(void)wr_id_t_check;
}
-/* If raw send operations are available, use them since they are faster. */
-#ifdef SEND_RAW_WR_SUPPORT
-typedef struct ibv_send_wr_raw mlx4_send_wr_t;
-#define mlx4_post_send ibv_post_send_raw
-#else
-typedef struct ibv_send_wr mlx4_send_wr_t;
-#define mlx4_post_send ibv_post_send
-#endif
+/* Transpose flags. Useful to convert IBV to DPDK flags. */
+#define TRANSPOSE(val, from, to) \
+ (((from) >= (to)) ? \
+ (((val) & (from)) / ((from) / (to))) : \
+ (((val) & (from)) * ((to) / (from))))
struct mlx4_rxq_stats {
unsigned int idx; /**< Mapping index. */
struct rxq_elt (*no_sp)[]; /* RX elements. */
} elts;
unsigned int sp:1; /* Use scattered RX elements. */
+ unsigned int csum:1; /* Enable checksum offloading. */
+ unsigned int csum_l2tun:1; /* Same for L2 tunnels. */
uint32_t mb_len; /* Length of a mp-issued mbuf. */
struct mlx4_rxq_stats stats; /* RX queue counters. */
unsigned int socket; /* CPU socket ID for allocations. */
+ struct ibv_exp_res_domain *rd; /* Resource Domain. */
};
/* TX element. */
struct txq_elt {
- mlx4_send_wr_t wr; /* Work Request. */
- struct ibv_sge sges[MLX4_PMD_SGE_WR_N]; /* Scatter/Gather Elements. */
- /* mbuf pointer is derived from WR_ID(wr.wr_id).offset. */
+ struct rte_mbuf *buf;
};
/* Linear buffer type. It is used when transmitting buffers with too many
} mp2mr[MLX4_PMD_TX_MP_CACHE]; /* MP to MR translation table. */
struct ibv_cq *cq; /* Completion Queue. */
struct ibv_qp *qp; /* Queue Pair. */
+ struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
+ struct ibv_exp_cq_family *if_cq; /* CQ interface. */
#if MLX4_PMD_MAX_INLINE > 0
uint32_t max_inline; /* Max inline send size <= MLX4_PMD_MAX_INLINE. */
#endif
linear_t (*elts_linear)[]; /* Linearized buffers. */
struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
unsigned int socket; /* CPU socket ID for allocations. */
+ struct ibv_exp_res_domain *rd; /* Resource Domain. */
};
struct priv {
unsigned int hw_qpg:1; /* QP groups are supported. */
unsigned int hw_tss:1; /* TSS is supported. */
unsigned int hw_rss:1; /* RSS is supported. */
+ unsigned int hw_csum:1; /* Checksum offload is supported. */
+ unsigned int hw_csum_l2tun:1; /* Same for L2 tunnels. */
unsigned int rss:1; /* RSS is enabled. */
unsigned int vf:1; /* This is a VF device. */
#ifdef INLINE_RECV
}
for (i = 0; (i != elts_n); ++i) {
struct txq_elt *elt = &(*elts)[i];
- mlx4_send_wr_t *wr = &elt->wr;
- /* Configure WR. */
- WR_ID(wr->wr_id).id = i;
- WR_ID(wr->wr_id).offset = 0;
- wr->sg_list = &elt->sges[0];
- wr->opcode = IBV_WR_SEND;
- /* Other fields are updated during TX. */
+ elt->buf = NULL;
}
DEBUG("%p: allocated and configured %u WRs", (void *)txq, elts_n);
txq->elts_n = elts_n;
for (i = 0; (i != elemof(*elts)); ++i) {
struct txq_elt *elt = &(*elts)[i];
- if (WR_ID(elt->wr.wr_id).offset == 0)
+ if (elt->buf == NULL)
continue;
- rte_pktmbuf_free((void *)((uintptr_t)elt->sges[0].addr -
- WR_ID(elt->wr.wr_id).offset));
+ rte_pktmbuf_free(elt->buf);
}
rte_free(elts);
}
static void
txq_cleanup(struct txq *txq)
{
+ struct ibv_exp_release_intf_params params;
size_t i;
DEBUG("cleaning up %p", (void *)txq);
txq_free_elts(txq);
+ if (txq->if_qp != NULL) {
+ assert(txq->priv != NULL);
+ assert(txq->priv->ctx != NULL);
+ assert(txq->qp != NULL);
+ params = (struct ibv_exp_release_intf_params){
+ .comp_mask = 0,
+ };
+ claim_zero(ibv_exp_release_intf(txq->priv->ctx,
+ txq->if_qp,
+ ¶ms));
+ }
+ if (txq->if_cq != NULL) {
+ assert(txq->priv != NULL);
+ assert(txq->priv->ctx != NULL);
+ assert(txq->cq != NULL);
+ params = (struct ibv_exp_release_intf_params){
+ .comp_mask = 0,
+ };
+ claim_zero(ibv_exp_release_intf(txq->priv->ctx,
+ txq->if_cq,
+ ¶ms));
+ }
if (txq->qp != NULL)
claim_zero(ibv_destroy_qp(txq->qp));
if (txq->cq != NULL)
claim_zero(ibv_destroy_cq(txq->cq));
+ if (txq->rd != NULL) {
+ struct ibv_exp_destroy_res_domain_attr attr = {
+ .comp_mask = 0,
+ };
+
+ assert(txq->priv != NULL);
+ assert(txq->priv->ctx != NULL);
+ claim_zero(ibv_exp_destroy_res_domain(txq->priv->ctx,
+ txq->rd,
+ &attr));
+ }
for (i = 0; (i != elemof(txq->mp2mr)); ++i) {
if (txq->mp2mr[i].mp == NULL)
break;
unsigned int elts_comp = txq->elts_comp;
unsigned int elts_tail = txq->elts_tail;
const unsigned int elts_n = txq->elts_n;
- struct ibv_wc wcs[elts_comp];
int wcs_n;
if (unlikely(elts_comp == 0))
DEBUG("%p: processing %u work requests completions",
(void *)txq, elts_comp);
#endif
- wcs_n = ibv_poll_cq(txq->cq, elts_comp, wcs);
+ wcs_n = txq->if_cq->poll_cnt(txq->cq, elts_comp);
if (unlikely(wcs_n == 0))
return 0;
if (unlikely(wcs_n < 0)) {
return txq->mp2mr[i].lkey;
}
+#if MLX4_PMD_SGE_WR_N > 1
+
/**
* Copy scattered mbuf contents to a single linear buffer.
*
return size;
}
+/**
+ * Handle scattered buffers for mlx4_tx_burst().
+ *
+ * @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.
+ *
+ * @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.
+ */
+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)[MLX4_PMD_SGE_WR_N])
+{
+ 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 > elemof(*sges))) {
+ segs = (elemof(*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, buf->pool);
+ 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);
+ }
+ /* 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 == (elemof(*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 MLX4_PMD_SGE_WR_N is 1, free mbuf immediately. */
+ if (elemof(*sges) == 1) {
+ do {
+ struct rte_mbuf *next = NEXT(buf);
+
+ 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;
+ }
+ return (struct tx_burst_sg_ret){
+ .length = sent_size,
+ .num = segs,
+ };
+stop:
+ return (struct tx_burst_sg_ret){
+ .length = -1,
+ .num = -1,
+ };
+}
+
+#endif /* MLX4_PMD_SGE_WR_N > 1 */
+
/**
* DPDK callback for TX.
*
mlx4_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
{
struct txq *txq = (struct txq *)dpdk_txq;
- mlx4_send_wr_t head;
- mlx4_send_wr_t **wr_next = &head.next;
- mlx4_send_wr_t *bad_wr;
unsigned int elts_head = txq->elts_head;
const unsigned int elts_tail = txq->elts_tail;
const unsigned int elts_n = txq->elts_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];
- mlx4_send_wr_t *wr = &elt->wr;
unsigned int segs = NB_SEGS(buf);
-#if (MLX4_PMD_MAX_INLINE > 0) || defined(MLX4_PMD_SOFT_COUNTERS)
+#ifdef MLX4_PMD_SOFT_COUNTERS
unsigned int sent_size = 0;
#endif
- unsigned int j;
- int linearize = 0;
+ uint32_t send_flags = 0;
/* Clean up old buffer. */
- if (likely(WR_ID(wr->wr_id).offset != 0)) {
- struct rte_mbuf *tmp = (void *)
- ((uintptr_t)elt->sges[0].addr -
- WR_ID(wr->wr_id).offset);
+ if (likely(elt->buf != NULL)) {
+ struct rte_mbuf *tmp = elt->buf;
/* Faster than rte_pktmbuf_free(). */
do {
tmp = next;
} while (tmp != NULL);
}
-#ifndef NDEBUG
- /* For assert(). */
- WR_ID(wr->wr_id).offset = 0;
- for (j = 0; ((int)j < wr->num_sge); ++j) {
- elt->sges[j].addr = 0;
- elt->sges[j].length = 0;
- elt->sges[j].lkey = 0;
+ /* 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;
}
- wr->next = NULL;
- wr->num_sge = 0;
-#endif
- /* Sanity checks, most of which are only relevant with
- * debugging enabled. */
- assert(WR_ID(wr->wr_id).id == elts_head);
- assert(WR_ID(wr->wr_id).offset == 0);
- assert(wr->next == NULL);
- assert(wr->sg_list == &elt->sges[0]);
- assert(wr->num_sge == 0);
- assert(wr->opcode == IBV_WR_SEND);
- /* When there are too many segments, extra segments are
- * linearized in the last SGE. */
- if (unlikely(segs > elemof(elt->sges))) {
- segs = (elemof(elt->sges) - 1);
- linearize = 1;
+ /* 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.
+ *
+ * FIXME: since PKT_TX_UDP_TUNNEL_PKT has been removed,
+ * the outer packet type is unknown. All we know is
+ * that the L2 header is of unusual length (not
+ * ETHER_HDR_LEN with or without 802.1Q header). */
+ if ((buf->l2_len != ETHER_HDR_LEN) &&
+ (buf->l2_len != (ETHER_HDR_LEN + 4)))
+ send_flags |= IBV_EXP_QP_BURST_TUNNEL;
}
- /* Set WR fields. */
- assert((rte_pktmbuf_mtod(buf, uintptr_t) -
- (uintptr_t)buf) <= 0xffff);
- WR_ID(wr->wr_id).offset =
- (rte_pktmbuf_mtod(buf, uintptr_t) -
- (uintptr_t)buf);
- wr->num_sge = segs;
- /* Register segments as SGEs. */
- for (j = 0; (j != segs); ++j) {
- struct ibv_sge *sge = &elt->sges[j];
+ if (likely(segs == 1)) {
+ uintptr_t addr;
+ uint32_t length;
uint32_t lkey;
+ /* 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, buf->pool);
if (unlikely(lkey == (uint32_t)-1)) {
DEBUG("%p: unable to get MP <-> MR"
" association", (void *)txq);
/* Clean up TX element. */
- WR_ID(elt->wr.wr_id).offset = 0;
-#ifndef NDEBUG
- /* For assert(). */
- while (j) {
- --j;
- --sge;
- sge->addr = 0;
- sge->length = 0;
- sge->lkey = 0;
- }
- wr->num_sge = 0;
-#endif
+ elt->buf = NULL;
goto stop;
}
- /* Sanity checks, only relevant with debugging
- * enabled. */
- assert(sge->addr == 0);
- assert(sge->length == 0);
- assert(sge->lkey == 0);
- /* Update SGE. */
- sge->addr = rte_pktmbuf_mtod(buf, uintptr_t);
+ /* Update element. */
+ elt->buf = buf;
if (txq->priv->vf)
rte_prefetch0((volatile void *)
- (uintptr_t)sge->addr);
- sge->length = DATA_LEN(buf);
- sge->lkey = lkey;
-#if (MLX4_PMD_MAX_INLINE > 0) || defined(MLX4_PMD_SOFT_COUNTERS)
- sent_size += sge->length;
-#endif
- buf = NEXT(buf);
- }
- /* 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 = &elt->sges[segs];
- linear_t *linear = &(*txq->elts_linear)[elts_head];
- unsigned int size = linearize_mbuf(linear, buf);
-
- assert(segs == (elemof(elt->sges) - 1));
- if (size == 0) {
- /* Invalid packet. */
- DEBUG("%p: packet too large to be linearized.",
- (void *)txq);
- /* Clean up TX element. */
- WR_ID(elt->wr.wr_id).offset = 0;
-#ifndef NDEBUG
- /* For assert(). */
- while (j) {
- --j;
- --sge;
- sge->addr = 0;
- sge->length = 0;
- sge->lkey = 0;
- }
- wr->num_sge = 0;
+ (uintptr_t)addr);
+ RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
+ /* Put packet into send queue. */
+#if MLX4_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;
- }
- /* If MLX4_PMD_SGE_WR_N is 1, free mbuf immediately
- * and clear offset from WR ID. */
- if (elemof(elt->sges) == 1) {
- do {
- struct rte_mbuf *next = NEXT(buf);
-
- rte_pktmbuf_free_seg(buf);
- buf = next;
- } while (buf != NULL);
- WR_ID(wr->wr_id).offset = 0;
- }
- /* Set WR fields and fill SGE with linear buffer. */
- ++wr->num_sge;
- /* Sanity checks, only relevant with debugging
- * enabled. */
- assert(sge->addr == 0);
- assert(sge->length == 0);
- assert(sge->lkey == 0);
- /* Update SGE. */
- sge->addr = (uintptr_t)&(*linear)[0];
- sge->length = size;
- sge->lkey = txq->mr_linear->lkey;
-#if (MLX4_PMD_MAX_INLINE > 0) || defined(MLX4_PMD_SOFT_COUNTERS)
- sent_size += size;
+#ifdef MLX4_PMD_SOFT_COUNTERS
+ sent_size += length;
#endif
- }
- /* Link WRs together for ibv_post_send(). */
- *wr_next = wr;
- wr_next = &wr->next;
-#if MLX4_PMD_MAX_INLINE > 0
- if (sent_size <= txq->max_inline)
- wr->send_flags = IBV_SEND_INLINE;
- else
+ } else {
+#if MLX4_PMD_SGE_WR_N > 1
+ struct ibv_sge sges[MLX4_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;
+#ifdef MLX4_PMD_SOFT_COUNTERS
+ sent_size += ret.length;
#endif
- wr->send_flags = 0;
- /* Request TX completion. */
- if (unlikely(--elts_comp_cd == 0)) {
- elts_comp_cd = txq->elts_comp_cd_init;
- ++elts_comp;
- wr->send_flags |= IBV_SEND_SIGNALED;
+#else /* MLX4_PMD_SGE_WR_N > 1 */
+ DEBUG("%p: TX scattered buffers support not"
+ " compiled in", (void *)txq);
+ goto stop;
+#endif /* MLX4_PMD_SGE_WR_N > 1 */
}
- if (++elts_head >= elts_n)
- elts_head = 0;
+ elts_head = elts_head_next;
#ifdef MLX4_PMD_SOFT_COUNTERS
/* Increment sent bytes counter. */
txq->stats.obytes += sent_size;
/* Increment sent packets counter. */
txq->stats.opackets += i;
#endif
- *wr_next = NULL;
- err = mlx4_post_send(txq->qp, head.next, &bad_wr);
+ /* Ring QP doorbell. */
+ err = txq->if_qp->send_flush(txq->qp);
if (unlikely(err)) {
- unsigned int unsent = 0;
-
- /* An error occurred, fix counters. */
- while (bad_wr != NULL) {
- struct txq_elt *elt =
- containerof(bad_wr, struct txq_elt, wr);
- mlx4_send_wr_t *wr = &elt->wr;
- mlx4_send_wr_t *next = wr->next;
-#if defined(MLX4_PMD_SOFT_COUNTERS) || !defined(NDEBUG)
- unsigned int j;
-#endif
-
- assert(wr == bad_wr);
- /* Clean up TX element without freeing it, caller
- * should take care of this. */
- WR_ID(elt->wr.wr_id).offset = 0;
-#ifdef MLX4_PMD_SOFT_COUNTERS
- for (j = 0; ((int)j < wr->num_sge); ++j)
- txq->stats.obytes -= wr->sg_list[j].length;
-#endif
- ++unsent;
- if (wr->send_flags & IBV_SEND_SIGNALED) {
- assert(elts_comp != 0);
- --elts_comp;
- }
- if (elts_comp_cd == txq->elts_comp_cd_init)
- elts_comp_cd = 1;
- else
- ++elts_comp_cd;
-#ifndef NDEBUG
- /* For assert(). */
- for (j = 0; ((int)j < wr->num_sge); ++j) {
- elt->sges[j].addr = 0;
- elt->sges[j].length = 0;
- elt->sges[j].lkey = 0;
- }
- wr->next = NULL;
- wr->num_sge = 0;
-#endif
- bad_wr = next;
- }
-#ifdef MLX4_PMD_SOFT_COUNTERS
- txq->stats.opackets -= unsent;
-#endif
- assert(i >= unsent);
- i -= unsent;
- /* "Unsend" remaining packets. */
- elts_head -= unsent;
- if (elts_head >= elts_n)
- elts_head += elts_n;
- assert(elts_head < elts_n);
- DEBUG("%p: mlx4_post_send() failed, %u unprocessed WRs: %s",
- (void *)txq, unsent,
- ((err <= -1) ? "Internal error" : strerror(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);
}
txq->elts_head = elts_head;
txq->elts_comp += elts_comp;
.socket = socket
};
union {
- struct ibv_qp_init_attr init;
+ struct ibv_exp_query_intf_params params;
+ struct ibv_exp_qp_init_attr init;
+ struct ibv_exp_res_domain_init_attr rd;
+ struct ibv_exp_cq_init_attr cq;
struct ibv_exp_qp_attr mod;
} attr;
+ enum ibv_exp_query_intf_status status;
int ret = 0;
(void)conf; /* Thresholds configuration (ignored). */
}
desc /= MLX4_PMD_SGE_WR_N;
/* MRs will be registered in mp2mr[] later. */
- tmpl.cq = ibv_create_cq(priv->ctx, desc, NULL, NULL, 0);
+ attr.rd = (struct ibv_exp_res_domain_init_attr){
+ .comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
+ IBV_EXP_RES_DOMAIN_MSG_MODEL),
+ .thread_model = IBV_EXP_THREAD_SINGLE,
+ .msg_model = IBV_EXP_MSG_HIGH_BW,
+ };
+ tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd);
+ if (tmpl.rd == NULL) {
+ ret = ENOMEM;
+ ERROR("%p: RD creation failure: %s",
+ (void *)dev, strerror(ret));
+ goto error;
+ }
+ attr.cq = (struct ibv_exp_cq_init_attr){
+ .comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN,
+ .res_domain = tmpl.rd,
+ };
+ tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, NULL, 0, &attr.cq);
if (tmpl.cq == NULL) {
ret = ENOMEM;
ERROR("%p: CQ creation failure: %s",
priv->device_attr.max_qp_wr);
DEBUG("priv->device_attr.max_sge is %d",
priv->device_attr.max_sge);
- attr.init = (struct ibv_qp_init_attr){
+ attr.init = (struct ibv_exp_qp_init_attr){
/* CQ to be associated with the send queue. */
.send_cq = tmpl.cq,
/* CQ to be associated with the receive queue. */
.qp_type = IBV_QPT_RAW_PACKET,
/* Do *NOT* enable this, completions events are managed per
* TX burst. */
- .sq_sig_all = 0
+ .sq_sig_all = 0,
+ .pd = priv->pd,
+ .res_domain = tmpl.rd,
+ .comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+ IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
};
- tmpl.qp = ibv_create_qp(priv->pd, &attr.init);
+ tmpl.qp = ibv_exp_create_qp(priv->ctx, &attr.init);
if (tmpl.qp == NULL) {
ret = (errno ? errno : EINVAL);
ERROR("%p: QP creation failure: %s",
(void *)dev, strerror(ret));
goto error;
}
+ attr.params = (struct ibv_exp_query_intf_params){
+ .intf_scope = IBV_EXP_INTF_GLOBAL,
+ .intf = IBV_EXP_INTF_CQ,
+ .obj = tmpl.cq,
+ };
+ tmpl.if_cq = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+ if (tmpl.if_cq == NULL) {
+ ERROR("%p: CQ interface family query failed with status %d",
+ (void *)dev, status);
+ goto error;
+ }
+ attr.params = (struct ibv_exp_query_intf_params){
+ .intf_scope = IBV_EXP_INTF_GLOBAL,
+ .intf = IBV_EXP_INTF_QP_BURST,
+ .obj = tmpl.qp,
+#ifdef HAVE_EXP_QP_BURST_CREATE_DISABLE_ETH_LOOPBACK
+ /* MC loopback must be disabled when not using a VF. */
+ .family_flags =
+ (!priv->vf ?
+ IBV_EXP_QP_BURST_CREATE_DISABLE_ETH_LOOPBACK :
+ 0),
+#endif
+ };
+ tmpl.if_qp = ibv_exp_query_intf(priv->ctx, &attr.params, &status);
+ if (tmpl.if_qp == NULL) {
+ ERROR("%p: QP interface family query failed with status %d",
+ (void *)dev, status);
+ goto error;
+ }
/* Clean up txq in case we're reinitializing it. */
DEBUG("%p: cleaning-up old txq just in case", (void *)txq);
txq_cleanup(txq);
}
if (rxq->cq != NULL)
claim_zero(ibv_destroy_cq(rxq->cq));
+ if (rxq->rd != NULL) {
+ struct ibv_exp_destroy_res_domain_attr attr = {
+ .comp_mask = 0,
+ };
+
+ assert(rxq->priv != NULL);
+ assert(rxq->priv->ctx != NULL);
+ claim_zero(ibv_exp_destroy_res_domain(rxq->priv->ctx,
+ rxq->rd,
+ &attr));
+ }
if (rxq->mr != NULL)
claim_zero(ibv_dereg_mr(rxq->mr));
memset(rxq, 0, sizeof(*rxq));
}
+/**
+ * Translate RX completion flags to offload flags.
+ *
+ * @param[in] rxq
+ * Pointer to RX queue structure.
+ * @param flags
+ * RX completion flags returned by poll_length_flags().
+ *
+ * @return
+ * Offload flags (ol_flags) for struct rte_mbuf.
+ */
+static inline uint32_t
+rxq_cq_to_ol_flags(const struct rxq *rxq, uint32_t flags)
+{
+ uint32_t ol_flags;
+
+ ol_flags =
+ TRANSPOSE(flags, IBV_EXP_CQ_RX_IPV4_PACKET, PKT_RX_IPV4_HDR) |
+ TRANSPOSE(flags, IBV_EXP_CQ_RX_IPV6_PACKET, PKT_RX_IPV6_HDR);
+ 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);
+ /*
+ * 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).
+ */
+ if ((flags & IBV_EXP_CQ_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
+ ol_flags |=
+ TRANSPOSE(flags,
+ IBV_EXP_CQ_RX_OUTER_IPV4_PACKET,
+ PKT_RX_TUNNEL_IPV4_HDR) |
+ TRANSPOSE(flags,
+ IBV_EXP_CQ_RX_OUTER_IPV6_PACKET,
+ PKT_RX_TUNNEL_IPV6_HDR) |
+ 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;
+}
+
static uint16_t
mlx4_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n);
struct rte_mbuf **pkt_buf_next = &pkt_buf;
unsigned int seg_headroom = RTE_PKTMBUF_HEADROOM;
unsigned int j = 0;
+ uint32_t flags;
/* Sanity checks. */
#ifdef NDEBUG
assert(wr->num_sge == elemof(elt->sges));
assert(elts_head < rxq->elts_n);
assert(rxq->elts_head < rxq->elts_n);
- ret = rxq->if_cq->poll_length(rxq->cq, NULL, NULL);
+ ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL,
+ &flags);
if (unlikely(ret < 0)) {
struct ibv_wc wc;
int wcs_n;
NB_SEGS(pkt_buf) = j;
PORT(pkt_buf) = rxq->port_id;
PKT_LEN(pkt_buf) = pkt_buf_len;
- pkt_buf->ol_flags = 0;
+ pkt_buf->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
/* Return packet. */
*(pkts++) = pkt_buf;
struct rte_mbuf *seg = (void *)((uintptr_t)elt->sge.addr -
WR_ID(wr_id).offset);
struct rte_mbuf *rep;
+ uint32_t flags;
/* Sanity checks. */
assert(WR_ID(wr_id).id < rxq->elts_n);
assert(wr->num_sge == 1);
assert(elts_head < rxq->elts_n);
assert(rxq->elts_head < rxq->elts_n);
- ret = rxq->if_cq->poll_length(rxq->cq, NULL, NULL);
+ ret = rxq->if_cq->poll_length_flags(rxq->cq, NULL, NULL,
+ &flags);
if (unlikely(ret < 0)) {
struct ibv_wc wc;
int wcs_n;
NEXT(seg) = NULL;
PKT_LEN(seg) = len;
DATA_LEN(seg) = len;
- seg->ol_flags = 0;
+ seg->ol_flags = rxq_cq_to_ol_flags(rxq, flags);
/* Return packet. */
*(pkts++) = seg;
* QP pointer or NULL in case of error.
*/
static struct ibv_qp *
-rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc)
+rxq_setup_qp(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
+ struct ibv_exp_res_domain *rd)
{
struct ibv_exp_qp_init_attr attr = {
/* CQ to be associated with the send queue. */
MLX4_PMD_SGE_WR_N),
},
.qp_type = IBV_QPT_RAW_PACKET,
- .comp_mask = IBV_EXP_QP_INIT_ATTR_PD,
+ .comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+ IBV_EXP_QP_INIT_ATTR_RES_DOMAIN),
.pd = priv->pd,
+ .res_domain = rd,
};
#ifdef INLINE_RECV
*/
static struct ibv_qp *
rxq_setup_qp_rss(struct priv *priv, struct ibv_cq *cq, uint16_t desc,
- int parent)
+ int parent, struct ibv_exp_res_domain *rd)
{
struct ibv_exp_qp_init_attr attr = {
/* CQ to be associated with the send queue. */
},
.qp_type = IBV_QPT_RAW_PACKET,
.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
+ IBV_EXP_QP_INIT_ATTR_RES_DOMAIN |
IBV_EXP_QP_INIT_ATTR_QPG),
- .pd = priv->pd
+ .pd = priv->pd,
+ .res_domain = rd,
};
#ifdef INLINE_RECV
/* Number of descriptors and mbufs currently allocated. */
desc_n = (tmpl.elts_n * (tmpl.sp ? MLX4_PMD_SGE_WR_N : 1));
mbuf_n = desc_n;
+ /* Toggle RX checksum offload if hardware supports it. */
+ if (priv->hw_csum) {
+ tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
+ rxq->csum = tmpl.csum;
+ }
+ if (priv->hw_csum_l2tun) {
+ tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
+ rxq->csum_l2tun = tmpl.csum_l2tun;
+ }
/* Enable scattered packets support for this queue if necessary. */
if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
(dev->data->dev_conf.rxmode.max_rx_pkt_len >
struct ibv_exp_qp_attr mod;
union {
struct ibv_exp_query_intf_params params;
+ struct ibv_exp_cq_init_attr cq;
+ struct ibv_exp_res_domain_init_attr rd;
} attr;
enum ibv_exp_query_intf_status status;
struct ibv_recv_wr *bad_wr;
rte_pktmbuf_tailroom(buf)) == tmpl.mb_len);
assert(rte_pktmbuf_headroom(buf) == RTE_PKTMBUF_HEADROOM);
rte_pktmbuf_free(buf);
+ /* Toggle RX checksum offload if hardware supports it. */
+ if (priv->hw_csum)
+ tmpl.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
+ if (priv->hw_csum_l2tun)
+ tmpl.csum_l2tun = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
/* Enable scattered packets support for this queue if necessary. */
if ((dev->data->dev_conf.rxmode.jumbo_frame) &&
(dev->data->dev_conf.rxmode.max_rx_pkt_len >
goto error;
}
skip_mr:
- tmpl.cq = ibv_create_cq(priv->ctx, desc, NULL, NULL, 0);
+ attr.rd = (struct ibv_exp_res_domain_init_attr){
+ .comp_mask = (IBV_EXP_RES_DOMAIN_THREAD_MODEL |
+ IBV_EXP_RES_DOMAIN_MSG_MODEL),
+ .thread_model = IBV_EXP_THREAD_SINGLE,
+ .msg_model = IBV_EXP_MSG_HIGH_BW,
+ };
+ tmpl.rd = ibv_exp_create_res_domain(priv->ctx, &attr.rd);
+ if (tmpl.rd == NULL) {
+ ret = ENOMEM;
+ ERROR("%p: RD creation failure: %s",
+ (void *)dev, strerror(ret));
+ goto error;
+ }
+ attr.cq = (struct ibv_exp_cq_init_attr){
+ .comp_mask = IBV_EXP_CQ_INIT_ATTR_RES_DOMAIN,
+ .res_domain = tmpl.rd,
+ };
+ tmpl.cq = ibv_exp_create_cq(priv->ctx, desc, NULL, NULL, 0, &attr.cq);
if (tmpl.cq == NULL) {
ret = ENOMEM;
ERROR("%p: CQ creation failure: %s",
priv->device_attr.max_sge);
#ifdef RSS_SUPPORT
if (priv->rss)
- tmpl.qp = rxq_setup_qp_rss(priv, tmpl.cq, desc, parent);
+ tmpl.qp = rxq_setup_qp_rss(priv, tmpl.cq, desc, parent,
+ tmpl.rd);
else
#endif /* RSS_SUPPORT */
- tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc);
+ tmpl.qp = rxq_setup_qp(priv, tmpl.cq, desc, tmpl.rd);
if (tmpl.qp == NULL) {
ret = (errno ? errno : EINVAL);
ERROR("%p: QP creation failure: %s",
info->max_rx_queues = max;
info->max_tx_queues = max;
info->max_mac_addrs = elemof(priv->mac);
+ info->rx_offload_capa =
+ (priv->hw_csum ?
+ (DEV_RX_OFFLOAD_IPV4_CKSUM |
+ DEV_RX_OFFLOAD_UDP_CKSUM |
+ DEV_RX_OFFLOAD_TCP_CKSUM) :
+ 0);
+ info->tx_offload_capa =
+ (priv->hw_csum ?
+ (DEV_TX_OFFLOAD_IPV4_CKSUM |
+ DEV_TX_OFFLOAD_UDP_CKSUM |
+ DEV_TX_OFFLOAD_TCP_CKSUM) :
+ 0);
priv_unlock(priv);
}
.mac_addr_remove = mlx4_mac_addr_remove,
.mac_addr_add = mlx4_mac_addr_add,
.mtu_set = mlx4_dev_set_mtu,
+ .udp_tunnel_add = NULL,
+ .udp_tunnel_del = NULL,
.fdir_add_signature_filter = NULL,
.fdir_update_signature_filter = NULL,
.fdir_remove_signature_filter = NULL,
exp_device_attr.max_rss_tbl_sz);
#endif /* RSS_SUPPORT */
+ priv->hw_csum =
+ ((exp_device_attr.exp_device_cap_flags &
+ IBV_EXP_DEVICE_RX_CSUM_TCP_UDP_PKT) &&
+ (exp_device_attr.exp_device_cap_flags &
+ IBV_EXP_DEVICE_RX_CSUM_IP_PKT));
+ DEBUG("checksum offloading is %ssupported",
+ (priv->hw_csum ? "" : "not "));
+
+ priv->hw_csum_l2tun = !!(exp_device_attr.exp_device_cap_flags &
+ IBV_EXP_DEVICE_VXLAN_SUPPORT);
+ DEBUG("L2 tunnel checksum offloads are %ssupported",
+ (priv->hw_csum_l2tun ? "" : "not "));
+
#ifdef INLINE_RECV
priv->inl_recv_size = mlx4_getenv_int("MLX4_INLINE_RECV_SIZE");
git.droids-corp.org / dpdk.git / blobdiff