4 * Copyright(c) 2010-2015 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 "base/i40e_prototype.h"
56 #include "base/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_TX_MAX_BURST 32
69 #define I40E_DMA_MEM_ALIGN 4096
71 #define I40E_SIMPLE_FLAGS ((uint32_t)ETH_TXQ_FLAGS_NOMULTSEGS | \
72 ETH_TXQ_FLAGS_NOOFFLOADS)
74 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
76 #define I40E_TX_CKSUM_OFFLOAD_MASK ( \
79 PKT_TX_OUTER_IP_CKSUM)
81 #define RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb) \
82 (uint64_t) ((mb)->buf_physaddr + RTE_PKTMBUF_HEADROOM)
84 #define RTE_MBUF_DATA_DMA_ADDR(mb) \
85 ((uint64_t)((mb)->buf_physaddr + (mb)->data_off))
87 static const struct rte_memzone *
88 i40e_ring_dma_zone_reserve(struct rte_eth_dev *dev,
89 const char *ring_name,
93 static uint16_t i40e_xmit_pkts_simple(void *tx_queue,
94 struct rte_mbuf **tx_pkts,
97 /* Translate the rx descriptor status to pkt flags */
98 static inline uint64_t
99 i40e_rxd_status_to_pkt_flags(uint64_t qword)
103 /* Check if VLAN packet */
104 flags = qword & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) ?
107 /* Check if RSS_HASH */
108 flags |= (((qword >> I40E_RX_DESC_STATUS_FLTSTAT_SHIFT) &
109 I40E_RX_DESC_FLTSTAT_RSS_HASH) ==
110 I40E_RX_DESC_FLTSTAT_RSS_HASH) ? PKT_RX_RSS_HASH : 0;
112 /* Check if FDIR Match */
113 flags |= (qword & (1 << I40E_RX_DESC_STATUS_FLM_SHIFT) ?
119 static inline uint64_t
120 i40e_rxd_error_to_pkt_flags(uint64_t qword)
123 uint64_t error_bits = (qword >> I40E_RXD_QW1_ERROR_SHIFT);
125 #define I40E_RX_ERR_BITS 0x3f
126 if (likely((error_bits & I40E_RX_ERR_BITS) == 0))
128 /* If RXE bit set, all other status bits are meaningless */
129 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
130 flags |= PKT_RX_MAC_ERR;
134 /* If RECIPE bit set, all other status indications should be ignored */
135 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_RECIPE_SHIFT))) {
136 flags |= PKT_RX_RECIP_ERR;
139 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT)))
140 flags |= PKT_RX_HBUF_OVERFLOW;
141 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_IPE_SHIFT)))
142 flags |= PKT_RX_IP_CKSUM_BAD;
143 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_L4E_SHIFT)))
144 flags |= PKT_RX_L4_CKSUM_BAD;
145 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT)))
146 flags |= PKT_RX_EIP_CKSUM_BAD;
147 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_OVERSIZE_SHIFT)))
148 flags |= PKT_RX_OVERSIZE;
153 /* Translate pkt types to pkt flags */
154 static inline uint64_t
155 i40e_rxd_ptype_to_pkt_flags(uint64_t qword)
157 uint8_t ptype = (uint8_t)((qword & I40E_RXD_QW1_PTYPE_MASK) >>
158 I40E_RXD_QW1_PTYPE_SHIFT);
159 static const uint64_t ip_ptype_map[I40E_MAX_PKT_TYPE] = {
182 PKT_RX_IPV4_HDR, /* PTYPE 22 */
183 PKT_RX_IPV4_HDR, /* PTYPE 23 */
184 PKT_RX_IPV4_HDR, /* PTYPE 24 */
186 PKT_RX_IPV4_HDR, /* PTYPE 26 */
187 PKT_RX_IPV4_HDR, /* PTYPE 27 */
188 PKT_RX_IPV4_HDR, /* PTYPE 28 */
189 PKT_RX_IPV4_HDR_EXT, /* PTYPE 29 */
190 PKT_RX_IPV4_HDR_EXT, /* PTYPE 30 */
191 PKT_RX_IPV4_HDR_EXT, /* PTYPE 31 */
193 PKT_RX_IPV4_HDR_EXT, /* PTYPE 33 */
194 PKT_RX_IPV4_HDR_EXT, /* PTYPE 34 */
195 PKT_RX_IPV4_HDR_EXT, /* PTYPE 35 */
196 PKT_RX_IPV4_HDR_EXT, /* PTYPE 36 */
197 PKT_RX_IPV4_HDR_EXT, /* PTYPE 37 */
198 PKT_RX_IPV4_HDR_EXT, /* PTYPE 38 */
200 PKT_RX_IPV4_HDR_EXT, /* PTYPE 40 */
201 PKT_RX_IPV4_HDR_EXT, /* PTYPE 41 */
202 PKT_RX_IPV4_HDR_EXT, /* PTYPE 42 */
203 PKT_RX_IPV4_HDR_EXT, /* PTYPE 43 */
204 PKT_RX_IPV4_HDR_EXT, /* PTYPE 44 */
205 PKT_RX_IPV4_HDR_EXT, /* PTYPE 45 */
206 PKT_RX_IPV4_HDR_EXT, /* PTYPE 46 */
208 PKT_RX_IPV4_HDR_EXT, /* PTYPE 48 */
209 PKT_RX_IPV4_HDR_EXT, /* PTYPE 49 */
210 PKT_RX_IPV4_HDR_EXT, /* PTYPE 50 */
211 PKT_RX_IPV4_HDR_EXT, /* PTYPE 51 */
212 PKT_RX_IPV4_HDR_EXT, /* PTYPE 52 */
213 PKT_RX_IPV4_HDR_EXT, /* PTYPE 53 */
215 PKT_RX_IPV4_HDR_EXT, /* PTYPE 55 */
216 PKT_RX_IPV4_HDR_EXT, /* PTYPE 56 */
217 PKT_RX_IPV4_HDR_EXT, /* PTYPE 57 */
218 PKT_RX_IPV4_HDR_EXT, /* PTYPE 58 */
219 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 59 */
220 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 60 */
221 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 61 */
223 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 63 */
224 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 64 */
225 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 65 */
226 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 66 */
227 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 67 */
228 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 68 */
230 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 70 */
231 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 71 */
232 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 72 */
233 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 73 */
234 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 74 */
235 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 75 */
236 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 76 */
238 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 78 */
239 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 79 */
240 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 80 */
241 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 81 */
242 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 82 */
243 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 83 */
245 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 85 */
246 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 86 */
247 PKT_RX_IPV4_HDR_EXT, /* PTYPE 87 */
248 PKT_RX_IPV6_HDR, /* PTYPE 88 */
249 PKT_RX_IPV6_HDR, /* PTYPE 89 */
250 PKT_RX_IPV6_HDR, /* PTYPE 90 */
252 PKT_RX_IPV6_HDR, /* PTYPE 92 */
253 PKT_RX_IPV6_HDR, /* PTYPE 93 */
254 PKT_RX_IPV6_HDR, /* PTYPE 94 */
255 PKT_RX_IPV6_HDR_EXT, /* PTYPE 95 */
256 PKT_RX_IPV6_HDR_EXT, /* PTYPE 96 */
257 PKT_RX_IPV6_HDR_EXT, /* PTYPE 97 */
259 PKT_RX_IPV6_HDR_EXT, /* PTYPE 99 */
260 PKT_RX_IPV6_HDR_EXT, /* PTYPE 100 */
261 PKT_RX_IPV6_HDR_EXT, /* PTYPE 101 */
262 PKT_RX_IPV6_HDR_EXT, /* PTYPE 102 */
263 PKT_RX_IPV6_HDR_EXT, /* PTYPE 103 */
264 PKT_RX_IPV6_HDR_EXT, /* PTYPE 104 */
266 PKT_RX_IPV6_HDR_EXT, /* PTYPE 106 */
267 PKT_RX_IPV6_HDR_EXT, /* PTYPE 107 */
268 PKT_RX_IPV6_HDR_EXT, /* PTYPE 108 */
269 PKT_RX_IPV6_HDR_EXT, /* PTYPE 109 */
270 PKT_RX_IPV6_HDR_EXT, /* PTYPE 110 */
271 PKT_RX_IPV6_HDR_EXT, /* PTYPE 111 */
272 PKT_RX_IPV6_HDR_EXT, /* PTYPE 112 */
274 PKT_RX_IPV6_HDR_EXT, /* PTYPE 114 */
275 PKT_RX_IPV6_HDR_EXT, /* PTYPE 115 */
276 PKT_RX_IPV6_HDR_EXT, /* PTYPE 116 */
277 PKT_RX_IPV6_HDR_EXT, /* PTYPE 117 */
278 PKT_RX_IPV6_HDR_EXT, /* PTYPE 118 */
279 PKT_RX_IPV6_HDR_EXT, /* PTYPE 119 */
281 PKT_RX_IPV6_HDR_EXT, /* PTYPE 121 */
282 PKT_RX_IPV6_HDR_EXT, /* PTYPE 122 */
283 PKT_RX_IPV6_HDR_EXT, /* PTYPE 123 */
284 PKT_RX_IPV6_HDR_EXT, /* PTYPE 124 */
285 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 125 */
286 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 126 */
287 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 127 */
289 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 129 */
290 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 130 */
291 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 131 */
292 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 132 */
293 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 133 */
294 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 134 */
296 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 136 */
297 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 137 */
298 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 138 */
299 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 139 */
300 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 140 */
301 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 141 */
302 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 142 */
304 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 144 */
305 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 145 */
306 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 146 */
307 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 147 */
308 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 148 */
309 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 149 */
311 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 151 */
312 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 152 */
313 PKT_RX_IPV6_HDR_EXT, /* PTYPE 153 */
418 return ip_ptype_map[ptype];
421 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_MASK 0x03
422 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_FD_ID 0x01
423 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_FLEX 0x02
424 #define I40E_RX_DESC_EXT_STATUS_FLEXBL_MASK 0x03
425 #define I40E_RX_DESC_EXT_STATUS_FLEXBL_FLEX 0x01
427 static inline uint64_t
428 i40e_rxd_build_fdir(volatile union i40e_rx_desc *rxdp, struct rte_mbuf *mb)
431 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
432 uint16_t flexbh, flexbl;
434 flexbh = (rte_le_to_cpu_32(rxdp->wb.qword2.ext_status) >>
435 I40E_RX_DESC_EXT_STATUS_FLEXBH_SHIFT) &
436 I40E_RX_DESC_EXT_STATUS_FLEXBH_MASK;
437 flexbl = (rte_le_to_cpu_32(rxdp->wb.qword2.ext_status) >>
438 I40E_RX_DESC_EXT_STATUS_FLEXBL_SHIFT) &
439 I40E_RX_DESC_EXT_STATUS_FLEXBL_MASK;
442 if (flexbh == I40E_RX_DESC_EXT_STATUS_FLEXBH_FD_ID) {
444 rte_le_to_cpu_32(rxdp->wb.qword3.hi_dword.fd_id);
445 flags |= PKT_RX_FDIR_ID;
446 } else if (flexbh == I40E_RX_DESC_EXT_STATUS_FLEXBH_FLEX) {
448 rte_le_to_cpu_32(rxdp->wb.qword3.hi_dword.flex_bytes_hi);
449 flags |= PKT_RX_FDIR_FLX;
451 if (flexbl == I40E_RX_DESC_EXT_STATUS_FLEXBL_FLEX) {
453 rte_le_to_cpu_32(rxdp->wb.qword3.lo_dword.flex_bytes_lo);
454 flags |= PKT_RX_FDIR_FLX;
458 rte_le_to_cpu_32(rxdp->wb.qword0.hi_dword.fd_id);
459 flags |= PKT_RX_FDIR_ID;
464 i40e_txd_enable_checksum(uint64_t ol_flags,
467 union i40e_tx_offload tx_offload,
468 uint32_t *cd_tunneling)
470 /* UDP tunneling packet TX checksum offload */
471 if (ol_flags & PKT_TX_OUTER_IP_CKSUM) {
473 *td_offset |= (tx_offload.outer_l2_len >> 1)
474 << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
476 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
477 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
478 else if (ol_flags & PKT_TX_OUTER_IPV4)
479 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
480 else if (ol_flags & PKT_TX_OUTER_IPV6)
481 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
483 /* Now set the ctx descriptor fields */
484 *cd_tunneling |= (tx_offload.outer_l3_len >> 2) <<
485 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
486 (tx_offload.l2_len >> 1) <<
487 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
490 *td_offset |= (tx_offload.l2_len >> 1)
491 << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
493 /* Enable L3 checksum offloads */
494 if (ol_flags & PKT_TX_IP_CKSUM) {
495 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
496 *td_offset |= (tx_offload.l3_len >> 2)
497 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
498 } else if (ol_flags & PKT_TX_IPV4) {
499 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
500 *td_offset |= (tx_offload.l3_len >> 2)
501 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
502 } else if (ol_flags & PKT_TX_IPV6) {
503 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
504 *td_offset |= (tx_offload.l3_len >> 2)
505 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
508 if (ol_flags & PKT_TX_TCP_SEG) {
509 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
510 *td_offset |= (tx_offload.l4_len >> 2)
511 << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
515 /* Enable L4 checksum offloads */
516 switch (ol_flags & PKT_TX_L4_MASK) {
517 case PKT_TX_TCP_CKSUM:
518 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
519 *td_offset |= (sizeof(struct tcp_hdr) >> 2) <<
520 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
522 case PKT_TX_SCTP_CKSUM:
523 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
524 *td_offset |= (sizeof(struct sctp_hdr) >> 2) <<
525 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
527 case PKT_TX_UDP_CKSUM:
528 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
529 *td_offset |= (sizeof(struct udp_hdr) >> 2) <<
530 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
537 static inline struct rte_mbuf *
538 rte_rxmbuf_alloc(struct rte_mempool *mp)
542 m = __rte_mbuf_raw_alloc(mp);
543 __rte_mbuf_sanity_check_raw(m, 0);
548 /* Construct the tx flags */
549 static inline uint64_t
550 i40e_build_ctob(uint32_t td_cmd,
555 return rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DATA |
556 ((uint64_t)td_cmd << I40E_TXD_QW1_CMD_SHIFT) |
557 ((uint64_t)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
558 ((uint64_t)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
559 ((uint64_t)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT));
563 i40e_xmit_cleanup(struct i40e_tx_queue *txq)
565 struct i40e_tx_entry *sw_ring = txq->sw_ring;
566 volatile struct i40e_tx_desc *txd = txq->tx_ring;
567 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
568 uint16_t nb_tx_desc = txq->nb_tx_desc;
569 uint16_t desc_to_clean_to;
570 uint16_t nb_tx_to_clean;
572 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
573 if (desc_to_clean_to >= nb_tx_desc)
574 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
576 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
577 if (!(txd[desc_to_clean_to].cmd_type_offset_bsz &
578 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE))) {
579 PMD_TX_FREE_LOG(DEBUG, "TX descriptor %4u is not done "
580 "(port=%d queue=%d)", desc_to_clean_to,
581 txq->port_id, txq->queue_id);
585 if (last_desc_cleaned > desc_to_clean_to)
586 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
589 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
592 txd[desc_to_clean_to].cmd_type_offset_bsz = 0;
594 txq->last_desc_cleaned = desc_to_clean_to;
595 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
601 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
602 check_rx_burst_bulk_alloc_preconditions(struct i40e_rx_queue *rxq)
604 check_rx_burst_bulk_alloc_preconditions(__rte_unused struct i40e_rx_queue *rxq)
609 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
610 if (!(rxq->rx_free_thresh >= RTE_PMD_I40E_RX_MAX_BURST)) {
611 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
612 "rxq->rx_free_thresh=%d, "
613 "RTE_PMD_I40E_RX_MAX_BURST=%d",
614 rxq->rx_free_thresh, RTE_PMD_I40E_RX_MAX_BURST);
616 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
617 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
618 "rxq->rx_free_thresh=%d, "
619 "rxq->nb_rx_desc=%d",
620 rxq->rx_free_thresh, rxq->nb_rx_desc);
622 } else if (rxq->nb_rx_desc % rxq->rx_free_thresh != 0) {
623 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
624 "rxq->nb_rx_desc=%d, "
625 "rxq->rx_free_thresh=%d",
626 rxq->nb_rx_desc, rxq->rx_free_thresh);
628 } else if (!(rxq->nb_rx_desc < (I40E_MAX_RING_DESC -
629 RTE_PMD_I40E_RX_MAX_BURST))) {
630 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
631 "rxq->nb_rx_desc=%d, "
632 "I40E_MAX_RING_DESC=%d, "
633 "RTE_PMD_I40E_RX_MAX_BURST=%d",
634 rxq->nb_rx_desc, I40E_MAX_RING_DESC,
635 RTE_PMD_I40E_RX_MAX_BURST);
645 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
646 #define I40E_LOOK_AHEAD 8
647 #if (I40E_LOOK_AHEAD != 8)
648 #error "PMD I40E: I40E_LOOK_AHEAD must be 8\n"
651 i40e_rx_scan_hw_ring(struct i40e_rx_queue *rxq)
653 volatile union i40e_rx_desc *rxdp;
654 struct i40e_rx_entry *rxep;
659 int32_t s[I40E_LOOK_AHEAD], nb_dd;
660 int32_t i, j, nb_rx = 0;
663 rxdp = &rxq->rx_ring[rxq->rx_tail];
664 rxep = &rxq->sw_ring[rxq->rx_tail];
666 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
667 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
668 I40E_RXD_QW1_STATUS_SHIFT;
670 /* Make sure there is at least 1 packet to receive */
671 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
675 * Scan LOOK_AHEAD descriptors at a time to determine which
676 * descriptors reference packets that are ready to be received.
678 for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; i+=I40E_LOOK_AHEAD,
679 rxdp += I40E_LOOK_AHEAD, rxep += I40E_LOOK_AHEAD) {
680 /* Read desc statuses backwards to avoid race condition */
681 for (j = I40E_LOOK_AHEAD - 1; j >= 0; j--) {
682 qword1 = rte_le_to_cpu_64(\
683 rxdp[j].wb.qword1.status_error_len);
684 s[j] = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
685 I40E_RXD_QW1_STATUS_SHIFT;
688 /* Compute how many status bits were set */
689 for (j = 0, nb_dd = 0; j < I40E_LOOK_AHEAD; j++)
690 nb_dd += s[j] & (1 << I40E_RX_DESC_STATUS_DD_SHIFT);
694 /* Translate descriptor info to mbuf parameters */
695 for (j = 0; j < nb_dd; j++) {
697 qword1 = rte_le_to_cpu_64(\
698 rxdp[j].wb.qword1.status_error_len);
699 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
700 I40E_RXD_QW1_STATUS_SHIFT;
701 pkt_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
702 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
703 mb->data_len = pkt_len;
704 mb->pkt_len = pkt_len;
705 mb->vlan_tci = rx_status &
706 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) ?
708 rxdp[j].wb.qword0.lo_dword.l2tag1) : 0;
709 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
710 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
711 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
713 mb->packet_type = (uint16_t)((qword1 &
714 I40E_RXD_QW1_PTYPE_MASK) >>
715 I40E_RXD_QW1_PTYPE_SHIFT);
716 if (pkt_flags & PKT_RX_RSS_HASH)
717 mb->hash.rss = rte_le_to_cpu_32(\
718 rxdp[j].wb.qword0.hi_dword.rss);
719 if (pkt_flags & PKT_RX_FDIR)
720 pkt_flags |= i40e_rxd_build_fdir(&rxdp[j], mb);
722 mb->ol_flags = pkt_flags;
725 for (j = 0; j < I40E_LOOK_AHEAD; j++)
726 rxq->rx_stage[i + j] = rxep[j].mbuf;
728 if (nb_dd != I40E_LOOK_AHEAD)
732 /* Clear software ring entries */
733 for (i = 0; i < nb_rx; i++)
734 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
739 static inline uint16_t
740 i40e_rx_fill_from_stage(struct i40e_rx_queue *rxq,
741 struct rte_mbuf **rx_pkts,
745 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
747 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
749 for (i = 0; i < nb_pkts; i++)
750 rx_pkts[i] = stage[i];
752 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
753 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
759 i40e_rx_alloc_bufs(struct i40e_rx_queue *rxq)
761 volatile union i40e_rx_desc *rxdp;
762 struct i40e_rx_entry *rxep;
764 uint16_t alloc_idx, i;
768 /* Allocate buffers in bulk */
769 alloc_idx = (uint16_t)(rxq->rx_free_trigger -
770 (rxq->rx_free_thresh - 1));
771 rxep = &(rxq->sw_ring[alloc_idx]);
772 diag = rte_mempool_get_bulk(rxq->mp, (void *)rxep,
773 rxq->rx_free_thresh);
774 if (unlikely(diag != 0)) {
775 PMD_DRV_LOG(ERR, "Failed to get mbufs in bulk");
779 rxdp = &rxq->rx_ring[alloc_idx];
780 for (i = 0; i < rxq->rx_free_thresh; i++) {
782 rte_mbuf_refcnt_set(mb, 1);
784 mb->data_off = RTE_PKTMBUF_HEADROOM;
786 mb->port = rxq->port_id;
787 dma_addr = rte_cpu_to_le_64(\
788 RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb));
789 rxdp[i].read.hdr_addr = dma_addr;
790 rxdp[i].read.pkt_addr = dma_addr;
793 /* Update rx tail regsiter */
795 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->rx_free_trigger);
797 rxq->rx_free_trigger =
798 (uint16_t)(rxq->rx_free_trigger + rxq->rx_free_thresh);
799 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
800 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
805 static inline uint16_t
806 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
808 struct i40e_rx_queue *rxq = (struct i40e_rx_queue *)rx_queue;
814 if (rxq->rx_nb_avail)
815 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
817 nb_rx = (uint16_t)i40e_rx_scan_hw_ring(rxq);
818 rxq->rx_next_avail = 0;
819 rxq->rx_nb_avail = nb_rx;
820 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
822 if (rxq->rx_tail > rxq->rx_free_trigger) {
823 if (i40e_rx_alloc_bufs(rxq) != 0) {
826 PMD_RX_LOG(DEBUG, "Rx mbuf alloc failed for "
827 "port_id=%u, queue_id=%u",
828 rxq->port_id, rxq->queue_id);
829 rxq->rx_nb_avail = 0;
830 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
831 for (i = 0, j = rxq->rx_tail; i < nb_rx; i++, j++)
832 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
838 if (rxq->rx_tail >= rxq->nb_rx_desc)
841 if (rxq->rx_nb_avail)
842 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
848 i40e_recv_pkts_bulk_alloc(void *rx_queue,
849 struct rte_mbuf **rx_pkts,
852 uint16_t nb_rx = 0, n, count;
854 if (unlikely(nb_pkts == 0))
857 if (likely(nb_pkts <= RTE_PMD_I40E_RX_MAX_BURST))
858 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
861 n = RTE_MIN(nb_pkts, RTE_PMD_I40E_RX_MAX_BURST);
862 count = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
863 nb_rx = (uint16_t)(nb_rx + count);
864 nb_pkts = (uint16_t)(nb_pkts - count);
871 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
874 i40e_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
876 struct i40e_rx_queue *rxq;
877 volatile union i40e_rx_desc *rx_ring;
878 volatile union i40e_rx_desc *rxdp;
879 union i40e_rx_desc rxd;
880 struct i40e_rx_entry *sw_ring;
881 struct i40e_rx_entry *rxe;
882 struct rte_mbuf *rxm;
883 struct rte_mbuf *nmb;
887 uint16_t rx_packet_len;
888 uint16_t rx_id, nb_hold;
895 rx_id = rxq->rx_tail;
896 rx_ring = rxq->rx_ring;
897 sw_ring = rxq->sw_ring;
899 while (nb_rx < nb_pkts) {
900 rxdp = &rx_ring[rx_id];
901 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
902 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK)
903 >> I40E_RXD_QW1_STATUS_SHIFT;
904 /* Check the DD bit first */
905 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
908 nmb = rte_rxmbuf_alloc(rxq->mp);
914 rxe = &sw_ring[rx_id];
916 if (unlikely(rx_id == rxq->nb_rx_desc))
919 /* Prefetch next mbuf */
920 rte_prefetch0(sw_ring[rx_id].mbuf);
923 * When next RX descriptor is on a cache line boundary,
924 * prefetch the next 4 RX descriptors and next 8 pointers
927 if ((rx_id & 0x3) == 0) {
928 rte_prefetch0(&rx_ring[rx_id]);
929 rte_prefetch0(&sw_ring[rx_id]);
934 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
935 rxdp->read.hdr_addr = dma_addr;
936 rxdp->read.pkt_addr = dma_addr;
938 rx_packet_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
939 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
941 rxm->data_off = RTE_PKTMBUF_HEADROOM;
942 rte_prefetch0(RTE_PTR_ADD(rxm->buf_addr, RTE_PKTMBUF_HEADROOM));
945 rxm->pkt_len = rx_packet_len;
946 rxm->data_len = rx_packet_len;
947 rxm->port = rxq->port_id;
949 rxm->vlan_tci = rx_status &
950 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) ?
951 rte_le_to_cpu_16(rxd.wb.qword0.lo_dword.l2tag1) : 0;
952 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
953 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
954 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
955 rxm->packet_type = (uint16_t)((qword1 & I40E_RXD_QW1_PTYPE_MASK) >>
956 I40E_RXD_QW1_PTYPE_SHIFT);
957 if (pkt_flags & PKT_RX_RSS_HASH)
959 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
960 if (pkt_flags & PKT_RX_FDIR)
961 pkt_flags |= i40e_rxd_build_fdir(&rxd, rxm);
963 rxm->ol_flags = pkt_flags;
965 rx_pkts[nb_rx++] = rxm;
967 rxq->rx_tail = rx_id;
970 * If the number of free RX descriptors is greater than the RX free
971 * threshold of the queue, advance the receive tail register of queue.
972 * Update that register with the value of the last processed RX
973 * descriptor minus 1.
975 nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
976 if (nb_hold > rxq->rx_free_thresh) {
977 rx_id = (uint16_t) ((rx_id == 0) ?
978 (rxq->nb_rx_desc - 1) : (rx_id - 1));
979 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
982 rxq->nb_rx_hold = nb_hold;
988 i40e_recv_scattered_pkts(void *rx_queue,
989 struct rte_mbuf **rx_pkts,
992 struct i40e_rx_queue *rxq = rx_queue;
993 volatile union i40e_rx_desc *rx_ring = rxq->rx_ring;
994 volatile union i40e_rx_desc *rxdp;
995 union i40e_rx_desc rxd;
996 struct i40e_rx_entry *sw_ring = rxq->sw_ring;
997 struct i40e_rx_entry *rxe;
998 struct rte_mbuf *first_seg = rxq->pkt_first_seg;
999 struct rte_mbuf *last_seg = rxq->pkt_last_seg;
1000 struct rte_mbuf *nmb, *rxm;
1001 uint16_t rx_id = rxq->rx_tail;
1002 uint16_t nb_rx = 0, nb_hold = 0, rx_packet_len;
1008 while (nb_rx < nb_pkts) {
1009 rxdp = &rx_ring[rx_id];
1010 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
1011 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
1012 I40E_RXD_QW1_STATUS_SHIFT;
1013 /* Check the DD bit */
1014 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
1017 nmb = rte_rxmbuf_alloc(rxq->mp);
1022 rxe = &sw_ring[rx_id];
1024 if (rx_id == rxq->nb_rx_desc)
1027 /* Prefetch next mbuf */
1028 rte_prefetch0(sw_ring[rx_id].mbuf);
1031 * When next RX descriptor is on a cache line boundary,
1032 * prefetch the next 4 RX descriptors and next 8 pointers
1035 if ((rx_id & 0x3) == 0) {
1036 rte_prefetch0(&rx_ring[rx_id]);
1037 rte_prefetch0(&sw_ring[rx_id]);
1043 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1045 /* Set data buffer address and data length of the mbuf */
1046 rxdp->read.hdr_addr = dma_addr;
1047 rxdp->read.pkt_addr = dma_addr;
1048 rx_packet_len = (qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1049 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1050 rxm->data_len = rx_packet_len;
1051 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1054 * If this is the first buffer of the received packet, set the
1055 * pointer to the first mbuf of the packet and initialize its
1056 * context. Otherwise, update the total length and the number
1057 * of segments of the current scattered packet, and update the
1058 * pointer to the last mbuf of the current packet.
1062 first_seg->nb_segs = 1;
1063 first_seg->pkt_len = rx_packet_len;
1065 first_seg->pkt_len =
1066 (uint16_t)(first_seg->pkt_len +
1068 first_seg->nb_segs++;
1069 last_seg->next = rxm;
1073 * If this is not the last buffer of the received packet,
1074 * update the pointer to the last mbuf of the current scattered
1075 * packet and continue to parse the RX ring.
1077 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT))) {
1083 * This is the last buffer of the received packet. If the CRC
1084 * is not stripped by the hardware:
1085 * - Subtract the CRC length from the total packet length.
1086 * - If the last buffer only contains the whole CRC or a part
1087 * of it, free the mbuf associated to the last buffer. If part
1088 * of the CRC is also contained in the previous mbuf, subtract
1089 * the length of that CRC part from the data length of the
1093 if (unlikely(rxq->crc_len > 0)) {
1094 first_seg->pkt_len -= ETHER_CRC_LEN;
1095 if (rx_packet_len <= ETHER_CRC_LEN) {
1096 rte_pktmbuf_free_seg(rxm);
1097 first_seg->nb_segs--;
1098 last_seg->data_len =
1099 (uint16_t)(last_seg->data_len -
1100 (ETHER_CRC_LEN - rx_packet_len));
1101 last_seg->next = NULL;
1103 rxm->data_len = (uint16_t)(rx_packet_len -
1107 first_seg->port = rxq->port_id;
1108 first_seg->vlan_tci = (rx_status &
1109 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
1110 rte_le_to_cpu_16(rxd.wb.qword0.lo_dword.l2tag1) : 0;
1111 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
1112 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
1113 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
1114 first_seg->packet_type = (uint16_t)((qword1 &
1115 I40E_RXD_QW1_PTYPE_MASK) >>
1116 I40E_RXD_QW1_PTYPE_SHIFT);
1117 if (pkt_flags & PKT_RX_RSS_HASH)
1119 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
1120 if (pkt_flags & PKT_RX_FDIR)
1121 pkt_flags |= i40e_rxd_build_fdir(&rxd, rxm);
1123 first_seg->ol_flags = pkt_flags;
1125 /* Prefetch data of first segment, if configured to do so. */
1126 rte_prefetch0(RTE_PTR_ADD(first_seg->buf_addr,
1127 first_seg->data_off));
1128 rx_pkts[nb_rx++] = first_seg;
1132 /* Record index of the next RX descriptor to probe. */
1133 rxq->rx_tail = rx_id;
1134 rxq->pkt_first_seg = first_seg;
1135 rxq->pkt_last_seg = last_seg;
1138 * If the number of free RX descriptors is greater than the RX free
1139 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1140 * register. Update the RDT with the value of the last processed RX
1141 * descriptor minus 1, to guarantee that the RDT register is never
1142 * equal to the RDH register, which creates a "full" ring situtation
1143 * from the hardware point of view.
1145 nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
1146 if (nb_hold > rxq->rx_free_thresh) {
1147 rx_id = (uint16_t)(rx_id == 0 ?
1148 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1149 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
1152 rxq->nb_rx_hold = nb_hold;
1157 /* Check if the context descriptor is needed for TX offloading */
1158 static inline uint16_t
1159 i40e_calc_context_desc(uint64_t flags)
1161 uint64_t mask = 0ULL;
1163 mask |= (PKT_TX_OUTER_IP_CKSUM | PKT_TX_TCP_SEG);
1165 #ifdef RTE_LIBRTE_IEEE1588
1166 mask |= PKT_TX_IEEE1588_TMST;
1174 /* set i40e TSO context descriptor */
1175 static inline uint64_t
1176 i40e_set_tso_ctx(struct rte_mbuf *mbuf, union i40e_tx_offload tx_offload)
1178 uint64_t ctx_desc = 0;
1179 uint32_t cd_cmd, hdr_len, cd_tso_len;
1181 if (!tx_offload.l4_len) {
1182 PMD_DRV_LOG(DEBUG, "L4 length set to 0");
1187 * in case of tunneling packet, the outer_l2_len and
1188 * outer_l3_len must be 0.
1190 hdr_len = tx_offload.outer_l2_len +
1191 tx_offload.outer_l3_len +
1196 cd_cmd = I40E_TX_CTX_DESC_TSO;
1197 cd_tso_len = mbuf->pkt_len - hdr_len;
1198 ctx_desc |= ((uint64_t)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1199 ((uint64_t)cd_tso_len <<
1200 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1201 ((uint64_t)mbuf->tso_segsz <<
1202 I40E_TXD_CTX_QW1_MSS_SHIFT);
1208 i40e_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1210 struct i40e_tx_queue *txq;
1211 struct i40e_tx_entry *sw_ring;
1212 struct i40e_tx_entry *txe, *txn;
1213 volatile struct i40e_tx_desc *txd;
1214 volatile struct i40e_tx_desc *txr;
1215 struct rte_mbuf *tx_pkt;
1216 struct rte_mbuf *m_seg;
1217 uint32_t cd_tunneling_params;
1229 uint64_t buf_dma_addr;
1230 union i40e_tx_offload tx_offload = {0};
1233 sw_ring = txq->sw_ring;
1235 tx_id = txq->tx_tail;
1236 txe = &sw_ring[tx_id];
1238 /* Check if the descriptor ring needs to be cleaned. */
1239 if ((txq->nb_tx_desc - txq->nb_tx_free) > txq->tx_free_thresh)
1240 i40e_xmit_cleanup(txq);
1242 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
1248 tx_pkt = *tx_pkts++;
1249 RTE_MBUF_PREFETCH_TO_FREE(txe->mbuf);
1251 ol_flags = tx_pkt->ol_flags;
1252 tx_offload.l2_len = tx_pkt->l2_len;
1253 tx_offload.l3_len = tx_pkt->l3_len;
1254 tx_offload.outer_l2_len = tx_pkt->outer_l2_len;
1255 tx_offload.outer_l3_len = tx_pkt->outer_l3_len;
1256 tx_offload.l4_len = tx_pkt->l4_len;
1257 tx_offload.tso_segsz = tx_pkt->tso_segsz;
1259 /* Calculate the number of context descriptors needed. */
1260 nb_ctx = i40e_calc_context_desc(ol_flags);
1263 * The number of descriptors that must be allocated for
1264 * a packet equals to the number of the segments of that
1265 * packet plus 1 context descriptor if needed.
1267 nb_used = (uint16_t)(tx_pkt->nb_segs + nb_ctx);
1268 tx_last = (uint16_t)(tx_id + nb_used - 1);
1271 if (tx_last >= txq->nb_tx_desc)
1272 tx_last = (uint16_t)(tx_last - txq->nb_tx_desc);
1274 if (nb_used > txq->nb_tx_free) {
1275 if (i40e_xmit_cleanup(txq) != 0) {
1280 if (unlikely(nb_used > txq->tx_rs_thresh)) {
1281 while (nb_used > txq->nb_tx_free) {
1282 if (i40e_xmit_cleanup(txq) != 0) {
1291 /* Descriptor based VLAN insertion */
1292 if (ol_flags & PKT_TX_VLAN_PKT) {
1293 tx_flags |= tx_pkt->vlan_tci <<
1294 I40E_TX_FLAG_L2TAG1_SHIFT;
1295 tx_flags |= I40E_TX_FLAG_INSERT_VLAN;
1296 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1297 td_tag = (tx_flags & I40E_TX_FLAG_L2TAG1_MASK) >>
1298 I40E_TX_FLAG_L2TAG1_SHIFT;
1301 /* Always enable CRC offload insertion */
1302 td_cmd |= I40E_TX_DESC_CMD_ICRC;
1304 /* Enable checksum offloading */
1305 cd_tunneling_params = 0;
1306 if (unlikely(ol_flags & I40E_TX_CKSUM_OFFLOAD_MASK)) {
1307 i40e_txd_enable_checksum(ol_flags, &td_cmd, &td_offset,
1308 tx_offload, &cd_tunneling_params);
1311 if (unlikely(nb_ctx)) {
1312 /* Setup TX context descriptor if required */
1313 volatile struct i40e_tx_context_desc *ctx_txd =
1314 (volatile struct i40e_tx_context_desc *)\
1316 uint16_t cd_l2tag2 = 0;
1317 uint64_t cd_type_cmd_tso_mss =
1318 I40E_TX_DESC_DTYPE_CONTEXT;
1320 txn = &sw_ring[txe->next_id];
1321 RTE_MBUF_PREFETCH_TO_FREE(txn->mbuf);
1322 if (txe->mbuf != NULL) {
1323 rte_pktmbuf_free_seg(txe->mbuf);
1327 /* TSO enabled means no timestamp */
1328 if (ol_flags & PKT_TX_TCP_SEG)
1329 cd_type_cmd_tso_mss |=
1330 i40e_set_tso_ctx(tx_pkt, tx_offload);
1332 #ifdef RTE_LIBRTE_IEEE1588
1333 if (ol_flags & PKT_TX_IEEE1588_TMST)
1334 cd_type_cmd_tso_mss |=
1335 ((uint64_t)I40E_TX_CTX_DESC_TSYN <<
1336 I40E_TXD_CTX_QW1_CMD_SHIFT);
1340 ctx_txd->tunneling_params =
1341 rte_cpu_to_le_32(cd_tunneling_params);
1342 ctx_txd->l2tag2 = rte_cpu_to_le_16(cd_l2tag2);
1343 ctx_txd->type_cmd_tso_mss =
1344 rte_cpu_to_le_64(cd_type_cmd_tso_mss);
1346 PMD_TX_LOG(DEBUG, "mbuf: %p, TCD[%u]:\n"
1347 "tunneling_params: %#x;\n"
1350 "type_cmd_tso_mss: %#"PRIx64";\n",
1352 ctx_txd->tunneling_params,
1355 ctx_txd->type_cmd_tso_mss);
1357 txe->last_id = tx_last;
1358 tx_id = txe->next_id;
1365 txn = &sw_ring[txe->next_id];
1368 rte_pktmbuf_free_seg(txe->mbuf);
1371 /* Setup TX Descriptor */
1372 slen = m_seg->data_len;
1373 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
1375 PMD_TX_LOG(DEBUG, "mbuf: %p, TDD[%u]:\n"
1376 "buf_dma_addr: %#"PRIx64";\n"
1381 tx_pkt, tx_id, buf_dma_addr,
1382 td_cmd, td_offset, slen, td_tag);
1384 txd->buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
1385 txd->cmd_type_offset_bsz = i40e_build_ctob(td_cmd,
1386 td_offset, slen, td_tag);
1387 txe->last_id = tx_last;
1388 tx_id = txe->next_id;
1390 m_seg = m_seg->next;
1391 } while (m_seg != NULL);
1393 /* The last packet data descriptor needs End Of Packet (EOP) */
1394 td_cmd |= I40E_TX_DESC_CMD_EOP;
1395 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
1396 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
1398 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
1399 PMD_TX_FREE_LOG(DEBUG,
1400 "Setting RS bit on TXD id="
1401 "%4u (port=%d queue=%d)",
1402 tx_last, txq->port_id, txq->queue_id);
1404 td_cmd |= I40E_TX_DESC_CMD_RS;
1406 /* Update txq RS bit counters */
1407 txq->nb_tx_used = 0;
1410 txd->cmd_type_offset_bsz |=
1411 rte_cpu_to_le_64(((uint64_t)td_cmd) <<
1412 I40E_TXD_QW1_CMD_SHIFT);
1418 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
1419 (unsigned) txq->port_id, (unsigned) txq->queue_id,
1420 (unsigned) tx_id, (unsigned) nb_tx);
1422 I40E_PCI_REG_WRITE(txq->qtx_tail, tx_id);
1423 txq->tx_tail = tx_id;
1428 static inline int __attribute__((always_inline))
1429 i40e_tx_free_bufs(struct i40e_tx_queue *txq)
1431 struct i40e_tx_entry *txep;
1434 if (!(txq->tx_ring[txq->tx_next_dd].cmd_type_offset_bsz &
1435 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE)))
1438 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
1440 for (i = 0; i < txq->tx_rs_thresh; i++)
1441 rte_prefetch0((txep + i)->mbuf);
1443 if (!(txq->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT)) {
1444 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1445 rte_mempool_put(txep->mbuf->pool, txep->mbuf);
1449 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1450 rte_pktmbuf_free_seg(txep->mbuf);
1455 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
1456 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
1457 if (txq->tx_next_dd >= txq->nb_tx_desc)
1458 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
1460 return txq->tx_rs_thresh;
1463 #define I40E_TD_CMD (I40E_TX_DESC_CMD_ICRC |\
1464 I40E_TX_DESC_CMD_EOP)
1466 /* Populate 4 descriptors with data from 4 mbufs */
1468 tx4(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1473 for (i = 0; i < 4; i++, txdp++, pkts++) {
1474 dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
1475 txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1476 txdp->cmd_type_offset_bsz =
1477 i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1478 (*pkts)->data_len, 0);
1482 /* Populate 1 descriptor with data from 1 mbuf */
1484 tx1(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1488 dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
1489 txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1490 txdp->cmd_type_offset_bsz =
1491 i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1492 (*pkts)->data_len, 0);
1495 /* Fill hardware descriptor ring with mbuf data */
1497 i40e_tx_fill_hw_ring(struct i40e_tx_queue *txq,
1498 struct rte_mbuf **pkts,
1501 volatile struct i40e_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
1502 struct i40e_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
1503 const int N_PER_LOOP = 4;
1504 const int N_PER_LOOP_MASK = N_PER_LOOP - 1;
1505 int mainpart, leftover;
1508 mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
1509 leftover = (nb_pkts & ((uint32_t) N_PER_LOOP_MASK));
1510 for (i = 0; i < mainpart; i += N_PER_LOOP) {
1511 for (j = 0; j < N_PER_LOOP; ++j) {
1512 (txep + i + j)->mbuf = *(pkts + i + j);
1514 tx4(txdp + i, pkts + i);
1516 if (unlikely(leftover > 0)) {
1517 for (i = 0; i < leftover; ++i) {
1518 (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
1519 tx1(txdp + mainpart + i, pkts + mainpart + i);
1524 static inline uint16_t
1525 tx_xmit_pkts(struct i40e_tx_queue *txq,
1526 struct rte_mbuf **tx_pkts,
1529 volatile struct i40e_tx_desc *txr = txq->tx_ring;
1533 * Begin scanning the H/W ring for done descriptors when the number
1534 * of available descriptors drops below tx_free_thresh. For each done
1535 * descriptor, free the associated buffer.
1537 if (txq->nb_tx_free < txq->tx_free_thresh)
1538 i40e_tx_free_bufs(txq);
1540 /* Use available descriptor only */
1541 nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
1542 if (unlikely(!nb_pkts))
1545 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
1546 if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
1547 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
1548 i40e_tx_fill_hw_ring(txq, tx_pkts, n);
1549 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1550 rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1551 I40E_TXD_QW1_CMD_SHIFT);
1552 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1556 /* Fill hardware descriptor ring with mbuf data */
1557 i40e_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
1558 txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
1560 /* Determin if RS bit needs to be set */
1561 if (txq->tx_tail > txq->tx_next_rs) {
1562 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1563 rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1564 I40E_TXD_QW1_CMD_SHIFT);
1566 (uint16_t)(txq->tx_next_rs + txq->tx_rs_thresh);
1567 if (txq->tx_next_rs >= txq->nb_tx_desc)
1568 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1571 if (txq->tx_tail >= txq->nb_tx_desc)
1574 /* Update the tx tail register */
1576 I40E_PCI_REG_WRITE(txq->qtx_tail, txq->tx_tail);
1582 i40e_xmit_pkts_simple(void *tx_queue,
1583 struct rte_mbuf **tx_pkts,
1588 if (likely(nb_pkts <= I40E_TX_MAX_BURST))
1589 return tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1593 uint16_t ret, num = (uint16_t)RTE_MIN(nb_pkts,
1596 ret = tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1597 &tx_pkts[nb_tx], num);
1598 nb_tx = (uint16_t)(nb_tx + ret);
1599 nb_pkts = (uint16_t)(nb_pkts - ret);
1608 * Find the VSI the queue belongs to. 'queue_idx' is the queue index
1609 * application used, which assume having sequential ones. But from driver's
1610 * perspective, it's different. For example, q0 belongs to FDIR VSI, q1-q64
1611 * to MAIN VSI, , q65-96 to SRIOV VSIs, q97-128 to VMDQ VSIs. For application
1612 * running on host, q1-64 and q97-128 can be used, total 96 queues. They can
1613 * use queue_idx from 0 to 95 to access queues, while real queue would be
1614 * different. This function will do a queue mapping to find VSI the queue
1617 static struct i40e_vsi*
1618 i40e_pf_get_vsi_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1620 /* the queue in MAIN VSI range */
1621 if (queue_idx < pf->main_vsi->nb_qps)
1622 return pf->main_vsi;
1624 queue_idx -= pf->main_vsi->nb_qps;
1626 /* queue_idx is greater than VMDQ VSIs range */
1627 if (queue_idx > pf->nb_cfg_vmdq_vsi * pf->vmdq_nb_qps - 1) {
1628 PMD_INIT_LOG(ERR, "queue_idx out of range. VMDQ configured?");
1632 return pf->vmdq[queue_idx / pf->vmdq_nb_qps].vsi;
1636 i40e_get_queue_offset_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1638 /* the queue in MAIN VSI range */
1639 if (queue_idx < pf->main_vsi->nb_qps)
1642 /* It's VMDQ queues */
1643 queue_idx -= pf->main_vsi->nb_qps;
1645 if (pf->nb_cfg_vmdq_vsi)
1646 return queue_idx % pf->vmdq_nb_qps;
1648 PMD_INIT_LOG(ERR, "Fail to get queue offset");
1649 return (uint16_t)(-1);
1654 i40e_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1656 struct i40e_rx_queue *rxq;
1658 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1660 PMD_INIT_FUNC_TRACE();
1662 if (rx_queue_id < dev->data->nb_rx_queues) {
1663 rxq = dev->data->rx_queues[rx_queue_id];
1665 err = i40e_alloc_rx_queue_mbufs(rxq);
1667 PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
1673 /* Init the RX tail regieter. */
1674 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1676 err = i40e_switch_rx_queue(hw, rxq->reg_idx, TRUE);
1679 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
1682 i40e_rx_queue_release_mbufs(rxq);
1683 i40e_reset_rx_queue(rxq);
1691 i40e_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1693 struct i40e_rx_queue *rxq;
1695 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1697 if (rx_queue_id < dev->data->nb_rx_queues) {
1698 rxq = dev->data->rx_queues[rx_queue_id];
1701 * rx_queue_id is queue id aplication refers to, while
1702 * rxq->reg_idx is the real queue index.
1704 err = i40e_switch_rx_queue(hw, rxq->reg_idx, FALSE);
1707 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
1711 i40e_rx_queue_release_mbufs(rxq);
1712 i40e_reset_rx_queue(rxq);
1719 i40e_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1722 struct i40e_tx_queue *txq;
1723 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1725 PMD_INIT_FUNC_TRACE();
1727 if (tx_queue_id < dev->data->nb_tx_queues) {
1728 txq = dev->data->tx_queues[tx_queue_id];
1731 * tx_queue_id is queue id aplication refers to, while
1732 * rxq->reg_idx is the real queue index.
1734 err = i40e_switch_tx_queue(hw, txq->reg_idx, TRUE);
1736 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
1744 i40e_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1746 struct i40e_tx_queue *txq;
1748 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1750 if (tx_queue_id < dev->data->nb_tx_queues) {
1751 txq = dev->data->tx_queues[tx_queue_id];
1754 * tx_queue_id is queue id aplication refers to, while
1755 * txq->reg_idx is the real queue index.
1757 err = i40e_switch_tx_queue(hw, txq->reg_idx, FALSE);
1760 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u of",
1765 i40e_tx_queue_release_mbufs(txq);
1766 i40e_reset_tx_queue(txq);
1773 i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
1776 unsigned int socket_id,
1777 const struct rte_eth_rxconf *rx_conf,
1778 struct rte_mempool *mp)
1780 struct i40e_vsi *vsi;
1781 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1782 struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
1783 struct i40e_rx_queue *rxq;
1784 const struct rte_memzone *rz;
1787 int use_def_burst_func = 1;
1789 if (hw->mac.type == I40E_MAC_VF) {
1790 struct i40e_vf *vf =
1791 I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1794 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
1797 PMD_DRV_LOG(ERR, "VSI not available or queue "
1798 "index exceeds the maximum");
1799 return I40E_ERR_PARAM;
1801 if (((nb_desc * sizeof(union i40e_rx_desc)) % I40E_ALIGN) != 0 ||
1802 (nb_desc > I40E_MAX_RING_DESC) ||
1803 (nb_desc < I40E_MIN_RING_DESC)) {
1804 PMD_DRV_LOG(ERR, "Number (%u) of receive descriptors is "
1805 "invalid", nb_desc);
1806 return I40E_ERR_PARAM;
1809 /* Free memory if needed */
1810 if (dev->data->rx_queues[queue_idx]) {
1811 i40e_dev_rx_queue_release(dev->data->rx_queues[queue_idx]);
1812 dev->data->rx_queues[queue_idx] = NULL;
1815 /* Allocate the rx queue data structure */
1816 rxq = rte_zmalloc_socket("i40e rx queue",
1817 sizeof(struct i40e_rx_queue),
1818 RTE_CACHE_LINE_SIZE,
1821 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
1822 "rx queue data structure");
1826 rxq->nb_rx_desc = nb_desc;
1827 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
1828 rxq->queue_id = queue_idx;
1829 if (hw->mac.type == I40E_MAC_VF)
1830 rxq->reg_idx = queue_idx;
1831 else /* PF device */
1832 rxq->reg_idx = vsi->base_queue +
1833 i40e_get_queue_offset_by_qindex(pf, queue_idx);
1835 rxq->port_id = dev->data->port_id;
1836 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
1838 rxq->drop_en = rx_conf->rx_drop_en;
1840 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
1842 /* Allocate the maximun number of RX ring hardware descriptor. */
1843 ring_size = sizeof(union i40e_rx_desc) * I40E_MAX_RING_DESC;
1844 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
1845 rz = i40e_ring_dma_zone_reserve(dev,
1851 i40e_dev_rx_queue_release(rxq);
1852 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX");
1856 /* Zero all the descriptors in the ring. */
1857 memset(rz->addr, 0, ring_size);
1859 #ifdef RTE_LIBRTE_XEN_DOM0
1860 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
1862 rxq->rx_ring_phys_addr = (uint64_t)rz->phys_addr;
1865 rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
1867 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1868 len = (uint16_t)(nb_desc + RTE_PMD_I40E_RX_MAX_BURST);
1873 /* Allocate the software ring. */
1875 rte_zmalloc_socket("i40e rx sw ring",
1876 sizeof(struct i40e_rx_entry) * len,
1877 RTE_CACHE_LINE_SIZE,
1879 if (!rxq->sw_ring) {
1880 i40e_dev_rx_queue_release(rxq);
1881 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW ring");
1885 i40e_reset_rx_queue(rxq);
1887 dev->data->rx_queues[queue_idx] = rxq;
1889 use_def_burst_func = check_rx_burst_bulk_alloc_preconditions(rxq);
1891 if (!use_def_burst_func && !dev->data->scattered_rx) {
1892 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1893 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
1894 "satisfied. Rx Burst Bulk Alloc function will be "
1895 "used on port=%d, queue=%d.",
1896 rxq->port_id, rxq->queue_id);
1897 dev->rx_pkt_burst = i40e_recv_pkts_bulk_alloc;
1898 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
1900 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
1901 "not satisfied, Scattered Rx is requested, "
1902 "or RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC is "
1903 "not enabled on port=%d, queue=%d.",
1904 rxq->port_id, rxq->queue_id);
1911 i40e_dev_rx_queue_release(void *rxq)
1913 struct i40e_rx_queue *q = (struct i40e_rx_queue *)rxq;
1916 PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL");
1920 i40e_rx_queue_release_mbufs(q);
1921 rte_free(q->sw_ring);
1926 i40e_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1928 #define I40E_RXQ_SCAN_INTERVAL 4
1929 volatile union i40e_rx_desc *rxdp;
1930 struct i40e_rx_queue *rxq;
1933 if (unlikely(rx_queue_id >= dev->data->nb_rx_queues)) {
1934 PMD_DRV_LOG(ERR, "Invalid RX queue id %u", rx_queue_id);
1938 rxq = dev->data->rx_queues[rx_queue_id];
1939 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
1940 while ((desc < rxq->nb_rx_desc) &&
1941 ((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1942 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
1943 (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) {
1945 * Check the DD bit of a rx descriptor of each 4 in a group,
1946 * to avoid checking too frequently and downgrading performance
1949 desc += I40E_RXQ_SCAN_INTERVAL;
1950 rxdp += I40E_RXQ_SCAN_INTERVAL;
1951 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
1952 rxdp = &(rxq->rx_ring[rxq->rx_tail +
1953 desc - rxq->nb_rx_desc]);
1960 i40e_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
1962 volatile union i40e_rx_desc *rxdp;
1963 struct i40e_rx_queue *rxq = rx_queue;
1967 if (unlikely(offset >= rxq->nb_rx_desc)) {
1968 PMD_DRV_LOG(ERR, "Invalid RX queue id %u", offset);
1972 desc = rxq->rx_tail + offset;
1973 if (desc >= rxq->nb_rx_desc)
1974 desc -= rxq->nb_rx_desc;
1976 rxdp = &(rxq->rx_ring[desc]);
1978 ret = !!(((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1979 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
1980 (1 << I40E_RX_DESC_STATUS_DD_SHIFT));
1986 i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
1989 unsigned int socket_id,
1990 const struct rte_eth_txconf *tx_conf)
1992 struct i40e_vsi *vsi;
1993 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1994 struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
1995 struct i40e_tx_queue *txq;
1996 const struct rte_memzone *tz;
1998 uint16_t tx_rs_thresh, tx_free_thresh;
2000 if (hw->mac.type == I40E_MAC_VF) {
2001 struct i40e_vf *vf =
2002 I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2005 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
2008 PMD_DRV_LOG(ERR, "VSI is NULL, or queue index (%u) "
2009 "exceeds the maximum", queue_idx);
2010 return I40E_ERR_PARAM;
2013 if (((nb_desc * sizeof(struct i40e_tx_desc)) % I40E_ALIGN) != 0 ||
2014 (nb_desc > I40E_MAX_RING_DESC) ||
2015 (nb_desc < I40E_MIN_RING_DESC)) {
2016 PMD_DRV_LOG(ERR, "Number (%u) of transmit descriptors is "
2017 "invalid", nb_desc);
2018 return I40E_ERR_PARAM;
2022 * The following two parameters control the setting of the RS bit on
2023 * transmit descriptors. TX descriptors will have their RS bit set
2024 * after txq->tx_rs_thresh descriptors have been used. The TX
2025 * descriptor ring will be cleaned after txq->tx_free_thresh
2026 * descriptors are used or if the number of descriptors required to
2027 * transmit a packet is greater than the number of free TX descriptors.
2029 * The following constraints must be satisfied:
2030 * - tx_rs_thresh must be greater than 0.
2031 * - tx_rs_thresh must be less than the size of the ring minus 2.
2032 * - tx_rs_thresh must be less than or equal to tx_free_thresh.
2033 * - tx_rs_thresh must be a divisor of the ring size.
2034 * - tx_free_thresh must be greater than 0.
2035 * - tx_free_thresh must be less than the size of the ring minus 3.
2037 * One descriptor in the TX ring is used as a sentinel to avoid a H/W
2038 * race condition, hence the maximum threshold constraints. When set
2039 * to zero use default values.
2041 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2042 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2043 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2044 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2045 if (tx_rs_thresh >= (nb_desc - 2)) {
2046 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2047 "number of TX descriptors minus 2. "
2048 "(tx_rs_thresh=%u port=%d queue=%d)",
2049 (unsigned int)tx_rs_thresh,
2050 (int)dev->data->port_id,
2052 return I40E_ERR_PARAM;
2054 if (tx_free_thresh >= (nb_desc - 3)) {
2055 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2056 "tx_free_thresh must be less than the "
2057 "number of TX descriptors minus 3. "
2058 "(tx_free_thresh=%u port=%d queue=%d)",
2059 (unsigned int)tx_free_thresh,
2060 (int)dev->data->port_id,
2062 return I40E_ERR_PARAM;
2064 if (tx_rs_thresh > tx_free_thresh) {
2065 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or "
2066 "equal to tx_free_thresh. (tx_free_thresh=%u"
2067 " tx_rs_thresh=%u port=%d queue=%d)",
2068 (unsigned int)tx_free_thresh,
2069 (unsigned int)tx_rs_thresh,
2070 (int)dev->data->port_id,
2072 return I40E_ERR_PARAM;
2074 if ((nb_desc % tx_rs_thresh) != 0) {
2075 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2076 "number of TX descriptors. (tx_rs_thresh=%u"
2077 " port=%d queue=%d)",
2078 (unsigned int)tx_rs_thresh,
2079 (int)dev->data->port_id,
2081 return I40E_ERR_PARAM;
2083 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2084 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2085 "tx_rs_thresh is greater than 1. "
2086 "(tx_rs_thresh=%u port=%d queue=%d)",
2087 (unsigned int)tx_rs_thresh,
2088 (int)dev->data->port_id,
2090 return I40E_ERR_PARAM;
2093 /* Free memory if needed. */
2094 if (dev->data->tx_queues[queue_idx]) {
2095 i40e_dev_tx_queue_release(dev->data->tx_queues[queue_idx]);
2096 dev->data->tx_queues[queue_idx] = NULL;
2099 /* Allocate the TX queue data structure. */
2100 txq = rte_zmalloc_socket("i40e tx queue",
2101 sizeof(struct i40e_tx_queue),
2102 RTE_CACHE_LINE_SIZE,
2105 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2106 "tx queue structure");
2110 /* Allocate TX hardware ring descriptors. */
2111 ring_size = sizeof(struct i40e_tx_desc) * I40E_MAX_RING_DESC;
2112 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2113 tz = i40e_ring_dma_zone_reserve(dev,
2119 i40e_dev_tx_queue_release(txq);
2120 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX");
2124 txq->nb_tx_desc = nb_desc;
2125 txq->tx_rs_thresh = tx_rs_thresh;
2126 txq->tx_free_thresh = tx_free_thresh;
2127 txq->pthresh = tx_conf->tx_thresh.pthresh;
2128 txq->hthresh = tx_conf->tx_thresh.hthresh;
2129 txq->wthresh = tx_conf->tx_thresh.wthresh;
2130 txq->queue_id = queue_idx;
2131 if (hw->mac.type == I40E_MAC_VF)
2132 txq->reg_idx = queue_idx;
2133 else /* PF device */
2134 txq->reg_idx = vsi->base_queue +
2135 i40e_get_queue_offset_by_qindex(pf, queue_idx);
2137 txq->port_id = dev->data->port_id;
2138 txq->txq_flags = tx_conf->txq_flags;
2140 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2142 #ifdef RTE_LIBRTE_XEN_DOM0
2143 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2145 txq->tx_ring_phys_addr = (uint64_t)tz->phys_addr;
2147 txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2149 /* Allocate software ring */
2151 rte_zmalloc_socket("i40e tx sw ring",
2152 sizeof(struct i40e_tx_entry) * nb_desc,
2153 RTE_CACHE_LINE_SIZE,
2155 if (!txq->sw_ring) {
2156 i40e_dev_tx_queue_release(txq);
2157 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW TX ring");
2161 i40e_reset_tx_queue(txq);
2163 dev->data->tx_queues[queue_idx] = txq;
2165 /* Use a simple TX queue without offloads or multi segs if possible */
2166 if (((txq->txq_flags & I40E_SIMPLE_FLAGS) == I40E_SIMPLE_FLAGS) &&
2167 (txq->tx_rs_thresh >= I40E_TX_MAX_BURST)) {
2168 PMD_INIT_LOG(INFO, "Using simple tx path");
2169 dev->tx_pkt_burst = i40e_xmit_pkts_simple;
2171 PMD_INIT_LOG(INFO, "Using full-featured tx path");
2172 dev->tx_pkt_burst = i40e_xmit_pkts;
2179 i40e_dev_tx_queue_release(void *txq)
2181 struct i40e_tx_queue *q = (struct i40e_tx_queue *)txq;
2184 PMD_DRV_LOG(DEBUG, "Pointer to TX queue is NULL");
2188 i40e_tx_queue_release_mbufs(q);
2189 rte_free(q->sw_ring);
2193 static const struct rte_memzone *
2194 i40e_ring_dma_zone_reserve(struct rte_eth_dev *dev,
2195 const char *ring_name,
2200 char z_name[RTE_MEMZONE_NAMESIZE];
2201 const struct rte_memzone *mz;
2203 snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
2204 dev->driver->pci_drv.name, ring_name,
2205 dev->data->port_id, queue_id);
2206 mz = rte_memzone_lookup(z_name);
2210 #ifdef RTE_LIBRTE_XEN_DOM0
2211 return rte_memzone_reserve_bounded(z_name, ring_size,
2212 socket_id, 0, I40E_ALIGN, RTE_PGSIZE_2M);
2214 return rte_memzone_reserve_aligned(z_name, ring_size,
2215 socket_id, 0, I40E_ALIGN);
2219 const struct rte_memzone *
2220 i40e_memzone_reserve(const char *name, uint32_t len, int socket_id)
2222 const struct rte_memzone *mz = NULL;
2224 mz = rte_memzone_lookup(name);
2227 #ifdef RTE_LIBRTE_XEN_DOM0
2228 mz = rte_memzone_reserve_bounded(name, len,
2229 socket_id, 0, I40E_ALIGN, RTE_PGSIZE_2M);
2231 mz = rte_memzone_reserve_aligned(name, len,
2232 socket_id, 0, I40E_ALIGN);
2238 i40e_rx_queue_release_mbufs(struct i40e_rx_queue *rxq)
2242 if (!rxq || !rxq->sw_ring) {
2243 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL");
2247 for (i = 0; i < rxq->nb_rx_desc; i++) {
2248 if (rxq->sw_ring[i].mbuf) {
2249 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2250 rxq->sw_ring[i].mbuf = NULL;
2253 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2254 if (rxq->rx_nb_avail == 0)
2256 for (i = 0; i < rxq->rx_nb_avail; i++) {
2257 struct rte_mbuf *mbuf;
2259 mbuf = rxq->rx_stage[rxq->rx_next_avail + i];
2260 rte_pktmbuf_free_seg(mbuf);
2262 rxq->rx_nb_avail = 0;
2263 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2267 i40e_reset_rx_queue(struct i40e_rx_queue *rxq)
2272 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2273 if (check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
2274 len = (uint16_t)(rxq->nb_rx_desc + RTE_PMD_I40E_RX_MAX_BURST);
2276 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2277 len = rxq->nb_rx_desc;
2279 for (i = 0; i < len * sizeof(union i40e_rx_desc); i++)
2280 ((volatile char *)rxq->rx_ring)[i] = 0;
2282 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2283 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2284 for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; ++i)
2285 rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
2287 rxq->rx_nb_avail = 0;
2288 rxq->rx_next_avail = 0;
2289 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2290 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2292 rxq->nb_rx_hold = 0;
2293 rxq->pkt_first_seg = NULL;
2294 rxq->pkt_last_seg = NULL;
2298 i40e_tx_queue_release_mbufs(struct i40e_tx_queue *txq)
2302 if (!txq || !txq->sw_ring) {
2303 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL");
2307 for (i = 0; i < txq->nb_tx_desc; i++) {
2308 if (txq->sw_ring[i].mbuf) {
2309 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2310 txq->sw_ring[i].mbuf = NULL;
2316 i40e_reset_tx_queue(struct i40e_tx_queue *txq)
2318 struct i40e_tx_entry *txe;
2319 uint16_t i, prev, size;
2322 PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
2327 size = sizeof(struct i40e_tx_desc) * txq->nb_tx_desc;
2328 for (i = 0; i < size; i++)
2329 ((volatile char *)txq->tx_ring)[i] = 0;
2331 prev = (uint16_t)(txq->nb_tx_desc - 1);
2332 for (i = 0; i < txq->nb_tx_desc; i++) {
2333 volatile struct i40e_tx_desc *txd = &txq->tx_ring[i];
2335 txd->cmd_type_offset_bsz =
2336 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE);
2339 txe[prev].next_id = i;
2343 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2344 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2347 txq->nb_tx_used = 0;
2349 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2350 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2353 /* Init the TX queue in hardware */
2355 i40e_tx_queue_init(struct i40e_tx_queue *txq)
2357 enum i40e_status_code err = I40E_SUCCESS;
2358 struct i40e_vsi *vsi = txq->vsi;
2359 struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2360 uint16_t pf_q = txq->reg_idx;
2361 struct i40e_hmc_obj_txq tx_ctx;
2364 /* clear the context structure first */
2365 memset(&tx_ctx, 0, sizeof(tx_ctx));
2366 tx_ctx.new_context = 1;
2367 tx_ctx.base = txq->tx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2368 tx_ctx.qlen = txq->nb_tx_desc;
2369 tx_ctx.rdylist = rte_le_to_cpu_16(vsi->info.qs_handle[0]);
2370 if (vsi->type == I40E_VSI_FDIR)
2371 tx_ctx.fd_ena = TRUE;
2373 err = i40e_clear_lan_tx_queue_context(hw, pf_q);
2374 if (err != I40E_SUCCESS) {
2375 PMD_DRV_LOG(ERR, "Failure of clean lan tx queue context");
2379 err = i40e_set_lan_tx_queue_context(hw, pf_q, &tx_ctx);
2380 if (err != I40E_SUCCESS) {
2381 PMD_DRV_LOG(ERR, "Failure of set lan tx queue context");
2385 /* Now associate this queue with this PCI function */
2386 qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
2387 qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
2388 I40E_QTX_CTL_PF_INDX_MASK);
2389 I40E_WRITE_REG(hw, I40E_QTX_CTL(pf_q), qtx_ctl);
2390 I40E_WRITE_FLUSH(hw);
2392 txq->qtx_tail = hw->hw_addr + I40E_QTX_TAIL(pf_q);
2398 i40e_alloc_rx_queue_mbufs(struct i40e_rx_queue *rxq)
2400 struct i40e_rx_entry *rxe = rxq->sw_ring;
2404 for (i = 0; i < rxq->nb_rx_desc; i++) {
2405 volatile union i40e_rx_desc *rxd;
2406 struct rte_mbuf *mbuf = rte_rxmbuf_alloc(rxq->mp);
2408 if (unlikely(!mbuf)) {
2409 PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
2413 rte_mbuf_refcnt_set(mbuf, 1);
2415 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
2417 mbuf->port = rxq->port_id;
2420 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf));
2422 rxd = &rxq->rx_ring[i];
2423 rxd->read.pkt_addr = dma_addr;
2424 rxd->read.hdr_addr = dma_addr;
2425 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2426 rxd->read.rsvd1 = 0;
2427 rxd->read.rsvd2 = 0;
2428 #endif /* RTE_LIBRTE_I40E_16BYTE_RX_DESC */
2437 * Calculate the buffer length, and check the jumbo frame
2438 * and maximum packet length.
2441 i40e_rx_queue_config(struct i40e_rx_queue *rxq)
2443 struct i40e_pf *pf = I40E_VSI_TO_PF(rxq->vsi);
2444 struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2445 struct rte_eth_dev_data *data = pf->dev_data;
2446 uint16_t buf_size, len;
2448 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
2449 RTE_PKTMBUF_HEADROOM);
2451 switch (pf->flags & (I40E_FLAG_HEADER_SPLIT_DISABLED |
2452 I40E_FLAG_HEADER_SPLIT_ENABLED)) {
2453 case I40E_FLAG_HEADER_SPLIT_ENABLED: /* Not supported */
2454 rxq->rx_hdr_len = RTE_ALIGN(I40E_RXBUF_SZ_1024,
2455 (1 << I40E_RXQ_CTX_HBUFF_SHIFT));
2456 rxq->rx_buf_len = RTE_ALIGN(I40E_RXBUF_SZ_2048,
2457 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2458 rxq->hs_mode = i40e_header_split_enabled;
2460 case I40E_FLAG_HEADER_SPLIT_DISABLED:
2462 rxq->rx_hdr_len = 0;
2463 rxq->rx_buf_len = RTE_ALIGN(buf_size,
2464 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2465 rxq->hs_mode = i40e_header_split_none;
2469 len = hw->func_caps.rx_buf_chain_len * rxq->rx_buf_len;
2470 rxq->max_pkt_len = RTE_MIN(len, data->dev_conf.rxmode.max_rx_pkt_len);
2471 if (data->dev_conf.rxmode.jumbo_frame == 1) {
2472 if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
2473 rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
2474 PMD_DRV_LOG(ERR, "maximum packet length must "
2475 "be larger than %u and smaller than %u,"
2476 "as jumbo frame is enabled",
2477 (uint32_t)ETHER_MAX_LEN,
2478 (uint32_t)I40E_FRAME_SIZE_MAX);
2479 return I40E_ERR_CONFIG;
2482 if (rxq->max_pkt_len < ETHER_MIN_LEN ||
2483 rxq->max_pkt_len > ETHER_MAX_LEN) {
2484 PMD_DRV_LOG(ERR, "maximum packet length must be "
2485 "larger than %u and smaller than %u, "
2486 "as jumbo frame is disabled",
2487 (uint32_t)ETHER_MIN_LEN,
2488 (uint32_t)ETHER_MAX_LEN);
2489 return I40E_ERR_CONFIG;
2496 /* Init the RX queue in hardware */
2498 i40e_rx_queue_init(struct i40e_rx_queue *rxq)
2500 int err = I40E_SUCCESS;
2501 struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2502 struct rte_eth_dev_data *dev_data = I40E_VSI_TO_DEV_DATA(rxq->vsi);
2503 struct rte_eth_dev *dev = I40E_VSI_TO_ETH_DEV(rxq->vsi);
2504 uint16_t pf_q = rxq->reg_idx;
2506 struct i40e_hmc_obj_rxq rx_ctx;
2508 err = i40e_rx_queue_config(rxq);
2510 PMD_DRV_LOG(ERR, "Failed to config RX queue");
2514 /* Clear the context structure first */
2515 memset(&rx_ctx, 0, sizeof(struct i40e_hmc_obj_rxq));
2516 rx_ctx.dbuff = rxq->rx_buf_len >> I40E_RXQ_CTX_DBUFF_SHIFT;
2517 rx_ctx.hbuff = rxq->rx_hdr_len >> I40E_RXQ_CTX_HBUFF_SHIFT;
2519 rx_ctx.base = rxq->rx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2520 rx_ctx.qlen = rxq->nb_rx_desc;
2521 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2524 rx_ctx.dtype = rxq->hs_mode;
2526 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_ALL;
2528 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_NONE;
2529 rx_ctx.rxmax = rxq->max_pkt_len;
2530 rx_ctx.tphrdesc_ena = 1;
2531 rx_ctx.tphwdesc_ena = 1;
2532 rx_ctx.tphdata_ena = 1;
2533 rx_ctx.tphhead_ena = 1;
2534 rx_ctx.lrxqthresh = 2;
2535 rx_ctx.crcstrip = (rxq->crc_len == 0) ? 1 : 0;
2540 err = i40e_clear_lan_rx_queue_context(hw, pf_q);
2541 if (err != I40E_SUCCESS) {
2542 PMD_DRV_LOG(ERR, "Failed to clear LAN RX queue context");
2545 err = i40e_set_lan_rx_queue_context(hw, pf_q, &rx_ctx);
2546 if (err != I40E_SUCCESS) {
2547 PMD_DRV_LOG(ERR, "Failed to set LAN RX queue context");
2551 rxq->qrx_tail = hw->hw_addr + I40E_QRX_TAIL(pf_q);
2553 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
2554 RTE_PKTMBUF_HEADROOM);
2556 /* Check if scattered RX needs to be used. */
2557 if ((rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
2558 dev_data->scattered_rx = 1;
2559 dev->rx_pkt_burst = i40e_recv_scattered_pkts;
2562 /* Init the RX tail regieter. */
2563 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
2569 i40e_dev_clear_queues(struct rte_eth_dev *dev)
2573 PMD_INIT_FUNC_TRACE();
2575 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2576 i40e_tx_queue_release_mbufs(dev->data->tx_queues[i]);
2577 i40e_reset_tx_queue(dev->data->tx_queues[i]);
2580 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2581 i40e_rx_queue_release_mbufs(dev->data->rx_queues[i]);
2582 i40e_reset_rx_queue(dev->data->rx_queues[i]);
2586 #define I40E_FDIR_NUM_TX_DESC I40E_MIN_RING_DESC
2587 #define I40E_FDIR_NUM_RX_DESC I40E_MIN_RING_DESC
2589 enum i40e_status_code
2590 i40e_fdir_setup_tx_resources(struct i40e_pf *pf)
2592 struct i40e_tx_queue *txq;
2593 const struct rte_memzone *tz = NULL;
2595 struct rte_eth_dev *dev = pf->adapter->eth_dev;
2598 PMD_DRV_LOG(ERR, "PF is not available");
2599 return I40E_ERR_BAD_PTR;
2602 /* Allocate the TX queue data structure. */
2603 txq = rte_zmalloc_socket("i40e fdir tx queue",
2604 sizeof(struct i40e_tx_queue),
2605 RTE_CACHE_LINE_SIZE,
2608 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2609 "tx queue structure.");
2610 return I40E_ERR_NO_MEMORY;
2613 /* Allocate TX hardware ring descriptors. */
2614 ring_size = sizeof(struct i40e_tx_desc) * I40E_FDIR_NUM_TX_DESC;
2615 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2617 tz = i40e_ring_dma_zone_reserve(dev,
2623 i40e_dev_tx_queue_release(txq);
2624 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX.");
2625 return I40E_ERR_NO_MEMORY;
2628 txq->nb_tx_desc = I40E_FDIR_NUM_TX_DESC;
2629 txq->queue_id = I40E_FDIR_QUEUE_ID;
2630 txq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2631 txq->vsi = pf->fdir.fdir_vsi;
2633 #ifdef RTE_LIBRTE_XEN_DOM0
2634 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2636 txq->tx_ring_phys_addr = (uint64_t)tz->phys_addr;
2638 txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2640 * don't need to allocate software ring and reset for the fdir
2641 * program queue just set the queue has been configured.
2646 return I40E_SUCCESS;
2649 enum i40e_status_code
2650 i40e_fdir_setup_rx_resources(struct i40e_pf *pf)
2652 struct i40e_rx_queue *rxq;
2653 const struct rte_memzone *rz = NULL;
2655 struct rte_eth_dev *dev = pf->adapter->eth_dev;
2658 PMD_DRV_LOG(ERR, "PF is not available");
2659 return I40E_ERR_BAD_PTR;
2662 /* Allocate the RX queue data structure. */
2663 rxq = rte_zmalloc_socket("i40e fdir rx queue",
2664 sizeof(struct i40e_rx_queue),
2665 RTE_CACHE_LINE_SIZE,
2668 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2669 "rx queue structure.");
2670 return I40E_ERR_NO_MEMORY;
2673 /* Allocate RX hardware ring descriptors. */
2674 ring_size = sizeof(union i40e_rx_desc) * I40E_FDIR_NUM_RX_DESC;
2675 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2677 rz = i40e_ring_dma_zone_reserve(dev,
2683 i40e_dev_rx_queue_release(rxq);
2684 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX.");
2685 return I40E_ERR_NO_MEMORY;
2688 rxq->nb_rx_desc = I40E_FDIR_NUM_RX_DESC;
2689 rxq->queue_id = I40E_FDIR_QUEUE_ID;
2690 rxq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2691 rxq->vsi = pf->fdir.fdir_vsi;
2693 #ifdef RTE_LIBRTE_XEN_DOM0
2694 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
2696 rxq->rx_ring_phys_addr = (uint64_t)rz->phys_addr;
2698 rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
2701 * Don't need to allocate software ring and reset for the fdir
2702 * rx queue, just set the queue has been configured.
2707 return I40E_SUCCESS;