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 I40E_TX_CKSUM_OFFLOAD_MASK ( \
80 PKT_TX_OUTER_IP_CKSUM)
82 #define RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb) \
83 (uint64_t) ((mb)->buf_physaddr + RTE_PKTMBUF_HEADROOM)
85 #define RTE_MBUF_DATA_DMA_ADDR(mb) \
86 ((uint64_t)((mb)->buf_physaddr + (mb)->data_off))
88 static const struct rte_memzone *
89 i40e_ring_dma_zone_reserve(struct rte_eth_dev *dev,
90 const char *ring_name,
94 static uint16_t i40e_xmit_pkts_simple(void *tx_queue,
95 struct rte_mbuf **tx_pkts,
98 /* Translate the rx descriptor status to pkt flags */
99 static inline uint64_t
100 i40e_rxd_status_to_pkt_flags(uint64_t qword)
104 /* Check if VLAN packet */
105 flags = qword & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) ?
108 /* Check if RSS_HASH */
109 flags |= (((qword >> I40E_RX_DESC_STATUS_FLTSTAT_SHIFT) &
110 I40E_RX_DESC_FLTSTAT_RSS_HASH) ==
111 I40E_RX_DESC_FLTSTAT_RSS_HASH) ? PKT_RX_RSS_HASH : 0;
113 /* Check if FDIR Match */
114 flags |= (qword & (1 << I40E_RX_DESC_STATUS_FLM_SHIFT) ?
120 static inline uint64_t
121 i40e_rxd_error_to_pkt_flags(uint64_t qword)
124 uint64_t error_bits = (qword >> I40E_RXD_QW1_ERROR_SHIFT);
126 #define I40E_RX_ERR_BITS 0x3f
127 if (likely((error_bits & I40E_RX_ERR_BITS) == 0))
129 /* If RXE bit set, all other status bits are meaningless */
130 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
131 flags |= PKT_RX_MAC_ERR;
135 /* If RECIPE bit set, all other status indications should be ignored */
136 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_RECIPE_SHIFT))) {
137 flags |= PKT_RX_RECIP_ERR;
140 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT)))
141 flags |= PKT_RX_HBUF_OVERFLOW;
142 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_IPE_SHIFT)))
143 flags |= PKT_RX_IP_CKSUM_BAD;
144 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_L4E_SHIFT)))
145 flags |= PKT_RX_L4_CKSUM_BAD;
146 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT)))
147 flags |= PKT_RX_EIP_CKSUM_BAD;
148 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_OVERSIZE_SHIFT)))
149 flags |= PKT_RX_OVERSIZE;
154 /* Translate pkt types to pkt flags */
155 static inline uint64_t
156 i40e_rxd_ptype_to_pkt_flags(uint64_t qword)
158 uint8_t ptype = (uint8_t)((qword & I40E_RXD_QW1_PTYPE_MASK) >>
159 I40E_RXD_QW1_PTYPE_SHIFT);
160 static const uint64_t ip_ptype_map[I40E_MAX_PKT_TYPE] = {
183 PKT_RX_IPV4_HDR, /* PTYPE 22 */
184 PKT_RX_IPV4_HDR, /* PTYPE 23 */
185 PKT_RX_IPV4_HDR, /* PTYPE 24 */
187 PKT_RX_IPV4_HDR, /* PTYPE 26 */
188 PKT_RX_IPV4_HDR, /* PTYPE 27 */
189 PKT_RX_IPV4_HDR, /* PTYPE 28 */
190 PKT_RX_IPV4_HDR_EXT, /* PTYPE 29 */
191 PKT_RX_IPV4_HDR_EXT, /* PTYPE 30 */
192 PKT_RX_IPV4_HDR_EXT, /* PTYPE 31 */
194 PKT_RX_IPV4_HDR_EXT, /* PTYPE 33 */
195 PKT_RX_IPV4_HDR_EXT, /* PTYPE 34 */
196 PKT_RX_IPV4_HDR_EXT, /* PTYPE 35 */
197 PKT_RX_IPV4_HDR_EXT, /* PTYPE 36 */
198 PKT_RX_IPV4_HDR_EXT, /* PTYPE 37 */
199 PKT_RX_IPV4_HDR_EXT, /* PTYPE 38 */
201 PKT_RX_IPV4_HDR_EXT, /* PTYPE 40 */
202 PKT_RX_IPV4_HDR_EXT, /* PTYPE 41 */
203 PKT_RX_IPV4_HDR_EXT, /* PTYPE 42 */
204 PKT_RX_IPV4_HDR_EXT, /* PTYPE 43 */
205 PKT_RX_IPV4_HDR_EXT, /* PTYPE 44 */
206 PKT_RX_IPV4_HDR_EXT, /* PTYPE 45 */
207 PKT_RX_IPV4_HDR_EXT, /* PTYPE 46 */
209 PKT_RX_IPV4_HDR_EXT, /* PTYPE 48 */
210 PKT_RX_IPV4_HDR_EXT, /* PTYPE 49 */
211 PKT_RX_IPV4_HDR_EXT, /* PTYPE 50 */
212 PKT_RX_IPV4_HDR_EXT, /* PTYPE 51 */
213 PKT_RX_IPV4_HDR_EXT, /* PTYPE 52 */
214 PKT_RX_IPV4_HDR_EXT, /* PTYPE 53 */
216 PKT_RX_IPV4_HDR_EXT, /* PTYPE 55 */
217 PKT_RX_IPV4_HDR_EXT, /* PTYPE 56 */
218 PKT_RX_IPV4_HDR_EXT, /* PTYPE 57 */
219 PKT_RX_IPV4_HDR_EXT, /* PTYPE 58 */
220 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 59 */
221 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 60 */
222 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 61 */
224 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 63 */
225 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 64 */
226 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 65 */
227 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 66 */
228 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 67 */
229 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 68 */
231 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 70 */
232 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 71 */
233 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 72 */
234 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 73 */
235 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 74 */
236 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 75 */
237 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 76 */
239 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 78 */
240 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 79 */
241 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 80 */
242 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 81 */
243 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 82 */
244 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 83 */
246 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 85 */
247 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 86 */
248 PKT_RX_IPV4_HDR_EXT, /* PTYPE 87 */
249 PKT_RX_IPV6_HDR, /* PTYPE 88 */
250 PKT_RX_IPV6_HDR, /* PTYPE 89 */
251 PKT_RX_IPV6_HDR, /* PTYPE 90 */
253 PKT_RX_IPV6_HDR, /* PTYPE 92 */
254 PKT_RX_IPV6_HDR, /* PTYPE 93 */
255 PKT_RX_IPV6_HDR, /* PTYPE 94 */
256 PKT_RX_IPV6_HDR_EXT, /* PTYPE 95 */
257 PKT_RX_IPV6_HDR_EXT, /* PTYPE 96 */
258 PKT_RX_IPV6_HDR_EXT, /* PTYPE 97 */
260 PKT_RX_IPV6_HDR_EXT, /* PTYPE 99 */
261 PKT_RX_IPV6_HDR_EXT, /* PTYPE 100 */
262 PKT_RX_IPV6_HDR_EXT, /* PTYPE 101 */
263 PKT_RX_IPV6_HDR_EXT, /* PTYPE 102 */
264 PKT_RX_IPV6_HDR_EXT, /* PTYPE 103 */
265 PKT_RX_IPV6_HDR_EXT, /* PTYPE 104 */
267 PKT_RX_IPV6_HDR_EXT, /* PTYPE 106 */
268 PKT_RX_IPV6_HDR_EXT, /* PTYPE 107 */
269 PKT_RX_IPV6_HDR_EXT, /* PTYPE 108 */
270 PKT_RX_IPV6_HDR_EXT, /* PTYPE 109 */
271 PKT_RX_IPV6_HDR_EXT, /* PTYPE 110 */
272 PKT_RX_IPV6_HDR_EXT, /* PTYPE 111 */
273 PKT_RX_IPV6_HDR_EXT, /* PTYPE 112 */
275 PKT_RX_IPV6_HDR_EXT, /* PTYPE 114 */
276 PKT_RX_IPV6_HDR_EXT, /* PTYPE 115 */
277 PKT_RX_IPV6_HDR_EXT, /* PTYPE 116 */
278 PKT_RX_IPV6_HDR_EXT, /* PTYPE 117 */
279 PKT_RX_IPV6_HDR_EXT, /* PTYPE 118 */
280 PKT_RX_IPV6_HDR_EXT, /* PTYPE 119 */
282 PKT_RX_IPV6_HDR_EXT, /* PTYPE 121 */
283 PKT_RX_IPV6_HDR_EXT, /* PTYPE 122 */
284 PKT_RX_IPV6_HDR_EXT, /* PTYPE 123 */
285 PKT_RX_IPV6_HDR_EXT, /* PTYPE 124 */
286 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 125 */
287 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 126 */
288 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 127 */
290 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 129 */
291 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 130 */
292 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 131 */
293 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 132 */
294 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 133 */
295 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 134 */
297 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 136 */
298 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 137 */
299 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 138 */
300 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 139 */
301 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 140 */
302 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 141 */
303 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 142 */
305 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 144 */
306 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 145 */
307 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 146 */
308 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 147 */
309 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 148 */
310 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 149 */
312 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 151 */
313 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 152 */
314 PKT_RX_IPV6_HDR_EXT, /* PTYPE 153 */
419 return ip_ptype_map[ptype];
422 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_MASK 0x03
423 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_FD_ID 0x01
424 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_FLEX 0x02
425 #define I40E_RX_DESC_EXT_STATUS_FLEXBL_MASK 0x03
426 #define I40E_RX_DESC_EXT_STATUS_FLEXBL_FLEX 0x01
428 static inline uint64_t
429 i40e_rxd_build_fdir(volatile union i40e_rx_desc *rxdp, struct rte_mbuf *mb)
432 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
433 uint16_t flexbh, flexbl;
435 flexbh = (rte_le_to_cpu_32(rxdp->wb.qword2.ext_status) >>
436 I40E_RX_DESC_EXT_STATUS_FLEXBH_SHIFT) &
437 I40E_RX_DESC_EXT_STATUS_FLEXBH_MASK;
438 flexbl = (rte_le_to_cpu_32(rxdp->wb.qword2.ext_status) >>
439 I40E_RX_DESC_EXT_STATUS_FLEXBL_SHIFT) &
440 I40E_RX_DESC_EXT_STATUS_FLEXBL_MASK;
443 if (flexbh == I40E_RX_DESC_EXT_STATUS_FLEXBH_FD_ID) {
445 rte_le_to_cpu_32(rxdp->wb.qword3.hi_dword.fd_id);
446 flags |= PKT_RX_FDIR_ID;
447 } else if (flexbh == I40E_RX_DESC_EXT_STATUS_FLEXBH_FLEX) {
449 rte_le_to_cpu_32(rxdp->wb.qword3.hi_dword.flex_bytes_hi);
450 flags |= PKT_RX_FDIR_FLX;
452 if (flexbl == I40E_RX_DESC_EXT_STATUS_FLEXBL_FLEX) {
454 rte_le_to_cpu_32(rxdp->wb.qword3.lo_dword.flex_bytes_lo);
455 flags |= PKT_RX_FDIR_FLX;
459 rte_le_to_cpu_32(rxdp->wb.qword0.hi_dword.fd_id);
460 flags |= PKT_RX_FDIR_ID;
465 i40e_txd_enable_checksum(uint64_t ol_flags,
468 union i40e_tx_offload tx_offload,
469 uint32_t *cd_tunneling)
471 /* UDP tunneling packet TX checksum offload */
472 if (ol_flags & PKT_TX_OUTER_IP_CKSUM) {
474 *td_offset |= (tx_offload.outer_l2_len >> 1)
475 << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
477 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
478 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
479 else if (ol_flags & PKT_TX_OUTER_IPV4)
480 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
481 else if (ol_flags & PKT_TX_OUTER_IPV6)
482 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
484 /* Now set the ctx descriptor fields */
485 *cd_tunneling |= (tx_offload.outer_l3_len >> 2) <<
486 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
487 (tx_offload.l2_len >> 1) <<
488 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
491 *td_offset |= (tx_offload.l2_len >> 1)
492 << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
494 /* Enable L3 checksum offloads */
495 if (ol_flags & PKT_TX_IP_CKSUM) {
496 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
497 *td_offset |= (tx_offload.l3_len >> 2)
498 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
499 } else if (ol_flags & PKT_TX_IPV4) {
500 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
501 *td_offset |= (tx_offload.l3_len >> 2)
502 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
503 } else if (ol_flags & PKT_TX_IPV6) {
504 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
505 *td_offset |= (tx_offload.l3_len >> 2)
506 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
509 /* Enable L4 checksum offloads */
510 switch (ol_flags & PKT_TX_L4_MASK) {
511 case PKT_TX_TCP_CKSUM:
512 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
513 *td_offset |= (sizeof(struct tcp_hdr) >> 2) <<
514 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
516 case PKT_TX_SCTP_CKSUM:
517 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
518 *td_offset |= (sizeof(struct sctp_hdr) >> 2) <<
519 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
521 case PKT_TX_UDP_CKSUM:
522 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
523 *td_offset |= (sizeof(struct udp_hdr) >> 2) <<
524 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
531 static inline struct rte_mbuf *
532 rte_rxmbuf_alloc(struct rte_mempool *mp)
536 m = __rte_mbuf_raw_alloc(mp);
537 __rte_mbuf_sanity_check_raw(m, 0);
542 /* Construct the tx flags */
543 static inline uint64_t
544 i40e_build_ctob(uint32_t td_cmd,
549 return rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DATA |
550 ((uint64_t)td_cmd << I40E_TXD_QW1_CMD_SHIFT) |
551 ((uint64_t)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
552 ((uint64_t)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
553 ((uint64_t)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT));
557 i40e_xmit_cleanup(struct i40e_tx_queue *txq)
559 struct i40e_tx_entry *sw_ring = txq->sw_ring;
560 volatile struct i40e_tx_desc *txd = txq->tx_ring;
561 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
562 uint16_t nb_tx_desc = txq->nb_tx_desc;
563 uint16_t desc_to_clean_to;
564 uint16_t nb_tx_to_clean;
566 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
567 if (desc_to_clean_to >= nb_tx_desc)
568 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
570 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
571 if (!(txd[desc_to_clean_to].cmd_type_offset_bsz &
572 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE))) {
573 PMD_TX_FREE_LOG(DEBUG, "TX descriptor %4u is not done "
574 "(port=%d queue=%d)", desc_to_clean_to,
575 txq->port_id, txq->queue_id);
579 if (last_desc_cleaned > desc_to_clean_to)
580 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
583 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
586 txd[desc_to_clean_to].cmd_type_offset_bsz = 0;
588 txq->last_desc_cleaned = desc_to_clean_to;
589 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
595 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
596 check_rx_burst_bulk_alloc_preconditions(struct i40e_rx_queue *rxq)
598 check_rx_burst_bulk_alloc_preconditions(__rte_unused struct i40e_rx_queue *rxq)
603 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
604 if (!(rxq->rx_free_thresh >= RTE_PMD_I40E_RX_MAX_BURST)) {
605 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
606 "rxq->rx_free_thresh=%d, "
607 "RTE_PMD_I40E_RX_MAX_BURST=%d",
608 rxq->rx_free_thresh, RTE_PMD_I40E_RX_MAX_BURST);
610 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
611 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
612 "rxq->rx_free_thresh=%d, "
613 "rxq->nb_rx_desc=%d",
614 rxq->rx_free_thresh, rxq->nb_rx_desc);
616 } else if (rxq->nb_rx_desc % rxq->rx_free_thresh != 0) {
617 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
618 "rxq->nb_rx_desc=%d, "
619 "rxq->rx_free_thresh=%d",
620 rxq->nb_rx_desc, rxq->rx_free_thresh);
622 } else if (!(rxq->nb_rx_desc < (I40E_MAX_RING_DESC -
623 RTE_PMD_I40E_RX_MAX_BURST))) {
624 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
625 "rxq->nb_rx_desc=%d, "
626 "I40E_MAX_RING_DESC=%d, "
627 "RTE_PMD_I40E_RX_MAX_BURST=%d",
628 rxq->nb_rx_desc, I40E_MAX_RING_DESC,
629 RTE_PMD_I40E_RX_MAX_BURST);
639 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
640 #define I40E_LOOK_AHEAD 8
641 #if (I40E_LOOK_AHEAD != 8)
642 #error "PMD I40E: I40E_LOOK_AHEAD must be 8\n"
645 i40e_rx_scan_hw_ring(struct i40e_rx_queue *rxq)
647 volatile union i40e_rx_desc *rxdp;
648 struct i40e_rx_entry *rxep;
653 int32_t s[I40E_LOOK_AHEAD], nb_dd;
654 int32_t i, j, nb_rx = 0;
657 rxdp = &rxq->rx_ring[rxq->rx_tail];
658 rxep = &rxq->sw_ring[rxq->rx_tail];
660 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
661 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
662 I40E_RXD_QW1_STATUS_SHIFT;
664 /* Make sure there is at least 1 packet to receive */
665 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
669 * Scan LOOK_AHEAD descriptors at a time to determine which
670 * descriptors reference packets that are ready to be received.
672 for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; i+=I40E_LOOK_AHEAD,
673 rxdp += I40E_LOOK_AHEAD, rxep += I40E_LOOK_AHEAD) {
674 /* Read desc statuses backwards to avoid race condition */
675 for (j = I40E_LOOK_AHEAD - 1; j >= 0; j--) {
676 qword1 = rte_le_to_cpu_64(\
677 rxdp[j].wb.qword1.status_error_len);
678 s[j] = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
679 I40E_RXD_QW1_STATUS_SHIFT;
682 /* Compute how many status bits were set */
683 for (j = 0, nb_dd = 0; j < I40E_LOOK_AHEAD; j++)
684 nb_dd += s[j] & (1 << I40E_RX_DESC_STATUS_DD_SHIFT);
688 /* Translate descriptor info to mbuf parameters */
689 for (j = 0; j < nb_dd; j++) {
691 qword1 = rte_le_to_cpu_64(\
692 rxdp[j].wb.qword1.status_error_len);
693 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
694 I40E_RXD_QW1_STATUS_SHIFT;
695 pkt_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
696 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
697 mb->data_len = pkt_len;
698 mb->pkt_len = pkt_len;
699 mb->vlan_tci = rx_status &
700 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) ?
702 rxdp[j].wb.qword0.lo_dword.l2tag1) : 0;
703 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
704 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
705 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
707 mb->packet_type = (uint16_t)((qword1 &
708 I40E_RXD_QW1_PTYPE_MASK) >>
709 I40E_RXD_QW1_PTYPE_SHIFT);
710 if (pkt_flags & PKT_RX_RSS_HASH)
711 mb->hash.rss = rte_le_to_cpu_32(\
712 rxdp[j].wb.qword0.hi_dword.rss);
713 if (pkt_flags & PKT_RX_FDIR)
714 pkt_flags |= i40e_rxd_build_fdir(&rxdp[j], mb);
716 mb->ol_flags = pkt_flags;
719 for (j = 0; j < I40E_LOOK_AHEAD; j++)
720 rxq->rx_stage[i + j] = rxep[j].mbuf;
722 if (nb_dd != I40E_LOOK_AHEAD)
726 /* Clear software ring entries */
727 for (i = 0; i < nb_rx; i++)
728 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
733 static inline uint16_t
734 i40e_rx_fill_from_stage(struct i40e_rx_queue *rxq,
735 struct rte_mbuf **rx_pkts,
739 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
741 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
743 for (i = 0; i < nb_pkts; i++)
744 rx_pkts[i] = stage[i];
746 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
747 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
753 i40e_rx_alloc_bufs(struct i40e_rx_queue *rxq)
755 volatile union i40e_rx_desc *rxdp;
756 struct i40e_rx_entry *rxep;
758 uint16_t alloc_idx, i;
762 /* Allocate buffers in bulk */
763 alloc_idx = (uint16_t)(rxq->rx_free_trigger -
764 (rxq->rx_free_thresh - 1));
765 rxep = &(rxq->sw_ring[alloc_idx]);
766 diag = rte_mempool_get_bulk(rxq->mp, (void *)rxep,
767 rxq->rx_free_thresh);
768 if (unlikely(diag != 0)) {
769 PMD_DRV_LOG(ERR, "Failed to get mbufs in bulk");
773 rxdp = &rxq->rx_ring[alloc_idx];
774 for (i = 0; i < rxq->rx_free_thresh; i++) {
776 rte_mbuf_refcnt_set(mb, 1);
778 mb->data_off = RTE_PKTMBUF_HEADROOM;
780 mb->port = rxq->port_id;
781 dma_addr = rte_cpu_to_le_64(\
782 RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb));
783 rxdp[i].read.hdr_addr = dma_addr;
784 rxdp[i].read.pkt_addr = dma_addr;
787 /* Update rx tail regsiter */
789 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->rx_free_trigger);
791 rxq->rx_free_trigger =
792 (uint16_t)(rxq->rx_free_trigger + rxq->rx_free_thresh);
793 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
794 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
799 static inline uint16_t
800 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
802 struct i40e_rx_queue *rxq = (struct i40e_rx_queue *)rx_queue;
808 if (rxq->rx_nb_avail)
809 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
811 nb_rx = (uint16_t)i40e_rx_scan_hw_ring(rxq);
812 rxq->rx_next_avail = 0;
813 rxq->rx_nb_avail = nb_rx;
814 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
816 if (rxq->rx_tail > rxq->rx_free_trigger) {
817 if (i40e_rx_alloc_bufs(rxq) != 0) {
820 PMD_RX_LOG(DEBUG, "Rx mbuf alloc failed for "
821 "port_id=%u, queue_id=%u",
822 rxq->port_id, rxq->queue_id);
823 rxq->rx_nb_avail = 0;
824 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
825 for (i = 0, j = rxq->rx_tail; i < nb_rx; i++, j++)
826 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
832 if (rxq->rx_tail >= rxq->nb_rx_desc)
835 if (rxq->rx_nb_avail)
836 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
842 i40e_recv_pkts_bulk_alloc(void *rx_queue,
843 struct rte_mbuf **rx_pkts,
846 uint16_t nb_rx = 0, n, count;
848 if (unlikely(nb_pkts == 0))
851 if (likely(nb_pkts <= RTE_PMD_I40E_RX_MAX_BURST))
852 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
855 n = RTE_MIN(nb_pkts, RTE_PMD_I40E_RX_MAX_BURST);
856 count = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
857 nb_rx = (uint16_t)(nb_rx + count);
858 nb_pkts = (uint16_t)(nb_pkts - count);
865 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
868 i40e_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
870 struct i40e_rx_queue *rxq;
871 volatile union i40e_rx_desc *rx_ring;
872 volatile union i40e_rx_desc *rxdp;
873 union i40e_rx_desc rxd;
874 struct i40e_rx_entry *sw_ring;
875 struct i40e_rx_entry *rxe;
876 struct rte_mbuf *rxm;
877 struct rte_mbuf *nmb;
881 uint16_t rx_packet_len;
882 uint16_t rx_id, nb_hold;
889 rx_id = rxq->rx_tail;
890 rx_ring = rxq->rx_ring;
891 sw_ring = rxq->sw_ring;
893 while (nb_rx < nb_pkts) {
894 rxdp = &rx_ring[rx_id];
895 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
896 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK)
897 >> I40E_RXD_QW1_STATUS_SHIFT;
898 /* Check the DD bit first */
899 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
902 nmb = rte_rxmbuf_alloc(rxq->mp);
908 rxe = &sw_ring[rx_id];
910 if (unlikely(rx_id == rxq->nb_rx_desc))
913 /* Prefetch next mbuf */
914 rte_prefetch0(sw_ring[rx_id].mbuf);
917 * When next RX descriptor is on a cache line boundary,
918 * prefetch the next 4 RX descriptors and next 8 pointers
921 if ((rx_id & 0x3) == 0) {
922 rte_prefetch0(&rx_ring[rx_id]);
923 rte_prefetch0(&sw_ring[rx_id]);
928 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
929 rxdp->read.hdr_addr = dma_addr;
930 rxdp->read.pkt_addr = dma_addr;
932 rx_packet_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
933 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
935 rxm->data_off = RTE_PKTMBUF_HEADROOM;
936 rte_prefetch0(RTE_PTR_ADD(rxm->buf_addr, RTE_PKTMBUF_HEADROOM));
939 rxm->pkt_len = rx_packet_len;
940 rxm->data_len = rx_packet_len;
941 rxm->port = rxq->port_id;
943 rxm->vlan_tci = rx_status &
944 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) ?
945 rte_le_to_cpu_16(rxd.wb.qword0.lo_dword.l2tag1) : 0;
946 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
947 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
948 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
949 rxm->packet_type = (uint16_t)((qword1 & I40E_RXD_QW1_PTYPE_MASK) >>
950 I40E_RXD_QW1_PTYPE_SHIFT);
951 if (pkt_flags & PKT_RX_RSS_HASH)
953 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
954 if (pkt_flags & PKT_RX_FDIR)
955 pkt_flags |= i40e_rxd_build_fdir(&rxd, rxm);
957 rxm->ol_flags = pkt_flags;
959 rx_pkts[nb_rx++] = rxm;
961 rxq->rx_tail = rx_id;
964 * If the number of free RX descriptors is greater than the RX free
965 * threshold of the queue, advance the receive tail register of queue.
966 * Update that register with the value of the last processed RX
967 * descriptor minus 1.
969 nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
970 if (nb_hold > rxq->rx_free_thresh) {
971 rx_id = (uint16_t) ((rx_id == 0) ?
972 (rxq->nb_rx_desc - 1) : (rx_id - 1));
973 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
976 rxq->nb_rx_hold = nb_hold;
982 i40e_recv_scattered_pkts(void *rx_queue,
983 struct rte_mbuf **rx_pkts,
986 struct i40e_rx_queue *rxq = rx_queue;
987 volatile union i40e_rx_desc *rx_ring = rxq->rx_ring;
988 volatile union i40e_rx_desc *rxdp;
989 union i40e_rx_desc rxd;
990 struct i40e_rx_entry *sw_ring = rxq->sw_ring;
991 struct i40e_rx_entry *rxe;
992 struct rte_mbuf *first_seg = rxq->pkt_first_seg;
993 struct rte_mbuf *last_seg = rxq->pkt_last_seg;
994 struct rte_mbuf *nmb, *rxm;
995 uint16_t rx_id = rxq->rx_tail;
996 uint16_t nb_rx = 0, nb_hold = 0, rx_packet_len;
1002 while (nb_rx < nb_pkts) {
1003 rxdp = &rx_ring[rx_id];
1004 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
1005 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
1006 I40E_RXD_QW1_STATUS_SHIFT;
1007 /* Check the DD bit */
1008 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
1011 nmb = rte_rxmbuf_alloc(rxq->mp);
1016 rxe = &sw_ring[rx_id];
1018 if (rx_id == rxq->nb_rx_desc)
1021 /* Prefetch next mbuf */
1022 rte_prefetch0(sw_ring[rx_id].mbuf);
1025 * When next RX descriptor is on a cache line boundary,
1026 * prefetch the next 4 RX descriptors and next 8 pointers
1029 if ((rx_id & 0x3) == 0) {
1030 rte_prefetch0(&rx_ring[rx_id]);
1031 rte_prefetch0(&sw_ring[rx_id]);
1037 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1039 /* Set data buffer address and data length of the mbuf */
1040 rxdp->read.hdr_addr = dma_addr;
1041 rxdp->read.pkt_addr = dma_addr;
1042 rx_packet_len = (qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1043 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1044 rxm->data_len = rx_packet_len;
1045 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1048 * If this is the first buffer of the received packet, set the
1049 * pointer to the first mbuf of the packet and initialize its
1050 * context. Otherwise, update the total length and the number
1051 * of segments of the current scattered packet, and update the
1052 * pointer to the last mbuf of the current packet.
1056 first_seg->nb_segs = 1;
1057 first_seg->pkt_len = rx_packet_len;
1059 first_seg->pkt_len =
1060 (uint16_t)(first_seg->pkt_len +
1062 first_seg->nb_segs++;
1063 last_seg->next = rxm;
1067 * If this is not the last buffer of the received packet,
1068 * update the pointer to the last mbuf of the current scattered
1069 * packet and continue to parse the RX ring.
1071 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT))) {
1077 * This is the last buffer of the received packet. If the CRC
1078 * is not stripped by the hardware:
1079 * - Subtract the CRC length from the total packet length.
1080 * - If the last buffer only contains the whole CRC or a part
1081 * of it, free the mbuf associated to the last buffer. If part
1082 * of the CRC is also contained in the previous mbuf, subtract
1083 * the length of that CRC part from the data length of the
1087 if (unlikely(rxq->crc_len > 0)) {
1088 first_seg->pkt_len -= ETHER_CRC_LEN;
1089 if (rx_packet_len <= ETHER_CRC_LEN) {
1090 rte_pktmbuf_free_seg(rxm);
1091 first_seg->nb_segs--;
1092 last_seg->data_len =
1093 (uint16_t)(last_seg->data_len -
1094 (ETHER_CRC_LEN - rx_packet_len));
1095 last_seg->next = NULL;
1097 rxm->data_len = (uint16_t)(rx_packet_len -
1101 first_seg->port = rxq->port_id;
1102 first_seg->vlan_tci = (rx_status &
1103 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
1104 rte_le_to_cpu_16(rxd.wb.qword0.lo_dword.l2tag1) : 0;
1105 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
1106 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
1107 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
1108 first_seg->packet_type = (uint16_t)((qword1 &
1109 I40E_RXD_QW1_PTYPE_MASK) >>
1110 I40E_RXD_QW1_PTYPE_SHIFT);
1111 if (pkt_flags & PKT_RX_RSS_HASH)
1113 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
1114 if (pkt_flags & PKT_RX_FDIR)
1115 pkt_flags |= i40e_rxd_build_fdir(&rxd, rxm);
1117 first_seg->ol_flags = pkt_flags;
1119 /* Prefetch data of first segment, if configured to do so. */
1120 rte_prefetch0(RTE_PTR_ADD(first_seg->buf_addr,
1121 first_seg->data_off));
1122 rx_pkts[nb_rx++] = first_seg;
1126 /* Record index of the next RX descriptor to probe. */
1127 rxq->rx_tail = rx_id;
1128 rxq->pkt_first_seg = first_seg;
1129 rxq->pkt_last_seg = last_seg;
1132 * If the number of free RX descriptors is greater than the RX free
1133 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1134 * register. Update the RDT with the value of the last processed RX
1135 * descriptor minus 1, to guarantee that the RDT register is never
1136 * equal to the RDH register, which creates a "full" ring situtation
1137 * from the hardware point of view.
1139 nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
1140 if (nb_hold > rxq->rx_free_thresh) {
1141 rx_id = (uint16_t)(rx_id == 0 ?
1142 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1143 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
1146 rxq->nb_rx_hold = nb_hold;
1151 /* Check if the context descriptor is needed for TX offloading */
1152 static inline uint16_t
1153 i40e_calc_context_desc(uint64_t flags)
1155 uint64_t mask = 0ULL;
1157 mask |= PKT_TX_OUTER_IP_CKSUM;
1159 #ifdef RTE_LIBRTE_IEEE1588
1160 mask |= PKT_TX_IEEE1588_TMST;
1169 i40e_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1171 struct i40e_tx_queue *txq;
1172 struct i40e_tx_entry *sw_ring;
1173 struct i40e_tx_entry *txe, *txn;
1174 volatile struct i40e_tx_desc *txd;
1175 volatile struct i40e_tx_desc *txr;
1176 struct rte_mbuf *tx_pkt;
1177 struct rte_mbuf *m_seg;
1178 uint32_t cd_tunneling_params;
1190 uint64_t buf_dma_addr;
1191 union i40e_tx_offload tx_offload = { .data = 0 };
1194 sw_ring = txq->sw_ring;
1196 tx_id = txq->tx_tail;
1197 txe = &sw_ring[tx_id];
1199 /* Check if the descriptor ring needs to be cleaned. */
1200 if ((txq->nb_tx_desc - txq->nb_tx_free) > txq->tx_free_thresh)
1201 i40e_xmit_cleanup(txq);
1203 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
1209 tx_pkt = *tx_pkts++;
1210 RTE_MBUF_PREFETCH_TO_FREE(txe->mbuf);
1212 ol_flags = tx_pkt->ol_flags;
1213 tx_offload.l2_len = tx_pkt->l2_len;
1214 tx_offload.l3_len = tx_pkt->l3_len;
1215 tx_offload.outer_l2_len = tx_pkt->outer_l2_len;
1216 tx_offload.outer_l3_len = tx_pkt->outer_l3_len;
1218 /* Calculate the number of context descriptors needed. */
1219 nb_ctx = i40e_calc_context_desc(ol_flags);
1222 * The number of descriptors that must be allocated for
1223 * a packet equals to the number of the segments of that
1224 * packet plus 1 context descriptor if needed.
1226 nb_used = (uint16_t)(tx_pkt->nb_segs + nb_ctx);
1227 tx_last = (uint16_t)(tx_id + nb_used - 1);
1230 if (tx_last >= txq->nb_tx_desc)
1231 tx_last = (uint16_t)(tx_last - txq->nb_tx_desc);
1233 if (nb_used > txq->nb_tx_free) {
1234 if (i40e_xmit_cleanup(txq) != 0) {
1239 if (unlikely(nb_used > txq->tx_rs_thresh)) {
1240 while (nb_used > txq->nb_tx_free) {
1241 if (i40e_xmit_cleanup(txq) != 0) {
1250 /* Descriptor based VLAN insertion */
1251 if (ol_flags & PKT_TX_VLAN_PKT) {
1252 tx_flags |= tx_pkt->vlan_tci <<
1253 I40E_TX_FLAG_L2TAG1_SHIFT;
1254 tx_flags |= I40E_TX_FLAG_INSERT_VLAN;
1255 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1256 td_tag = (tx_flags & I40E_TX_FLAG_L2TAG1_MASK) >>
1257 I40E_TX_FLAG_L2TAG1_SHIFT;
1260 /* Always enable CRC offload insertion */
1261 td_cmd |= I40E_TX_DESC_CMD_ICRC;
1263 /* Enable checksum offloading */
1264 cd_tunneling_params = 0;
1265 if (unlikely(ol_flags & I40E_TX_CKSUM_OFFLOAD_MASK)) {
1266 i40e_txd_enable_checksum(ol_flags, &td_cmd, &td_offset,
1267 tx_offload, &cd_tunneling_params);
1270 if (unlikely(nb_ctx)) {
1271 /* Setup TX context descriptor if required */
1272 volatile struct i40e_tx_context_desc *ctx_txd =
1273 (volatile struct i40e_tx_context_desc *)\
1275 uint16_t cd_l2tag2 = 0;
1276 uint64_t cd_type_cmd_tso_mss =
1277 I40E_TX_DESC_DTYPE_CONTEXT;
1279 txn = &sw_ring[txe->next_id];
1280 RTE_MBUF_PREFETCH_TO_FREE(txn->mbuf);
1281 if (txe->mbuf != NULL) {
1282 rte_pktmbuf_free_seg(txe->mbuf);
1285 #ifdef RTE_LIBRTE_IEEE1588
1286 if (ol_flags & PKT_TX_IEEE1588_TMST)
1287 cd_type_cmd_tso_mss |=
1288 ((uint64_t)I40E_TX_CTX_DESC_TSYN <<
1289 I40E_TXD_CTX_QW1_CMD_SHIFT);
1291 ctx_txd->tunneling_params =
1292 rte_cpu_to_le_32(cd_tunneling_params);
1293 ctx_txd->l2tag2 = rte_cpu_to_le_16(cd_l2tag2);
1294 ctx_txd->type_cmd_tso_mss =
1295 rte_cpu_to_le_64(cd_type_cmd_tso_mss);
1297 PMD_TX_LOG(DEBUG, "mbuf: %p, TCD[%u]:\n"
1298 "tunneling_params: %#x;\n"
1301 "type_cmd_tso_mss: %#lx;\n",
1303 ctx_txd->tunneling_params,
1306 ctx_txd->type_cmd_tso_mss);
1308 txe->last_id = tx_last;
1309 tx_id = txe->next_id;
1316 txn = &sw_ring[txe->next_id];
1319 rte_pktmbuf_free_seg(txe->mbuf);
1322 /* Setup TX Descriptor */
1323 slen = m_seg->data_len;
1324 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
1326 PMD_TX_LOG(DEBUG, "mbuf: %p, TDD[%u]:\n"
1327 "buf_dma_addr: %#"PRIx64";\n"
1332 tx_pkt, tx_id, buf_dma_addr,
1333 td_cmd, td_offset, slen, td_tag);
1335 txd->buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
1336 txd->cmd_type_offset_bsz = i40e_build_ctob(td_cmd,
1337 td_offset, slen, td_tag);
1338 txe->last_id = tx_last;
1339 tx_id = txe->next_id;
1341 m_seg = m_seg->next;
1342 } while (m_seg != NULL);
1344 /* The last packet data descriptor needs End Of Packet (EOP) */
1345 td_cmd |= I40E_TX_DESC_CMD_EOP;
1346 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
1347 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
1349 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
1350 PMD_TX_FREE_LOG(DEBUG,
1351 "Setting RS bit on TXD id="
1352 "%4u (port=%d queue=%d)",
1353 tx_last, txq->port_id, txq->queue_id);
1355 td_cmd |= I40E_TX_DESC_CMD_RS;
1357 /* Update txq RS bit counters */
1358 txq->nb_tx_used = 0;
1361 txd->cmd_type_offset_bsz |=
1362 rte_cpu_to_le_64(((uint64_t)td_cmd) <<
1363 I40E_TXD_QW1_CMD_SHIFT);
1369 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
1370 (unsigned) txq->port_id, (unsigned) txq->queue_id,
1371 (unsigned) tx_id, (unsigned) nb_tx);
1373 I40E_PCI_REG_WRITE(txq->qtx_tail, tx_id);
1374 txq->tx_tail = tx_id;
1379 static inline int __attribute__((always_inline))
1380 i40e_tx_free_bufs(struct i40e_tx_queue *txq)
1382 struct i40e_tx_entry *txep;
1385 if (!(txq->tx_ring[txq->tx_next_dd].cmd_type_offset_bsz &
1386 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE)))
1389 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
1391 for (i = 0; i < txq->tx_rs_thresh; i++)
1392 rte_prefetch0((txep + i)->mbuf);
1394 if (!(txq->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT)) {
1395 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1396 rte_mempool_put(txep->mbuf->pool, txep->mbuf);
1400 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1401 rte_pktmbuf_free_seg(txep->mbuf);
1406 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
1407 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
1408 if (txq->tx_next_dd >= txq->nb_tx_desc)
1409 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
1411 return txq->tx_rs_thresh;
1414 #define I40E_TD_CMD (I40E_TX_DESC_CMD_ICRC |\
1415 I40E_TX_DESC_CMD_EOP)
1417 /* Populate 4 descriptors with data from 4 mbufs */
1419 tx4(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1424 for (i = 0; i < 4; i++, txdp++, pkts++) {
1425 dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
1426 txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1427 txdp->cmd_type_offset_bsz =
1428 i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1429 (*pkts)->data_len, 0);
1433 /* Populate 1 descriptor with data from 1 mbuf */
1435 tx1(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1439 dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
1440 txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1441 txdp->cmd_type_offset_bsz =
1442 i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1443 (*pkts)->data_len, 0);
1446 /* Fill hardware descriptor ring with mbuf data */
1448 i40e_tx_fill_hw_ring(struct i40e_tx_queue *txq,
1449 struct rte_mbuf **pkts,
1452 volatile struct i40e_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
1453 struct i40e_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
1454 const int N_PER_LOOP = 4;
1455 const int N_PER_LOOP_MASK = N_PER_LOOP - 1;
1456 int mainpart, leftover;
1459 mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
1460 leftover = (nb_pkts & ((uint32_t) N_PER_LOOP_MASK));
1461 for (i = 0; i < mainpart; i += N_PER_LOOP) {
1462 for (j = 0; j < N_PER_LOOP; ++j) {
1463 (txep + i + j)->mbuf = *(pkts + i + j);
1465 tx4(txdp + i, pkts + i);
1467 if (unlikely(leftover > 0)) {
1468 for (i = 0; i < leftover; ++i) {
1469 (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
1470 tx1(txdp + mainpart + i, pkts + mainpart + i);
1475 static inline uint16_t
1476 tx_xmit_pkts(struct i40e_tx_queue *txq,
1477 struct rte_mbuf **tx_pkts,
1480 volatile struct i40e_tx_desc *txr = txq->tx_ring;
1484 * Begin scanning the H/W ring for done descriptors when the number
1485 * of available descriptors drops below tx_free_thresh. For each done
1486 * descriptor, free the associated buffer.
1488 if (txq->nb_tx_free < txq->tx_free_thresh)
1489 i40e_tx_free_bufs(txq);
1491 /* Use available descriptor only */
1492 nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
1493 if (unlikely(!nb_pkts))
1496 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
1497 if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
1498 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
1499 i40e_tx_fill_hw_ring(txq, tx_pkts, n);
1500 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1501 rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1502 I40E_TXD_QW1_CMD_SHIFT);
1503 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1507 /* Fill hardware descriptor ring with mbuf data */
1508 i40e_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
1509 txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
1511 /* Determin if RS bit needs to be set */
1512 if (txq->tx_tail > txq->tx_next_rs) {
1513 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1514 rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1515 I40E_TXD_QW1_CMD_SHIFT);
1517 (uint16_t)(txq->tx_next_rs + txq->tx_rs_thresh);
1518 if (txq->tx_next_rs >= txq->nb_tx_desc)
1519 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1522 if (txq->tx_tail >= txq->nb_tx_desc)
1525 /* Update the tx tail register */
1527 I40E_PCI_REG_WRITE(txq->qtx_tail, txq->tx_tail);
1533 i40e_xmit_pkts_simple(void *tx_queue,
1534 struct rte_mbuf **tx_pkts,
1539 if (likely(nb_pkts <= I40E_TX_MAX_BURST))
1540 return tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1544 uint16_t ret, num = (uint16_t)RTE_MIN(nb_pkts,
1547 ret = tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1548 &tx_pkts[nb_tx], num);
1549 nb_tx = (uint16_t)(nb_tx + ret);
1550 nb_pkts = (uint16_t)(nb_pkts - ret);
1559 * Find the VSI the queue belongs to. 'queue_idx' is the queue index
1560 * application used, which assume having sequential ones. But from driver's
1561 * perspective, it's different. For example, q0 belongs to FDIR VSI, q1-q64
1562 * to MAIN VSI, , q65-96 to SRIOV VSIs, q97-128 to VMDQ VSIs. For application
1563 * running on host, q1-64 and q97-128 can be used, total 96 queues. They can
1564 * use queue_idx from 0 to 95 to access queues, while real queue would be
1565 * different. This function will do a queue mapping to find VSI the queue
1568 static struct i40e_vsi*
1569 i40e_pf_get_vsi_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1571 /* the queue in MAIN VSI range */
1572 if (queue_idx < pf->main_vsi->nb_qps)
1573 return pf->main_vsi;
1575 queue_idx -= pf->main_vsi->nb_qps;
1577 /* queue_idx is greater than VMDQ VSIs range */
1578 if (queue_idx > pf->nb_cfg_vmdq_vsi * pf->vmdq_nb_qps - 1) {
1579 PMD_INIT_LOG(ERR, "queue_idx out of range. VMDQ configured?");
1583 return pf->vmdq[queue_idx / pf->vmdq_nb_qps].vsi;
1587 i40e_get_queue_offset_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1589 /* the queue in MAIN VSI range */
1590 if (queue_idx < pf->main_vsi->nb_qps)
1593 /* It's VMDQ queues */
1594 queue_idx -= pf->main_vsi->nb_qps;
1596 if (pf->nb_cfg_vmdq_vsi)
1597 return queue_idx % pf->vmdq_nb_qps;
1599 PMD_INIT_LOG(ERR, "Fail to get queue offset");
1600 return (uint16_t)(-1);
1605 i40e_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1607 struct i40e_rx_queue *rxq;
1609 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1611 PMD_INIT_FUNC_TRACE();
1613 if (rx_queue_id < dev->data->nb_rx_queues) {
1614 rxq = dev->data->rx_queues[rx_queue_id];
1616 err = i40e_alloc_rx_queue_mbufs(rxq);
1618 PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
1624 /* Init the RX tail regieter. */
1625 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1627 err = i40e_switch_rx_queue(hw, rxq->reg_idx, TRUE);
1630 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
1633 i40e_rx_queue_release_mbufs(rxq);
1634 i40e_reset_rx_queue(rxq);
1642 i40e_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1644 struct i40e_rx_queue *rxq;
1646 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1648 if (rx_queue_id < dev->data->nb_rx_queues) {
1649 rxq = dev->data->rx_queues[rx_queue_id];
1652 * rx_queue_id is queue id aplication refers to, while
1653 * rxq->reg_idx is the real queue index.
1655 err = i40e_switch_rx_queue(hw, rxq->reg_idx, FALSE);
1658 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
1662 i40e_rx_queue_release_mbufs(rxq);
1663 i40e_reset_rx_queue(rxq);
1670 i40e_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1673 struct i40e_tx_queue *txq;
1674 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1676 PMD_INIT_FUNC_TRACE();
1678 if (tx_queue_id < dev->data->nb_tx_queues) {
1679 txq = dev->data->tx_queues[tx_queue_id];
1682 * tx_queue_id is queue id aplication refers to, while
1683 * rxq->reg_idx is the real queue index.
1685 err = i40e_switch_tx_queue(hw, txq->reg_idx, TRUE);
1687 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
1695 i40e_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1697 struct i40e_tx_queue *txq;
1699 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1701 if (tx_queue_id < dev->data->nb_tx_queues) {
1702 txq = dev->data->tx_queues[tx_queue_id];
1705 * tx_queue_id is queue id aplication refers to, while
1706 * txq->reg_idx is the real queue index.
1708 err = i40e_switch_tx_queue(hw, txq->reg_idx, FALSE);
1711 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u of",
1716 i40e_tx_queue_release_mbufs(txq);
1717 i40e_reset_tx_queue(txq);
1724 i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
1727 unsigned int socket_id,
1728 const struct rte_eth_rxconf *rx_conf,
1729 struct rte_mempool *mp)
1731 struct i40e_vsi *vsi;
1732 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1733 struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
1734 struct i40e_rx_queue *rxq;
1735 const struct rte_memzone *rz;
1738 int use_def_burst_func = 1;
1740 if (hw->mac.type == I40E_MAC_VF) {
1741 struct i40e_vf *vf =
1742 I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1745 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
1748 PMD_DRV_LOG(ERR, "VSI not available or queue "
1749 "index exceeds the maximum");
1750 return I40E_ERR_PARAM;
1752 if (((nb_desc * sizeof(union i40e_rx_desc)) % I40E_ALIGN) != 0 ||
1753 (nb_desc > I40E_MAX_RING_DESC) ||
1754 (nb_desc < I40E_MIN_RING_DESC)) {
1755 PMD_DRV_LOG(ERR, "Number (%u) of receive descriptors is "
1756 "invalid", nb_desc);
1757 return I40E_ERR_PARAM;
1760 /* Free memory if needed */
1761 if (dev->data->rx_queues[queue_idx]) {
1762 i40e_dev_rx_queue_release(dev->data->rx_queues[queue_idx]);
1763 dev->data->rx_queues[queue_idx] = NULL;
1766 /* Allocate the rx queue data structure */
1767 rxq = rte_zmalloc_socket("i40e rx queue",
1768 sizeof(struct i40e_rx_queue),
1769 RTE_CACHE_LINE_SIZE,
1772 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
1773 "rx queue data structure");
1777 rxq->nb_rx_desc = nb_desc;
1778 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
1779 rxq->queue_id = queue_idx;
1780 if (hw->mac.type == I40E_MAC_VF)
1781 rxq->reg_idx = queue_idx;
1782 else /* PF device */
1783 rxq->reg_idx = vsi->base_queue +
1784 i40e_get_queue_offset_by_qindex(pf, queue_idx);
1786 rxq->port_id = dev->data->port_id;
1787 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
1789 rxq->drop_en = rx_conf->rx_drop_en;
1791 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
1793 /* Allocate the maximun number of RX ring hardware descriptor. */
1794 ring_size = sizeof(union i40e_rx_desc) * I40E_MAX_RING_DESC;
1795 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
1796 rz = i40e_ring_dma_zone_reserve(dev,
1802 i40e_dev_rx_queue_release(rxq);
1803 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX");
1807 /* Zero all the descriptors in the ring. */
1808 memset(rz->addr, 0, ring_size);
1810 #ifdef RTE_LIBRTE_XEN_DOM0
1811 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
1813 rxq->rx_ring_phys_addr = (uint64_t)rz->phys_addr;
1816 rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
1818 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1819 len = (uint16_t)(nb_desc + RTE_PMD_I40E_RX_MAX_BURST);
1824 /* Allocate the software ring. */
1826 rte_zmalloc_socket("i40e rx sw ring",
1827 sizeof(struct i40e_rx_entry) * len,
1828 RTE_CACHE_LINE_SIZE,
1830 if (!rxq->sw_ring) {
1831 i40e_dev_rx_queue_release(rxq);
1832 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW ring");
1836 i40e_reset_rx_queue(rxq);
1838 dev->data->rx_queues[queue_idx] = rxq;
1840 use_def_burst_func = check_rx_burst_bulk_alloc_preconditions(rxq);
1842 if (!use_def_burst_func && !dev->data->scattered_rx) {
1843 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1844 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
1845 "satisfied. Rx Burst Bulk Alloc function will be "
1846 "used on port=%d, queue=%d.",
1847 rxq->port_id, rxq->queue_id);
1848 dev->rx_pkt_burst = i40e_recv_pkts_bulk_alloc;
1849 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
1851 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
1852 "not satisfied, Scattered Rx is requested, "
1853 "or RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC is "
1854 "not enabled on port=%d, queue=%d.",
1855 rxq->port_id, rxq->queue_id);
1862 i40e_dev_rx_queue_release(void *rxq)
1864 struct i40e_rx_queue *q = (struct i40e_rx_queue *)rxq;
1867 PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL");
1871 i40e_rx_queue_release_mbufs(q);
1872 rte_free(q->sw_ring);
1877 i40e_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1879 #define I40E_RXQ_SCAN_INTERVAL 4
1880 volatile union i40e_rx_desc *rxdp;
1881 struct i40e_rx_queue *rxq;
1884 if (unlikely(rx_queue_id >= dev->data->nb_rx_queues)) {
1885 PMD_DRV_LOG(ERR, "Invalid RX queue id %u", rx_queue_id);
1889 rxq = dev->data->rx_queues[rx_queue_id];
1890 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
1891 while ((desc < rxq->nb_rx_desc) &&
1892 ((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1893 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
1894 (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) {
1896 * Check the DD bit of a rx descriptor of each 4 in a group,
1897 * to avoid checking too frequently and downgrading performance
1900 desc += I40E_RXQ_SCAN_INTERVAL;
1901 rxdp += I40E_RXQ_SCAN_INTERVAL;
1902 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
1903 rxdp = &(rxq->rx_ring[rxq->rx_tail +
1904 desc - rxq->nb_rx_desc]);
1911 i40e_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
1913 volatile union i40e_rx_desc *rxdp;
1914 struct i40e_rx_queue *rxq = rx_queue;
1918 if (unlikely(offset >= rxq->nb_rx_desc)) {
1919 PMD_DRV_LOG(ERR, "Invalid RX queue id %u", offset);
1923 desc = rxq->rx_tail + offset;
1924 if (desc >= rxq->nb_rx_desc)
1925 desc -= rxq->nb_rx_desc;
1927 rxdp = &(rxq->rx_ring[desc]);
1929 ret = !!(((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1930 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
1931 (1 << I40E_RX_DESC_STATUS_DD_SHIFT));
1937 i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
1940 unsigned int socket_id,
1941 const struct rte_eth_txconf *tx_conf)
1943 struct i40e_vsi *vsi;
1944 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1945 struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
1946 struct i40e_tx_queue *txq;
1947 const struct rte_memzone *tz;
1949 uint16_t tx_rs_thresh, tx_free_thresh;
1951 if (hw->mac.type == I40E_MAC_VF) {
1952 struct i40e_vf *vf =
1953 I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1956 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
1959 PMD_DRV_LOG(ERR, "VSI is NULL, or queue index (%u) "
1960 "exceeds the maximum", queue_idx);
1961 return I40E_ERR_PARAM;
1964 if (((nb_desc * sizeof(struct i40e_tx_desc)) % I40E_ALIGN) != 0 ||
1965 (nb_desc > I40E_MAX_RING_DESC) ||
1966 (nb_desc < I40E_MIN_RING_DESC)) {
1967 PMD_DRV_LOG(ERR, "Number (%u) of transmit descriptors is "
1968 "invalid", nb_desc);
1969 return I40E_ERR_PARAM;
1973 * The following two parameters control the setting of the RS bit on
1974 * transmit descriptors. TX descriptors will have their RS bit set
1975 * after txq->tx_rs_thresh descriptors have been used. The TX
1976 * descriptor ring will be cleaned after txq->tx_free_thresh
1977 * descriptors are used or if the number of descriptors required to
1978 * transmit a packet is greater than the number of free TX descriptors.
1980 * The following constraints must be satisfied:
1981 * - tx_rs_thresh must be greater than 0.
1982 * - tx_rs_thresh must be less than the size of the ring minus 2.
1983 * - tx_rs_thresh must be less than or equal to tx_free_thresh.
1984 * - tx_rs_thresh must be a divisor of the ring size.
1985 * - tx_free_thresh must be greater than 0.
1986 * - tx_free_thresh must be less than the size of the ring minus 3.
1988 * One descriptor in the TX ring is used as a sentinel to avoid a H/W
1989 * race condition, hence the maximum threshold constraints. When set
1990 * to zero use default values.
1992 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
1993 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
1994 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
1995 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
1996 if (tx_rs_thresh >= (nb_desc - 2)) {
1997 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
1998 "number of TX descriptors minus 2. "
1999 "(tx_rs_thresh=%u port=%d queue=%d)",
2000 (unsigned int)tx_rs_thresh,
2001 (int)dev->data->port_id,
2003 return I40E_ERR_PARAM;
2005 if (tx_free_thresh >= (nb_desc - 3)) {
2006 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2007 "tx_free_thresh must be less than the "
2008 "number of TX descriptors minus 3. "
2009 "(tx_free_thresh=%u port=%d queue=%d)",
2010 (unsigned int)tx_free_thresh,
2011 (int)dev->data->port_id,
2013 return I40E_ERR_PARAM;
2015 if (tx_rs_thresh > tx_free_thresh) {
2016 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or "
2017 "equal to tx_free_thresh. (tx_free_thresh=%u"
2018 " tx_rs_thresh=%u port=%d queue=%d)",
2019 (unsigned int)tx_free_thresh,
2020 (unsigned int)tx_rs_thresh,
2021 (int)dev->data->port_id,
2023 return I40E_ERR_PARAM;
2025 if ((nb_desc % tx_rs_thresh) != 0) {
2026 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2027 "number of TX descriptors. (tx_rs_thresh=%u"
2028 " port=%d queue=%d)",
2029 (unsigned int)tx_rs_thresh,
2030 (int)dev->data->port_id,
2032 return I40E_ERR_PARAM;
2034 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2035 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2036 "tx_rs_thresh is greater than 1. "
2037 "(tx_rs_thresh=%u port=%d queue=%d)",
2038 (unsigned int)tx_rs_thresh,
2039 (int)dev->data->port_id,
2041 return I40E_ERR_PARAM;
2044 /* Free memory if needed. */
2045 if (dev->data->tx_queues[queue_idx]) {
2046 i40e_dev_tx_queue_release(dev->data->tx_queues[queue_idx]);
2047 dev->data->tx_queues[queue_idx] = NULL;
2050 /* Allocate the TX queue data structure. */
2051 txq = rte_zmalloc_socket("i40e tx queue",
2052 sizeof(struct i40e_tx_queue),
2053 RTE_CACHE_LINE_SIZE,
2056 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2057 "tx queue structure");
2061 /* Allocate TX hardware ring descriptors. */
2062 ring_size = sizeof(struct i40e_tx_desc) * I40E_MAX_RING_DESC;
2063 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2064 tz = i40e_ring_dma_zone_reserve(dev,
2070 i40e_dev_tx_queue_release(txq);
2071 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX");
2075 txq->nb_tx_desc = nb_desc;
2076 txq->tx_rs_thresh = tx_rs_thresh;
2077 txq->tx_free_thresh = tx_free_thresh;
2078 txq->pthresh = tx_conf->tx_thresh.pthresh;
2079 txq->hthresh = tx_conf->tx_thresh.hthresh;
2080 txq->wthresh = tx_conf->tx_thresh.wthresh;
2081 txq->queue_id = queue_idx;
2082 if (hw->mac.type == I40E_MAC_VF)
2083 txq->reg_idx = queue_idx;
2084 else /* PF device */
2085 txq->reg_idx = vsi->base_queue +
2086 i40e_get_queue_offset_by_qindex(pf, queue_idx);
2088 txq->port_id = dev->data->port_id;
2089 txq->txq_flags = tx_conf->txq_flags;
2091 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2093 #ifdef RTE_LIBRTE_XEN_DOM0
2094 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2096 txq->tx_ring_phys_addr = (uint64_t)tz->phys_addr;
2098 txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2100 /* Allocate software ring */
2102 rte_zmalloc_socket("i40e tx sw ring",
2103 sizeof(struct i40e_tx_entry) * nb_desc,
2104 RTE_CACHE_LINE_SIZE,
2106 if (!txq->sw_ring) {
2107 i40e_dev_tx_queue_release(txq);
2108 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW TX ring");
2112 i40e_reset_tx_queue(txq);
2114 dev->data->tx_queues[queue_idx] = txq;
2116 /* Use a simple TX queue without offloads or multi segs if possible */
2117 if (((txq->txq_flags & I40E_SIMPLE_FLAGS) == I40E_SIMPLE_FLAGS) &&
2118 (txq->tx_rs_thresh >= I40E_TX_MAX_BURST)) {
2119 PMD_INIT_LOG(INFO, "Using simple tx path");
2120 dev->tx_pkt_burst = i40e_xmit_pkts_simple;
2122 PMD_INIT_LOG(INFO, "Using full-featured tx path");
2123 dev->tx_pkt_burst = i40e_xmit_pkts;
2130 i40e_dev_tx_queue_release(void *txq)
2132 struct i40e_tx_queue *q = (struct i40e_tx_queue *)txq;
2135 PMD_DRV_LOG(DEBUG, "Pointer to TX queue is NULL");
2139 i40e_tx_queue_release_mbufs(q);
2140 rte_free(q->sw_ring);
2144 static const struct rte_memzone *
2145 i40e_ring_dma_zone_reserve(struct rte_eth_dev *dev,
2146 const char *ring_name,
2151 char z_name[RTE_MEMZONE_NAMESIZE];
2152 const struct rte_memzone *mz;
2154 snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
2155 dev->driver->pci_drv.name, ring_name,
2156 dev->data->port_id, queue_id);
2157 mz = rte_memzone_lookup(z_name);
2161 #ifdef RTE_LIBRTE_XEN_DOM0
2162 return rte_memzone_reserve_bounded(z_name, ring_size,
2163 socket_id, 0, I40E_ALIGN, RTE_PGSIZE_2M);
2165 return rte_memzone_reserve_aligned(z_name, ring_size,
2166 socket_id, 0, I40E_ALIGN);
2170 const struct rte_memzone *
2171 i40e_memzone_reserve(const char *name, uint32_t len, int socket_id)
2173 const struct rte_memzone *mz = NULL;
2175 mz = rte_memzone_lookup(name);
2178 #ifdef RTE_LIBRTE_XEN_DOM0
2179 mz = rte_memzone_reserve_bounded(name, len,
2180 socket_id, 0, I40E_ALIGN, RTE_PGSIZE_2M);
2182 mz = rte_memzone_reserve_aligned(name, len,
2183 socket_id, 0, I40E_ALIGN);
2189 i40e_rx_queue_release_mbufs(struct i40e_rx_queue *rxq)
2193 if (!rxq || !rxq->sw_ring) {
2194 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL");
2198 for (i = 0; i < rxq->nb_rx_desc; i++) {
2199 if (rxq->sw_ring[i].mbuf) {
2200 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2201 rxq->sw_ring[i].mbuf = NULL;
2204 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2205 if (rxq->rx_nb_avail == 0)
2207 for (i = 0; i < rxq->rx_nb_avail; i++) {
2208 struct rte_mbuf *mbuf;
2210 mbuf = rxq->rx_stage[rxq->rx_next_avail + i];
2211 rte_pktmbuf_free_seg(mbuf);
2213 rxq->rx_nb_avail = 0;
2214 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2218 i40e_reset_rx_queue(struct i40e_rx_queue *rxq)
2223 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2224 if (check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
2225 len = (uint16_t)(rxq->nb_rx_desc + RTE_PMD_I40E_RX_MAX_BURST);
2227 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2228 len = rxq->nb_rx_desc;
2230 for (i = 0; i < len * sizeof(union i40e_rx_desc); i++)
2231 ((volatile char *)rxq->rx_ring)[i] = 0;
2233 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2234 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2235 for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; ++i)
2236 rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
2238 rxq->rx_nb_avail = 0;
2239 rxq->rx_next_avail = 0;
2240 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2241 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2243 rxq->nb_rx_hold = 0;
2244 rxq->pkt_first_seg = NULL;
2245 rxq->pkt_last_seg = NULL;
2249 i40e_tx_queue_release_mbufs(struct i40e_tx_queue *txq)
2253 if (!txq || !txq->sw_ring) {
2254 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL");
2258 for (i = 0; i < txq->nb_tx_desc; i++) {
2259 if (txq->sw_ring[i].mbuf) {
2260 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2261 txq->sw_ring[i].mbuf = NULL;
2267 i40e_reset_tx_queue(struct i40e_tx_queue *txq)
2269 struct i40e_tx_entry *txe;
2270 uint16_t i, prev, size;
2273 PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
2278 size = sizeof(struct i40e_tx_desc) * txq->nb_tx_desc;
2279 for (i = 0; i < size; i++)
2280 ((volatile char *)txq->tx_ring)[i] = 0;
2282 prev = (uint16_t)(txq->nb_tx_desc - 1);
2283 for (i = 0; i < txq->nb_tx_desc; i++) {
2284 volatile struct i40e_tx_desc *txd = &txq->tx_ring[i];
2286 txd->cmd_type_offset_bsz =
2287 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE);
2290 txe[prev].next_id = i;
2294 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2295 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2298 txq->nb_tx_used = 0;
2300 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2301 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2304 /* Init the TX queue in hardware */
2306 i40e_tx_queue_init(struct i40e_tx_queue *txq)
2308 enum i40e_status_code err = I40E_SUCCESS;
2309 struct i40e_vsi *vsi = txq->vsi;
2310 struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2311 uint16_t pf_q = txq->reg_idx;
2312 struct i40e_hmc_obj_txq tx_ctx;
2315 /* clear the context structure first */
2316 memset(&tx_ctx, 0, sizeof(tx_ctx));
2317 tx_ctx.new_context = 1;
2318 tx_ctx.base = txq->tx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2319 tx_ctx.qlen = txq->nb_tx_desc;
2320 tx_ctx.rdylist = rte_le_to_cpu_16(vsi->info.qs_handle[0]);
2321 if (vsi->type == I40E_VSI_FDIR)
2322 tx_ctx.fd_ena = TRUE;
2324 err = i40e_clear_lan_tx_queue_context(hw, pf_q);
2325 if (err != I40E_SUCCESS) {
2326 PMD_DRV_LOG(ERR, "Failure of clean lan tx queue context");
2330 err = i40e_set_lan_tx_queue_context(hw, pf_q, &tx_ctx);
2331 if (err != I40E_SUCCESS) {
2332 PMD_DRV_LOG(ERR, "Failure of set lan tx queue context");
2336 /* Now associate this queue with this PCI function */
2337 qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
2338 qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
2339 I40E_QTX_CTL_PF_INDX_MASK);
2340 I40E_WRITE_REG(hw, I40E_QTX_CTL(pf_q), qtx_ctl);
2341 I40E_WRITE_FLUSH(hw);
2343 txq->qtx_tail = hw->hw_addr + I40E_QTX_TAIL(pf_q);
2349 i40e_alloc_rx_queue_mbufs(struct i40e_rx_queue *rxq)
2351 struct i40e_rx_entry *rxe = rxq->sw_ring;
2355 for (i = 0; i < rxq->nb_rx_desc; i++) {
2356 volatile union i40e_rx_desc *rxd;
2357 struct rte_mbuf *mbuf = rte_rxmbuf_alloc(rxq->mp);
2359 if (unlikely(!mbuf)) {
2360 PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
2364 rte_mbuf_refcnt_set(mbuf, 1);
2366 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
2368 mbuf->port = rxq->port_id;
2371 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf));
2373 rxd = &rxq->rx_ring[i];
2374 rxd->read.pkt_addr = dma_addr;
2375 rxd->read.hdr_addr = dma_addr;
2376 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2377 rxd->read.rsvd1 = 0;
2378 rxd->read.rsvd2 = 0;
2379 #endif /* RTE_LIBRTE_I40E_16BYTE_RX_DESC */
2388 * Calculate the buffer length, and check the jumbo frame
2389 * and maximum packet length.
2392 i40e_rx_queue_config(struct i40e_rx_queue *rxq)
2394 struct i40e_pf *pf = I40E_VSI_TO_PF(rxq->vsi);
2395 struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2396 struct rte_eth_dev_data *data = pf->dev_data;
2397 struct rte_pktmbuf_pool_private *mbp_priv =
2398 rte_mempool_get_priv(rxq->mp);
2399 uint16_t buf_size = (uint16_t)(mbp_priv->mbuf_data_room_size -
2400 RTE_PKTMBUF_HEADROOM);
2403 switch (pf->flags & (I40E_FLAG_HEADER_SPLIT_DISABLED |
2404 I40E_FLAG_HEADER_SPLIT_ENABLED)) {
2405 case I40E_FLAG_HEADER_SPLIT_ENABLED: /* Not supported */
2406 rxq->rx_hdr_len = RTE_ALIGN(I40E_RXBUF_SZ_1024,
2407 (1 << I40E_RXQ_CTX_HBUFF_SHIFT));
2408 rxq->rx_buf_len = RTE_ALIGN(I40E_RXBUF_SZ_2048,
2409 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2410 rxq->hs_mode = i40e_header_split_enabled;
2412 case I40E_FLAG_HEADER_SPLIT_DISABLED:
2414 rxq->rx_hdr_len = 0;
2415 rxq->rx_buf_len = RTE_ALIGN(buf_size,
2416 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2417 rxq->hs_mode = i40e_header_split_none;
2421 len = hw->func_caps.rx_buf_chain_len * rxq->rx_buf_len;
2422 rxq->max_pkt_len = RTE_MIN(len, data->dev_conf.rxmode.max_rx_pkt_len);
2423 if (data->dev_conf.rxmode.jumbo_frame == 1) {
2424 if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
2425 rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
2426 PMD_DRV_LOG(ERR, "maximum packet length must "
2427 "be larger than %u and smaller than %u,"
2428 "as jumbo frame is enabled",
2429 (uint32_t)ETHER_MAX_LEN,
2430 (uint32_t)I40E_FRAME_SIZE_MAX);
2431 return I40E_ERR_CONFIG;
2434 if (rxq->max_pkt_len < ETHER_MIN_LEN ||
2435 rxq->max_pkt_len > ETHER_MAX_LEN) {
2436 PMD_DRV_LOG(ERR, "maximum packet length must be "
2437 "larger than %u and smaller than %u, "
2438 "as jumbo frame is disabled",
2439 (uint32_t)ETHER_MIN_LEN,
2440 (uint32_t)ETHER_MAX_LEN);
2441 return I40E_ERR_CONFIG;
2448 /* Init the RX queue in hardware */
2450 i40e_rx_queue_init(struct i40e_rx_queue *rxq)
2452 int err = I40E_SUCCESS;
2453 struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2454 struct rte_eth_dev_data *dev_data = I40E_VSI_TO_DEV_DATA(rxq->vsi);
2455 struct rte_eth_dev *dev = I40E_VSI_TO_ETH_DEV(rxq->vsi);
2456 uint16_t pf_q = rxq->reg_idx;
2458 struct i40e_hmc_obj_rxq rx_ctx;
2459 struct rte_pktmbuf_pool_private *mbp_priv;
2461 err = i40e_rx_queue_config(rxq);
2463 PMD_DRV_LOG(ERR, "Failed to config RX queue");
2467 /* Clear the context structure first */
2468 memset(&rx_ctx, 0, sizeof(struct i40e_hmc_obj_rxq));
2469 rx_ctx.dbuff = rxq->rx_buf_len >> I40E_RXQ_CTX_DBUFF_SHIFT;
2470 rx_ctx.hbuff = rxq->rx_hdr_len >> I40E_RXQ_CTX_HBUFF_SHIFT;
2472 rx_ctx.base = rxq->rx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2473 rx_ctx.qlen = rxq->nb_rx_desc;
2474 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2477 rx_ctx.dtype = rxq->hs_mode;
2479 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_ALL;
2481 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_NONE;
2482 rx_ctx.rxmax = rxq->max_pkt_len;
2483 rx_ctx.tphrdesc_ena = 1;
2484 rx_ctx.tphwdesc_ena = 1;
2485 rx_ctx.tphdata_ena = 1;
2486 rx_ctx.tphhead_ena = 1;
2487 rx_ctx.lrxqthresh = 2;
2488 rx_ctx.crcstrip = (rxq->crc_len == 0) ? 1 : 0;
2493 err = i40e_clear_lan_rx_queue_context(hw, pf_q);
2494 if (err != I40E_SUCCESS) {
2495 PMD_DRV_LOG(ERR, "Failed to clear LAN RX queue context");
2498 err = i40e_set_lan_rx_queue_context(hw, pf_q, &rx_ctx);
2499 if (err != I40E_SUCCESS) {
2500 PMD_DRV_LOG(ERR, "Failed to set LAN RX queue context");
2504 rxq->qrx_tail = hw->hw_addr + I40E_QRX_TAIL(pf_q);
2506 mbp_priv = rte_mempool_get_priv(rxq->mp);
2507 buf_size = (uint16_t)(mbp_priv->mbuf_data_room_size -
2508 RTE_PKTMBUF_HEADROOM);
2510 /* Check if scattered RX needs to be used. */
2511 if ((rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
2512 dev_data->scattered_rx = 1;
2513 dev->rx_pkt_burst = i40e_recv_scattered_pkts;
2516 /* Init the RX tail regieter. */
2517 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
2523 i40e_dev_clear_queues(struct rte_eth_dev *dev)
2527 PMD_INIT_FUNC_TRACE();
2529 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2530 i40e_tx_queue_release_mbufs(dev->data->tx_queues[i]);
2531 i40e_reset_tx_queue(dev->data->tx_queues[i]);
2534 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2535 i40e_rx_queue_release_mbufs(dev->data->rx_queues[i]);
2536 i40e_reset_rx_queue(dev->data->rx_queues[i]);
2540 #define I40E_FDIR_NUM_TX_DESC I40E_MIN_RING_DESC
2541 #define I40E_FDIR_NUM_RX_DESC I40E_MIN_RING_DESC
2543 enum i40e_status_code
2544 i40e_fdir_setup_tx_resources(struct i40e_pf *pf)
2546 struct i40e_tx_queue *txq;
2547 const struct rte_memzone *tz = NULL;
2549 struct rte_eth_dev *dev = pf->adapter->eth_dev;
2552 PMD_DRV_LOG(ERR, "PF is not available");
2553 return I40E_ERR_BAD_PTR;
2556 /* Allocate the TX queue data structure. */
2557 txq = rte_zmalloc_socket("i40e fdir tx queue",
2558 sizeof(struct i40e_tx_queue),
2559 RTE_CACHE_LINE_SIZE,
2562 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2563 "tx queue structure.");
2564 return I40E_ERR_NO_MEMORY;
2567 /* Allocate TX hardware ring descriptors. */
2568 ring_size = sizeof(struct i40e_tx_desc) * I40E_FDIR_NUM_TX_DESC;
2569 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2571 tz = i40e_ring_dma_zone_reserve(dev,
2577 i40e_dev_tx_queue_release(txq);
2578 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX.");
2579 return I40E_ERR_NO_MEMORY;
2582 txq->nb_tx_desc = I40E_FDIR_NUM_TX_DESC;
2583 txq->queue_id = I40E_FDIR_QUEUE_ID;
2584 txq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2585 txq->vsi = pf->fdir.fdir_vsi;
2587 #ifdef RTE_LIBRTE_XEN_DOM0
2588 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2590 txq->tx_ring_phys_addr = (uint64_t)tz->phys_addr;
2592 txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2594 * don't need to allocate software ring and reset for the fdir
2595 * program queue just set the queue has been configured.
2600 return I40E_SUCCESS;
2603 enum i40e_status_code
2604 i40e_fdir_setup_rx_resources(struct i40e_pf *pf)
2606 struct i40e_rx_queue *rxq;
2607 const struct rte_memzone *rz = NULL;
2609 struct rte_eth_dev *dev = pf->adapter->eth_dev;
2612 PMD_DRV_LOG(ERR, "PF is not available");
2613 return I40E_ERR_BAD_PTR;
2616 /* Allocate the RX queue data structure. */
2617 rxq = rte_zmalloc_socket("i40e fdir rx queue",
2618 sizeof(struct i40e_rx_queue),
2619 RTE_CACHE_LINE_SIZE,
2622 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2623 "rx queue structure.");
2624 return I40E_ERR_NO_MEMORY;
2627 /* Allocate RX hardware ring descriptors. */
2628 ring_size = sizeof(union i40e_rx_desc) * I40E_FDIR_NUM_RX_DESC;
2629 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2631 rz = i40e_ring_dma_zone_reserve(dev,
2637 i40e_dev_rx_queue_release(rxq);
2638 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX.");
2639 return I40E_ERR_NO_MEMORY;
2642 rxq->nb_rx_desc = I40E_FDIR_NUM_RX_DESC;
2643 rxq->queue_id = I40E_FDIR_QUEUE_ID;
2644 rxq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2645 rxq->vsi = pf->fdir.fdir_vsi;
2647 #ifdef RTE_LIBRTE_XEN_DOM0
2648 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
2650 rxq->rx_ring_phys_addr = (uint64_t)rz->phys_addr;
2652 rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
2655 * Don't need to allocate software ring and reset for the fdir
2656 * rx queue, just set the queue has been configured.
2661 return I40E_SUCCESS;