4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
42 #include <sys/queue.h>
44 #include <rte_string_fns.h>
45 #include <rte_memzone.h>
47 #include <rte_malloc.h>
48 #include <rte_ether.h>
49 #include <rte_ethdev.h>
54 #include "i40e_logs.h"
55 #include "i40e/i40e_prototype.h"
56 #include "i40e/i40e_type.h"
57 #include "i40e_ethdev.h"
58 #include "i40e_rxtx.h"
60 #define I40E_MIN_RING_DESC 64
61 #define I40E_MAX_RING_DESC 4096
62 #define I40E_ALIGN 128
63 #define DEFAULT_TX_RS_THRESH 32
64 #define DEFAULT_TX_FREE_THRESH 32
65 #define I40E_MAX_PKT_TYPE 256
67 #define I40E_VLAN_TAG_SIZE 4
68 #define I40E_TX_MAX_BURST 32
70 #define I40E_DMA_MEM_ALIGN 4096
72 #define I40E_SIMPLE_FLAGS ((uint32_t)ETH_TXQ_FLAGS_NOMULTSEGS | \
73 ETH_TXQ_FLAGS_NOOFFLOADS)
75 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
77 #define RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb) \
78 (uint64_t) ((mb)->buf_physaddr + RTE_PKTMBUF_HEADROOM)
80 #define RTE_MBUF_DATA_DMA_ADDR(mb) \
81 ((uint64_t)((mb)->buf_physaddr + \
82 (uint64_t)((char *)((mb)->pkt.data) - \
83 (char *)(mb)->buf_addr)))
85 static const struct rte_memzone *
86 i40e_ring_dma_zone_reserve(struct rte_eth_dev *dev,
87 const char *ring_name,
91 static uint16_t i40e_xmit_pkts_simple(void *tx_queue,
92 struct rte_mbuf **tx_pkts,
95 /* Translate the rx descriptor status to pkt flags */
96 static inline uint16_t
97 i40e_rxd_status_to_pkt_flags(uint64_t qword)
101 /* Check if VLAN packet */
102 flags = (uint16_t)(qword & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) ?
103 PKT_RX_VLAN_PKT : 0);
105 /* Check if RSS_HASH */
106 flags |= (uint16_t)((((qword >> I40E_RX_DESC_STATUS_FLTSTAT_SHIFT) &
107 I40E_RX_DESC_FLTSTAT_RSS_HASH) ==
108 I40E_RX_DESC_FLTSTAT_RSS_HASH) ? PKT_RX_RSS_HASH : 0);
113 static inline uint16_t
114 i40e_rxd_error_to_pkt_flags(uint64_t qword)
117 uint64_t error_bits = (qword >> I40E_RXD_QW1_ERROR_SHIFT);
119 #define I40E_RX_ERR_BITS 0x3f
120 if (likely((error_bits & I40E_RX_ERR_BITS) == 0))
122 /* If RXE bit set, all other status bits are meaningless */
123 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
124 flags |= PKT_RX_MAC_ERR;
128 /* If RECIPE bit set, all other status indications should be ignored */
129 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_RECIPE_SHIFT))) {
130 flags |= PKT_RX_RECIP_ERR;
133 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT)))
134 flags |= PKT_RX_HBUF_OVERFLOW;
135 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_IPE_SHIFT)))
136 flags |= PKT_RX_IP_CKSUM_BAD;
137 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_L4E_SHIFT)))
138 flags |= PKT_RX_L4_CKSUM_BAD;
139 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT)))
140 flags |= PKT_RX_EIP_CKSUM_BAD;
141 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_OVERSIZE_SHIFT)))
142 flags |= PKT_RX_OVERSIZE;
147 /* Translate pkt types to pkt flags */
148 static inline uint16_t
149 i40e_rxd_ptype_to_pkt_flags(uint64_t qword)
151 uint8_t ptype = (uint8_t)((qword & I40E_RXD_QW1_PTYPE_MASK) >>
152 I40E_RXD_QW1_PTYPE_SHIFT);
153 static const uint16_t ip_ptype_map[I40E_MAX_PKT_TYPE] = {
176 PKT_RX_IPV4_HDR, /* PTYPE 22 */
177 PKT_RX_IPV4_HDR, /* PTYPE 23 */
178 PKT_RX_IPV4_HDR, /* PTYPE 24 */
180 PKT_RX_IPV4_HDR, /* PTYPE 26 */
181 PKT_RX_IPV4_HDR, /* PTYPE 27 */
182 PKT_RX_IPV4_HDR, /* PTYPE 28 */
183 PKT_RX_IPV4_HDR_EXT, /* PTYPE 29 */
184 PKT_RX_IPV4_HDR_EXT, /* PTYPE 30 */
185 PKT_RX_IPV4_HDR_EXT, /* PTYPE 31 */
187 PKT_RX_IPV4_HDR_EXT, /* PTYPE 33 */
188 PKT_RX_IPV4_HDR_EXT, /* PTYPE 34 */
189 PKT_RX_IPV4_HDR_EXT, /* PTYPE 35 */
190 PKT_RX_IPV4_HDR_EXT, /* PTYPE 36 */
191 PKT_RX_IPV4_HDR_EXT, /* PTYPE 37 */
192 PKT_RX_IPV4_HDR_EXT, /* PTYPE 38 */
194 PKT_RX_IPV4_HDR_EXT, /* PTYPE 40 */
195 PKT_RX_IPV4_HDR_EXT, /* PTYPE 41 */
196 PKT_RX_IPV4_HDR_EXT, /* PTYPE 42 */
197 PKT_RX_IPV4_HDR_EXT, /* PTYPE 43 */
198 PKT_RX_IPV4_HDR_EXT, /* PTYPE 44 */
199 PKT_RX_IPV4_HDR_EXT, /* PTYPE 45 */
200 PKT_RX_IPV4_HDR_EXT, /* PTYPE 46 */
202 PKT_RX_IPV4_HDR_EXT, /* PTYPE 48 */
203 PKT_RX_IPV4_HDR_EXT, /* PTYPE 49 */
204 PKT_RX_IPV4_HDR_EXT, /* PTYPE 50 */
205 PKT_RX_IPV4_HDR_EXT, /* PTYPE 51 */
206 PKT_RX_IPV4_HDR_EXT, /* PTYPE 52 */
207 PKT_RX_IPV4_HDR_EXT, /* PTYPE 53 */
209 PKT_RX_IPV4_HDR_EXT, /* PTYPE 55 */
210 PKT_RX_IPV4_HDR_EXT, /* PTYPE 56 */
211 PKT_RX_IPV4_HDR_EXT, /* PTYPE 57 */
212 PKT_RX_IPV4_HDR_EXT, /* PTYPE 58 */
213 PKT_RX_IPV4_HDR_EXT, /* PTYPE 59 */
214 PKT_RX_IPV4_HDR_EXT, /* PTYPE 60 */
215 PKT_RX_IPV4_HDR_EXT, /* PTYPE 61 */
217 PKT_RX_IPV4_HDR_EXT, /* PTYPE 63 */
218 PKT_RX_IPV4_HDR_EXT, /* PTYPE 64 */
219 PKT_RX_IPV4_HDR_EXT, /* PTYPE 65 */
220 PKT_RX_IPV4_HDR_EXT, /* PTYPE 66 */
221 PKT_RX_IPV4_HDR_EXT, /* PTYPE 67 */
222 PKT_RX_IPV4_HDR_EXT, /* PTYPE 68 */
224 PKT_RX_IPV4_HDR_EXT, /* PTYPE 70 */
225 PKT_RX_IPV4_HDR_EXT, /* PTYPE 71 */
226 PKT_RX_IPV4_HDR_EXT, /* PTYPE 72 */
227 PKT_RX_IPV4_HDR_EXT, /* PTYPE 73 */
228 PKT_RX_IPV4_HDR_EXT, /* PTYPE 74 */
229 PKT_RX_IPV4_HDR_EXT, /* PTYPE 75 */
230 PKT_RX_IPV4_HDR_EXT, /* PTYPE 76 */
232 PKT_RX_IPV4_HDR_EXT, /* PTYPE 78 */
233 PKT_RX_IPV4_HDR_EXT, /* PTYPE 79 */
234 PKT_RX_IPV4_HDR_EXT, /* PTYPE 80 */
235 PKT_RX_IPV4_HDR_EXT, /* PTYPE 81 */
236 PKT_RX_IPV4_HDR_EXT, /* PTYPE 82 */
237 PKT_RX_IPV4_HDR_EXT, /* PTYPE 83 */
239 PKT_RX_IPV4_HDR_EXT, /* PTYPE 85 */
240 PKT_RX_IPV4_HDR_EXT, /* PTYPE 86 */
241 PKT_RX_IPV4_HDR_EXT, /* PTYPE 87 */
242 PKT_RX_IPV6_HDR, /* PTYPE 88 */
243 PKT_RX_IPV6_HDR, /* PTYPE 89 */
244 PKT_RX_IPV6_HDR, /* PTYPE 90 */
246 PKT_RX_IPV6_HDR, /* PTYPE 92 */
247 PKT_RX_IPV6_HDR, /* PTYPE 93 */
248 PKT_RX_IPV6_HDR, /* PTYPE 94 */
249 PKT_RX_IPV6_HDR_EXT, /* PTYPE 95 */
250 PKT_RX_IPV6_HDR_EXT, /* PTYPE 96 */
251 PKT_RX_IPV6_HDR_EXT, /* PTYPE 97 */
253 PKT_RX_IPV6_HDR_EXT, /* PTYPE 99 */
254 PKT_RX_IPV6_HDR_EXT, /* PTYPE 100 */
255 PKT_RX_IPV6_HDR_EXT, /* PTYPE 101 */
256 PKT_RX_IPV6_HDR_EXT, /* PTYPE 102 */
257 PKT_RX_IPV6_HDR_EXT, /* PTYPE 103 */
258 PKT_RX_IPV6_HDR_EXT, /* PTYPE 104 */
260 PKT_RX_IPV6_HDR_EXT, /* PTYPE 106 */
261 PKT_RX_IPV6_HDR_EXT, /* PTYPE 107 */
262 PKT_RX_IPV6_HDR_EXT, /* PTYPE 108 */
263 PKT_RX_IPV6_HDR_EXT, /* PTYPE 109 */
264 PKT_RX_IPV6_HDR_EXT, /* PTYPE 110 */
265 PKT_RX_IPV6_HDR_EXT, /* PTYPE 111 */
266 PKT_RX_IPV6_HDR_EXT, /* PTYPE 112 */
268 PKT_RX_IPV6_HDR_EXT, /* PTYPE 114 */
269 PKT_RX_IPV6_HDR_EXT, /* PTYPE 115 */
270 PKT_RX_IPV6_HDR_EXT, /* PTYPE 116 */
271 PKT_RX_IPV6_HDR_EXT, /* PTYPE 117 */
272 PKT_RX_IPV6_HDR_EXT, /* PTYPE 118 */
273 PKT_RX_IPV6_HDR_EXT, /* PTYPE 119 */
275 PKT_RX_IPV6_HDR_EXT, /* PTYPE 121 */
276 PKT_RX_IPV6_HDR_EXT, /* PTYPE 122 */
277 PKT_RX_IPV6_HDR_EXT, /* PTYPE 123 */
278 PKT_RX_IPV6_HDR_EXT, /* PTYPE 124 */
279 PKT_RX_IPV6_HDR_EXT, /* PTYPE 125 */
280 PKT_RX_IPV6_HDR_EXT, /* PTYPE 126 */
281 PKT_RX_IPV6_HDR_EXT, /* PTYPE 127 */
283 PKT_RX_IPV6_HDR_EXT, /* PTYPE 129 */
284 PKT_RX_IPV6_HDR_EXT, /* PTYPE 130 */
285 PKT_RX_IPV6_HDR_EXT, /* PTYPE 131 */
286 PKT_RX_IPV6_HDR_EXT, /* PTYPE 132 */
287 PKT_RX_IPV6_HDR_EXT, /* PTYPE 133 */
288 PKT_RX_IPV6_HDR_EXT, /* PTYPE 134 */
290 PKT_RX_IPV6_HDR_EXT, /* PTYPE 136 */
291 PKT_RX_IPV6_HDR_EXT, /* PTYPE 137 */
292 PKT_RX_IPV6_HDR_EXT, /* PTYPE 138 */
293 PKT_RX_IPV6_HDR_EXT, /* PTYPE 139 */
294 PKT_RX_IPV6_HDR_EXT, /* PTYPE 140 */
295 PKT_RX_IPV6_HDR_EXT, /* PTYPE 141 */
296 PKT_RX_IPV6_HDR_EXT, /* PTYPE 142 */
298 PKT_RX_IPV6_HDR_EXT, /* PTYPE 144 */
299 PKT_RX_IPV6_HDR_EXT, /* PTYPE 145 */
300 PKT_RX_IPV6_HDR_EXT, /* PTYPE 146 */
301 PKT_RX_IPV6_HDR_EXT, /* PTYPE 147 */
302 PKT_RX_IPV6_HDR_EXT, /* PTYPE 148 */
303 PKT_RX_IPV6_HDR_EXT, /* PTYPE 149 */
305 PKT_RX_IPV6_HDR_EXT, /* PTYPE 151 */
306 PKT_RX_IPV6_HDR_EXT, /* PTYPE 152 */
307 PKT_RX_IPV6_HDR_EXT, /* PTYPE 153 */
412 return ip_ptype_map[ptype];
416 i40e_txd_enable_checksum(uint32_t ol_flags,
423 PMD_DRV_LOG(DEBUG, "L2 length set to 0\n");
426 *td_offset |= (l2_len >> 1) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
429 PMD_DRV_LOG(DEBUG, "L3 length set to 0\n");
433 /* Enable L3 checksum offloads */
434 if (ol_flags & PKT_TX_IPV4_CSUM) {
435 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
436 *td_offset |= (l3_len >> 2) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
437 } else if (ol_flags & PKT_TX_IPV4) {
438 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
439 *td_offset |= (l3_len >> 2) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
440 } else if (ol_flags & PKT_TX_IPV6) {
441 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
442 *td_offset |= (l3_len >> 2) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
445 /* Enable L4 checksum offloads */
446 switch (ol_flags & PKT_TX_L4_MASK) {
447 case PKT_TX_TCP_CKSUM:
448 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
449 *td_offset |= (sizeof(struct tcp_hdr) >> 2) <<
450 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
452 case PKT_TX_SCTP_CKSUM:
453 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
454 *td_offset |= (sizeof(struct sctp_hdr) >> 2) <<
455 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
457 case PKT_TX_UDP_CKSUM:
458 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
459 *td_offset |= (sizeof(struct udp_hdr) >> 2) <<
460 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
467 static inline struct rte_mbuf *
468 rte_rxmbuf_alloc(struct rte_mempool *mp)
472 m = __rte_mbuf_raw_alloc(mp);
473 __rte_mbuf_sanity_check_raw(m, RTE_MBUF_PKT, 0);
478 /* Construct the tx flags */
479 static inline uint64_t
480 i40e_build_ctob(uint32_t td_cmd,
485 return rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DATA |
486 ((uint64_t)td_cmd << I40E_TXD_QW1_CMD_SHIFT) |
487 ((uint64_t)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
488 ((uint64_t)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
489 ((uint64_t)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT));
493 i40e_xmit_cleanup(struct i40e_tx_queue *txq)
495 struct i40e_tx_entry *sw_ring = txq->sw_ring;
496 volatile struct i40e_tx_desc *txd = txq->tx_ring;
497 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
498 uint16_t nb_tx_desc = txq->nb_tx_desc;
499 uint16_t desc_to_clean_to;
500 uint16_t nb_tx_to_clean;
502 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
503 if (desc_to_clean_to >= nb_tx_desc)
504 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
506 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
507 if (!(txd[desc_to_clean_to].cmd_type_offset_bsz &
508 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE))) {
509 PMD_TX_FREE_LOG(DEBUG, "TX descriptor %4u is not done "
510 "(port=%d queue=%d)", desc_to_clean_to,
511 txq->port_id, txq->queue_id);
515 if (last_desc_cleaned > desc_to_clean_to)
516 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
519 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
522 txd[desc_to_clean_to].cmd_type_offset_bsz = 0;
524 txq->last_desc_cleaned = desc_to_clean_to;
525 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
531 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
532 check_rx_burst_bulk_alloc_preconditions(struct i40e_rx_queue *rxq)
534 check_rx_burst_bulk_alloc_preconditions(__rte_unused struct i40e_rx_queue *rxq)
539 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
540 if (!(rxq->rx_free_thresh >= RTE_PMD_I40E_RX_MAX_BURST))
542 else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc))
544 else if (!(rxq->nb_rx_desc % rxq->rx_free_thresh) == 0)
546 else if (!(rxq->nb_rx_desc < (I40E_MAX_RING_DESC -
547 RTE_PMD_I40E_RX_MAX_BURST)))
556 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
557 #define I40E_LOOK_AHEAD 8
558 #if (I40E_LOOK_AHEAD != 8)
559 #error "PMD I40E: I40E_LOOK_AHEAD must be 8\n"
562 i40e_rx_scan_hw_ring(struct i40e_rx_queue *rxq)
564 volatile union i40e_rx_desc *rxdp;
565 struct i40e_rx_entry *rxep;
570 int32_t s[I40E_LOOK_AHEAD], nb_dd;
571 int32_t i, j, nb_rx = 0;
574 rxdp = &rxq->rx_ring[rxq->rx_tail];
575 rxep = &rxq->sw_ring[rxq->rx_tail];
577 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
578 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
579 I40E_RXD_QW1_STATUS_SHIFT;
581 /* Make sure there is at least 1 packet to receive */
582 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
586 * Scan LOOK_AHEAD descriptors at a time to determine which
587 * descriptors reference packets that are ready to be received.
589 for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; i+=I40E_LOOK_AHEAD,
590 rxdp += I40E_LOOK_AHEAD, rxep += I40E_LOOK_AHEAD) {
591 /* Read desc statuses backwards to avoid race condition */
592 for (j = I40E_LOOK_AHEAD - 1; j >= 0; j--) {
593 qword1 = rte_le_to_cpu_64(\
594 rxdp[j].wb.qword1.status_error_len);
595 s[j] = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
596 I40E_RXD_QW1_STATUS_SHIFT;
599 /* Compute how many status bits were set */
600 for (j = 0, nb_dd = 0; j < I40E_LOOK_AHEAD; j++)
601 nb_dd += s[j] & (1 << I40E_RX_DESC_STATUS_DD_SHIFT);
605 /* Translate descriptor info to mbuf parameters */
606 for (j = 0; j < nb_dd; j++) {
608 qword1 = rte_le_to_cpu_64(\
609 rxdp[j].wb.qword1.status_error_len);
610 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
611 I40E_RXD_QW1_STATUS_SHIFT;
612 pkt_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
613 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
614 mb->pkt.data_len = pkt_len;
615 mb->pkt.pkt_len = pkt_len;
616 mb->pkt.vlan_macip.f.vlan_tci = rx_status &
617 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) ?
619 rxdp[j].wb.qword0.lo_dword.l2tag1) : 0;
620 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
621 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
622 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
623 mb->ol_flags = pkt_flags;
624 if (pkt_flags & PKT_RX_RSS_HASH)
625 mb->pkt.hash.rss = rte_le_to_cpu_32(\
626 rxdp->wb.qword0.hi_dword.rss);
629 for (j = 0; j < I40E_LOOK_AHEAD; j++)
630 rxq->rx_stage[i + j] = rxep[j].mbuf;
632 if (nb_dd != I40E_LOOK_AHEAD)
636 /* Clear software ring entries */
637 for (i = 0; i < nb_rx; i++)
638 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
643 static inline uint16_t
644 i40e_rx_fill_from_stage(struct i40e_rx_queue *rxq,
645 struct rte_mbuf **rx_pkts,
649 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
651 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
653 for (i = 0; i < nb_pkts; i++)
654 rx_pkts[i] = stage[i];
656 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
657 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
663 i40e_rx_alloc_bufs(struct i40e_rx_queue *rxq)
665 volatile union i40e_rx_desc *rxdp;
666 struct i40e_rx_entry *rxep;
668 uint16_t alloc_idx, i;
672 /* Allocate buffers in bulk */
673 alloc_idx = (uint16_t)(rxq->rx_free_trigger -
674 (rxq->rx_free_thresh - 1));
675 rxep = &(rxq->sw_ring[alloc_idx]);
676 diag = rte_mempool_get_bulk(rxq->mp, (void *)rxep,
677 rxq->rx_free_thresh);
678 if (unlikely(diag != 0)) {
679 PMD_DRV_LOG(ERR, "Failed to get mbufs in bulk\n");
683 rxdp = &rxq->rx_ring[alloc_idx];
684 for (i = 0; i < rxq->rx_free_thresh; i++) {
686 rte_mbuf_refcnt_set(mb, 1);
687 mb->type = RTE_MBUF_PKT;
689 mb->pkt.data = (char *)mb->buf_addr + RTE_PKTMBUF_HEADROOM;
691 mb->pkt.in_port = rxq->port_id;
692 dma_addr = rte_cpu_to_le_64(\
693 RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb));
694 rxdp[i].read.hdr_addr = dma_addr;
695 rxdp[i].read.pkt_addr = dma_addr;
698 /* Update rx tail regsiter */
700 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->rx_free_trigger);
702 rxq->rx_free_trigger =
703 (uint16_t)(rxq->rx_free_trigger + rxq->rx_free_thresh);
704 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
705 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
710 static inline uint16_t
711 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
713 struct i40e_rx_queue *rxq = (struct i40e_rx_queue *)rx_queue;
719 if (rxq->rx_nb_avail)
720 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
722 nb_rx = (uint16_t)i40e_rx_scan_hw_ring(rxq);
723 rxq->rx_next_avail = 0;
724 rxq->rx_nb_avail = nb_rx;
725 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
727 if (rxq->rx_tail > rxq->rx_free_trigger) {
728 if (i40e_rx_alloc_bufs(rxq) != 0) {
731 PMD_RX_LOG(DEBUG, "Rx mbuf alloc failed for "
732 "port_id=%u, queue_id=%u\n",
733 rxq->port_id, rxq->queue_id);
734 rxq->rx_nb_avail = 0;
735 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
736 for (i = 0, j = rxq->rx_tail; i < nb_rx; i++, j++)
737 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
743 if (rxq->rx_tail >= rxq->nb_rx_desc)
746 if (rxq->rx_nb_avail)
747 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
753 i40e_recv_pkts_bulk_alloc(void *rx_queue,
754 struct rte_mbuf **rx_pkts,
757 uint16_t nb_rx = 0, n, count;
759 if (unlikely(nb_pkts == 0))
762 if (likely(nb_pkts <= RTE_PMD_I40E_RX_MAX_BURST))
763 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
766 n = RTE_MIN(nb_pkts, RTE_PMD_I40E_RX_MAX_BURST);
767 count = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
768 nb_rx = (uint16_t)(nb_rx + count);
769 nb_pkts = (uint16_t)(nb_pkts - count);
776 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
779 i40e_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
781 struct i40e_rx_queue *rxq;
782 volatile union i40e_rx_desc *rx_ring;
783 volatile union i40e_rx_desc *rxdp;
784 union i40e_rx_desc rxd;
785 struct i40e_rx_entry *sw_ring;
786 struct i40e_rx_entry *rxe;
787 struct rte_mbuf *rxm;
788 struct rte_mbuf *nmb;
792 uint16_t rx_packet_len;
793 uint16_t rx_id, nb_hold;
800 rx_id = rxq->rx_tail;
801 rx_ring = rxq->rx_ring;
802 sw_ring = rxq->sw_ring;
804 while (nb_rx < nb_pkts) {
805 rxdp = &rx_ring[rx_id];
806 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
807 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK)
808 >> I40E_RXD_QW1_STATUS_SHIFT;
809 /* Check the DD bit first */
810 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
813 nmb = rte_rxmbuf_alloc(rxq->mp);
819 rxe = &sw_ring[rx_id];
821 if (unlikely(rx_id == rxq->nb_rx_desc))
824 /* Prefetch next mbuf */
825 rte_prefetch0(sw_ring[rx_id].mbuf);
828 * When next RX descriptor is on a cache line boundary,
829 * prefetch the next 4 RX descriptors and next 8 pointers
832 if ((rx_id & 0x3) == 0) {
833 rte_prefetch0(&rx_ring[rx_id]);
834 rte_prefetch0(&sw_ring[rx_id]);
839 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
840 rxdp->read.hdr_addr = dma_addr;
841 rxdp->read.pkt_addr = dma_addr;
843 rx_packet_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
844 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
846 rxm->pkt.data = (char *)rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
847 rte_prefetch0(rxm->pkt.data);
848 rxm->pkt.nb_segs = 1;
849 rxm->pkt.next = NULL;
850 rxm->pkt.pkt_len = rx_packet_len;
851 rxm->pkt.data_len = rx_packet_len;
852 rxm->pkt.in_port = rxq->port_id;
854 rxm->pkt.vlan_macip.f.vlan_tci = rx_status &
855 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) ?
856 rte_le_to_cpu_16(rxd.wb.qword0.lo_dword.l2tag1) : 0;
857 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
858 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
859 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
860 rxm->ol_flags = pkt_flags;
861 if (pkt_flags & PKT_RX_RSS_HASH)
863 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
865 rx_pkts[nb_rx++] = rxm;
867 rxq->rx_tail = rx_id;
870 * If the number of free RX descriptors is greater than the RX free
871 * threshold of the queue, advance the receive tail register of queue.
872 * Update that register with the value of the last processed RX
873 * descriptor minus 1.
875 nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
876 if (nb_hold > rxq->rx_free_thresh) {
877 rx_id = (uint16_t) ((rx_id == 0) ?
878 (rxq->nb_rx_desc - 1) : (rx_id - 1));
879 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
882 rxq->nb_rx_hold = nb_hold;
888 i40e_recv_scattered_pkts(void *rx_queue,
889 struct rte_mbuf **rx_pkts,
892 struct i40e_rx_queue *rxq = rx_queue;
893 volatile union i40e_rx_desc *rx_ring = rxq->rx_ring;
894 volatile union i40e_rx_desc *rxdp;
895 union i40e_rx_desc rxd;
896 struct i40e_rx_entry *sw_ring = rxq->sw_ring;
897 struct i40e_rx_entry *rxe;
898 struct rte_mbuf *first_seg = rxq->pkt_first_seg;
899 struct rte_mbuf *last_seg = rxq->pkt_last_seg;
900 struct rte_mbuf *nmb, *rxm;
901 uint16_t rx_id = rxq->rx_tail;
902 uint16_t nb_rx = 0, nb_hold = 0, rx_packet_len, pkt_flags;
907 while (nb_rx < nb_pkts) {
908 rxdp = &rx_ring[rx_id];
909 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
910 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
911 I40E_RXD_QW1_STATUS_SHIFT;
912 /* Check the DD bit */
913 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
916 nmb = rte_rxmbuf_alloc(rxq->mp);
921 rxe = &sw_ring[rx_id];
923 if (rx_id == rxq->nb_rx_desc)
926 /* Prefetch next mbuf */
927 rte_prefetch0(sw_ring[rx_id].mbuf);
930 * When next RX descriptor is on a cache line boundary,
931 * prefetch the next 4 RX descriptors and next 8 pointers
934 if ((rx_id & 0x3) == 0) {
935 rte_prefetch0(&rx_ring[rx_id]);
936 rte_prefetch0(&sw_ring[rx_id]);
942 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
944 /* Set data buffer address and data length of the mbuf */
945 rxdp->read.hdr_addr = dma_addr;
946 rxdp->read.pkt_addr = dma_addr;
947 rx_packet_len = (qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
948 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
949 rxm->pkt.data_len = rx_packet_len;
950 rxm->pkt.data = (char *)rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
953 * If this is the first buffer of the received packet, set the
954 * pointer to the first mbuf of the packet and initialize its
955 * context. Otherwise, update the total length and the number
956 * of segments of the current scattered packet, and update the
957 * pointer to the last mbuf of the current packet.
961 first_seg->pkt.nb_segs = 1;
962 first_seg->pkt.pkt_len = rx_packet_len;
964 first_seg->pkt.pkt_len =
965 (uint16_t)(first_seg->pkt.pkt_len +
967 first_seg->pkt.nb_segs++;
968 last_seg->pkt.next = rxm;
972 * If this is not the last buffer of the received packet,
973 * update the pointer to the last mbuf of the current scattered
974 * packet and continue to parse the RX ring.
976 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT))) {
982 * This is the last buffer of the received packet. If the CRC
983 * is not stripped by the hardware:
984 * - Subtract the CRC length from the total packet length.
985 * - If the last buffer only contains the whole CRC or a part
986 * of it, free the mbuf associated to the last buffer. If part
987 * of the CRC is also contained in the previous mbuf, subtract
988 * the length of that CRC part from the data length of the
991 rxm->pkt.next = NULL;
992 if (unlikely(rxq->crc_len > 0)) {
993 first_seg->pkt.pkt_len -= ETHER_CRC_LEN;
994 if (rx_packet_len <= ETHER_CRC_LEN) {
995 rte_pktmbuf_free_seg(rxm);
996 first_seg->pkt.nb_segs--;
997 last_seg->pkt.data_len =
998 (uint16_t)(last_seg->pkt.data_len -
999 (ETHER_CRC_LEN - rx_packet_len));
1000 last_seg->pkt.next = NULL;
1002 rxm->pkt.data_len = (uint16_t)(rx_packet_len -
1006 first_seg->pkt.in_port = rxq->port_id;
1007 first_seg->pkt.vlan_macip.f.vlan_tci = (rx_status &
1008 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
1009 rte_le_to_cpu_16(rxd.wb.qword0.lo_dword.l2tag1) : 0;
1010 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
1011 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
1012 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
1013 first_seg->ol_flags = pkt_flags;
1014 if (pkt_flags & PKT_RX_RSS_HASH)
1016 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
1018 /* Prefetch data of first segment, if configured to do so. */
1019 rte_prefetch0(first_seg->pkt.data);
1020 rx_pkts[nb_rx++] = first_seg;
1024 /* Record index of the next RX descriptor to probe. */
1025 rxq->rx_tail = rx_id;
1026 rxq->pkt_first_seg = first_seg;
1027 rxq->pkt_last_seg = last_seg;
1030 * If the number of free RX descriptors is greater than the RX free
1031 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1032 * register. Update the RDT with the value of the last processed RX
1033 * descriptor minus 1, to guarantee that the RDT register is never
1034 * equal to the RDH register, which creates a "full" ring situtation
1035 * from the hardware point of view.
1037 nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
1038 if (nb_hold > rxq->rx_free_thresh) {
1039 rx_id = (uint16_t)(rx_id == 0 ?
1040 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1041 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
1044 rxq->nb_rx_hold = nb_hold;
1049 /* Check if the context descriptor is needed for TX offloading */
1050 static inline uint16_t
1051 i40e_calc_context_desc(uint16_t flags)
1055 #ifdef RTE_LIBRTE_IEEE1588
1056 mask |= PKT_TX_IEEE1588_TMST;
1065 i40e_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1067 struct i40e_tx_queue *txq;
1068 struct i40e_tx_entry *sw_ring;
1069 struct i40e_tx_entry *txe, *txn;
1070 volatile struct i40e_tx_desc *txd;
1071 volatile struct i40e_tx_desc *txr;
1072 struct rte_mbuf *tx_pkt;
1073 struct rte_mbuf *m_seg;
1087 uint64_t buf_dma_addr;
1090 sw_ring = txq->sw_ring;
1092 tx_id = txq->tx_tail;
1093 txe = &sw_ring[tx_id];
1095 /* Check if the descriptor ring needs to be cleaned. */
1096 if ((txq->nb_tx_desc - txq->nb_tx_free) > txq->tx_free_thresh)
1097 i40e_xmit_cleanup(txq);
1099 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
1105 tx_pkt = *tx_pkts++;
1106 RTE_MBUF_PREFETCH_TO_FREE(txe->mbuf);
1108 ol_flags = tx_pkt->ol_flags;
1109 l2_len = tx_pkt->pkt.vlan_macip.f.l2_len;
1110 l3_len = tx_pkt->pkt.vlan_macip.f.l3_len;
1112 /* Calculate the number of context descriptors needed. */
1113 nb_ctx = i40e_calc_context_desc(ol_flags);
1116 * The number of descriptors that must be allocated for
1117 * a packet equals to the number of the segments of that
1118 * packet plus 1 context descriptor if needed.
1120 nb_used = (uint16_t)(tx_pkt->pkt.nb_segs + nb_ctx);
1121 tx_last = (uint16_t)(tx_id + nb_used - 1);
1124 if (tx_last >= txq->nb_tx_desc)
1125 tx_last = (uint16_t)(tx_last - txq->nb_tx_desc);
1127 if (nb_used > txq->nb_tx_free) {
1128 if (i40e_xmit_cleanup(txq) != 0) {
1133 if (unlikely(nb_used > txq->tx_rs_thresh)) {
1134 while (nb_used > txq->nb_tx_free) {
1135 if (i40e_xmit_cleanup(txq) != 0) {
1144 /* Descriptor based VLAN insertion */
1145 if (ol_flags & PKT_TX_VLAN_PKT) {
1146 tx_flags |= tx_pkt->pkt.vlan_macip.f.vlan_tci <<
1147 I40E_TX_FLAG_L2TAG1_SHIFT;
1148 tx_flags |= I40E_TX_FLAG_INSERT_VLAN;
1149 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1150 td_tag = (tx_flags & I40E_TX_FLAG_L2TAG1_MASK) >>
1151 I40E_TX_FLAG_L2TAG1_SHIFT;
1154 /* Always enable CRC offload insertion */
1155 td_cmd |= I40E_TX_DESC_CMD_ICRC;
1157 /* Enable checksum offloading */
1158 i40e_txd_enable_checksum(ol_flags, &td_cmd, &td_offset,
1161 if (unlikely(nb_ctx)) {
1162 /* Setup TX context descriptor if required */
1163 volatile struct i40e_tx_context_desc *ctx_txd =
1164 (volatile struct i40e_tx_context_desc *)\
1166 uint32_t cd_tunneling_params = 0;
1167 uint16_t cd_l2tag2 = 0;
1168 uint64_t cd_type_cmd_tso_mss =
1169 I40E_TX_DESC_DTYPE_CONTEXT;
1171 txn = &sw_ring[txe->next_id];
1172 RTE_MBUF_PREFETCH_TO_FREE(txn->mbuf);
1173 if (txe->mbuf != NULL) {
1174 rte_pktmbuf_free_seg(txe->mbuf);
1177 #ifdef RTE_LIBRTE_IEEE1588
1178 if (ol_flags & PKT_TX_IEEE1588_TMST)
1179 cd_type_cmd_tso_mss |=
1180 ((uint64_t)I40E_TX_CTX_DESC_TSYN <<
1181 I40E_TXD_CTX_QW1_CMD_SHIFT);
1183 ctx_txd->tunneling_params =
1184 rte_cpu_to_le_32(cd_tunneling_params);
1185 ctx_txd->l2tag2 = rte_cpu_to_le_16(cd_l2tag2);
1186 ctx_txd->type_cmd_tso_mss =
1187 rte_cpu_to_le_64(cd_type_cmd_tso_mss);
1188 txe->last_id = tx_last;
1189 tx_id = txe->next_id;
1196 txn = &sw_ring[txe->next_id];
1199 rte_pktmbuf_free_seg(txe->mbuf);
1202 /* Setup TX Descriptor */
1203 slen = m_seg->pkt.data_len;
1204 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
1205 txd->buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
1206 txd->cmd_type_offset_bsz = i40e_build_ctob(td_cmd,
1207 td_offset, slen, td_tag);
1208 txe->last_id = tx_last;
1209 tx_id = txe->next_id;
1211 m_seg = m_seg->pkt.next;
1212 } while (m_seg != NULL);
1214 /* The last packet data descriptor needs End Of Packet (EOP) */
1215 td_cmd |= I40E_TX_DESC_CMD_EOP;
1216 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
1217 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
1219 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
1220 PMD_TX_FREE_LOG(DEBUG,
1221 "Setting RS bit on TXD id="
1222 "%4u (port=%d queue=%d)",
1223 tx_last, txq->port_id, txq->queue_id);
1225 td_cmd |= I40E_TX_DESC_CMD_RS;
1227 /* Update txq RS bit counters */
1228 txq->nb_tx_used = 0;
1231 txd->cmd_type_offset_bsz |=
1232 rte_cpu_to_le_64(((uint64_t)td_cmd) <<
1233 I40E_TXD_QW1_CMD_SHIFT);
1239 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
1240 (unsigned) txq->port_id, (unsigned) txq->queue_id,
1241 (unsigned) tx_id, (unsigned) nb_tx);
1243 I40E_PCI_REG_WRITE(txq->qtx_tail, tx_id);
1244 txq->tx_tail = tx_id;
1249 static inline int __attribute__((always_inline))
1250 i40e_tx_free_bufs(struct i40e_tx_queue *txq)
1252 struct i40e_tx_entry *txep;
1255 if (!(txq->tx_ring[txq->tx_next_dd].cmd_type_offset_bsz &
1256 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE)))
1259 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
1261 for (i = 0; i < txq->tx_rs_thresh; i++)
1262 rte_prefetch0((txep + i)->mbuf);
1264 if (!(txq->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT)) {
1265 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1266 rte_mempool_put(txep->mbuf->pool, txep->mbuf);
1270 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1271 rte_pktmbuf_free_seg(txep->mbuf);
1276 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
1277 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
1278 if (txq->tx_next_dd >= txq->nb_tx_desc)
1279 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
1281 return txq->tx_rs_thresh;
1284 #define I40E_TD_CMD (I40E_TX_DESC_CMD_ICRC |\
1285 I40E_TX_DESC_CMD_EOP)
1287 /* Populate 4 descriptors with data from 4 mbufs */
1289 tx4(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1294 for (i = 0; i < 4; i++, txdp++, pkts++) {
1295 dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
1296 txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1297 txdp->cmd_type_offset_bsz =
1298 i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1299 (*pkts)->pkt.data_len, 0);
1303 /* Populate 1 descriptor with data from 1 mbuf */
1305 tx1(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1309 dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
1310 txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1311 txdp->cmd_type_offset_bsz =
1312 i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1313 (*pkts)->pkt.data_len, 0);
1316 /* Fill hardware descriptor ring with mbuf data */
1318 i40e_tx_fill_hw_ring(struct i40e_tx_queue *txq,
1319 struct rte_mbuf **pkts,
1322 volatile struct i40e_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
1323 struct i40e_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
1324 const int N_PER_LOOP = 4;
1325 const int N_PER_LOOP_MASK = N_PER_LOOP - 1;
1326 int mainpart, leftover;
1329 mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
1330 leftover = (nb_pkts & ((uint32_t) N_PER_LOOP_MASK));
1331 for (i = 0; i < mainpart; i += N_PER_LOOP) {
1332 for (j = 0; j < N_PER_LOOP; ++j) {
1333 (txep + i + j)->mbuf = *(pkts + i + j);
1335 tx4(txdp + i, pkts + i);
1337 if (unlikely(leftover > 0)) {
1338 for (i = 0; i < leftover; ++i) {
1339 (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
1340 tx1(txdp + mainpart + i, pkts + mainpart + i);
1345 static inline uint16_t
1346 tx_xmit_pkts(struct i40e_tx_queue *txq,
1347 struct rte_mbuf **tx_pkts,
1350 volatile struct i40e_tx_desc *txr = txq->tx_ring;
1354 * Begin scanning the H/W ring for done descriptors when the number
1355 * of available descriptors drops below tx_free_thresh. For each done
1356 * descriptor, free the associated buffer.
1358 if (txq->nb_tx_free < txq->tx_free_thresh)
1359 i40e_tx_free_bufs(txq);
1361 /* Use available descriptor only */
1362 nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
1363 if (unlikely(!nb_pkts))
1366 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
1367 if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
1368 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
1369 i40e_tx_fill_hw_ring(txq, tx_pkts, n);
1370 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1371 rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1372 I40E_TXD_QW1_CMD_SHIFT);
1373 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1377 /* Fill hardware descriptor ring with mbuf data */
1378 i40e_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
1379 txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
1381 /* Determin if RS bit needs to be set */
1382 if (txq->tx_tail > txq->tx_next_rs) {
1383 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1384 rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1385 I40E_TXD_QW1_CMD_SHIFT);
1387 (uint16_t)(txq->tx_next_rs + txq->tx_rs_thresh);
1388 if (txq->tx_next_rs >= txq->nb_tx_desc)
1389 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1392 if (txq->tx_tail >= txq->nb_tx_desc)
1395 /* Update the tx tail register */
1397 I40E_PCI_REG_WRITE(txq->qtx_tail, txq->tx_tail);
1403 i40e_xmit_pkts_simple(void *tx_queue,
1404 struct rte_mbuf **tx_pkts,
1409 if (likely(nb_pkts <= I40E_TX_MAX_BURST))
1410 return tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1414 uint16_t ret, num = (uint16_t)RTE_MIN(nb_pkts,
1417 ret = tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1418 &tx_pkts[nb_tx], num);
1419 nb_tx = (uint16_t)(nb_tx + ret);
1420 nb_pkts = (uint16_t)(nb_pkts - ret);
1429 i40e_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1431 struct i40e_vsi *vsi = I40E_DEV_PRIVATE_TO_VSI(dev->data->dev_private);
1432 struct i40e_rx_queue *rxq;
1434 struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
1435 uint16_t q_base = vsi->base_queue;
1437 PMD_INIT_FUNC_TRACE();
1439 if (rx_queue_id < dev->data->nb_rx_queues) {
1440 rxq = dev->data->rx_queues[rx_queue_id];
1442 err = i40e_alloc_rx_queue_mbufs(rxq);
1444 PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf\n");
1450 /* Init the RX tail regieter. */
1451 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1453 err = i40e_switch_rx_queue(hw, rx_queue_id + q_base, TRUE);
1456 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on\n",
1459 i40e_rx_queue_release_mbufs(rxq);
1460 i40e_reset_rx_queue(rxq);
1468 i40e_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1470 struct i40e_vsi *vsi = I40E_DEV_PRIVATE_TO_VSI(dev->data->dev_private);
1471 struct i40e_rx_queue *rxq;
1473 struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
1474 uint16_t q_base = vsi->base_queue;
1476 if (rx_queue_id < dev->data->nb_rx_queues) {
1477 rxq = dev->data->rx_queues[rx_queue_id];
1479 err = i40e_switch_rx_queue(hw, rx_queue_id + q_base, FALSE);
1482 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off\n",
1486 i40e_rx_queue_release_mbufs(rxq);
1487 i40e_reset_rx_queue(rxq);
1494 i40e_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1496 struct i40e_vsi *vsi = I40E_DEV_PRIVATE_TO_VSI(dev->data->dev_private);
1498 struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
1499 uint16_t q_base = vsi->base_queue;
1501 PMD_INIT_FUNC_TRACE();
1503 if (tx_queue_id < dev->data->nb_tx_queues) {
1504 err = i40e_switch_tx_queue(hw, tx_queue_id + q_base, TRUE);
1506 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on\n",
1514 i40e_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1516 struct i40e_vsi *vsi = I40E_DEV_PRIVATE_TO_VSI(dev->data->dev_private);
1517 struct i40e_tx_queue *txq;
1519 struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
1520 uint16_t q_base = vsi->base_queue;
1522 if (tx_queue_id < dev->data->nb_tx_queues) {
1523 txq = dev->data->tx_queues[tx_queue_id];
1525 err = i40e_switch_tx_queue(hw, tx_queue_id + q_base, FALSE);
1528 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u of\n",
1533 i40e_tx_queue_release_mbufs(txq);
1534 i40e_reset_tx_queue(txq);
1541 i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
1544 unsigned int socket_id,
1545 const struct rte_eth_rxconf *rx_conf,
1546 struct rte_mempool *mp)
1548 struct i40e_vsi *vsi = I40E_DEV_PRIVATE_TO_VSI(dev->data->dev_private);
1549 struct i40e_rx_queue *rxq;
1550 const struct rte_memzone *rz;
1553 int use_def_burst_func = 1;
1555 if (!vsi || queue_idx >= vsi->nb_qps) {
1556 PMD_DRV_LOG(ERR, "VSI not available or queue "
1557 "index exceeds the maximum\n");
1558 return I40E_ERR_PARAM;
1560 if (((nb_desc * sizeof(union i40e_rx_desc)) % I40E_ALIGN) != 0 ||
1561 (nb_desc > I40E_MAX_RING_DESC) ||
1562 (nb_desc < I40E_MIN_RING_DESC)) {
1563 PMD_DRV_LOG(ERR, "Number (%u) of receive descriptors is "
1564 "invalid\n", nb_desc);
1565 return I40E_ERR_PARAM;
1568 /* Free memory if needed */
1569 if (dev->data->rx_queues[queue_idx]) {
1570 i40e_dev_rx_queue_release(dev->data->rx_queues[queue_idx]);
1571 dev->data->rx_queues[queue_idx] = NULL;
1574 /* Allocate the rx queue data structure */
1575 rxq = rte_zmalloc_socket("i40e rx queue",
1576 sizeof(struct i40e_rx_queue),
1580 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
1581 "rx queue data structure\n");
1585 rxq->nb_rx_desc = nb_desc;
1586 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
1587 rxq->queue_id = queue_idx;
1588 rxq->reg_idx = vsi->base_queue + queue_idx;
1589 rxq->port_id = dev->data->port_id;
1590 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
1592 rxq->drop_en = rx_conf->rx_drop_en;
1594 rxq->start_rx_per_q = rx_conf->start_rx_per_q;
1596 /* Allocate the maximun number of RX ring hardware descriptor. */
1597 ring_size = sizeof(union i40e_rx_desc) * I40E_MAX_RING_DESC;
1598 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
1599 rz = i40e_ring_dma_zone_reserve(dev,
1605 i40e_dev_rx_queue_release(rxq);
1606 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX\n");
1610 /* Zero all the descriptors in the ring. */
1611 memset(rz->addr, 0, ring_size);
1613 #ifdef RTE_LIBRTE_XEN_DOM0
1614 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
1616 rxq->rx_ring_phys_addr = (uint64_t)rz->phys_addr;
1619 rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
1621 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1622 len = (uint16_t)(nb_desc + RTE_PMD_I40E_RX_MAX_BURST);
1627 /* Allocate the software ring. */
1629 rte_zmalloc_socket("i40e rx sw ring",
1630 sizeof(struct i40e_rx_entry) * len,
1633 if (!rxq->sw_ring) {
1634 i40e_dev_rx_queue_release(rxq);
1635 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW ring\n");
1639 i40e_reset_rx_queue(rxq);
1641 dev->data->rx_queues[queue_idx] = rxq;
1643 use_def_burst_func = check_rx_burst_bulk_alloc_preconditions(rxq);
1645 if (!use_def_burst_func && !dev->data->scattered_rx) {
1646 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1647 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
1648 "satisfied. Rx Burst Bulk Alloc function will be "
1649 "used on port=%d, queue=%d.\n",
1650 rxq->port_id, rxq->queue_id);
1651 dev->rx_pkt_burst = i40e_recv_pkts_bulk_alloc;
1652 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
1654 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
1655 "not satisfied, Scattered Rx is requested, "
1656 "or RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC is "
1657 "not enabled on port=%d, queue=%d.\n",
1658 rxq->port_id, rxq->queue_id);
1665 i40e_dev_rx_queue_release(void *rxq)
1667 struct i40e_rx_queue *q = (struct i40e_rx_queue *)rxq;
1670 PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL\n");
1674 i40e_rx_queue_release_mbufs(q);
1675 rte_free(q->sw_ring);
1680 i40e_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1682 #define I40E_RXQ_SCAN_INTERVAL 4
1683 volatile union i40e_rx_desc *rxdp;
1684 struct i40e_rx_queue *rxq;
1687 if (unlikely(rx_queue_id >= dev->data->nb_rx_queues)) {
1688 PMD_DRV_LOG(ERR, "Invalid RX queue id %u\n", rx_queue_id);
1692 rxq = dev->data->rx_queues[rx_queue_id];
1693 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
1694 while ((desc < rxq->nb_rx_desc) &&
1695 ((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1696 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
1697 (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) {
1699 * Check the DD bit of a rx descriptor of each 4 in a group,
1700 * to avoid checking too frequently and downgrading performance
1703 desc += I40E_RXQ_SCAN_INTERVAL;
1704 rxdp += I40E_RXQ_SCAN_INTERVAL;
1705 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
1706 rxdp = &(rxq->rx_ring[rxq->rx_tail +
1707 desc - rxq->nb_rx_desc]);
1714 i40e_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
1716 volatile union i40e_rx_desc *rxdp;
1717 struct i40e_rx_queue *rxq = rx_queue;
1721 if (unlikely(offset >= rxq->nb_rx_desc)) {
1722 PMD_DRV_LOG(ERR, "Invalid RX queue id %u\n", offset);
1726 desc = rxq->rx_tail + offset;
1727 if (desc >= rxq->nb_rx_desc)
1728 desc -= rxq->nb_rx_desc;
1730 rxdp = &(rxq->rx_ring[desc]);
1732 ret = !!(((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1733 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
1734 (1 << I40E_RX_DESC_STATUS_DD_SHIFT));
1740 i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
1743 unsigned int socket_id,
1744 const struct rte_eth_txconf *tx_conf)
1746 struct i40e_vsi *vsi = I40E_DEV_PRIVATE_TO_VSI(dev->data->dev_private);
1747 struct i40e_tx_queue *txq;
1748 const struct rte_memzone *tz;
1750 uint16_t tx_rs_thresh, tx_free_thresh;
1752 if (!vsi || queue_idx >= vsi->nb_qps) {
1753 PMD_DRV_LOG(ERR, "VSI is NULL, or queue index (%u) "
1754 "exceeds the maximum\n", queue_idx);
1755 return I40E_ERR_PARAM;
1758 if (((nb_desc * sizeof(struct i40e_tx_desc)) % I40E_ALIGN) != 0 ||
1759 (nb_desc > I40E_MAX_RING_DESC) ||
1760 (nb_desc < I40E_MIN_RING_DESC)) {
1761 PMD_DRV_LOG(ERR, "Number (%u) of transmit descriptors is "
1762 "invalid\n", nb_desc);
1763 return I40E_ERR_PARAM;
1767 * The following two parameters control the setting of the RS bit on
1768 * transmit descriptors. TX descriptors will have their RS bit set
1769 * after txq->tx_rs_thresh descriptors have been used. The TX
1770 * descriptor ring will be cleaned after txq->tx_free_thresh
1771 * descriptors are used or if the number of descriptors required to
1772 * transmit a packet is greater than the number of free TX descriptors.
1774 * The following constraints must be satisfied:
1775 * - tx_rs_thresh must be greater than 0.
1776 * - tx_rs_thresh must be less than the size of the ring minus 2.
1777 * - tx_rs_thresh must be less than or equal to tx_free_thresh.
1778 * - tx_rs_thresh must be a divisor of the ring size.
1779 * - tx_free_thresh must be greater than 0.
1780 * - tx_free_thresh must be less than the size of the ring minus 3.
1782 * One descriptor in the TX ring is used as a sentinel to avoid a H/W
1783 * race condition, hence the maximum threshold constraints. When set
1784 * to zero use default values.
1786 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
1787 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
1788 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
1789 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
1790 if (tx_rs_thresh >= (nb_desc - 2)) {
1791 RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than the "
1792 "number of TX descriptors minus 2. "
1793 "(tx_rs_thresh=%u port=%d queue=%d)\n",
1794 (unsigned int)tx_rs_thresh,
1795 (int)dev->data->port_id,
1797 return I40E_ERR_PARAM;
1799 if (tx_free_thresh >= (nb_desc - 3)) {
1800 RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than the "
1801 "tx_free_thresh must be less than the "
1802 "number of TX descriptors minus 3. "
1803 "(tx_free_thresh=%u port=%d queue=%d)\n",
1804 (unsigned int)tx_free_thresh,
1805 (int)dev->data->port_id,
1807 return I40E_ERR_PARAM;
1809 if (tx_rs_thresh > tx_free_thresh) {
1810 RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than or "
1811 "equal to tx_free_thresh. (tx_free_thresh=%u"
1812 " tx_rs_thresh=%u port=%d queue=%d)\n",
1813 (unsigned int)tx_free_thresh,
1814 (unsigned int)tx_rs_thresh,
1815 (int)dev->data->port_id,
1817 return I40E_ERR_PARAM;
1819 if ((nb_desc % tx_rs_thresh) != 0) {
1820 RTE_LOG(ERR, PMD, "tx_rs_thresh must be a divisor of the "
1821 "number of TX descriptors. (tx_rs_thresh=%u"
1822 " port=%d queue=%d)\n",
1823 (unsigned int)tx_rs_thresh,
1824 (int)dev->data->port_id,
1826 return I40E_ERR_PARAM;
1828 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
1829 RTE_LOG(ERR, PMD, "TX WTHRESH must be set to 0 if "
1830 "tx_rs_thresh is greater than 1. "
1831 "(tx_rs_thresh=%u port=%d queue=%d)\n",
1832 (unsigned int)tx_rs_thresh,
1833 (int)dev->data->port_id,
1835 return I40E_ERR_PARAM;
1838 /* Free memory if needed. */
1839 if (dev->data->tx_queues[queue_idx]) {
1840 i40e_dev_tx_queue_release(dev->data->tx_queues[queue_idx]);
1841 dev->data->tx_queues[queue_idx] = NULL;
1844 /* Allocate the TX queue data structure. */
1845 txq = rte_zmalloc_socket("i40e tx queue",
1846 sizeof(struct i40e_tx_queue),
1850 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
1851 "tx queue structure\n");
1855 /* Allocate TX hardware ring descriptors. */
1856 ring_size = sizeof(struct i40e_tx_desc) * I40E_MAX_RING_DESC;
1857 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
1858 tz = i40e_ring_dma_zone_reserve(dev,
1864 i40e_dev_tx_queue_release(txq);
1865 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX\n");
1869 txq->nb_tx_desc = nb_desc;
1870 txq->tx_rs_thresh = tx_rs_thresh;
1871 txq->tx_free_thresh = tx_free_thresh;
1872 txq->pthresh = tx_conf->tx_thresh.pthresh;
1873 txq->hthresh = tx_conf->tx_thresh.hthresh;
1874 txq->wthresh = tx_conf->tx_thresh.wthresh;
1875 txq->queue_id = queue_idx;
1876 txq->reg_idx = vsi->base_queue + queue_idx;
1877 txq->port_id = dev->data->port_id;
1878 txq->txq_flags = tx_conf->txq_flags;
1880 txq->start_tx_per_q = tx_conf->start_tx_per_q;
1882 #ifdef RTE_LIBRTE_XEN_DOM0
1883 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
1885 txq->tx_ring_phys_addr = (uint64_t)tz->phys_addr;
1887 txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
1889 /* Allocate software ring */
1891 rte_zmalloc_socket("i40e tx sw ring",
1892 sizeof(struct i40e_tx_entry) * nb_desc,
1895 if (!txq->sw_ring) {
1896 i40e_dev_tx_queue_release(txq);
1897 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW TX ring\n");
1901 i40e_reset_tx_queue(txq);
1903 dev->data->tx_queues[queue_idx] = txq;
1905 /* Use a simple TX queue without offloads or multi segs if possible */
1906 if (((txq->txq_flags & I40E_SIMPLE_FLAGS) == I40E_SIMPLE_FLAGS) &&
1907 (txq->tx_rs_thresh >= I40E_TX_MAX_BURST)) {
1908 PMD_INIT_LOG(INFO, "Using simple tx path\n");
1909 dev->tx_pkt_burst = i40e_xmit_pkts_simple;
1911 PMD_INIT_LOG(INFO, "Using full-featured tx path\n");
1912 dev->tx_pkt_burst = i40e_xmit_pkts;
1919 i40e_dev_tx_queue_release(void *txq)
1921 struct i40e_tx_queue *q = (struct i40e_tx_queue *)txq;
1924 PMD_DRV_LOG(DEBUG, "Pointer to TX queue is NULL\n");
1928 i40e_tx_queue_release_mbufs(q);
1929 rte_free(q->sw_ring);
1933 static const struct rte_memzone *
1934 i40e_ring_dma_zone_reserve(struct rte_eth_dev *dev,
1935 const char *ring_name,
1940 char z_name[RTE_MEMZONE_NAMESIZE];
1941 const struct rte_memzone *mz;
1943 snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
1944 dev->driver->pci_drv.name, ring_name,
1945 dev->data->port_id, queue_id);
1946 mz = rte_memzone_lookup(z_name);
1950 #ifdef RTE_LIBRTE_XEN_DOM0
1951 return rte_memzone_reserve_bounded(z_name, ring_size,
1952 socket_id, 0, I40E_ALIGN, RTE_PGSIZE_2M);
1954 return rte_memzone_reserve_aligned(z_name, ring_size,
1955 socket_id, 0, I40E_ALIGN);
1960 i40e_rx_queue_release_mbufs(struct i40e_rx_queue *rxq)
1964 if (!rxq || !rxq->sw_ring) {
1965 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL\n");
1969 for (i = 0; i < rxq->nb_rx_desc; i++) {
1970 if (rxq->sw_ring[i].mbuf) {
1971 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
1972 rxq->sw_ring[i].mbuf = NULL;
1975 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1976 if (rxq->rx_nb_avail == 0)
1978 for (i = 0; i < rxq->rx_nb_avail; i++) {
1979 struct rte_mbuf *mbuf;
1981 mbuf = rxq->rx_stage[rxq->rx_next_avail + i];
1982 rte_pktmbuf_free_seg(mbuf);
1984 rxq->rx_nb_avail = 0;
1985 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
1989 i40e_reset_rx_queue(struct i40e_rx_queue *rxq)
1994 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1995 if (check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
1996 len = (uint16_t)(rxq->nb_rx_desc + RTE_PMD_I40E_RX_MAX_BURST);
1998 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
1999 len = rxq->nb_rx_desc;
2001 for (i = 0; i < len * sizeof(union i40e_rx_desc); i++)
2002 ((volatile char *)rxq->rx_ring)[i] = 0;
2004 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2005 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2006 for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; ++i)
2007 rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
2009 rxq->rx_nb_avail = 0;
2010 rxq->rx_next_avail = 0;
2011 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2012 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2014 rxq->nb_rx_hold = 0;
2015 rxq->pkt_first_seg = NULL;
2016 rxq->pkt_last_seg = NULL;
2020 i40e_tx_queue_release_mbufs(struct i40e_tx_queue *txq)
2024 if (!txq || !txq->sw_ring) {
2025 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL\n");
2029 for (i = 0; i < txq->nb_tx_desc; i++) {
2030 if (txq->sw_ring[i].mbuf) {
2031 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2032 txq->sw_ring[i].mbuf = NULL;
2038 i40e_reset_tx_queue(struct i40e_tx_queue *txq)
2040 struct i40e_tx_entry *txe;
2041 uint16_t i, prev, size;
2044 PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL\n");
2049 size = sizeof(struct i40e_tx_desc) * txq->nb_tx_desc;
2050 for (i = 0; i < size; i++)
2051 ((volatile char *)txq->tx_ring)[i] = 0;
2053 prev = (uint16_t)(txq->nb_tx_desc - 1);
2054 for (i = 0; i < txq->nb_tx_desc; i++) {
2055 volatile struct i40e_tx_desc *txd = &txq->tx_ring[i];
2057 txd[i].cmd_type_offset_bsz =
2058 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE);
2061 txe[prev].next_id = i;
2065 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2066 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2069 txq->nb_tx_used = 0;
2071 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2072 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2075 /* Init the TX queue in hardware */
2077 i40e_tx_queue_init(struct i40e_tx_queue *txq)
2079 enum i40e_status_code err = I40E_SUCCESS;
2080 struct i40e_vsi *vsi = txq->vsi;
2081 struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2082 uint16_t pf_q = txq->reg_idx;
2083 struct i40e_hmc_obj_txq tx_ctx;
2086 /* clear the context structure first */
2087 memset(&tx_ctx, 0, sizeof(tx_ctx));
2088 tx_ctx.new_context = 1;
2089 tx_ctx.base = txq->tx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2090 tx_ctx.qlen = txq->nb_tx_desc;
2091 tx_ctx.rdylist = rte_le_to_cpu_16(vsi->info.qs_handle[0]);
2093 err = i40e_clear_lan_tx_queue_context(hw, pf_q);
2094 if (err != I40E_SUCCESS) {
2095 PMD_DRV_LOG(ERR, "Failure of clean lan tx queue context\n");
2099 err = i40e_set_lan_tx_queue_context(hw, pf_q, &tx_ctx);
2100 if (err != I40E_SUCCESS) {
2101 PMD_DRV_LOG(ERR, "Failure of set lan tx queue context\n");
2105 /* Now associate this queue with this PCI function */
2106 qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
2107 qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
2108 I40E_QTX_CTL_PF_INDX_MASK);
2109 I40E_WRITE_REG(hw, I40E_QTX_CTL(pf_q), qtx_ctl);
2110 I40E_WRITE_FLUSH(hw);
2112 txq->qtx_tail = hw->hw_addr + I40E_QTX_TAIL(pf_q);
2118 i40e_alloc_rx_queue_mbufs(struct i40e_rx_queue *rxq)
2120 struct i40e_rx_entry *rxe = rxq->sw_ring;
2124 for (i = 0; i < rxq->nb_rx_desc; i++) {
2125 volatile union i40e_rx_desc *rxd;
2126 struct rte_mbuf *mbuf = rte_rxmbuf_alloc(rxq->mp);
2128 if (unlikely(!mbuf)) {
2129 PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX\n");
2133 rte_mbuf_refcnt_set(mbuf, 1);
2134 mbuf->type = RTE_MBUF_PKT;
2135 mbuf->pkt.next = NULL;
2136 mbuf->pkt.data = (char *)mbuf->buf_addr + RTE_PKTMBUF_HEADROOM;
2137 mbuf->pkt.nb_segs = 1;
2138 mbuf->pkt.in_port = rxq->port_id;
2141 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf));
2143 rxd = &rxq->rx_ring[i];
2144 rxd->read.pkt_addr = dma_addr;
2145 rxd->read.hdr_addr = dma_addr;
2146 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2147 rxd->read.rsvd1 = 0;
2148 rxd->read.rsvd2 = 0;
2149 #endif /* RTE_LIBRTE_I40E_16BYTE_RX_DESC */
2158 * Calculate the buffer length, and check the jumbo frame
2159 * and maximum packet length.
2162 i40e_rx_queue_config(struct i40e_rx_queue *rxq)
2164 struct i40e_pf *pf = I40E_VSI_TO_PF(rxq->vsi);
2165 struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2166 struct rte_eth_dev_data *data = pf->dev_data;
2167 struct rte_pktmbuf_pool_private *mbp_priv =
2168 rte_mempool_get_priv(rxq->mp);
2169 uint16_t buf_size = (uint16_t)(mbp_priv->mbuf_data_room_size -
2170 RTE_PKTMBUF_HEADROOM);
2173 switch (pf->flags & (I40E_FLAG_HEADER_SPLIT_DISABLED |
2174 I40E_FLAG_HEADER_SPLIT_ENABLED)) {
2175 case I40E_FLAG_HEADER_SPLIT_ENABLED: /* Not supported */
2176 rxq->rx_hdr_len = RTE_ALIGN(I40E_RXBUF_SZ_1024,
2177 (1 << I40E_RXQ_CTX_HBUFF_SHIFT));
2178 rxq->rx_buf_len = RTE_ALIGN(I40E_RXBUF_SZ_2048,
2179 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2180 rxq->hs_mode = i40e_header_split_enabled;
2182 case I40E_FLAG_HEADER_SPLIT_DISABLED:
2184 rxq->rx_hdr_len = 0;
2185 rxq->rx_buf_len = RTE_ALIGN(buf_size,
2186 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2187 rxq->hs_mode = i40e_header_split_none;
2191 len = hw->func_caps.rx_buf_chain_len * rxq->rx_buf_len;
2192 rxq->max_pkt_len = RTE_MIN(len, data->dev_conf.rxmode.max_rx_pkt_len);
2193 if (data->dev_conf.rxmode.jumbo_frame == 1) {
2194 if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
2195 rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
2196 PMD_DRV_LOG(ERR, "maximum packet length must "
2197 "be larger than %u and smaller than %u,"
2198 "as jumbo frame is enabled\n",
2199 (uint32_t)ETHER_MAX_LEN,
2200 (uint32_t)I40E_FRAME_SIZE_MAX);
2201 return I40E_ERR_CONFIG;
2204 if (rxq->max_pkt_len < ETHER_MIN_LEN ||
2205 rxq->max_pkt_len > ETHER_MAX_LEN) {
2206 PMD_DRV_LOG(ERR, "maximum packet length must be "
2207 "larger than %u and smaller than %u, "
2208 "as jumbo frame is disabled\n",
2209 (uint32_t)ETHER_MIN_LEN,
2210 (uint32_t)ETHER_MAX_LEN);
2211 return I40E_ERR_CONFIG;
2218 /* Init the RX queue in hardware */
2220 i40e_rx_queue_init(struct i40e_rx_queue *rxq)
2222 int err = I40E_SUCCESS;
2223 struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2224 struct rte_eth_dev_data *dev_data = I40E_VSI_TO_DEV_DATA(rxq->vsi);
2225 struct rte_eth_dev *dev = I40E_VSI_TO_ETH_DEV(rxq->vsi);
2226 uint16_t pf_q = rxq->reg_idx;
2228 struct i40e_hmc_obj_rxq rx_ctx;
2229 struct rte_pktmbuf_pool_private *mbp_priv;
2231 err = i40e_rx_queue_config(rxq);
2233 PMD_DRV_LOG(ERR, "Failed to config RX queue\n");
2237 /* Clear the context structure first */
2238 memset(&rx_ctx, 0, sizeof(struct i40e_hmc_obj_rxq));
2239 rx_ctx.dbuff = rxq->rx_buf_len >> I40E_RXQ_CTX_DBUFF_SHIFT;
2240 rx_ctx.hbuff = rxq->rx_hdr_len >> I40E_RXQ_CTX_HBUFF_SHIFT;
2242 rx_ctx.base = rxq->rx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2243 rx_ctx.qlen = rxq->nb_rx_desc;
2244 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2247 rx_ctx.dtype = rxq->hs_mode;
2249 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_ALL;
2251 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_NONE;
2252 rx_ctx.rxmax = rxq->max_pkt_len;
2253 rx_ctx.tphrdesc_ena = 1;
2254 rx_ctx.tphwdesc_ena = 1;
2255 rx_ctx.tphdata_ena = 1;
2256 rx_ctx.tphhead_ena = 1;
2257 rx_ctx.lrxqthresh = 2;
2258 rx_ctx.crcstrip = (rxq->crc_len == 0) ? 1 : 0;
2263 err = i40e_clear_lan_rx_queue_context(hw, pf_q);
2264 if (err != I40E_SUCCESS) {
2265 PMD_DRV_LOG(ERR, "Failed to clear LAN RX queue context\n");
2268 err = i40e_set_lan_rx_queue_context(hw, pf_q, &rx_ctx);
2269 if (err != I40E_SUCCESS) {
2270 PMD_DRV_LOG(ERR, "Failed to set LAN RX queue context\n");
2274 rxq->qrx_tail = hw->hw_addr + I40E_QRX_TAIL(pf_q);
2276 mbp_priv = rte_mempool_get_priv(rxq->mp);
2277 buf_size = (uint16_t)(mbp_priv->mbuf_data_room_size -
2278 RTE_PKTMBUF_HEADROOM);
2280 /* Check if scattered RX needs to be used. */
2281 if ((rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
2282 dev_data->scattered_rx = 1;
2283 dev->rx_pkt_burst = i40e_recv_scattered_pkts;
2286 /* Init the RX tail regieter. */
2287 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
2293 i40e_dev_clear_queues(struct rte_eth_dev *dev)
2297 PMD_INIT_FUNC_TRACE();
2299 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2300 i40e_tx_queue_release_mbufs(dev->data->tx_queues[i]);
2301 i40e_reset_tx_queue(dev->data->tx_queues[i]);
2304 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2305 i40e_rx_queue_release_mbufs(dev->data->rx_queues[i]);
2306 i40e_reset_rx_queue(dev->data->rx_queues[i]);