;
+ if (unlikely(--max_wqe == 0))
+ break;
+ wqe = (volatile struct mlx5_wqe_v *)
+ tx_mlx5_wqe(txq, txq->wqe_ci);
+ rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
+ if (pkts_n > 1)
+ rte_prefetch0(*pkts);
+ addr = rte_pktmbuf_mtod(buf, uintptr_t);
+ length = DATA_LEN(buf);
+ ehdr = (((uint8_t *)addr)[1] << 8) |
+ ((uint8_t *)addr)[0];
+#ifdef MLX5_PMD_SOFT_COUNTERS
+ total_length = length;
+#endif
+ assert(length >= MLX5_WQE_DWORD_SIZE);
+ /* Update element. */
+ (*txq->elts)[elts_head] = buf;
+ elts_head = (elts_head + 1) & (elts_n - 1);
+ /* Prefetch next buffer data. */
+ if (pkts_n > 1) {
+ volatile void *pkt_addr;
- rte_pktmbuf_free_seg(tmp);
- tmp = next;
- } while (tmp != NULL);
- }
- /* Request TX completion. */
- if (unlikely(--elts_comp_cd == 0)) {
- elts_comp_cd = txq->elts_comp_cd_init;
- ++elts_comp;
- send_flags |= IBV_EXP_QP_BURST_SIGNALED;
+ pkt_addr = rte_pktmbuf_mtod(*pkts, volatile void *);
+ rte_prefetch0(pkt_addr);
}
/* Should we enable HW CKSUM offload */
if (buf->ol_flags &
(PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM)) {
- send_flags |= IBV_EXP_QP_BURST_IP_CSUM;
- /* HW does not support checksum offloads at arbitrary
- * offsets but automatically recognizes the packet
- * type. For inner L3/L4 checksums, only VXLAN (UDP)
- * tunnels are currently supported. */
- if (RTE_ETH_IS_TUNNEL_PKT(buf->packet_type))
- send_flags |= IBV_EXP_QP_BURST_TUNNEL;
+ const uint64_t is_tunneled = buf->ol_flags &
+ (PKT_TX_TUNNEL_GRE |
+ PKT_TX_TUNNEL_VXLAN);
+
+ if (is_tunneled && txq->tunnel_en) {
+ cs_flags = MLX5_ETH_WQE_L3_INNER_CSUM |
+ MLX5_ETH_WQE_L4_INNER_CSUM;
+ if (buf->ol_flags & PKT_TX_OUTER_IP_CKSUM)
+ cs_flags |= MLX5_ETH_WQE_L3_CSUM;
+ } else {
+ cs_flags = MLX5_ETH_WQE_L3_CSUM |
+ MLX5_ETH_WQE_L4_CSUM;
+ }
}
- if (likely(segs == 1)) {
- uintptr_t addr;
- uint32_t length;
- uint32_t lkey;
+ raw = ((uint8_t *)(uintptr_t)wqe) + 2 * MLX5_WQE_DWORD_SIZE;
+ /* Replace the Ethernet type by the VLAN if necessary. */
+ if (buf->ol_flags & PKT_TX_VLAN_PKT) {
+ uint32_t vlan = htonl(0x81000000 | buf->vlan_tci);
+ unsigned int len = 2 * ETHER_ADDR_LEN - 2;
- /* Retrieve buffer information. */
- addr = rte_pktmbuf_mtod(buf, uintptr_t);
- length = DATA_LEN(buf);
- /* Retrieve Memory Region key for this memory pool. */
- lkey = txq_mp2mr(txq, txq_mb2mp(buf));
- if (unlikely(lkey == (uint32_t)-1)) {
- /* MR does not exist. */
- DEBUG("%p: unable to get MP <-> MR"
- " association", (void *)txq);
- /* Clean up TX element. */
- elt->buf = NULL;
- goto stop;
+ addr += 2;
+ length -= 2;
+ /* Copy Destination and source mac address. */
+ memcpy((uint8_t *)raw, ((uint8_t *)addr), len);
+ /* Copy VLAN. */
+ memcpy((uint8_t *)raw + len, &vlan, sizeof(vlan));
+ /* Copy missing two bytes to end the DSeg. */
+ memcpy((uint8_t *)raw + len + sizeof(vlan),
+ ((uint8_t *)addr) + len, 2);
+ addr += len + 2;
+ length -= (len + 2);
+ } else {
+ memcpy((uint8_t *)raw, ((uint8_t *)addr) + 2,
+ MLX5_WQE_DWORD_SIZE);
+ length -= pkt_inline_sz;
+ addr += pkt_inline_sz;
+ }
+ if (txq->tso_en) {
+ tso = buf->ol_flags & PKT_TX_TCP_SEG;
+ if (tso) {
+ uintptr_t end = (uintptr_t)
+ (((uintptr_t)txq->wqes) +
+ (1 << txq->wqe_n) *
+ MLX5_WQE_SIZE);
+ unsigned int copy_b;
+ uint8_t vlan_sz = (buf->ol_flags &
+ PKT_TX_VLAN_PKT) ? 4 : 0;
+ const uint64_t is_tunneled =
+ buf->ol_flags &
+ (PKT_TX_TUNNEL_GRE |
+ PKT_TX_TUNNEL_VXLAN);
+
+ tso_header_sz = buf->l2_len + vlan_sz +
+ buf->l3_len + buf->l4_len;
+
+ if (is_tunneled && txq->tunnel_en) {
+ tso_header_sz += buf->outer_l2_len +
+ buf->outer_l3_len;
+ cs_flags |= MLX5_ETH_WQE_L4_INNER_CSUM;
+ } else {
+ cs_flags |= MLX5_ETH_WQE_L4_CSUM;
+ }
+ if (unlikely(tso_header_sz >
+ MLX5_MAX_TSO_HEADER))
+ break;
+ copy_b = tso_header_sz - pkt_inline_sz;
+ /* First seg must contain all headers. */
+ assert(copy_b <= length);
+ raw += MLX5_WQE_DWORD_SIZE;
+ if (copy_b &&
+ ((end - (uintptr_t)raw) > copy_b)) {
+ uint16_t n = (MLX5_WQE_DS(copy_b) -
+ 1 + 3) / 4;
+
+ if (unlikely(max_wqe < n))
+ break;
+ max_wqe -= n;
+ rte_memcpy((void *)raw,
+ (void *)addr, copy_b);
+ addr += copy_b;
+ length -= copy_b;
+ pkt_inline_sz += copy_b;
+ /*
+ * Another DWORD will be added
+ * in the inline part.
+ */
+ raw += MLX5_WQE_DS(copy_b) *
+ MLX5_WQE_DWORD_SIZE -
+ MLX5_WQE_DWORD_SIZE;
+ } else {
+ /* NOP WQE. */
+ wqe->ctrl = (rte_v128u32_t){
+ htonl(txq->wqe_ci << 8),
+ htonl(txq->qp_num_8s | 1),
+ 0,
+ 0,
+ };
+ ds = 1;
+ total_length = 0;
+ pkts--;
+ pkts_n++;
+ elts_head = (elts_head - 1) &
+ (elts_n - 1);
+ k++;
+ goto next_wqe;
+ }
+ }
+ }
+ /* Inline if enough room. */
+ if (txq->inline_en || tso) {
+ uintptr_t end = (uintptr_t)
+ (((uintptr_t)txq->wqes) +
+ (1 << txq->wqe_n) * MLX5_WQE_SIZE);
+ unsigned int max_inline = txq->max_inline *
+ RTE_CACHE_LINE_SIZE -
+ (pkt_inline_sz - 2);
+ uintptr_t addr_end = (addr + max_inline) &
+ ~(RTE_CACHE_LINE_SIZE - 1);
+ unsigned int copy_b = (addr_end > addr) ?
+ RTE_MIN((addr_end - addr), length) :
+ 0;
+
+ raw += MLX5_WQE_DWORD_SIZE;
+ if (copy_b && ((end - (uintptr_t)raw) > copy_b)) {
+ /*
+ * One Dseg remains in the current WQE. To
+ * keep the computation positive, it is
+ * removed after the bytes to Dseg conversion.
+ */
+ uint16_t n = (MLX5_WQE_DS(copy_b) - 1 + 3) / 4;
+
+ if (unlikely(max_wqe < n))
+ break;
+ max_wqe -= n;
+ if (tso) {
+ uint32_t inl =
+ htonl(copy_b | MLX5_INLINE_SEG);
+
+ pkt_inline_sz =
+ MLX5_WQE_DS(tso_header_sz) *
+ MLX5_WQE_DWORD_SIZE;
+ rte_memcpy((void *)raw,
+ (void *)&inl, sizeof(inl));
+ raw += sizeof(inl);
+ pkt_inline_sz += sizeof(inl);
+ }
+ rte_memcpy((void *)raw, (void *)addr, copy_b);
+ addr += copy_b;
+ length -= copy_b;
+ pkt_inline_sz += copy_b;
+ }
+ /*
+ * 2 DWORDs consumed by the WQE header + ETH segment +
+ * the size of the inline part of the packet.
+ */
+ ds = 2 + MLX5_WQE_DS(pkt_inline_sz - 2);
+ if (length > 0) {
+ if (ds % (MLX5_WQE_SIZE /
+ MLX5_WQE_DWORD_SIZE) == 0) {
+ if (unlikely(--max_wqe == 0))
+ break;
+ dseg = (volatile rte_v128u32_t *)
+ tx_mlx5_wqe(txq, txq->wqe_ci +
+ ds / 4);
+ } else {
+ dseg = (volatile rte_v128u32_t *)
+ ((uintptr_t)wqe +
+ (ds * MLX5_WQE_DWORD_SIZE));
+ }
+ goto use_dseg;
+ } else if (!segs_n) {
+ goto next_pkt;
+ } else {
+ /* dseg will be advance as part of next_seg */
+ dseg = (volatile rte_v128u32_t *)
+ ((uintptr_t)wqe +
+ ((ds - 1) * MLX5_WQE_DWORD_SIZE));
+ goto next_seg;
}
- /* Update element. */
- elt->buf = buf;
- if (txq->priv->vf)
- rte_prefetch0((volatile void *)
- (uintptr_t)addr);
- RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
- /* Put packet into send queue. */
-#if MLX5_PMD_MAX_INLINE > 0
- if (length <= txq->max_inline)
- err = txq->if_qp->send_pending_inline
- (txq->qp,
- (void *)addr,
- length,
- send_flags);
- else
-#endif
- err = txq->if_qp->send_pending
- (txq->qp,
- addr,
- length,
- lkey,
- send_flags);
- if (unlikely(err))
- goto stop;
-#ifdef MLX5_PMD_SOFT_COUNTERS
- sent_size += length;
-#endif
} else {
-#if MLX5_PMD_SGE_WR_N > 1
- struct ibv_sge sges[MLX5_PMD_SGE_WR_N];
- struct tx_burst_sg_ret ret;
-
- ret = tx_burst_sg(txq, segs, elt, buf, elts_head,
- &sges);
- if (ret.length == (unsigned int)-1)
- goto stop;
- RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
- /* Put SG list into send queue. */
- err = txq->if_qp->send_pending_sg_list
- (txq->qp,
- sges,
- ret.num,
- send_flags);
- if (unlikely(err))
- goto stop;
+ /*
+ * No inline has been done in the packet, only the
+ * Ethernet Header as been stored.
+ */
+ dseg = (volatile rte_v128u32_t *)
+ ((uintptr_t)wqe + (3 * MLX5_WQE_DWORD_SIZE));
+ ds = 3;
+use_dseg:
+ /* Add the remaining packet as a simple ds. */
+ naddr = htonll(addr);
+ *dseg = (rte_v128u32_t){
+ htonl(length),
+ txq_mp2mr(txq, txq_mb2mp(buf)),
+ naddr,
+ naddr >> 32,
+ };
+ ++ds;
+ if (!segs_n)
+ goto next_pkt;
+ }
+next_seg:
+ assert(buf);
+ assert(ds);
+ assert(wqe);
+ /*
+ * Spill on next WQE when the current one does not have
+ * enough room left. Size of WQE must a be a multiple
+ * of data segment size.
+ */
+ assert(!(MLX5_WQE_SIZE % MLX5_WQE_DWORD_SIZE));
+ if (!(ds % (MLX5_WQE_SIZE / MLX5_WQE_DWORD_SIZE))) {
+ if (unlikely(--max_wqe == 0))
+ break;
+ dseg = (volatile rte_v128u32_t *)
+ tx_mlx5_wqe(txq, txq->wqe_ci + ds / 4);
+ rte_prefetch0(tx_mlx5_wqe(txq,
+ txq->wqe_ci + ds / 4 + 1));
+ } else {
+ ++dseg;
+ }
+ ++ds;
+ buf = buf->next;
+ assert(buf);
+ length = DATA_LEN(buf);
#ifdef MLX5_PMD_SOFT_COUNTERS
- sent_size += ret.length;
+ total_length += length;
#endif
-#else /* MLX5_PMD_SGE_WR_N > 1 */
- DEBUG("%p: TX scattered buffers support not"
- " compiled in", (void *)txq);
- goto stop;
-#endif /* MLX5_PMD_SGE_WR_N > 1 */
+ /* Store segment information. */
+ naddr = htonll(rte_pktmbuf_mtod(buf, uintptr_t));
+ *dseg = (rte_v128u32_t){
+ htonl(length),
+ txq_mp2mr(txq, txq_mb2mp(buf)),
+ naddr,
+ naddr >> 32,
+ };
+ (*txq->elts)[elts_head] = buf;
+ elts_head = (elts_head + 1) & (elts_n - 1);
+ ++j;
+ --segs_n;
+ if (segs_n)
+ goto next_seg;
+ else
+ --pkts_n;
+next_pkt:
+ ++i;
+ /* Initialize known and common part of the WQE structure. */
+ if (tso) {
+ wqe->ctrl = (rte_v128u32_t){
+ htonl((txq->wqe_ci << 8) | MLX5_OPCODE_TSO),
+ htonl(txq->qp_num_8s | ds),
+ 0,
+ 0,
+ };
+ wqe->eseg = (rte_v128u32_t){
+ 0,
+ cs_flags | (htons(buf->tso_segsz) << 16),
+ 0,
+ (ehdr << 16) | htons(tso_header_sz),
+ };
+ } else {
+ wqe->ctrl = (rte_v128u32_t){
+ htonl((txq->wqe_ci << 8) | MLX5_OPCODE_SEND),
+ htonl(txq->qp_num_8s | ds),
+ 0,
+ 0,
+ };
+ wqe->eseg = (rte_v128u32_t){
+ 0,
+ cs_flags,
+ 0,
+ (ehdr << 16) | htons(pkt_inline_sz),
+ };
}
- elts_head = elts_head_next;
+next_wqe:
+ txq->wqe_ci += (ds + 3) / 4;
#ifdef MLX5_PMD_SOFT_COUNTERS
/* Increment sent bytes counter. */
- txq->stats.obytes += sent_size;
+ txq->stats.obytes += total_length;
#endif
- }
-stop:
+ } while (pkts_n);
/* Take a shortcut if nothing must be sent. */
- if (unlikely(i == 0))
+ if (unlikely((i + k) == 0))
return 0;
+ /* Check whether completion threshold has been reached. */
+ comp = txq->elts_comp + i + j + k;
+ if (comp >= MLX5_TX_COMP_THRESH) {
+ volatile struct mlx5_wqe_ctrl *w =
+ (volatile struct mlx5_wqe_ctrl *)wqe;
+
+ /* Request completion on last WQE. */
+ w->ctrl2 = htonl(8);
+ /* Save elts_head in unused "immediate" field of WQE. */
+ w->ctrl3 = elts_head;
+ txq->elts_comp = 0;
+ } else {
+ txq->elts_comp = comp;
+ }
#ifdef MLX5_PMD_SOFT_COUNTERS
/* Increment sent packets counter. */
txq->stats.opackets += i;
#endif
/* Ring QP doorbell. */
- err = txq->if_qp->send_flush(txq->qp);
- if (unlikely(err)) {
- /* A nonzero value is not supposed to be returned.
- * Nothing can be done about it. */
- DEBUG("%p: send_flush() failed with error %d",
- (void *)txq, err);
- }
+ mlx5_tx_dbrec(txq, (volatile struct mlx5_wqe *)wqe);
txq->elts_head = elts_head;
- txq->elts_comp += elts_comp;
- txq->elts_comp_cd = elts_comp_cd;
return i;
}
/**
- * Translate RX completion flags to packet type.
+ * Open a MPW session.
*
- * @param flags
- * RX completion flags returned by poll_length_flags().
+ * @param txq
+ * Pointer to TX queue structure.
+ * @param mpw
+ * Pointer to MPW session structure.
+ * @param length
+ * Packet length.
+ */
+static inline void
+mlx5_mpw_new(struct txq *txq, struct mlx5_mpw *mpw, uint32_t length)
+{
+ uint16_t idx = txq->wqe_ci & ((1 << txq->wqe_n) - 1);
+ volatile struct mlx5_wqe_data_seg (*dseg)[MLX5_MPW_DSEG_MAX] =
+ (volatile struct mlx5_wqe_data_seg (*)[])
+ tx_mlx5_wqe(txq, idx + 1);
+
+ mpw->state = MLX5_MPW_STATE_OPENED;
+ mpw->pkts_n = 0;
+ mpw->len = length;
+ mpw->total_len = 0;
+ mpw->wqe = (volatile struct mlx5_wqe *)tx_mlx5_wqe(txq, idx);
+ mpw->wqe->eseg.mss = htons(length);
+ mpw->wqe->eseg.inline_hdr_sz = 0;
+ mpw->wqe->eseg.rsvd0 = 0;
+ mpw->wqe->eseg.rsvd1 = 0;
+ mpw->wqe->eseg.rsvd2 = 0;
+ mpw->wqe->ctrl[0] = htonl((MLX5_OPC_MOD_MPW << 24) |
+ (txq->wqe_ci << 8) | MLX5_OPCODE_TSO);
+ mpw->wqe->ctrl[2] = 0;
+ mpw->wqe->ctrl[3] = 0;
+ mpw->data.dseg[0] = (volatile struct mlx5_wqe_data_seg *)
+ (((uintptr_t)mpw->wqe) + (2 * MLX5_WQE_DWORD_SIZE));
+ mpw->data.dseg[1] = (volatile struct mlx5_wqe_data_seg *)
+ (((uintptr_t)mpw->wqe) + (3 * MLX5_WQE_DWORD_SIZE));
+ mpw->data.dseg[2] = &(*dseg)[0];
+ mpw->data.dseg[3] = &(*dseg)[1];
+ mpw->data.dseg[4] = &(*dseg)[2];
+}
+
+/**
+ * Close a MPW session.
*
- * @return
- * Packet type for struct rte_mbuf.
+ * @param txq
+ * Pointer to TX queue structure.
+ * @param mpw
+ * Pointer to MPW session structure.
*/
-static inline uint32_t
-rxq_cq_to_pkt_type(uint32_t flags)
+static inline void
+mlx5_mpw_close(struct txq *txq, struct mlx5_mpw *mpw)
{
- uint32_t pkt_type;
+ unsigned int num = mpw->pkts_n;
- if (flags & IBV_EXP_CQ_RX_TUNNEL_PACKET)
- pkt_type =
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_OUTER_IPV4_PACKET,
- RTE_PTYPE_L3_IPV4) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_OUTER_IPV6_PACKET,
- RTE_PTYPE_L3_IPV6) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_IPV4_PACKET,
- RTE_PTYPE_INNER_L3_IPV4) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_IPV6_PACKET,
- RTE_PTYPE_INNER_L3_IPV6);
+ /*
+ * Store size in multiple of 16 bytes. Control and Ethernet segments
+ * count as 2.
+ */
+ mpw->wqe->ctrl[1] = htonl(txq->qp_num_8s | (2 + num));
+ mpw->state = MLX5_MPW_STATE_CLOSED;
+ if (num < 3)
+ ++txq->wqe_ci;
else
- pkt_type =
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_IPV4_PACKET,
- RTE_PTYPE_L3_IPV4) |
- TRANSPOSE(flags,
- IBV_EXP_CQ_RX_IPV6_PACKET,
- RTE_PTYPE_L3_IPV6);
- return pkt_type;
+ txq->wqe_ci += 2;
+ rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci));
+ rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
}
/**
- * Translate RX completion flags to offload flags.
+ * DPDK callback for TX with MPW support.
*
- * @param[in] rxq
- * Pointer to RX queue structure.
- * @param flags
- * RX completion flags returned by poll_length_flags().
+ * @param dpdk_txq
+ * Generic pointer to TX queue structure.
+ * @param[in] pkts
+ * Packets to transmit.
+ * @param pkts_n
+ * Number of packets in array.
*
* @return
- * Offload flags (ol_flags) for struct rte_mbuf.
+ * Number of packets successfully transmitted (<= pkts_n).
*/
-static inline uint32_t
-rxq_cq_to_ol_flags(const struct rxq *rxq, uint32_t flags)
+uint16_t
+mlx5_tx_burst_mpw(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
{
- uint32_t ol_flags = 0;
+ struct txq *txq = (struct txq *)dpdk_txq;
+ uint16_t elts_head = txq->elts_head;
+ const unsigned int elts_n = 1 << txq->elts_n;
+ unsigned int i = 0;
+ unsigned int j = 0;
+ unsigned int max;
+ uint16_t max_wqe;
+ unsigned int comp;
+ struct mlx5_mpw mpw = {
+ .state = MLX5_MPW_STATE_CLOSED,
+ };
- if (rxq->csum)
- ol_flags |=
- TRANSPOSE(~flags,
- IBV_EXP_CQ_RX_IP_CSUM_OK,
- PKT_RX_IP_CKSUM_BAD) |
- TRANSPOSE(~flags,
- IBV_EXP_CQ_RX_TCP_UDP_CSUM_OK,
- PKT_RX_L4_CKSUM_BAD);
+ if (unlikely(!pkts_n))
+ return 0;
+ /* Prefetch first packet cacheline. */
+ rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci));
+ rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
+ /* Start processing. */
+ txq_complete(txq);
+ max = (elts_n - (elts_head - txq->elts_tail));
+ if (max > elts_n)
+ max -= elts_n;
+ max_wqe = (1u << txq->wqe_n) - (txq->wqe_ci - txq->wqe_pi);
+ if (unlikely(!max_wqe))
+ return 0;
+ do {
+ struct rte_mbuf *buf = *(pkts++);
+ unsigned int elts_head_next;
+ uint32_t length;
+ unsigned int segs_n = buf->nb_segs;
+ uint32_t cs_flags = 0;
+
+ /*
+ * Make sure there is enough room to store this packet and
+ * that one ring entry remains unused.
+ */
+ assert(segs_n);
+ if (max < segs_n + 1)
+ break;
+ /* Do not bother with large packets MPW cannot handle. */
+ if (segs_n > MLX5_MPW_DSEG_MAX)
+ break;
+ max -= segs_n;
+ --pkts_n;
+ /* Should we enable HW CKSUM offload */
+ if (buf->ol_flags &
+ (PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM))
+ cs_flags = MLX5_ETH_WQE_L3_CSUM | MLX5_ETH_WQE_L4_CSUM;
+ /* Retrieve packet information. */
+ length = PKT_LEN(buf);
+ assert(length);
+ /* Start new session if packet differs. */
+ if ((mpw.state == MLX5_MPW_STATE_OPENED) &&
+ ((mpw.len != length) ||
+ (segs_n != 1) ||
+ (mpw.wqe->eseg.cs_flags != cs_flags)))
+ mlx5_mpw_close(txq, &mpw);
+ if (mpw.state == MLX5_MPW_STATE_CLOSED) {
+ /*
+ * Multi-Packet WQE consumes at most two WQE.
+ * mlx5_mpw_new() expects to be able to use such
+ * resources.
+ */
+ if (unlikely(max_wqe < 2))
+ break;
+ max_wqe -= 2;
+ mlx5_mpw_new(txq, &mpw, length);
+ mpw.wqe->eseg.cs_flags = cs_flags;
+ }
+ /* Multi-segment packets must be alone in their MPW. */
+ assert((segs_n == 1) || (mpw.pkts_n == 0));
+#if defined(MLX5_PMD_SOFT_COUNTERS) || !defined(NDEBUG)
+ length = 0;
+#endif
+ do {
+ volatile struct mlx5_wqe_data_seg *dseg;
+ uintptr_t addr;
+
+ elts_head_next = (elts_head + 1) & (elts_n - 1);
+ assert(buf);
+ (*txq->elts)[elts_head] = buf;
+ dseg = mpw.data.dseg[mpw.pkts_n];
+ addr = rte_pktmbuf_mtod(buf, uintptr_t);
+ *dseg = (struct mlx5_wqe_data_seg){
+ .byte_count = htonl(DATA_LEN(buf)),
+ .lkey = txq_mp2mr(txq, txq_mb2mp(buf)),
+ .addr = htonll(addr),
+ };
+ elts_head = elts_head_next;
+#if defined(MLX5_PMD_SOFT_COUNTERS) || !defined(NDEBUG)
+ length += DATA_LEN(buf);
+#endif
+ buf = buf->next;
+ ++mpw.pkts_n;
+ ++j;
+ } while (--segs_n);
+ assert(length == mpw.len);
+ if (mpw.pkts_n == MLX5_MPW_DSEG_MAX)
+ mlx5_mpw_close(txq, &mpw);
+ elts_head = elts_head_next;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+ /* Increment sent bytes counter. */
+ txq->stats.obytes += length;
+#endif
+ ++i;
+ } while (pkts_n);
+ /* Take a shortcut if nothing must be sent. */
+ if (unlikely(i == 0))
+ return 0;
+ /* Check whether completion threshold has been reached. */
+ /* "j" includes both packets and segments. */
+ comp = txq->elts_comp + j;
+ if (comp >= MLX5_TX_COMP_THRESH) {
+ volatile struct mlx5_wqe *wqe = mpw.wqe;
+
+ /* Request completion on last WQE. */
+ wqe->ctrl[2] = htonl(8);
+ /* Save elts_head in unused "immediate" field of WQE. */
+ wqe->ctrl[3] = elts_head;
+ txq->elts_comp = 0;
+ } else {
+ txq->elts_comp = comp;
+ }
+#ifdef MLX5_PMD_SOFT_COUNTERS
+ /* Increment sent packets counter. */
+ txq->stats.opackets += i;
+#endif
+ /* Ring QP doorbell. */
+ if (mpw.state == MLX5_MPW_STATE_OPENED)
+ mlx5_mpw_close(txq, &mpw);
+ mlx5_tx_dbrec(txq, mpw.wqe);
+ txq->elts_head = elts_head;
+ return i;
+}
+
+/**
+ * Open a MPW inline session.
+ *
+ * @param txq
+ * Pointer to TX queue structure.
+ * @param mpw
+ * Pointer to MPW session structure.
+ * @param length
+ * Packet length.
+ */
+static inline void
+mlx5_mpw_inline_new(struct txq *txq, struct mlx5_mpw *mpw, uint32_t length)
+{
+ uint16_t idx = txq->wqe_ci & ((1 << txq->wqe_n) - 1);
+ struct mlx5_wqe_inl_small *inl;
+
+ mpw->state = MLX5_MPW_INL_STATE_OPENED;
+ mpw->pkts_n = 0;
+ mpw->len = length;
+ mpw->total_len = 0;
+ mpw->wqe = (volatile struct mlx5_wqe *)tx_mlx5_wqe(txq, idx);
+ mpw->wqe->ctrl[0] = htonl((MLX5_OPC_MOD_MPW << 24) |
+ (txq->wqe_ci << 8) |
+ MLX5_OPCODE_TSO);
+ mpw->wqe->ctrl[2] = 0;
+ mpw->wqe->ctrl[3] = 0;
+ mpw->wqe->eseg.mss = htons(length);
+ mpw->wqe->eseg.inline_hdr_sz = 0;
+ mpw->wqe->eseg.cs_flags = 0;
+ mpw->wqe->eseg.rsvd0 = 0;
+ mpw->wqe->eseg.rsvd1 = 0;
+ mpw->wqe->eseg.rsvd2 = 0;
+ inl = (struct mlx5_wqe_inl_small *)
+ (((uintptr_t)mpw->wqe) + 2 * MLX5_WQE_DWORD_SIZE);
+ mpw->data.raw = (uint8_t *)&inl->raw;
+}
+
+/**
+ * Close a MPW inline session.
+ *
+ * @param txq
+ * Pointer to TX queue structure.
+ * @param mpw
+ * Pointer to MPW session structure.
+ */
+static inline void
+mlx5_mpw_inline_close(struct txq *txq, struct mlx5_mpw *mpw)
+{
+ unsigned int size;
+ struct mlx5_wqe_inl_small *inl = (struct mlx5_wqe_inl_small *)
+ (((uintptr_t)mpw->wqe) + (2 * MLX5_WQE_DWORD_SIZE));
+
+ size = MLX5_WQE_SIZE - MLX5_MWQE64_INL_DATA + mpw->total_len;
/*
- * PKT_RX_IP_CKSUM_BAD and PKT_RX_L4_CKSUM_BAD are used in place
- * of PKT_RX_EIP_CKSUM_BAD because the latter is not functional
- * (its value is 0).
+ * Store size in multiple of 16 bytes. Control and Ethernet segments
+ * count as 2.
*/
- if ((flags & IBV_EXP_CQ_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
- ol_flags |=
- TRANSPOSE(~flags,
- IBV_EXP_CQ_RX_OUTER_IP_CSUM_OK,
- PKT_RX_IP_CKSUM_BAD) |
- TRANSPOSE(~flags,
- IBV_EXP_CQ_RX_OUTER_TCP_UDP_CSUM_OK,
- PKT_RX_L4_CKSUM_BAD);
- return ol_flags;
+ mpw->wqe->ctrl[1] = htonl(txq->qp_num_8s | MLX5_WQE_DS(size));
+ mpw->state = MLX5_MPW_STATE_CLOSED;
+ inl->byte_cnt = htonl(mpw->total_len | MLX5_INLINE_SEG);
+ txq->wqe_ci += (size + (MLX5_WQE_SIZE - 1)) / MLX5_WQE_SIZE;
}
/**
- * DPDK callback for RX with scattered packets support.
+ * DPDK callback for TX with MPW inline support.
*
- * @param dpdk_rxq
- * Generic pointer to RX queue structure.
- * @param[out] pkts
- * Array to store received packets.
+ * @param dpdk_txq
+ * Generic pointer to TX queue structure.
+ * @param[in] pkts
+ * Packets to transmit.
* @param pkts_n
- * Maximum number of packets in array.
+ * Number of packets in array.
*
* @return
- * Number of packets successfully received (<= pkts_n).
+ * Number of packets successfully transmitted (<= pkts_n).
*/
uint16_t
-mlx5_rx_burst_sp(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
+mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
+ uint16_t pkts_n)
{
- struct rxq *rxq = (struct rxq *)dpdk_rxq;
- struct rxq_elt_sp (*elts)[rxq->elts_n] = rxq->elts.sp;
- const unsigned int elts_n = rxq->elts_n;
- unsigned int elts_head = rxq->elts_head;
- unsigned int i;
- unsigned int pkts_ret = 0;
- int ret;
+ struct txq *txq = (struct txq *)dpdk_txq;
+ uint16_t elts_head = txq->elts_head;
+ const unsigned int elts_n = 1 << txq->elts_n;
+ unsigned int i = 0;
+ unsigned int j = 0;
+ unsigned int max;
+ uint16_t max_wqe;
+ unsigned int comp;
+ unsigned int inline_room = txq->max_inline * RTE_CACHE_LINE_SIZE;
+ struct mlx5_mpw mpw = {
+ .state = MLX5_MPW_STATE_CLOSED,
+ };
+ /*
+ * Compute the maximum number of WQE which can be consumed by inline
+ * code.
+ * - 2 DSEG for:
+ * - 1 control segment,
+ * - 1 Ethernet segment,
+ * - N Dseg from the inline request.
+ */
+ const unsigned int wqe_inl_n =
+ ((2 * MLX5_WQE_DWORD_SIZE +
+ txq->max_inline * RTE_CACHE_LINE_SIZE) +
+ RTE_CACHE_LINE_SIZE - 1) / RTE_CACHE_LINE_SIZE;
- if (unlikely(!rxq->sp))
- return mlx5_rx_burst(dpdk_rxq, pkts, pkts_n);
- if (unlikely(elts == NULL)) /* See RTE_DEV_CMD_SET_MTU. */
+ if (unlikely(!pkts_n))
return 0;
- for (i = 0; (i != pkts_n); ++i) {
- struct rxq_elt_sp *elt = &(*elts)[elts_head];
- unsigned int len;
- unsigned int pkt_buf_len;
- struct rte_mbuf *pkt_buf = NULL; /* Buffer returned in pkts. */
- struct rte_mbuf **pkt_buf_next = &pkt_buf;
- unsigned int seg_headroom = RTE_PKTMBUF_HEADROOM;
- unsigned int j = 0;
- uint32_t flags;
- uint16_t vlan_tci;
-
- /* Sanity checks. */
- assert(elts_head < rxq->elts_n);
- assert(rxq->elts_head < rxq->elts_n);
-#ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS
- ret = rxq->if_cq->poll_length_flags_cvlan(rxq->cq, NULL, NULL,
- &flags, &vlan_tci);
-#else /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
- ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL,
- &flags);
- (void)vlan_tci;
-#endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
- if (unlikely(ret < 0)) {
- struct ibv_wc wc;
- int wcs_n;
-
- DEBUG("rxq=%p, poll_length() failed (ret=%d)",
- (void *)rxq, ret);
- /* ibv_poll_cq() must be used in case of failure. */
- wcs_n = ibv_poll_cq(rxq->cq, 1, &wc);
- if (unlikely(wcs_n == 0))
- break;
- if (unlikely(wcs_n < 0)) {
- DEBUG("rxq=%p, ibv_poll_cq() failed (wcs_n=%d)",
- (void *)rxq, wcs_n);
- break;
- }
- assert(wcs_n == 1);
- if (unlikely(wc.status != IBV_WC_SUCCESS)) {
- /* Whatever, just repost the offending WR. */
- DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work"
- " completion status (%d): %s",
- (void *)rxq, wc.wr_id, wc.status,
- ibv_wc_status_str(wc.status));
-#ifdef MLX5_PMD_SOFT_COUNTERS
- /* Increment dropped packets counter. */
- ++rxq->stats.idropped;
-#endif
- goto repost;
- }
- ret = wc.byte_len;
- }
- if (ret == 0)
+ /* Prefetch first packet cacheline. */
+ rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci));
+ rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
+ /* Start processing. */
+ txq_complete(txq);
+ max = (elts_n - (elts_head - txq->elts_tail));
+ if (max > elts_n)
+ max -= elts_n;
+ do {
+ struct rte_mbuf *buf = *(pkts++);
+ unsigned int elts_head_next;
+ uintptr_t addr;
+ uint32_t length;
+ unsigned int segs_n = buf->nb_segs;
+ uint32_t cs_flags = 0;
+
+ /*
+ * Make sure there is enough room to store this packet and
+ * that one ring entry remains unused.
+ */
+ assert(segs_n);
+ if (max < segs_n + 1)
+ break;
+ /* Do not bother with large packets MPW cannot handle. */
+ if (segs_n > MLX5_MPW_DSEG_MAX)
break;
- len = ret;
- pkt_buf_len = len;
+ max -= segs_n;
+ --pkts_n;
/*
- * Replace spent segments with new ones, concatenate and
- * return them as pkt_buf.
+ * Compute max_wqe in case less WQE were consumed in previous
+ * iteration.
*/
- while (1) {
- struct ibv_sge *sge = &elt->sges[j];
- struct rte_mbuf *seg = elt->bufs[j];
- struct rte_mbuf *rep;
- unsigned int seg_tailroom;
-
- assert(seg != NULL);
- /*
- * Fetch initial bytes of packet descriptor into a
- * cacheline while allocating rep.
- */
- rte_prefetch0(seg);
- rep = __rte_mbuf_raw_alloc(rxq->mp);
- if (unlikely(rep == NULL)) {
+ max_wqe = (1u << txq->wqe_n) - (txq->wqe_ci - txq->wqe_pi);
+ /* Should we enable HW CKSUM offload */
+ if (buf->ol_flags &
+ (PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM))
+ cs_flags = MLX5_ETH_WQE_L3_CSUM | MLX5_ETH_WQE_L4_CSUM;
+ /* Retrieve packet information. */
+ length = PKT_LEN(buf);
+ /* Start new session if packet differs. */
+ if (mpw.state == MLX5_MPW_STATE_OPENED) {
+ if ((mpw.len != length) ||
+ (segs_n != 1) ||
+ (mpw.wqe->eseg.cs_flags != cs_flags))
+ mlx5_mpw_close(txq, &mpw);
+ } else if (mpw.state == MLX5_MPW_INL_STATE_OPENED) {
+ if ((mpw.len != length) ||
+ (segs_n != 1) ||
+ (length > inline_room) ||
+ (mpw.wqe->eseg.cs_flags != cs_flags)) {
+ mlx5_mpw_inline_close(txq, &mpw);
+ inline_room =
+ txq->max_inline * RTE_CACHE_LINE_SIZE;
+ }
+ }
+ if (mpw.state == MLX5_MPW_STATE_CLOSED) {
+ if ((segs_n != 1) ||
+ (length > inline_room)) {
/*
- * Unable to allocate a replacement mbuf,
- * repost WR.
+ * Multi-Packet WQE consumes at most two WQE.
+ * mlx5_mpw_new() expects to be able to use
+ * such resources.
*/
- DEBUG("rxq=%p: can't allocate a new mbuf",
- (void *)rxq);
- if (pkt_buf != NULL) {
- *pkt_buf_next = NULL;
- rte_pktmbuf_free(pkt_buf);
- }
- /* Increment out of memory counters. */
- ++rxq->stats.rx_nombuf;
- ++rxq->priv->dev->data->rx_mbuf_alloc_failed;
- goto repost;
+ if (unlikely(max_wqe < 2))
+ break;
+ max_wqe -= 2;
+ mlx5_mpw_new(txq, &mpw, length);
+ mpw.wqe->eseg.cs_flags = cs_flags;
+ } else {
+ if (unlikely(max_wqe < wqe_inl_n))
+ break;
+ max_wqe -= wqe_inl_n;
+ mlx5_mpw_inline_new(txq, &mpw, length);
+ mpw.wqe->eseg.cs_flags = cs_flags;
}
-#ifndef NDEBUG
- /* Poison user-modifiable fields in rep. */
- NEXT(rep) = (void *)((uintptr_t)-1);
- SET_DATA_OFF(rep, 0xdead);
- DATA_LEN(rep) = 0xd00d;
- PKT_LEN(rep) = 0xdeadd00d;
- NB_SEGS(rep) = 0x2a;
- PORT(rep) = 0x2a;
- rep->ol_flags = -1;
+ }
+ /* Multi-segment packets must be alone in their MPW. */
+ assert((segs_n == 1) || (mpw.pkts_n == 0));
+ if (mpw.state == MLX5_MPW_STATE_OPENED) {
+ assert(inline_room ==
+ txq->max_inline * RTE_CACHE_LINE_SIZE);
+#if defined(MLX5_PMD_SOFT_COUNTERS) || !defined(NDEBUG)
+ length = 0;
+#endif
+ do {
+ volatile struct mlx5_wqe_data_seg *dseg;
+
+ elts_head_next =
+ (elts_head + 1) & (elts_n - 1);
+ assert(buf);
+ (*txq->elts)[elts_head] = buf;
+ dseg = mpw.data.dseg[mpw.pkts_n];
+ addr = rte_pktmbuf_mtod(buf, uintptr_t);
+ *dseg = (struct mlx5_wqe_data_seg){
+ .byte_count = htonl(DATA_LEN(buf)),
+ .lkey = txq_mp2mr(txq, txq_mb2mp(buf)),
+ .addr = htonll(addr),
+ };
+ elts_head = elts_head_next;
+#if defined(MLX5_PMD_SOFT_COUNTERS) || !defined(NDEBUG)
+ length += DATA_LEN(buf);
#endif
- assert(rep->buf_len == seg->buf_len);
- assert(rep->buf_len == rxq->mb_len);
- /* Reconfigure sge to use rep instead of seg. */
- assert(sge->lkey == rxq->mr->lkey);
- sge->addr = ((uintptr_t)rep->buf_addr + seg_headroom);
- elt->bufs[j] = rep;
+ buf = buf->next;
+ ++mpw.pkts_n;
+ ++j;
+ } while (--segs_n);
+ assert(length == mpw.len);
+ if (mpw.pkts_n == MLX5_MPW_DSEG_MAX)
+ mlx5_mpw_close(txq, &mpw);
+ } else {
+ unsigned int max;
+
+ assert(mpw.state == MLX5_MPW_INL_STATE_OPENED);
+ assert(length <= inline_room);
+ assert(length == DATA_LEN(buf));
+ elts_head_next = (elts_head + 1) & (elts_n - 1);
+ addr = rte_pktmbuf_mtod(buf, uintptr_t);
+ (*txq->elts)[elts_head] = buf;
+ /* Maximum number of bytes before wrapping. */
+ max = ((((uintptr_t)(txq->wqes)) +
+ (1 << txq->wqe_n) *
+ MLX5_WQE_SIZE) -
+ (uintptr_t)mpw.data.raw);
+ if (length > max) {
+ rte_memcpy((void *)(uintptr_t)mpw.data.raw,
+ (void *)addr,
+ max);
+ mpw.data.raw = (volatile void *)txq->wqes;
+ rte_memcpy((void *)(uintptr_t)mpw.data.raw,
+ (void *)(addr + max),
+ length - max);
+ mpw.data.raw += length - max;
+ } else {
+ rte_memcpy((void *)(uintptr_t)mpw.data.raw,
+ (void *)addr,
+ length);
+
+ if (length == max)
+ mpw.data.raw =
+ (volatile void *)txq->wqes;
+ else
+ mpw.data.raw += length;
+ }
+ ++mpw.pkts_n;
+ mpw.total_len += length;
++j;
- /* Update pkt_buf if it's the first segment, or link
- * seg to the previous one and update pkt_buf_next. */
- *pkt_buf_next = seg;
- pkt_buf_next = &NEXT(seg);
- /* Update seg information. */
- seg_tailroom = (seg->buf_len - seg_headroom);
- assert(sge->length == seg_tailroom);
- SET_DATA_OFF(seg, seg_headroom);
- if (likely(len <= seg_tailroom)) {
- /* Last segment. */
- DATA_LEN(seg) = len;
- PKT_LEN(seg) = len;
- /* Sanity check. */
- assert(rte_pktmbuf_headroom(seg) ==
- seg_headroom);
- assert(rte_pktmbuf_tailroom(seg) ==
- (seg_tailroom - len));
- break;
+ if (mpw.pkts_n == MLX5_MPW_DSEG_MAX) {
+ mlx5_mpw_inline_close(txq, &mpw);
+ inline_room =
+ txq->max_inline * RTE_CACHE_LINE_SIZE;
+ } else {
+ inline_room -= length;
}
- DATA_LEN(seg) = seg_tailroom;
- PKT_LEN(seg) = seg_tailroom;
- /* Sanity check. */
- assert(rte_pktmbuf_headroom(seg) == seg_headroom);
- assert(rte_pktmbuf_tailroom(seg) == 0);
- /* Fix len and clear headroom for next segments. */
- len -= seg_tailroom;
- seg_headroom = 0;
- }
- /* Update head and tail segments. */
- *pkt_buf_next = NULL;
- assert(pkt_buf != NULL);
- assert(j != 0);
- NB_SEGS(pkt_buf) = j;
- PORT(pkt_buf) = rxq->port_id;
- PKT_LEN(pkt_buf) = pkt_buf_len;
- pkt_buf->packet_type = rxq_cq_to_pkt_type(flags);
- pkt_buf->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
-#ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS
- if (flags & IBV_EXP_CQ_RX_CVLAN_STRIPPED_V1) {
- pkt_buf->ol_flags |= PKT_RX_VLAN_PKT;
- pkt_buf->vlan_tci = vlan_tci;
}
-#endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
-
- /* Return packet. */
- *(pkts++) = pkt_buf;
- ++pkts_ret;
+ elts_head = elts_head_next;
#ifdef MLX5_PMD_SOFT_COUNTERS
- /* Increment bytes counter. */
- rxq->stats.ibytes += pkt_buf_len;
+ /* Increment sent bytes counter. */
+ txq->stats.obytes += length;
#endif
-repost:
- ret = rxq->if_wq->recv_sg_list(rxq->wq,
- elt->sges,
- RTE_DIM(elt->sges));
- if (unlikely(ret)) {
- /* Inability to repost WRs is fatal. */
- DEBUG("%p: recv_sg_list(): failed (ret=%d)",
- (void *)rxq->priv,
- ret);
- abort();
- }
- if (++elts_head >= elts_n)
- elts_head = 0;
- continue;
- }
+ ++i;
+ } while (pkts_n);
+ /* Take a shortcut if nothing must be sent. */
if (unlikely(i == 0))
return 0;
- rxq->elts_head = elts_head;
+ /* Check whether completion threshold has been reached. */
+ /* "j" includes both packets and segments. */
+ comp = txq->elts_comp + j;
+ if (comp >= MLX5_TX_COMP_THRESH) {
+ volatile struct mlx5_wqe *wqe = mpw.wqe;
+
+ /* Request completion on last WQE. */
+ wqe->ctrl[2] = htonl(8);
+ /* Save elts_head in unused "immediate" field of WQE. */
+ wqe->ctrl[3] = elts_head;
+ txq->elts_comp = 0;
+ } else {
+ txq->elts_comp = comp;
+ }
#ifdef MLX5_PMD_SOFT_COUNTERS
- /* Increment packets counter. */
- rxq->stats.ipackets += pkts_ret;
+ /* Increment sent packets counter. */
+ txq->stats.opackets += i;
#endif
- return pkts_ret;
+ /* Ring QP doorbell. */
+ if (mpw.state == MLX5_MPW_INL_STATE_OPENED)
+ mlx5_mpw_inline_close(txq, &mpw);
+ else if (mpw.state == MLX5_MPW_STATE_OPENED)
+ mlx5_mpw_close(txq, &mpw);
+ mlx5_tx_dbrec(txq, mpw.wqe);
+ txq->elts_head = elts_head;
+ return i;
}
/**
- * DPDK callback for RX.
+ * Translate RX completion flags to packet type.
+ *
+ * @param[in] cqe
+ * Pointer to CQE.
+ *
+ * @note: fix mlx5_dev_supported_ptypes_get() if any change here.
+ *
+ * @return
+ * Packet type for struct rte_mbuf.
+ */
+static inline uint32_t
+rxq_cq_to_pkt_type(volatile struct mlx5_cqe *cqe)
+{
+ uint32_t pkt_type;
+ uint16_t flags = ntohs(cqe->hdr_type_etc);
+
+ if (cqe->pkt_info & MLX5_CQE_RX_TUNNEL_PACKET) {
+ pkt_type =
+ TRANSPOSE(flags,
+ MLX5_CQE_RX_IPV4_PACKET,
+ RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN) |
+ TRANSPOSE(flags,
+ MLX5_CQE_RX_IPV6_PACKET,
+ RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN);
+ pkt_type |= ((cqe->pkt_info & MLX5_CQE_RX_OUTER_PACKET) ?
+ RTE_PTYPE_L3_IPV6_EXT_UNKNOWN :
+ RTE_PTYPE_L3_IPV4_EXT_UNKNOWN);
+ } else {
+ pkt_type =
+ TRANSPOSE(flags,
+ MLX5_CQE_L3_HDR_TYPE_IPV6,
+ RTE_PTYPE_L3_IPV6_EXT_UNKNOWN) |
+ TRANSPOSE(flags,
+ MLX5_CQE_L3_HDR_TYPE_IPV4,
+ RTE_PTYPE_L3_IPV4_EXT_UNKNOWN);
+ }
+ return pkt_type;
+}
+
+/**
+ * Get size of the next packet for a given CQE. For compressed CQEs, the
+ * consumer index is updated only once all packets of the current one have
+ * been processed.
+ *
+ * @param rxq
+ * Pointer to RX queue.
+ * @param cqe
+ * CQE to process.
+ * @param[out] rss_hash
+ * Packet RSS Hash result.
+ *
+ * @return
+ * Packet size in bytes (0 if there is none), -1 in case of completion
+ * with error.
+ */
+static inline int
+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;
+ uint16_t cqe_n = cqe_cnt + 1;
+ int len = 0;
+ uint16_t idx, end;
+
+ /* Process compressed data in the CQE and mini arrays. */
+ if (zip->ai) {
+ volatile struct mlx5_mini_cqe8 (*mc)[8] =
+ (volatile struct mlx5_mini_cqe8 (*)[8])
+ (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);
+ if ((++zip->ai & 7) == 0) {
+ /* Invalidate consumed CQEs */
+ idx = zip->ca;
+ end = zip->na;
+ while (idx != end) {
+ (*rxq->cqes)[idx & cqe_cnt].op_own =
+ MLX5_CQE_INVALIDATE;
+ ++idx;
+ }
+ /*
+ * Increment consumer index to skip the number of
+ * CQEs consumed. Hardware leaves holes in the CQ
+ * ring for software use.
+ */
+ zip->ca = zip->na;
+ zip->na += 8;
+ }
+ if (unlikely(rxq->zip.ai == rxq->zip.cqe_cnt)) {
+ /* Invalidate the rest */
+ idx = zip->ca;
+ end = zip->cq_ci;
+
+ while (idx != end) {
+ (*rxq->cqes)[idx & cqe_cnt].op_own =
+ MLX5_CQE_INVALIDATE;
+ ++idx;
+ }
+ rxq->cq_ci = zip->cq_ci;
+ zip->ai = 0;
+ }
+ /* No compressed data, get next CQE and verify if it is compressed. */
+ } else {
+ int ret;
+ int8_t op_own;
+
+ ret = check_cqe(cqe, cqe_n, rxq->cq_ci);
+ if (unlikely(ret == 1))
+ return 0;
+ ++rxq->cq_ci;
+ op_own = cqe->op_own;
+ if (MLX5_CQE_FORMAT(op_own) == MLX5_COMPRESSED) {
+ volatile struct mlx5_mini_cqe8 (*mc)[8] =
+ (volatile struct mlx5_mini_cqe8 (*)[8])
+ (uintptr_t)(&(*rxq->cqes)[rxq->cq_ci &
+ cqe_cnt]);
+
+ /* Fix endianness. */
+ zip->cqe_cnt = ntohl(cqe->byte_cnt);
+ /*
+ * Current mini array position is the one returned by
+ * check_cqe64().
+ *
+ * If completion comprises several mini arrays, as a
+ * special case the second one is located 7 CQEs after
+ * the initial CQE instead of 8 for subsequent ones.
+ */
+ zip->ca = rxq->cq_ci;
+ zip->na = zip->ca + 7;
+ /* Compute the next non compressed CQE. */
+ --rxq->cq_ci;
+ zip->cq_ci = rxq->cq_ci + zip->cqe_cnt;
+ /* Get packet size to return. */
+ len = ntohl((*mc)[0].byte_cnt);
+ *rss_hash = ntohl((*mc)[0].rx_hash_result);
+ zip->ai = 1;
+ /* Prefetch all the entries to be invalidated */
+ idx = zip->ca;
+ end = zip->cq_ci;
+ while (idx != end) {
+ rte_prefetch0(&(*rxq->cqes)[(idx) & cqe_cnt]);
+ ++idx;
+ }
+ } else {
+ len = ntohl(cqe->byte_cnt);
+ *rss_hash = ntohl(cqe->rx_hash_res);
+ }
+ /* Error while receiving packet. */
+ if (unlikely(MLX5_CQE_OPCODE(op_own) == MLX5_CQE_RESP_ERR))
+ return -1;
+ }
+ return len;
+}
+
+/**
+ * Translate RX completion flags to offload flags.
+ *
+ * @param[in] rxq
+ * Pointer to RX queue structure.
+ * @param[in] cqe
+ * Pointer to CQE.
*
- * The following function is the same as mlx5_rx_burst_sp(), except it doesn't
- * manage scattered packets. Improves performance when MRU is lower than the
- * size of the first segment.
+ * @return
+ * Offload flags (ol_flags) for struct rte_mbuf.
+ */
+static inline uint32_t
+rxq_cq_to_ol_flags(struct rxq *rxq, volatile struct mlx5_cqe *cqe)
+{
+ uint32_t ol_flags = 0;
+ uint16_t flags = ntohs(cqe->hdr_type_etc);
+
+ ol_flags =
+ TRANSPOSE(flags,
+ MLX5_CQE_RX_L3_HDR_VALID,
+ PKT_RX_IP_CKSUM_GOOD) |
+ TRANSPOSE(flags,
+ MLX5_CQE_RX_L4_HDR_VALID,
+ PKT_RX_L4_CKSUM_GOOD);
+ if ((cqe->pkt_info & MLX5_CQE_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
+ ol_flags |=
+ TRANSPOSE(flags,
+ MLX5_CQE_RX_L3_HDR_VALID,
+ PKT_RX_IP_CKSUM_GOOD) |
+ TRANSPOSE(flags,
+ MLX5_CQE_RX_L4_HDR_VALID,
+ PKT_RX_L4_CKSUM_GOOD);
+ return ol_flags;
+}
+
+/**
+ * DPDK callback for RX.
*
* @param dpdk_rxq
* Generic pointer to RX queue structure.
uint16_t
mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
{
- struct rxq *rxq = (struct rxq *)dpdk_rxq;
- struct rxq_elt (*elts)[rxq->elts_n] = rxq->elts.no_sp;
- const unsigned int elts_n = rxq->elts_n;
- unsigned int elts_head = rxq->elts_head;
- struct ibv_sge sges[pkts_n];
- unsigned int i;
- unsigned int pkts_ret = 0;
- int ret;
-
- if (unlikely(rxq->sp))
- return mlx5_rx_burst_sp(dpdk_rxq, pkts, pkts_n);
- for (i = 0; (i != pkts_n); ++i) {
- struct rxq_elt *elt = &(*elts)[elts_head];
- unsigned int len;
- struct rte_mbuf *seg = elt->buf;
- struct rte_mbuf *rep;
- uint32_t flags;
- uint16_t vlan_tci;
-
- /* Sanity checks. */
- assert(seg != NULL);
- assert(elts_head < rxq->elts_n);
- assert(rxq->elts_head < rxq->elts_n);
- /*
- * Fetch initial bytes of packet descriptor into a
- * cacheline while allocating rep.
- */
+ struct rxq *rxq = dpdk_rxq;
+ const unsigned int wqe_cnt = (1 << rxq->elts_n) - 1;
+ const unsigned int cqe_cnt = (1 << rxq->cqe_n) - 1;
+ const unsigned int sges_n = rxq->sges_n;
+ struct rte_mbuf *pkt = NULL;
+ struct rte_mbuf *seg = NULL;
+ 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. */
+
+ while (pkts_n) {
+ unsigned int idx = rq_ci & wqe_cnt;
+ volatile struct mlx5_wqe_data_seg *wqe = &(*rxq->wqes)[idx];
+ struct rte_mbuf *rep = (*rxq->elts)[idx];
+ uint32_t rss_hash_res = 0;
+
+ if (pkt)
+ NEXT(seg) = rep;
+ seg = rep;
rte_prefetch0(seg);
- rte_prefetch0(&seg->cacheline1);
-#ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS
- ret = rxq->if_cq->poll_length_flags_cvlan(rxq->cq, NULL, NULL,
- &flags, &vlan_tci);
-#else /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
- ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL,
- &flags);
- (void)vlan_tci;
-#endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
- if (unlikely(ret < 0)) {
- struct ibv_wc wc;
- int wcs_n;
-
- DEBUG("rxq=%p, poll_length() failed (ret=%d)",
- (void *)rxq, ret);
- /* ibv_poll_cq() must be used in case of failure. */
- wcs_n = ibv_poll_cq(rxq->cq, 1, &wc);
- if (unlikely(wcs_n == 0))
+ rte_prefetch0(cqe);
+ rte_prefetch0(wqe);
+ rep = rte_mbuf_raw_alloc(rxq->mp);
+ if (unlikely(rep == NULL)) {
+ ++rxq->stats.rx_nombuf;
+ if (!pkt) {
+ /*
+ * no buffers before we even started,
+ * bail out silently.
+ */
break;
- if (unlikely(wcs_n < 0)) {
- DEBUG("rxq=%p, ibv_poll_cq() failed (wcs_n=%d)",
- (void *)rxq, wcs_n);
+ }
+ while (pkt != seg) {
+ assert(pkt != (*rxq->elts)[idx]);
+ rep = NEXT(pkt);
+ rte_mbuf_refcnt_set(pkt, 0);
+ __rte_mbuf_raw_free(pkt);
+ pkt = rep;
+ }
+ break;
+ }
+ if (!pkt) {
+ cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_cnt];
+ len = mlx5_rx_poll_len(rxq, cqe, cqe_cnt,
+ &rss_hash_res);
+ if (!len) {
+ rte_mbuf_refcnt_set(rep, 0);
+ __rte_mbuf_raw_free(rep);
break;
}
- assert(wcs_n == 1);
- if (unlikely(wc.status != IBV_WC_SUCCESS)) {
- /* Whatever, just repost the offending WR. */
- DEBUG("rxq=%p, wr_id=%" PRIu64 ": bad work"
- " completion status (%d): %s",
- (void *)rxq, wc.wr_id, wc.status,
- ibv_wc_status_str(wc.status));
-#ifdef MLX5_PMD_SOFT_COUNTERS
- /* Increment dropped packets counter. */
+ if (unlikely(len == -1)) {
+ /* RX error, packet is likely too large. */
+ rte_mbuf_refcnt_set(rep, 0);
+ __rte_mbuf_raw_free(rep);
++rxq->stats.idropped;
-#endif
- /* Add SGE to array for repost. */
- sges[i] = elt->sge;
- goto repost;
+ goto skip;
+ }
+ pkt = seg;
+ assert(len >= (rxq->crc_present << 2));
+ /* Update packet information. */
+ pkt->packet_type = 0;
+ pkt->ol_flags = 0;
+ if (rss_hash_res && rxq->rss_hash) {
+ pkt->hash.rss = rss_hash_res;
+ pkt->ol_flags = PKT_RX_RSS_HASH;
+ }
+ if (rxq->mark && (cqe->sop_drop_qpn !=
+ htonl(MLX5_FLOW_MARK_INVALID))) {
+ pkt->ol_flags |= PKT_RX_FDIR;
+ if (cqe->sop_drop_qpn !=
+ htonl(MLX5_FLOW_MARK_DEFAULT)) {
+ uint32_t mark = cqe->sop_drop_qpn;
+
+ pkt->ol_flags |= PKT_RX_FDIR_ID;
+ pkt->hash.fdir.hi =
+ mlx5_flow_mark_get(mark);
+ }
+ }
+ if (rxq->csum | rxq->csum_l2tun | rxq->vlan_strip |
+ rxq->crc_present) {
+ if (rxq->csum) {
+ pkt->packet_type =
+ rxq_cq_to_pkt_type(cqe);
+ pkt->ol_flags |=
+ rxq_cq_to_ol_flags(rxq, cqe);
+ }
+ if (ntohs(cqe->hdr_type_etc) &
+ MLX5_CQE_VLAN_STRIPPED) {
+ pkt->ol_flags |= PKT_RX_VLAN_PKT |
+ PKT_RX_VLAN_STRIPPED;
+ pkt->vlan_tci = ntohs(cqe->vlan_info);
+ }
+ if (rxq->crc_present)
+ len -= ETHER_CRC_LEN;
}
- ret = wc.byte_len;
+ PKT_LEN(pkt) = len;
}
- if (ret == 0)
- break;
- len = ret;
- rep = __rte_mbuf_raw_alloc(rxq->mp);
- if (unlikely(rep == NULL)) {
- /*
- * Unable to allocate a replacement mbuf,
- * repost WR.
- */
- DEBUG("rxq=%p: can't allocate a new mbuf",
- (void *)rxq);
- /* Increment out of memory counters. */
- ++rxq->stats.rx_nombuf;
- ++rxq->priv->dev->data->rx_mbuf_alloc_failed;
- goto repost;
+ DATA_LEN(rep) = DATA_LEN(seg);
+ PKT_LEN(rep) = PKT_LEN(seg);
+ SET_DATA_OFF(rep, DATA_OFF(seg));
+ NB_SEGS(rep) = NB_SEGS(seg);
+ PORT(rep) = PORT(seg);
+ NEXT(rep) = NULL;
+ (*rxq->elts)[idx] = rep;
+ /*
+ * Fill NIC descriptor with the new buffer. The lkey and size
+ * of the buffers are already known, only the buffer address
+ * changes.
+ */
+ wqe->addr = htonll(rte_pktmbuf_mtod(rep, uintptr_t));
+ if (len > DATA_LEN(seg)) {
+ len -= DATA_LEN(seg);
+ ++NB_SEGS(pkt);
+ ++rq_ci;
+ continue;
}
-
- /* Reconfigure sge to use rep instead of seg. */
- elt->sge.addr = (uintptr_t)rep->buf_addr + RTE_PKTMBUF_HEADROOM;
- assert(elt->sge.lkey == rxq->mr->lkey);
- elt->buf = rep;
-
- /* Add SGE to array for repost. */
- sges[i] = elt->sge;
-
- /* Update seg information. */
- SET_DATA_OFF(seg, RTE_PKTMBUF_HEADROOM);
- NB_SEGS(seg) = 1;
- PORT(seg) = rxq->port_id;
- NEXT(seg) = NULL;
- PKT_LEN(seg) = len;
DATA_LEN(seg) = len;
- seg->packet_type = rxq_cq_to_pkt_type(flags);
- seg->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
-#ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS
- if (flags & IBV_EXP_CQ_RX_CVLAN_STRIPPED_V1) {
- seg->ol_flags |= PKT_RX_VLAN_PKT;
- seg->vlan_tci = vlan_tci;
- }
-#endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
-
- /* Return packet. */
- *(pkts++) = seg;
- ++pkts_ret;
#ifdef MLX5_PMD_SOFT_COUNTERS
/* Increment bytes counter. */
- rxq->stats.ibytes += len;
+ rxq->stats.ibytes += PKT_LEN(pkt);
#endif
-repost:
- if (++elts_head >= elts_n)
- elts_head = 0;
- continue;
+ /* Return packet. */
+ *(pkts++) = pkt;
+ pkt = NULL;
+ --pkts_n;
+ ++i;
+skip:
+ /* Align consumer index to the next stride. */
+ rq_ci >>= sges_n;
+ ++rq_ci;
+ rq_ci <<= sges_n;
}
- if (unlikely(i == 0))
+ if (unlikely((i == 0) && ((rq_ci >> sges_n) == rxq->rq_ci)))
return 0;
- /* Repost WRs. */
-#ifdef DEBUG_RECV
- DEBUG("%p: reposting %u WRs", (void *)rxq, i);
-#endif
- ret = rxq->if_wq->recv_burst(rxq->wq, sges, i);
- if (unlikely(ret)) {
- /* Inability to repost WRs is fatal. */
- DEBUG("%p: recv_burst(): failed (ret=%d)",
- (void *)rxq->priv,
- ret);
- abort();
- }
- rxq->elts_head = elts_head;
+ /* Update the consumer index. */
+ rxq->rq_ci = rq_ci >> sges_n;
+ rte_wmb();
+ *rxq->cq_db = htonl(rxq->cq_ci);
+ rte_wmb();
+ *rxq->rq_db = htonl(rxq->rq_ci);
#ifdef MLX5_PMD_SOFT_COUNTERS
/* Increment packets counter. */
- rxq->stats.ipackets += pkts_ret;
+ rxq->stats.ipackets += i;
#endif
- return pkts_ret;
+ return i;
}
/**