#ifndef RTE_PMD_MLX5_RXTX_VEC_SSE_H_
#define RTE_PMD_MLX5_RXTX_VEC_SSE_H_
-#include <assert.h>
#include <stdint.h>
#include <string.h>
#include <stdlib.h>
#include <rte_mempool.h>
#include <rte_prefetch.h>
+#include <mlx5_prm.h>
+
+#include "mlx5_defs.h"
#include "mlx5.h"
#include "mlx5_utils.h"
#include "mlx5_rxtx.h"
#include "mlx5_rxtx_vec.h"
#include "mlx5_autoconf.h"
-#include "mlx5_defs.h"
-#include "mlx5_prm.h"
#ifndef __INTEL_COMPILER
#pragma GCC diagnostic ignored "-Wcast-qual"
/**
* Store free buffers to RX SW ring.
*
- * @param rxq
- * Pointer to RX queue structure.
+ * @param elts
+ * Pointer to SW ring to be filled.
* @param pkts
* Pointer to array of packets to be stored.
* @param pkts_n
* Number of packets to be stored.
*/
static inline void
-rxq_copy_mbuf_v(struct mlx5_rxq_data *rxq, struct rte_mbuf **pkts, uint16_t n)
+rxq_copy_mbuf_v(struct rte_mbuf **elts, struct rte_mbuf **pkts, uint16_t n)
{
- const uint16_t q_mask = (1 << rxq->elts_n) - 1;
- struct rte_mbuf **elts = &(*rxq->elts)[rxq->rq_pi & q_mask];
unsigned int pos;
uint16_t p = n & -2;
14, 15, 6, 7,
10, 11, 2, 3);
#endif
-
/*
* A. load mCQEs into a 128bit register.
* B. store rearm data to mbuf.
__m128i byte_cnt, invalid_mask;
#endif
- if (!(pos & 0x7) && pos + 8 < mcqe_n)
- rte_prefetch0((void *)(cq + pos + 8));
+ for (i = 0; i < MLX5_VPMD_DESCS_PER_LOOP; ++i)
+ if (likely(pos + i < mcqe_n))
+ rte_prefetch0((void *)(cq + pos + i));
/* A.1 load mCQEs into a 128bit register. */
mcqe1 = _mm_loadu_si128((__m128i *)&mcq[pos % 8]);
mcqe2 = _mm_loadu_si128((__m128i *)&mcq[pos % 8 + 2]);
elts[pos + 2]->hash.fdir.hi = flow_tag;
elts[pos + 3]->hash.fdir.hi = flow_tag;
}
+ if (rxq->dynf_meta) {
+ int32_t offs = rxq->flow_meta_offset;
+ const uint32_t meta =
+ *RTE_MBUF_DYNFIELD(t_pkt, offs, uint32_t *);
+
+ /* Check if title packet has valid metadata. */
+ if (meta) {
+ MLX5_ASSERT(t_pkt->ol_flags &
+ rxq->flow_meta_mask);
+ *RTE_MBUF_DYNFIELD(elts[pos], offs,
+ uint32_t *) = meta;
+ *RTE_MBUF_DYNFIELD(elts[pos + 1], offs,
+ uint32_t *) = meta;
+ *RTE_MBUF_DYNFIELD(elts[pos + 2], offs,
+ uint32_t *) = meta;
+ *RTE_MBUF_DYNFIELD(elts[pos + 3], offs,
+ uint32_t *) = meta;
+ }
+ }
pos += MLX5_VPMD_DESCS_PER_LOOP;
/* Move to next CQE and invalidate consumed CQEs. */
if (!(pos & 0x7) && pos < mcqe_n) {
+ if (pos + 8 < mcqe_n)
+ rte_prefetch0((void *)(cq + pos + 8));
mcq = (void *)(cq + pos);
for (i = 0; i < 8; ++i)
cq[inv++].op_own = MLX5_CQE_INVALIDATE;
rxq->stats.ipackets += mcqe_n;
rxq->stats.ibytes += rcvd_byte;
#endif
- rxq->cq_ci += mcqe_n;
return mcqe_n;
}
__m128i pinfo0, pinfo1;
__m128i pinfo, ptype;
__m128i ol_flags = _mm_set1_epi32(rxq->rss_hash * PKT_RX_RSS_HASH |
- rxq->hw_timestamp * PKT_RX_TIMESTAMP);
+ rxq->hw_timestamp * rxq->timestamp_rx_flag);
__m128i cv_flags;
const __m128i zero = _mm_setzero_si128();
const __m128i ptype_mask =
PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_GOOD |
PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
const __m128i mbuf_init =
- _mm_loadl_epi64((__m128i *)&rxq->mbuf_initializer);
+ _mm_load_si128((__m128i *)&rxq->mbuf_initializer);
__m128i rearm0, rearm1, rearm2, rearm3;
uint8_t pt_idx0, pt_idx1, pt_idx2, pt_idx3;
pinfo1 = _mm_unpackhi_epi32(cqes[2], cqes[3]);
ptype = _mm_unpacklo_epi64(pinfo0, pinfo1);
if (rxq->mark) {
- const __m128i pinfo_ft_mask =
- _mm_set_epi32(0xffffff00, 0xffffff00,
- 0xffffff00, 0xffffff00);
+ const __m128i pinfo_ft_mask = _mm_set1_epi32(0xffffff00);
const __m128i fdir_flags = _mm_set1_epi32(PKT_RX_FDIR);
__m128i fdir_id_flags = _mm_set1_epi32(PKT_RX_FDIR_ID);
__m128i flow_tag, invalid_mask;
}
/**
- * Receive burst of packets. An errored completion also consumes a mbuf, but the
- * packet_type is set to be RTE_PTYPE_ALL_MASK. Marked mbufs should be freed
- * before returning to application.
+ * Process a non-compressed completion and fill in mbufs in RX SW ring
+ * with data extracted from the title completion descriptor.
*
* @param rxq
* Pointer to RX queue structure.
+ * @param cq
+ * Pointer to completion array having a non-compressed completion at first.
+ * @param elts
+ * Pointer to SW ring to be filled. The first mbuf has to be pre-built from
+ * the title completion descriptor to be copied to the rest of mbufs.
* @param[out] pkts
* Array to store received packets.
* @param pkts_n
* @param[out] err
* Pointer to a flag. Set non-zero value if pkts array has at least one error
* packet to handle.
+ * @param[out] comp
+ * Pointer to a index. Set it to the first compressed completion if any.
*
* @return
- * Number of packets received including errors (<= pkts_n).
+ * Number of CQEs successfully processed.
*/
static inline uint16_t
-rxq_burst_v(struct mlx5_rxq_data *rxq, struct rte_mbuf **pkts, uint16_t pkts_n,
- uint64_t *err)
+rxq_cq_process_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,
+ struct rte_mbuf **elts, struct rte_mbuf **pkts,
+ uint16_t pkts_n, uint64_t *err, uint64_t *comp)
{
const uint16_t q_n = 1 << rxq->cqe_n;
const uint16_t q_mask = q_n - 1;
- volatile struct mlx5_cqe *cq;
- struct rte_mbuf **elts;
unsigned int pos;
- uint64_t n;
- uint16_t repl_n;
+ uint64_t n = 0;
uint64_t comp_idx = MLX5_VPMD_DESCS_PER_LOOP;
uint16_t nocmp_n = 0;
- uint16_t rcvd_pkt = 0;
- unsigned int cq_idx = rxq->cq_ci & q_mask;
- unsigned int elts_idx;
unsigned int ownership = !!(rxq->cq_ci & (q_mask + 1));
- const __m128i owner_check =
- _mm_set_epi64x(0x0100000001000000LL, 0x0100000001000000LL);
- const __m128i opcode_check =
- _mm_set_epi64x(0xf0000000f0000000LL, 0xf0000000f0000000LL);
- const __m128i format_check =
- _mm_set_epi64x(0x0c0000000c000000LL, 0x0c0000000c000000LL);
- const __m128i resp_err_check =
- _mm_set_epi64x(0xe0000000e0000000LL, 0xe0000000e0000000LL);
+ const __m128i owner_check = _mm_set1_epi64x(0x0100000001000000LL);
+ const __m128i opcode_check = _mm_set1_epi64x(0xf0000000f0000000LL);
+ const __m128i format_check = _mm_set1_epi64x(0x0c0000000c000000LL);
+ const __m128i resp_err_check = _mm_set1_epi64x(0xe0000000e0000000LL);
#ifdef MLX5_PMD_SOFT_COUNTERS
uint32_t rcvd_byte = 0;
/* Mask to shuffle byte_cnt to add up stats. Do bswap16 for all. */
0,
rxq->crc_present * RTE_ETHER_CRC_LEN);
const __m128i flow_mark_adj = _mm_set_epi32(rxq->mark * (-1), 0, 0, 0);
-
- assert(rxq->sges_n == 0);
- assert(rxq->cqe_n == rxq->elts_n);
- cq = &(*rxq->cqes)[cq_idx];
- rte_prefetch0(cq);
- rte_prefetch0(cq + 1);
- rte_prefetch0(cq + 2);
- rte_prefetch0(cq + 3);
- pkts_n = RTE_MIN(pkts_n, MLX5_VPMD_RX_MAX_BURST);
- repl_n = q_n - (rxq->rq_ci - rxq->rq_pi);
- if (repl_n >= rxq->rq_repl_thresh)
- mlx5_rx_replenish_bulk_mbuf(rxq, repl_n);
- /* See if there're unreturned mbufs from compressed CQE. */
- rcvd_pkt = rxq->decompressed;
- if (rcvd_pkt > 0) {
- rcvd_pkt = RTE_MIN(rcvd_pkt, pkts_n);
- rxq_copy_mbuf_v(rxq, pkts, rcvd_pkt);
- rxq->rq_pi += rcvd_pkt;
- rxq->decompressed -= rcvd_pkt;
- pkts += rcvd_pkt;
- }
- elts_idx = rxq->rq_pi & q_mask;
- elts = &(*rxq->elts)[elts_idx];
- /* Not to overflow pkts array. */
- pkts_n = RTE_ALIGN_FLOOR(pkts_n - rcvd_pkt, MLX5_VPMD_DESCS_PER_LOOP);
- /* Not to cross queue end. */
- pkts_n = RTE_MIN(pkts_n, q_n - elts_idx);
- pkts_n = RTE_MIN(pkts_n, q_n - cq_idx);
- if (!pkts_n)
- return rcvd_pkt;
- /* At this point, there shouldn't be any remained packets. */
- assert(rxq->decompressed == 0);
/*
* A. load first Qword (8bytes) in one loop.
* B. copy 4 mbuf pointers from elts ring to returing pkts.
/* B.2 copy mbuf pointers. */
_mm_storeu_si128((__m128i *)&pkts[pos], mbp1);
_mm_storeu_si128((__m128i *)&pkts[pos + 2], mbp2);
- rte_cio_rmb();
+ rte_io_rmb();
/* C.1 load remained CQE data and extract necessary fields. */
cqe_tmp2 = _mm_load_si128((__m128i *)&cq[pos + p3]);
cqe_tmp1 = _mm_load_si128((__m128i *)&cq[pos + p2]);
cqes[3] = _mm_blendv_epi8(cqes[3], cqe_tmp2, blend_mask);
cqes[2] = _mm_blendv_epi8(cqes[2], cqe_tmp1, blend_mask);
- cqe_tmp2 = _mm_loadu_si128((__m128i *)&cq[pos + p3].rsvd1[3]);
- cqe_tmp1 = _mm_loadu_si128((__m128i *)&cq[pos + p2].rsvd1[3]);
+ cqe_tmp2 = _mm_loadu_si128((__m128i *)&cq[pos + p3].csum);
+ cqe_tmp1 = _mm_loadu_si128((__m128i *)&cq[pos + p2].csum);
cqes[3] = _mm_blend_epi16(cqes[3], cqe_tmp2, 0x30);
cqes[2] = _mm_blend_epi16(cqes[2], cqe_tmp1, 0x30);
- cqe_tmp2 = _mm_loadl_epi64((__m128i *)&cq[pos + p3].rsvd2[10]);
- cqe_tmp1 = _mm_loadl_epi64((__m128i *)&cq[pos + p2].rsvd2[10]);
+ cqe_tmp2 = _mm_loadl_epi64((__m128i *)&cq[pos + p3].rsvd4[2]);
+ cqe_tmp1 = _mm_loadl_epi64((__m128i *)&cq[pos + p2].rsvd4[2]);
cqes[3] = _mm_blend_epi16(cqes[3], cqe_tmp2, 0x04);
cqes[2] = _mm_blend_epi16(cqes[2], cqe_tmp1, 0x04);
/* C.2 generate final structure for mbuf with swapping bytes. */
cqe_tmp1 = _mm_load_si128((__m128i *)&cq[pos]);
cqes[1] = _mm_blendv_epi8(cqes[1], cqe_tmp2, blend_mask);
cqes[0] = _mm_blendv_epi8(cqes[0], cqe_tmp1, blend_mask);
- cqe_tmp2 = _mm_loadu_si128((__m128i *)&cq[pos + p1].rsvd1[3]);
- cqe_tmp1 = _mm_loadu_si128((__m128i *)&cq[pos].rsvd1[3]);
+ cqe_tmp2 = _mm_loadu_si128((__m128i *)&cq[pos + p1].csum);
+ cqe_tmp1 = _mm_loadu_si128((__m128i *)&cq[pos].csum);
cqes[1] = _mm_blend_epi16(cqes[1], cqe_tmp2, 0x30);
cqes[0] = _mm_blend_epi16(cqes[0], cqe_tmp1, 0x30);
- cqe_tmp2 = _mm_loadl_epi64((__m128i *)&cq[pos + p1].rsvd2[10]);
- cqe_tmp1 = _mm_loadl_epi64((__m128i *)&cq[pos].rsvd2[10]);
+ cqe_tmp2 = _mm_loadl_epi64((__m128i *)&cq[pos + p1].rsvd4[2]);
+ cqe_tmp1 = _mm_loadl_epi64((__m128i *)&cq[pos].rsvd4[2]);
cqes[1] = _mm_blend_epi16(cqes[1], cqe_tmp2, 0x04);
cqes[0] = _mm_blend_epi16(cqes[0], cqe_tmp1, 0x04);
/* C.2 generate final structure for mbuf with swapping bytes. */
/* D.5 fill in mbuf - rearm_data and packet_type. */
rxq_cq_to_ptype_oflags_v(rxq, cqes, opcode, &pkts[pos]);
if (rxq->hw_timestamp) {
- pkts[pos]->timestamp =
- rte_be_to_cpu_64(cq[pos].timestamp);
- pkts[pos + 1]->timestamp =
- rte_be_to_cpu_64(cq[pos + p1].timestamp);
- pkts[pos + 2]->timestamp =
- rte_be_to_cpu_64(cq[pos + p2].timestamp);
- pkts[pos + 3]->timestamp =
- rte_be_to_cpu_64(cq[pos + p3].timestamp);
+ int offset = rxq->timestamp_offset;
+ if (rxq->rt_timestamp) {
+ struct mlx5_dev_ctx_shared *sh = rxq->sh;
+ uint64_t ts;
+
+ ts = rte_be_to_cpu_64(cq[pos].timestamp);
+ mlx5_timestamp_set(pkts[pos], offset,
+ mlx5_txpp_convert_rx_ts(sh, ts));
+ ts = rte_be_to_cpu_64(cq[pos + p1].timestamp);
+ mlx5_timestamp_set(pkts[pos + 1], offset,
+ mlx5_txpp_convert_rx_ts(sh, ts));
+ ts = rte_be_to_cpu_64(cq[pos + p2].timestamp);
+ mlx5_timestamp_set(pkts[pos + 2], offset,
+ mlx5_txpp_convert_rx_ts(sh, ts));
+ ts = rte_be_to_cpu_64(cq[pos + p3].timestamp);
+ mlx5_timestamp_set(pkts[pos + 3], offset,
+ mlx5_txpp_convert_rx_ts(sh, ts));
+ } else {
+ mlx5_timestamp_set(pkts[pos], offset,
+ rte_be_to_cpu_64(cq[pos].timestamp));
+ mlx5_timestamp_set(pkts[pos + 1], offset,
+ rte_be_to_cpu_64(cq[pos + p1].timestamp));
+ mlx5_timestamp_set(pkts[pos + 2], offset,
+ rte_be_to_cpu_64(cq[pos + p2].timestamp));
+ mlx5_timestamp_set(pkts[pos + 3], offset,
+ rte_be_to_cpu_64(cq[pos + p3].timestamp));
+ }
+ }
+ if (rxq->dynf_meta) {
+ /* This code is subject for futher optimization. */
+ int32_t offs = rxq->flow_meta_offset;
+
+ *RTE_MBUF_DYNFIELD(pkts[pos], offs, uint32_t *) =
+ cq[pos].flow_table_metadata;
+ *RTE_MBUF_DYNFIELD(pkts[pos + 1], offs, uint32_t *) =
+ cq[pos + p1].flow_table_metadata;
+ *RTE_MBUF_DYNFIELD(pkts[pos + 2], offs, uint32_t *) =
+ cq[pos + p2].flow_table_metadata;
+ *RTE_MBUF_DYNFIELD(pkts[pos + 3], offs, uint32_t *) =
+ cq[pos + p3].flow_table_metadata;
+ if (*RTE_MBUF_DYNFIELD(pkts[pos], offs, uint32_t *))
+ pkts[pos]->ol_flags |= rxq->flow_meta_mask;
+ if (*RTE_MBUF_DYNFIELD(pkts[pos + 1], offs, uint32_t *))
+ pkts[pos + 1]->ol_flags |= rxq->flow_meta_mask;
+ if (*RTE_MBUF_DYNFIELD(pkts[pos + 2], offs, uint32_t *))
+ pkts[pos + 2]->ol_flags |= rxq->flow_meta_mask;
+ if (*RTE_MBUF_DYNFIELD(pkts[pos + 3], offs, uint32_t *))
+ pkts[pos + 3]->ol_flags |= rxq->flow_meta_mask;
}
#ifdef MLX5_PMD_SOFT_COUNTERS
/* Add up received bytes count. */
if (n != MLX5_VPMD_DESCS_PER_LOOP)
break;
}
- /* If no new CQE seen, return without updating cq_db. */
- if (unlikely(!nocmp_n && comp_idx == MLX5_VPMD_DESCS_PER_LOOP))
- return rcvd_pkt;
- /* Update the consumer indexes for non-compressed CQEs. */
- assert(nocmp_n <= pkts_n);
- rxq->cq_ci += nocmp_n;
- rxq->rq_pi += nocmp_n;
- rcvd_pkt += nocmp_n;
#ifdef MLX5_PMD_SOFT_COUNTERS
rxq->stats.ipackets += nocmp_n;
rxq->stats.ibytes += rcvd_byte;
#endif
- /* Decompress the last CQE if compressed. */
- if (comp_idx < MLX5_VPMD_DESCS_PER_LOOP && comp_idx == n) {
- assert(comp_idx == (nocmp_n % MLX5_VPMD_DESCS_PER_LOOP));
- rxq->decompressed = rxq_cq_decompress_v(rxq, &cq[nocmp_n],
- &elts[nocmp_n]);
- /* Return more packets if needed. */
- if (nocmp_n < pkts_n) {
- uint16_t n = rxq->decompressed;
-
- n = RTE_MIN(n, pkts_n - nocmp_n);
- rxq_copy_mbuf_v(rxq, &pkts[nocmp_n], n);
- rxq->rq_pi += n;
- rcvd_pkt += n;
- rxq->decompressed -= n;
- }
- }
- rte_compiler_barrier();
- *rxq->cq_db = rte_cpu_to_be_32(rxq->cq_ci);
- return rcvd_pkt;
+ if (comp_idx == n)
+ *comp = comp_idx;
+ return nocmp_n;
}
#endif /* RTE_PMD_MLX5_RXTX_VEC_SSE_H_ */
git.droids-corp.org / dpdk.git / blobdiff