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,
98 i40e_rxd_to_vlan_tci(struct rte_mbuf *mb, volatile union i40e_rx_desc *rxdp)
100 if (rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
101 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) {
102 mb->ol_flags |= PKT_RX_VLAN_PKT;
104 rte_le_to_cpu_16(rxdp->wb.qword0.lo_dword.l2tag1);
105 PMD_RX_LOG(DEBUG, "Descriptor l2tag1: %u",
106 rte_le_to_cpu_16(rxdp->wb.qword0.lo_dword.l2tag1));
110 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
111 if (rte_le_to_cpu_16(rxdp->wb.qword2.ext_status) &
112 (1 << I40E_RX_DESC_EXT_STATUS_L2TAG2P_SHIFT)) {
113 mb->ol_flags |= PKT_RX_QINQ_PKT;
114 mb->vlan_tci_outer = mb->vlan_tci;
115 mb->vlan_tci = rte_le_to_cpu_16(rxdp->wb.qword2.l2tag2_2);
116 PMD_RX_LOG(DEBUG, "Descriptor l2tag2_1: %u, l2tag2_2: %u",
117 rte_le_to_cpu_16(rxdp->wb.qword2.l2tag2_1),
118 rte_le_to_cpu_16(rxdp->wb.qword2.l2tag2_2));
120 mb->vlan_tci_outer = 0;
123 PMD_RX_LOG(DEBUG, "Mbuf vlan_tci: %u, vlan_tci_outer: %u",
124 mb->vlan_tci, mb->vlan_tci_outer);
127 /* Translate the rx descriptor status to pkt flags */
128 static inline uint64_t
129 i40e_rxd_status_to_pkt_flags(uint64_t qword)
133 /* Check if RSS_HASH */
134 flags = (((qword >> I40E_RX_DESC_STATUS_FLTSTAT_SHIFT) &
135 I40E_RX_DESC_FLTSTAT_RSS_HASH) ==
136 I40E_RX_DESC_FLTSTAT_RSS_HASH) ? PKT_RX_RSS_HASH : 0;
138 /* Check if FDIR Match */
139 flags |= (qword & (1 << I40E_RX_DESC_STATUS_FLM_SHIFT) ?
145 static inline uint64_t
146 i40e_rxd_error_to_pkt_flags(uint64_t qword)
149 uint64_t error_bits = (qword >> I40E_RXD_QW1_ERROR_SHIFT);
151 #define I40E_RX_ERR_BITS 0x3f
152 if (likely((error_bits & I40E_RX_ERR_BITS) == 0))
154 /* If RXE bit set, all other status bits are meaningless */
155 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
156 flags |= PKT_RX_MAC_ERR;
160 /* If RECIPE bit set, all other status indications should be ignored */
161 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_RECIPE_SHIFT))) {
162 flags |= PKT_RX_RECIP_ERR;
165 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT)))
166 flags |= PKT_RX_HBUF_OVERFLOW;
167 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_IPE_SHIFT)))
168 flags |= PKT_RX_IP_CKSUM_BAD;
169 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_L4E_SHIFT)))
170 flags |= PKT_RX_L4_CKSUM_BAD;
171 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT)))
172 flags |= PKT_RX_EIP_CKSUM_BAD;
173 if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_OVERSIZE_SHIFT)))
174 flags |= PKT_RX_OVERSIZE;
179 /* Translate pkt types to pkt flags */
180 static inline uint64_t
181 i40e_rxd_ptype_to_pkt_flags(uint64_t qword)
183 uint8_t ptype = (uint8_t)((qword & I40E_RXD_QW1_PTYPE_MASK) >>
184 I40E_RXD_QW1_PTYPE_SHIFT);
185 static const uint64_t ip_ptype_map[I40E_MAX_PKT_TYPE] = {
208 PKT_RX_IPV4_HDR, /* PTYPE 22 */
209 PKT_RX_IPV4_HDR, /* PTYPE 23 */
210 PKT_RX_IPV4_HDR, /* PTYPE 24 */
212 PKT_RX_IPV4_HDR, /* PTYPE 26 */
213 PKT_RX_IPV4_HDR, /* PTYPE 27 */
214 PKT_RX_IPV4_HDR, /* PTYPE 28 */
215 PKT_RX_IPV4_HDR_EXT, /* PTYPE 29 */
216 PKT_RX_IPV4_HDR_EXT, /* PTYPE 30 */
217 PKT_RX_IPV4_HDR_EXT, /* PTYPE 31 */
219 PKT_RX_IPV4_HDR_EXT, /* PTYPE 33 */
220 PKT_RX_IPV4_HDR_EXT, /* PTYPE 34 */
221 PKT_RX_IPV4_HDR_EXT, /* PTYPE 35 */
222 PKT_RX_IPV4_HDR_EXT, /* PTYPE 36 */
223 PKT_RX_IPV4_HDR_EXT, /* PTYPE 37 */
224 PKT_RX_IPV4_HDR_EXT, /* PTYPE 38 */
226 PKT_RX_IPV4_HDR_EXT, /* PTYPE 40 */
227 PKT_RX_IPV4_HDR_EXT, /* PTYPE 41 */
228 PKT_RX_IPV4_HDR_EXT, /* PTYPE 42 */
229 PKT_RX_IPV4_HDR_EXT, /* PTYPE 43 */
230 PKT_RX_IPV4_HDR_EXT, /* PTYPE 44 */
231 PKT_RX_IPV4_HDR_EXT, /* PTYPE 45 */
232 PKT_RX_IPV4_HDR_EXT, /* PTYPE 46 */
234 PKT_RX_IPV4_HDR_EXT, /* PTYPE 48 */
235 PKT_RX_IPV4_HDR_EXT, /* PTYPE 49 */
236 PKT_RX_IPV4_HDR_EXT, /* PTYPE 50 */
237 PKT_RX_IPV4_HDR_EXT, /* PTYPE 51 */
238 PKT_RX_IPV4_HDR_EXT, /* PTYPE 52 */
239 PKT_RX_IPV4_HDR_EXT, /* PTYPE 53 */
241 PKT_RX_IPV4_HDR_EXT, /* PTYPE 55 */
242 PKT_RX_IPV4_HDR_EXT, /* PTYPE 56 */
243 PKT_RX_IPV4_HDR_EXT, /* PTYPE 57 */
244 PKT_RX_IPV4_HDR_EXT, /* PTYPE 58 */
245 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 59 */
246 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 60 */
247 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 61 */
249 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 63 */
250 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 64 */
251 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 65 */
252 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 66 */
253 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 67 */
254 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 68 */
256 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 70 */
257 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 71 */
258 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 72 */
259 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 73 */
260 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 74 */
261 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 75 */
262 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 76 */
264 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 78 */
265 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 79 */
266 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 80 */
267 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 81 */
268 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 82 */
269 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 83 */
271 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 85 */
272 PKT_RX_TUNNEL_IPV4_HDR, /* PTYPE 86 */
273 PKT_RX_IPV4_HDR_EXT, /* PTYPE 87 */
274 PKT_RX_IPV6_HDR, /* PTYPE 88 */
275 PKT_RX_IPV6_HDR, /* PTYPE 89 */
276 PKT_RX_IPV6_HDR, /* PTYPE 90 */
278 PKT_RX_IPV6_HDR, /* PTYPE 92 */
279 PKT_RX_IPV6_HDR, /* PTYPE 93 */
280 PKT_RX_IPV6_HDR, /* PTYPE 94 */
281 PKT_RX_IPV6_HDR_EXT, /* PTYPE 95 */
282 PKT_RX_IPV6_HDR_EXT, /* PTYPE 96 */
283 PKT_RX_IPV6_HDR_EXT, /* PTYPE 97 */
285 PKT_RX_IPV6_HDR_EXT, /* PTYPE 99 */
286 PKT_RX_IPV6_HDR_EXT, /* PTYPE 100 */
287 PKT_RX_IPV6_HDR_EXT, /* PTYPE 101 */
288 PKT_RX_IPV6_HDR_EXT, /* PTYPE 102 */
289 PKT_RX_IPV6_HDR_EXT, /* PTYPE 103 */
290 PKT_RX_IPV6_HDR_EXT, /* PTYPE 104 */
292 PKT_RX_IPV6_HDR_EXT, /* PTYPE 106 */
293 PKT_RX_IPV6_HDR_EXT, /* PTYPE 107 */
294 PKT_RX_IPV6_HDR_EXT, /* PTYPE 108 */
295 PKT_RX_IPV6_HDR_EXT, /* PTYPE 109 */
296 PKT_RX_IPV6_HDR_EXT, /* PTYPE 110 */
297 PKT_RX_IPV6_HDR_EXT, /* PTYPE 111 */
298 PKT_RX_IPV6_HDR_EXT, /* PTYPE 112 */
300 PKT_RX_IPV6_HDR_EXT, /* PTYPE 114 */
301 PKT_RX_IPV6_HDR_EXT, /* PTYPE 115 */
302 PKT_RX_IPV6_HDR_EXT, /* PTYPE 116 */
303 PKT_RX_IPV6_HDR_EXT, /* PTYPE 117 */
304 PKT_RX_IPV6_HDR_EXT, /* PTYPE 118 */
305 PKT_RX_IPV6_HDR_EXT, /* PTYPE 119 */
307 PKT_RX_IPV6_HDR_EXT, /* PTYPE 121 */
308 PKT_RX_IPV6_HDR_EXT, /* PTYPE 122 */
309 PKT_RX_IPV6_HDR_EXT, /* PTYPE 123 */
310 PKT_RX_IPV6_HDR_EXT, /* PTYPE 124 */
311 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 125 */
312 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 126 */
313 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 127 */
315 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 129 */
316 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 130 */
317 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 131 */
318 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 132 */
319 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 133 */
320 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 134 */
322 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 136 */
323 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 137 */
324 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 138 */
325 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 139 */
326 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 140 */
327 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 141 */
328 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 142 */
330 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 144 */
331 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 145 */
332 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 146 */
333 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 147 */
334 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 148 */
335 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 149 */
337 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 151 */
338 PKT_RX_TUNNEL_IPV6_HDR, /* PTYPE 152 */
339 PKT_RX_IPV6_HDR_EXT, /* PTYPE 153 */
444 return ip_ptype_map[ptype];
447 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_MASK 0x03
448 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_FD_ID 0x01
449 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_FLEX 0x02
450 #define I40E_RX_DESC_EXT_STATUS_FLEXBL_MASK 0x03
451 #define I40E_RX_DESC_EXT_STATUS_FLEXBL_FLEX 0x01
453 static inline uint64_t
454 i40e_rxd_build_fdir(volatile union i40e_rx_desc *rxdp, struct rte_mbuf *mb)
457 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
458 uint16_t flexbh, flexbl;
460 flexbh = (rte_le_to_cpu_32(rxdp->wb.qword2.ext_status) >>
461 I40E_RX_DESC_EXT_STATUS_FLEXBH_SHIFT) &
462 I40E_RX_DESC_EXT_STATUS_FLEXBH_MASK;
463 flexbl = (rte_le_to_cpu_32(rxdp->wb.qword2.ext_status) >>
464 I40E_RX_DESC_EXT_STATUS_FLEXBL_SHIFT) &
465 I40E_RX_DESC_EXT_STATUS_FLEXBL_MASK;
468 if (flexbh == I40E_RX_DESC_EXT_STATUS_FLEXBH_FD_ID) {
470 rte_le_to_cpu_32(rxdp->wb.qword3.hi_dword.fd_id);
471 flags |= PKT_RX_FDIR_ID;
472 } else if (flexbh == I40E_RX_DESC_EXT_STATUS_FLEXBH_FLEX) {
474 rte_le_to_cpu_32(rxdp->wb.qword3.hi_dword.flex_bytes_hi);
475 flags |= PKT_RX_FDIR_FLX;
477 if (flexbl == I40E_RX_DESC_EXT_STATUS_FLEXBL_FLEX) {
479 rte_le_to_cpu_32(rxdp->wb.qword3.lo_dword.flex_bytes_lo);
480 flags |= PKT_RX_FDIR_FLX;
484 rte_le_to_cpu_32(rxdp->wb.qword0.hi_dword.fd_id);
485 flags |= PKT_RX_FDIR_ID;
490 i40e_txd_enable_checksum(uint64_t ol_flags,
493 union i40e_tx_offload tx_offload,
494 uint32_t *cd_tunneling)
496 /* UDP tunneling packet TX checksum offload */
497 if (ol_flags & PKT_TX_OUTER_IP_CKSUM) {
499 *td_offset |= (tx_offload.outer_l2_len >> 1)
500 << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
502 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
503 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
504 else if (ol_flags & PKT_TX_OUTER_IPV4)
505 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
506 else if (ol_flags & PKT_TX_OUTER_IPV6)
507 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
509 /* Now set the ctx descriptor fields */
510 *cd_tunneling |= (tx_offload.outer_l3_len >> 2) <<
511 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
512 (tx_offload.l2_len >> 1) <<
513 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
516 *td_offset |= (tx_offload.l2_len >> 1)
517 << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
519 /* Enable L3 checksum offloads */
520 if (ol_flags & PKT_TX_IP_CKSUM) {
521 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
522 *td_offset |= (tx_offload.l3_len >> 2)
523 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
524 } else if (ol_flags & PKT_TX_IPV4) {
525 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
526 *td_offset |= (tx_offload.l3_len >> 2)
527 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
528 } else if (ol_flags & PKT_TX_IPV6) {
529 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
530 *td_offset |= (tx_offload.l3_len >> 2)
531 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
534 if (ol_flags & PKT_TX_TCP_SEG) {
535 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
536 *td_offset |= (tx_offload.l4_len >> 2)
537 << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
541 /* Enable L4 checksum offloads */
542 switch (ol_flags & PKT_TX_L4_MASK) {
543 case PKT_TX_TCP_CKSUM:
544 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
545 *td_offset |= (sizeof(struct tcp_hdr) >> 2) <<
546 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
548 case PKT_TX_SCTP_CKSUM:
549 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
550 *td_offset |= (sizeof(struct sctp_hdr) >> 2) <<
551 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
553 case PKT_TX_UDP_CKSUM:
554 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
555 *td_offset |= (sizeof(struct udp_hdr) >> 2) <<
556 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
563 static inline struct rte_mbuf *
564 rte_rxmbuf_alloc(struct rte_mempool *mp)
568 m = __rte_mbuf_raw_alloc(mp);
569 __rte_mbuf_sanity_check_raw(m, 0);
574 /* Construct the tx flags */
575 static inline uint64_t
576 i40e_build_ctob(uint32_t td_cmd,
581 return rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DATA |
582 ((uint64_t)td_cmd << I40E_TXD_QW1_CMD_SHIFT) |
583 ((uint64_t)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
584 ((uint64_t)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
585 ((uint64_t)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT));
589 i40e_xmit_cleanup(struct i40e_tx_queue *txq)
591 struct i40e_tx_entry *sw_ring = txq->sw_ring;
592 volatile struct i40e_tx_desc *txd = txq->tx_ring;
593 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
594 uint16_t nb_tx_desc = txq->nb_tx_desc;
595 uint16_t desc_to_clean_to;
596 uint16_t nb_tx_to_clean;
598 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
599 if (desc_to_clean_to >= nb_tx_desc)
600 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
602 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
603 if (!(txd[desc_to_clean_to].cmd_type_offset_bsz &
604 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE))) {
605 PMD_TX_FREE_LOG(DEBUG, "TX descriptor %4u is not done "
606 "(port=%d queue=%d)", desc_to_clean_to,
607 txq->port_id, txq->queue_id);
611 if (last_desc_cleaned > desc_to_clean_to)
612 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
615 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
618 txd[desc_to_clean_to].cmd_type_offset_bsz = 0;
620 txq->last_desc_cleaned = desc_to_clean_to;
621 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
627 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
628 check_rx_burst_bulk_alloc_preconditions(struct i40e_rx_queue *rxq)
630 check_rx_burst_bulk_alloc_preconditions(__rte_unused struct i40e_rx_queue *rxq)
635 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
636 if (!(rxq->rx_free_thresh >= RTE_PMD_I40E_RX_MAX_BURST)) {
637 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
638 "rxq->rx_free_thresh=%d, "
639 "RTE_PMD_I40E_RX_MAX_BURST=%d",
640 rxq->rx_free_thresh, RTE_PMD_I40E_RX_MAX_BURST);
642 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
643 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
644 "rxq->rx_free_thresh=%d, "
645 "rxq->nb_rx_desc=%d",
646 rxq->rx_free_thresh, rxq->nb_rx_desc);
648 } else if (rxq->nb_rx_desc % rxq->rx_free_thresh != 0) {
649 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
650 "rxq->nb_rx_desc=%d, "
651 "rxq->rx_free_thresh=%d",
652 rxq->nb_rx_desc, rxq->rx_free_thresh);
654 } else if (!(rxq->nb_rx_desc < (I40E_MAX_RING_DESC -
655 RTE_PMD_I40E_RX_MAX_BURST))) {
656 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
657 "rxq->nb_rx_desc=%d, "
658 "I40E_MAX_RING_DESC=%d, "
659 "RTE_PMD_I40E_RX_MAX_BURST=%d",
660 rxq->nb_rx_desc, I40E_MAX_RING_DESC,
661 RTE_PMD_I40E_RX_MAX_BURST);
671 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
672 #define I40E_LOOK_AHEAD 8
673 #if (I40E_LOOK_AHEAD != 8)
674 #error "PMD I40E: I40E_LOOK_AHEAD must be 8\n"
677 i40e_rx_scan_hw_ring(struct i40e_rx_queue *rxq)
679 volatile union i40e_rx_desc *rxdp;
680 struct i40e_rx_entry *rxep;
685 int32_t s[I40E_LOOK_AHEAD], nb_dd;
686 int32_t i, j, nb_rx = 0;
689 rxdp = &rxq->rx_ring[rxq->rx_tail];
690 rxep = &rxq->sw_ring[rxq->rx_tail];
692 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
693 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
694 I40E_RXD_QW1_STATUS_SHIFT;
696 /* Make sure there is at least 1 packet to receive */
697 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
701 * Scan LOOK_AHEAD descriptors at a time to determine which
702 * descriptors reference packets that are ready to be received.
704 for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; i+=I40E_LOOK_AHEAD,
705 rxdp += I40E_LOOK_AHEAD, rxep += I40E_LOOK_AHEAD) {
706 /* Read desc statuses backwards to avoid race condition */
707 for (j = I40E_LOOK_AHEAD - 1; j >= 0; j--) {
708 qword1 = rte_le_to_cpu_64(\
709 rxdp[j].wb.qword1.status_error_len);
710 s[j] = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
711 I40E_RXD_QW1_STATUS_SHIFT;
714 /* Compute how many status bits were set */
715 for (j = 0, nb_dd = 0; j < I40E_LOOK_AHEAD; j++)
716 nb_dd += s[j] & (1 << I40E_RX_DESC_STATUS_DD_SHIFT);
720 /* Translate descriptor info to mbuf parameters */
721 for (j = 0; j < nb_dd; j++) {
723 qword1 = rte_le_to_cpu_64(\
724 rxdp[j].wb.qword1.status_error_len);
725 pkt_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
726 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
727 mb->data_len = pkt_len;
728 mb->pkt_len = pkt_len;
730 i40e_rxd_to_vlan_tci(mb, &rxdp[j]);
731 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
732 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
733 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
735 mb->packet_type = (uint16_t)((qword1 &
736 I40E_RXD_QW1_PTYPE_MASK) >>
737 I40E_RXD_QW1_PTYPE_SHIFT);
738 if (pkt_flags & PKT_RX_RSS_HASH)
739 mb->hash.rss = rte_le_to_cpu_32(\
740 rxdp[j].wb.qword0.hi_dword.rss);
741 if (pkt_flags & PKT_RX_FDIR)
742 pkt_flags |= i40e_rxd_build_fdir(&rxdp[j], mb);
744 mb->ol_flags |= pkt_flags;
747 for (j = 0; j < I40E_LOOK_AHEAD; j++)
748 rxq->rx_stage[i + j] = rxep[j].mbuf;
750 if (nb_dd != I40E_LOOK_AHEAD)
754 /* Clear software ring entries */
755 for (i = 0; i < nb_rx; i++)
756 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
761 static inline uint16_t
762 i40e_rx_fill_from_stage(struct i40e_rx_queue *rxq,
763 struct rte_mbuf **rx_pkts,
767 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
769 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
771 for (i = 0; i < nb_pkts; i++)
772 rx_pkts[i] = stage[i];
774 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
775 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
781 i40e_rx_alloc_bufs(struct i40e_rx_queue *rxq)
783 volatile union i40e_rx_desc *rxdp;
784 struct i40e_rx_entry *rxep;
786 uint16_t alloc_idx, i;
790 /* Allocate buffers in bulk */
791 alloc_idx = (uint16_t)(rxq->rx_free_trigger -
792 (rxq->rx_free_thresh - 1));
793 rxep = &(rxq->sw_ring[alloc_idx]);
794 diag = rte_mempool_get_bulk(rxq->mp, (void *)rxep,
795 rxq->rx_free_thresh);
796 if (unlikely(diag != 0)) {
797 PMD_DRV_LOG(ERR, "Failed to get mbufs in bulk");
801 rxdp = &rxq->rx_ring[alloc_idx];
802 for (i = 0; i < rxq->rx_free_thresh; i++) {
804 rte_mbuf_refcnt_set(mb, 1);
806 mb->data_off = RTE_PKTMBUF_HEADROOM;
808 mb->port = rxq->port_id;
809 dma_addr = rte_cpu_to_le_64(\
810 RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb));
811 rxdp[i].read.hdr_addr = dma_addr;
812 rxdp[i].read.pkt_addr = dma_addr;
815 /* Update rx tail regsiter */
817 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->rx_free_trigger);
819 rxq->rx_free_trigger =
820 (uint16_t)(rxq->rx_free_trigger + rxq->rx_free_thresh);
821 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
822 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
827 static inline uint16_t
828 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
830 struct i40e_rx_queue *rxq = (struct i40e_rx_queue *)rx_queue;
836 if (rxq->rx_nb_avail)
837 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
839 nb_rx = (uint16_t)i40e_rx_scan_hw_ring(rxq);
840 rxq->rx_next_avail = 0;
841 rxq->rx_nb_avail = nb_rx;
842 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
844 if (rxq->rx_tail > rxq->rx_free_trigger) {
845 if (i40e_rx_alloc_bufs(rxq) != 0) {
848 PMD_RX_LOG(DEBUG, "Rx mbuf alloc failed for "
849 "port_id=%u, queue_id=%u",
850 rxq->port_id, rxq->queue_id);
851 rxq->rx_nb_avail = 0;
852 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
853 for (i = 0, j = rxq->rx_tail; i < nb_rx; i++, j++)
854 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
860 if (rxq->rx_tail >= rxq->nb_rx_desc)
863 if (rxq->rx_nb_avail)
864 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
870 i40e_recv_pkts_bulk_alloc(void *rx_queue,
871 struct rte_mbuf **rx_pkts,
874 uint16_t nb_rx = 0, n, count;
876 if (unlikely(nb_pkts == 0))
879 if (likely(nb_pkts <= RTE_PMD_I40E_RX_MAX_BURST))
880 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
883 n = RTE_MIN(nb_pkts, RTE_PMD_I40E_RX_MAX_BURST);
884 count = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
885 nb_rx = (uint16_t)(nb_rx + count);
886 nb_pkts = (uint16_t)(nb_pkts - count);
893 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
896 i40e_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
898 struct i40e_rx_queue *rxq;
899 volatile union i40e_rx_desc *rx_ring;
900 volatile union i40e_rx_desc *rxdp;
901 union i40e_rx_desc rxd;
902 struct i40e_rx_entry *sw_ring;
903 struct i40e_rx_entry *rxe;
904 struct rte_mbuf *rxm;
905 struct rte_mbuf *nmb;
909 uint16_t rx_packet_len;
910 uint16_t rx_id, nb_hold;
917 rx_id = rxq->rx_tail;
918 rx_ring = rxq->rx_ring;
919 sw_ring = rxq->sw_ring;
921 while (nb_rx < nb_pkts) {
922 rxdp = &rx_ring[rx_id];
923 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
924 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK)
925 >> I40E_RXD_QW1_STATUS_SHIFT;
926 /* Check the DD bit first */
927 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
930 nmb = rte_rxmbuf_alloc(rxq->mp);
936 rxe = &sw_ring[rx_id];
938 if (unlikely(rx_id == rxq->nb_rx_desc))
941 /* Prefetch next mbuf */
942 rte_prefetch0(sw_ring[rx_id].mbuf);
945 * When next RX descriptor is on a cache line boundary,
946 * prefetch the next 4 RX descriptors and next 8 pointers
949 if ((rx_id & 0x3) == 0) {
950 rte_prefetch0(&rx_ring[rx_id]);
951 rte_prefetch0(&sw_ring[rx_id]);
956 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
957 rxdp->read.hdr_addr = dma_addr;
958 rxdp->read.pkt_addr = dma_addr;
960 rx_packet_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
961 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
963 rxm->data_off = RTE_PKTMBUF_HEADROOM;
964 rte_prefetch0(RTE_PTR_ADD(rxm->buf_addr, RTE_PKTMBUF_HEADROOM));
967 rxm->pkt_len = rx_packet_len;
968 rxm->data_len = rx_packet_len;
969 rxm->port = rxq->port_id;
971 i40e_rxd_to_vlan_tci(rxm, &rxd);
972 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
973 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
974 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
975 rxm->packet_type = (uint16_t)((qword1 & I40E_RXD_QW1_PTYPE_MASK) >>
976 I40E_RXD_QW1_PTYPE_SHIFT);
977 if (pkt_flags & PKT_RX_RSS_HASH)
979 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
980 if (pkt_flags & PKT_RX_FDIR)
981 pkt_flags |= i40e_rxd_build_fdir(&rxd, rxm);
983 rxm->ol_flags |= pkt_flags;
985 rx_pkts[nb_rx++] = rxm;
987 rxq->rx_tail = rx_id;
990 * If the number of free RX descriptors is greater than the RX free
991 * threshold of the queue, advance the receive tail register of queue.
992 * Update that register with the value of the last processed RX
993 * descriptor minus 1.
995 nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
996 if (nb_hold > rxq->rx_free_thresh) {
997 rx_id = (uint16_t) ((rx_id == 0) ?
998 (rxq->nb_rx_desc - 1) : (rx_id - 1));
999 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
1002 rxq->nb_rx_hold = nb_hold;
1008 i40e_recv_scattered_pkts(void *rx_queue,
1009 struct rte_mbuf **rx_pkts,
1012 struct i40e_rx_queue *rxq = rx_queue;
1013 volatile union i40e_rx_desc *rx_ring = rxq->rx_ring;
1014 volatile union i40e_rx_desc *rxdp;
1015 union i40e_rx_desc rxd;
1016 struct i40e_rx_entry *sw_ring = rxq->sw_ring;
1017 struct i40e_rx_entry *rxe;
1018 struct rte_mbuf *first_seg = rxq->pkt_first_seg;
1019 struct rte_mbuf *last_seg = rxq->pkt_last_seg;
1020 struct rte_mbuf *nmb, *rxm;
1021 uint16_t rx_id = rxq->rx_tail;
1022 uint16_t nb_rx = 0, nb_hold = 0, rx_packet_len;
1028 while (nb_rx < nb_pkts) {
1029 rxdp = &rx_ring[rx_id];
1030 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
1031 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
1032 I40E_RXD_QW1_STATUS_SHIFT;
1033 /* Check the DD bit */
1034 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
1037 nmb = rte_rxmbuf_alloc(rxq->mp);
1042 rxe = &sw_ring[rx_id];
1044 if (rx_id == rxq->nb_rx_desc)
1047 /* Prefetch next mbuf */
1048 rte_prefetch0(sw_ring[rx_id].mbuf);
1051 * When next RX descriptor is on a cache line boundary,
1052 * prefetch the next 4 RX descriptors and next 8 pointers
1055 if ((rx_id & 0x3) == 0) {
1056 rte_prefetch0(&rx_ring[rx_id]);
1057 rte_prefetch0(&sw_ring[rx_id]);
1063 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1065 /* Set data buffer address and data length of the mbuf */
1066 rxdp->read.hdr_addr = dma_addr;
1067 rxdp->read.pkt_addr = dma_addr;
1068 rx_packet_len = (qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1069 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1070 rxm->data_len = rx_packet_len;
1071 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1074 * If this is the first buffer of the received packet, set the
1075 * pointer to the first mbuf of the packet and initialize its
1076 * context. Otherwise, update the total length and the number
1077 * of segments of the current scattered packet, and update the
1078 * pointer to the last mbuf of the current packet.
1082 first_seg->nb_segs = 1;
1083 first_seg->pkt_len = rx_packet_len;
1085 first_seg->pkt_len =
1086 (uint16_t)(first_seg->pkt_len +
1088 first_seg->nb_segs++;
1089 last_seg->next = rxm;
1093 * If this is not the last buffer of the received packet,
1094 * update the pointer to the last mbuf of the current scattered
1095 * packet and continue to parse the RX ring.
1097 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT))) {
1103 * This is the last buffer of the received packet. If the CRC
1104 * is not stripped by the hardware:
1105 * - Subtract the CRC length from the total packet length.
1106 * - If the last buffer only contains the whole CRC or a part
1107 * of it, free the mbuf associated to the last buffer. If part
1108 * of the CRC is also contained in the previous mbuf, subtract
1109 * the length of that CRC part from the data length of the
1113 if (unlikely(rxq->crc_len > 0)) {
1114 first_seg->pkt_len -= ETHER_CRC_LEN;
1115 if (rx_packet_len <= ETHER_CRC_LEN) {
1116 rte_pktmbuf_free_seg(rxm);
1117 first_seg->nb_segs--;
1118 last_seg->data_len =
1119 (uint16_t)(last_seg->data_len -
1120 (ETHER_CRC_LEN - rx_packet_len));
1121 last_seg->next = NULL;
1123 rxm->data_len = (uint16_t)(rx_packet_len -
1127 first_seg->port = rxq->port_id;
1128 first_seg->ol_flags = 0;
1129 i40e_rxd_to_vlan_tci(first_seg, &rxd);
1130 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
1131 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
1132 pkt_flags |= i40e_rxd_ptype_to_pkt_flags(qword1);
1133 first_seg->packet_type = (uint16_t)((qword1 &
1134 I40E_RXD_QW1_PTYPE_MASK) >>
1135 I40E_RXD_QW1_PTYPE_SHIFT);
1136 if (pkt_flags & PKT_RX_RSS_HASH)
1138 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
1139 if (pkt_flags & PKT_RX_FDIR)
1140 pkt_flags |= i40e_rxd_build_fdir(&rxd, rxm);
1142 first_seg->ol_flags |= pkt_flags;
1144 /* Prefetch data of first segment, if configured to do so. */
1145 rte_prefetch0(RTE_PTR_ADD(first_seg->buf_addr,
1146 first_seg->data_off));
1147 rx_pkts[nb_rx++] = first_seg;
1151 /* Record index of the next RX descriptor to probe. */
1152 rxq->rx_tail = rx_id;
1153 rxq->pkt_first_seg = first_seg;
1154 rxq->pkt_last_seg = last_seg;
1157 * If the number of free RX descriptors is greater than the RX free
1158 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1159 * register. Update the RDT with the value of the last processed RX
1160 * descriptor minus 1, to guarantee that the RDT register is never
1161 * equal to the RDH register, which creates a "full" ring situtation
1162 * from the hardware point of view.
1164 nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
1165 if (nb_hold > rxq->rx_free_thresh) {
1166 rx_id = (uint16_t)(rx_id == 0 ?
1167 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1168 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
1171 rxq->nb_rx_hold = nb_hold;
1176 /* Check if the context descriptor is needed for TX offloading */
1177 static inline uint16_t
1178 i40e_calc_context_desc(uint64_t flags)
1180 static uint64_t mask = PKT_TX_OUTER_IP_CKSUM |
1184 #ifdef RTE_LIBRTE_IEEE1588
1185 mask |= PKT_TX_IEEE1588_TMST;
1188 return ((flags & mask) ? 1 : 0);
1191 /* set i40e TSO context descriptor */
1192 static inline uint64_t
1193 i40e_set_tso_ctx(struct rte_mbuf *mbuf, union i40e_tx_offload tx_offload)
1195 uint64_t ctx_desc = 0;
1196 uint32_t cd_cmd, hdr_len, cd_tso_len;
1198 if (!tx_offload.l4_len) {
1199 PMD_DRV_LOG(DEBUG, "L4 length set to 0");
1204 * in case of tunneling packet, the outer_l2_len and
1205 * outer_l3_len must be 0.
1207 hdr_len = tx_offload.outer_l2_len +
1208 tx_offload.outer_l3_len +
1213 cd_cmd = I40E_TX_CTX_DESC_TSO;
1214 cd_tso_len = mbuf->pkt_len - hdr_len;
1215 ctx_desc |= ((uint64_t)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1216 ((uint64_t)cd_tso_len <<
1217 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1218 ((uint64_t)mbuf->tso_segsz <<
1219 I40E_TXD_CTX_QW1_MSS_SHIFT);
1225 i40e_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1227 struct i40e_tx_queue *txq;
1228 struct i40e_tx_entry *sw_ring;
1229 struct i40e_tx_entry *txe, *txn;
1230 volatile struct i40e_tx_desc *txd;
1231 volatile struct i40e_tx_desc *txr;
1232 struct rte_mbuf *tx_pkt;
1233 struct rte_mbuf *m_seg;
1234 uint32_t cd_tunneling_params;
1246 uint64_t buf_dma_addr;
1247 union i40e_tx_offload tx_offload = {0};
1250 sw_ring = txq->sw_ring;
1252 tx_id = txq->tx_tail;
1253 txe = &sw_ring[tx_id];
1255 /* Check if the descriptor ring needs to be cleaned. */
1256 if (txq->nb_tx_free < txq->tx_free_thresh)
1257 i40e_xmit_cleanup(txq);
1259 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
1265 tx_pkt = *tx_pkts++;
1266 RTE_MBUF_PREFETCH_TO_FREE(txe->mbuf);
1268 ol_flags = tx_pkt->ol_flags;
1269 tx_offload.l2_len = tx_pkt->l2_len;
1270 tx_offload.l3_len = tx_pkt->l3_len;
1271 tx_offload.outer_l2_len = tx_pkt->outer_l2_len;
1272 tx_offload.outer_l3_len = tx_pkt->outer_l3_len;
1273 tx_offload.l4_len = tx_pkt->l4_len;
1274 tx_offload.tso_segsz = tx_pkt->tso_segsz;
1276 /* Calculate the number of context descriptors needed. */
1277 nb_ctx = i40e_calc_context_desc(ol_flags);
1280 * The number of descriptors that must be allocated for
1281 * a packet equals to the number of the segments of that
1282 * packet plus 1 context descriptor if needed.
1284 nb_used = (uint16_t)(tx_pkt->nb_segs + nb_ctx);
1285 tx_last = (uint16_t)(tx_id + nb_used - 1);
1288 if (tx_last >= txq->nb_tx_desc)
1289 tx_last = (uint16_t)(tx_last - txq->nb_tx_desc);
1291 if (nb_used > txq->nb_tx_free) {
1292 if (i40e_xmit_cleanup(txq) != 0) {
1297 if (unlikely(nb_used > txq->tx_rs_thresh)) {
1298 while (nb_used > txq->nb_tx_free) {
1299 if (i40e_xmit_cleanup(txq) != 0) {
1308 /* Descriptor based VLAN insertion */
1309 if (ol_flags & (PKT_TX_VLAN_PKT | PKT_TX_QINQ_PKT)) {
1310 tx_flags |= tx_pkt->vlan_tci <<
1311 I40E_TX_FLAG_L2TAG1_SHIFT;
1312 tx_flags |= I40E_TX_FLAG_INSERT_VLAN;
1313 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1314 td_tag = (tx_flags & I40E_TX_FLAG_L2TAG1_MASK) >>
1315 I40E_TX_FLAG_L2TAG1_SHIFT;
1318 /* Always enable CRC offload insertion */
1319 td_cmd |= I40E_TX_DESC_CMD_ICRC;
1321 /* Enable checksum offloading */
1322 cd_tunneling_params = 0;
1323 if (unlikely(ol_flags & I40E_TX_CKSUM_OFFLOAD_MASK)) {
1324 i40e_txd_enable_checksum(ol_flags, &td_cmd, &td_offset,
1325 tx_offload, &cd_tunneling_params);
1328 if (unlikely(nb_ctx)) {
1329 /* Setup TX context descriptor if required */
1330 volatile struct i40e_tx_context_desc *ctx_txd =
1331 (volatile struct i40e_tx_context_desc *)\
1333 uint16_t cd_l2tag2 = 0;
1334 uint64_t cd_type_cmd_tso_mss =
1335 I40E_TX_DESC_DTYPE_CONTEXT;
1337 txn = &sw_ring[txe->next_id];
1338 RTE_MBUF_PREFETCH_TO_FREE(txn->mbuf);
1339 if (txe->mbuf != NULL) {
1340 rte_pktmbuf_free_seg(txe->mbuf);
1344 /* TSO enabled means no timestamp */
1345 if (ol_flags & PKT_TX_TCP_SEG)
1346 cd_type_cmd_tso_mss |=
1347 i40e_set_tso_ctx(tx_pkt, tx_offload);
1349 #ifdef RTE_LIBRTE_IEEE1588
1350 if (ol_flags & PKT_TX_IEEE1588_TMST)
1351 cd_type_cmd_tso_mss |=
1352 ((uint64_t)I40E_TX_CTX_DESC_TSYN <<
1353 I40E_TXD_CTX_QW1_CMD_SHIFT);
1357 ctx_txd->tunneling_params =
1358 rte_cpu_to_le_32(cd_tunneling_params);
1359 if (ol_flags & PKT_TX_QINQ_PKT) {
1360 cd_l2tag2 = tx_pkt->vlan_tci_outer;
1361 cd_type_cmd_tso_mss |=
1362 ((uint64_t)I40E_TX_CTX_DESC_IL2TAG2 <<
1363 I40E_TXD_CTX_QW1_CMD_SHIFT);
1365 ctx_txd->l2tag2 = rte_cpu_to_le_16(cd_l2tag2);
1366 ctx_txd->type_cmd_tso_mss =
1367 rte_cpu_to_le_64(cd_type_cmd_tso_mss);
1369 PMD_TX_LOG(DEBUG, "mbuf: %p, TCD[%u]:\n"
1370 "tunneling_params: %#x;\n"
1373 "type_cmd_tso_mss: %#"PRIx64";\n",
1375 ctx_txd->tunneling_params,
1378 ctx_txd->type_cmd_tso_mss);
1380 txe->last_id = tx_last;
1381 tx_id = txe->next_id;
1388 txn = &sw_ring[txe->next_id];
1391 rte_pktmbuf_free_seg(txe->mbuf);
1394 /* Setup TX Descriptor */
1395 slen = m_seg->data_len;
1396 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
1398 PMD_TX_LOG(DEBUG, "mbuf: %p, TDD[%u]:\n"
1399 "buf_dma_addr: %#"PRIx64";\n"
1404 tx_pkt, tx_id, buf_dma_addr,
1405 td_cmd, td_offset, slen, td_tag);
1407 txd->buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
1408 txd->cmd_type_offset_bsz = i40e_build_ctob(td_cmd,
1409 td_offset, slen, td_tag);
1410 txe->last_id = tx_last;
1411 tx_id = txe->next_id;
1413 m_seg = m_seg->next;
1414 } while (m_seg != NULL);
1416 /* The last packet data descriptor needs End Of Packet (EOP) */
1417 td_cmd |= I40E_TX_DESC_CMD_EOP;
1418 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
1419 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
1421 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
1422 PMD_TX_FREE_LOG(DEBUG,
1423 "Setting RS bit on TXD id="
1424 "%4u (port=%d queue=%d)",
1425 tx_last, txq->port_id, txq->queue_id);
1427 td_cmd |= I40E_TX_DESC_CMD_RS;
1429 /* Update txq RS bit counters */
1430 txq->nb_tx_used = 0;
1433 txd->cmd_type_offset_bsz |=
1434 rte_cpu_to_le_64(((uint64_t)td_cmd) <<
1435 I40E_TXD_QW1_CMD_SHIFT);
1441 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
1442 (unsigned) txq->port_id, (unsigned) txq->queue_id,
1443 (unsigned) tx_id, (unsigned) nb_tx);
1445 I40E_PCI_REG_WRITE(txq->qtx_tail, tx_id);
1446 txq->tx_tail = tx_id;
1451 static inline int __attribute__((always_inline))
1452 i40e_tx_free_bufs(struct i40e_tx_queue *txq)
1454 struct i40e_tx_entry *txep;
1457 if (!(txq->tx_ring[txq->tx_next_dd].cmd_type_offset_bsz &
1458 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE)))
1461 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
1463 for (i = 0; i < txq->tx_rs_thresh; i++)
1464 rte_prefetch0((txep + i)->mbuf);
1466 if (!(txq->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT)) {
1467 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1468 rte_mempool_put(txep->mbuf->pool, txep->mbuf);
1472 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1473 rte_pktmbuf_free_seg(txep->mbuf);
1478 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
1479 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
1480 if (txq->tx_next_dd >= txq->nb_tx_desc)
1481 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
1483 return txq->tx_rs_thresh;
1486 #define I40E_TD_CMD (I40E_TX_DESC_CMD_ICRC |\
1487 I40E_TX_DESC_CMD_EOP)
1489 /* Populate 4 descriptors with data from 4 mbufs */
1491 tx4(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1496 for (i = 0; i < 4; i++, txdp++, pkts++) {
1497 dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
1498 txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1499 txdp->cmd_type_offset_bsz =
1500 i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1501 (*pkts)->data_len, 0);
1505 /* Populate 1 descriptor with data from 1 mbuf */
1507 tx1(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1511 dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
1512 txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1513 txdp->cmd_type_offset_bsz =
1514 i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1515 (*pkts)->data_len, 0);
1518 /* Fill hardware descriptor ring with mbuf data */
1520 i40e_tx_fill_hw_ring(struct i40e_tx_queue *txq,
1521 struct rte_mbuf **pkts,
1524 volatile struct i40e_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
1525 struct i40e_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
1526 const int N_PER_LOOP = 4;
1527 const int N_PER_LOOP_MASK = N_PER_LOOP - 1;
1528 int mainpart, leftover;
1531 mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
1532 leftover = (nb_pkts & ((uint32_t) N_PER_LOOP_MASK));
1533 for (i = 0; i < mainpart; i += N_PER_LOOP) {
1534 for (j = 0; j < N_PER_LOOP; ++j) {
1535 (txep + i + j)->mbuf = *(pkts + i + j);
1537 tx4(txdp + i, pkts + i);
1539 if (unlikely(leftover > 0)) {
1540 for (i = 0; i < leftover; ++i) {
1541 (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
1542 tx1(txdp + mainpart + i, pkts + mainpart + i);
1547 static inline uint16_t
1548 tx_xmit_pkts(struct i40e_tx_queue *txq,
1549 struct rte_mbuf **tx_pkts,
1552 volatile struct i40e_tx_desc *txr = txq->tx_ring;
1556 * Begin scanning the H/W ring for done descriptors when the number
1557 * of available descriptors drops below tx_free_thresh. For each done
1558 * descriptor, free the associated buffer.
1560 if (txq->nb_tx_free < txq->tx_free_thresh)
1561 i40e_tx_free_bufs(txq);
1563 /* Use available descriptor only */
1564 nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
1565 if (unlikely(!nb_pkts))
1568 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
1569 if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
1570 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
1571 i40e_tx_fill_hw_ring(txq, tx_pkts, n);
1572 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1573 rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1574 I40E_TXD_QW1_CMD_SHIFT);
1575 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1579 /* Fill hardware descriptor ring with mbuf data */
1580 i40e_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
1581 txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
1583 /* Determin if RS bit needs to be set */
1584 if (txq->tx_tail > txq->tx_next_rs) {
1585 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1586 rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1587 I40E_TXD_QW1_CMD_SHIFT);
1589 (uint16_t)(txq->tx_next_rs + txq->tx_rs_thresh);
1590 if (txq->tx_next_rs >= txq->nb_tx_desc)
1591 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1594 if (txq->tx_tail >= txq->nb_tx_desc)
1597 /* Update the tx tail register */
1599 I40E_PCI_REG_WRITE(txq->qtx_tail, txq->tx_tail);
1605 i40e_xmit_pkts_simple(void *tx_queue,
1606 struct rte_mbuf **tx_pkts,
1611 if (likely(nb_pkts <= I40E_TX_MAX_BURST))
1612 return tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1616 uint16_t ret, num = (uint16_t)RTE_MIN(nb_pkts,
1619 ret = tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1620 &tx_pkts[nb_tx], num);
1621 nb_tx = (uint16_t)(nb_tx + ret);
1622 nb_pkts = (uint16_t)(nb_pkts - ret);
1631 * Find the VSI the queue belongs to. 'queue_idx' is the queue index
1632 * application used, which assume having sequential ones. But from driver's
1633 * perspective, it's different. For example, q0 belongs to FDIR VSI, q1-q64
1634 * to MAIN VSI, , q65-96 to SRIOV VSIs, q97-128 to VMDQ VSIs. For application
1635 * running on host, q1-64 and q97-128 can be used, total 96 queues. They can
1636 * use queue_idx from 0 to 95 to access queues, while real queue would be
1637 * different. This function will do a queue mapping to find VSI the queue
1640 static struct i40e_vsi*
1641 i40e_pf_get_vsi_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1643 /* the queue in MAIN VSI range */
1644 if (queue_idx < pf->main_vsi->nb_qps)
1645 return pf->main_vsi;
1647 queue_idx -= pf->main_vsi->nb_qps;
1649 /* queue_idx is greater than VMDQ VSIs range */
1650 if (queue_idx > pf->nb_cfg_vmdq_vsi * pf->vmdq_nb_qps - 1) {
1651 PMD_INIT_LOG(ERR, "queue_idx out of range. VMDQ configured?");
1655 return pf->vmdq[queue_idx / pf->vmdq_nb_qps].vsi;
1659 i40e_get_queue_offset_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1661 /* the queue in MAIN VSI range */
1662 if (queue_idx < pf->main_vsi->nb_qps)
1665 /* It's VMDQ queues */
1666 queue_idx -= pf->main_vsi->nb_qps;
1668 if (pf->nb_cfg_vmdq_vsi)
1669 return queue_idx % pf->vmdq_nb_qps;
1671 PMD_INIT_LOG(ERR, "Fail to get queue offset");
1672 return (uint16_t)(-1);
1677 i40e_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1679 struct i40e_rx_queue *rxq;
1681 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1683 PMD_INIT_FUNC_TRACE();
1685 if (rx_queue_id < dev->data->nb_rx_queues) {
1686 rxq = dev->data->rx_queues[rx_queue_id];
1688 err = i40e_alloc_rx_queue_mbufs(rxq);
1690 PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
1696 /* Init the RX tail regieter. */
1697 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1699 err = i40e_switch_rx_queue(hw, rxq->reg_idx, TRUE);
1702 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
1705 i40e_rx_queue_release_mbufs(rxq);
1706 i40e_reset_rx_queue(rxq);
1714 i40e_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1716 struct i40e_rx_queue *rxq;
1718 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1720 if (rx_queue_id < dev->data->nb_rx_queues) {
1721 rxq = dev->data->rx_queues[rx_queue_id];
1724 * rx_queue_id is queue id aplication refers to, while
1725 * rxq->reg_idx is the real queue index.
1727 err = i40e_switch_rx_queue(hw, rxq->reg_idx, FALSE);
1730 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
1734 i40e_rx_queue_release_mbufs(rxq);
1735 i40e_reset_rx_queue(rxq);
1742 i40e_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1745 struct i40e_tx_queue *txq;
1746 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1748 PMD_INIT_FUNC_TRACE();
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 * rxq->reg_idx is the real queue index.
1757 err = i40e_switch_tx_queue(hw, txq->reg_idx, TRUE);
1759 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
1767 i40e_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1769 struct i40e_tx_queue *txq;
1771 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1773 if (tx_queue_id < dev->data->nb_tx_queues) {
1774 txq = dev->data->tx_queues[tx_queue_id];
1777 * tx_queue_id is queue id aplication refers to, while
1778 * txq->reg_idx is the real queue index.
1780 err = i40e_switch_tx_queue(hw, txq->reg_idx, FALSE);
1783 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u of",
1788 i40e_tx_queue_release_mbufs(txq);
1789 i40e_reset_tx_queue(txq);
1796 i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
1799 unsigned int socket_id,
1800 const struct rte_eth_rxconf *rx_conf,
1801 struct rte_mempool *mp)
1803 struct i40e_vsi *vsi;
1804 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1805 struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
1806 struct i40e_rx_queue *rxq;
1807 const struct rte_memzone *rz;
1810 int use_def_burst_func = 1;
1812 if (hw->mac.type == I40E_MAC_VF) {
1813 struct i40e_vf *vf =
1814 I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1817 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
1820 PMD_DRV_LOG(ERR, "VSI not available or queue "
1821 "index exceeds the maximum");
1822 return I40E_ERR_PARAM;
1824 if (((nb_desc * sizeof(union i40e_rx_desc)) % I40E_ALIGN) != 0 ||
1825 (nb_desc > I40E_MAX_RING_DESC) ||
1826 (nb_desc < I40E_MIN_RING_DESC)) {
1827 PMD_DRV_LOG(ERR, "Number (%u) of receive descriptors is "
1828 "invalid", nb_desc);
1829 return I40E_ERR_PARAM;
1832 /* Free memory if needed */
1833 if (dev->data->rx_queues[queue_idx]) {
1834 i40e_dev_rx_queue_release(dev->data->rx_queues[queue_idx]);
1835 dev->data->rx_queues[queue_idx] = NULL;
1838 /* Allocate the rx queue data structure */
1839 rxq = rte_zmalloc_socket("i40e rx queue",
1840 sizeof(struct i40e_rx_queue),
1841 RTE_CACHE_LINE_SIZE,
1844 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
1845 "rx queue data structure");
1849 rxq->nb_rx_desc = nb_desc;
1850 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
1851 rxq->queue_id = queue_idx;
1852 if (hw->mac.type == I40E_MAC_VF)
1853 rxq->reg_idx = queue_idx;
1854 else /* PF device */
1855 rxq->reg_idx = vsi->base_queue +
1856 i40e_get_queue_offset_by_qindex(pf, queue_idx);
1858 rxq->port_id = dev->data->port_id;
1859 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
1861 rxq->drop_en = rx_conf->rx_drop_en;
1863 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
1865 /* Allocate the maximun number of RX ring hardware descriptor. */
1866 ring_size = sizeof(union i40e_rx_desc) * I40E_MAX_RING_DESC;
1867 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
1868 rz = i40e_ring_dma_zone_reserve(dev,
1874 i40e_dev_rx_queue_release(rxq);
1875 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX");
1879 /* Zero all the descriptors in the ring. */
1880 memset(rz->addr, 0, ring_size);
1882 #ifdef RTE_LIBRTE_XEN_DOM0
1883 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
1885 rxq->rx_ring_phys_addr = (uint64_t)rz->phys_addr;
1888 rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
1890 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1891 len = (uint16_t)(nb_desc + RTE_PMD_I40E_RX_MAX_BURST);
1896 /* Allocate the software ring. */
1898 rte_zmalloc_socket("i40e rx sw ring",
1899 sizeof(struct i40e_rx_entry) * len,
1900 RTE_CACHE_LINE_SIZE,
1902 if (!rxq->sw_ring) {
1903 i40e_dev_rx_queue_release(rxq);
1904 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW ring");
1908 i40e_reset_rx_queue(rxq);
1910 dev->data->rx_queues[queue_idx] = rxq;
1912 use_def_burst_func = check_rx_burst_bulk_alloc_preconditions(rxq);
1914 if (!use_def_burst_func && !dev->data->scattered_rx) {
1915 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1916 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
1917 "satisfied. Rx Burst Bulk Alloc function will be "
1918 "used on port=%d, queue=%d.",
1919 rxq->port_id, rxq->queue_id);
1920 dev->rx_pkt_burst = i40e_recv_pkts_bulk_alloc;
1921 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
1923 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
1924 "not satisfied, Scattered Rx is requested, "
1925 "or RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC is "
1926 "not enabled on port=%d, queue=%d.",
1927 rxq->port_id, rxq->queue_id);
1934 i40e_dev_rx_queue_release(void *rxq)
1936 struct i40e_rx_queue *q = (struct i40e_rx_queue *)rxq;
1939 PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL");
1943 i40e_rx_queue_release_mbufs(q);
1944 rte_free(q->sw_ring);
1949 i40e_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1951 #define I40E_RXQ_SCAN_INTERVAL 4
1952 volatile union i40e_rx_desc *rxdp;
1953 struct i40e_rx_queue *rxq;
1956 if (unlikely(rx_queue_id >= dev->data->nb_rx_queues)) {
1957 PMD_DRV_LOG(ERR, "Invalid RX queue id %u", rx_queue_id);
1961 rxq = dev->data->rx_queues[rx_queue_id];
1962 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
1963 while ((desc < rxq->nb_rx_desc) &&
1964 ((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1965 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
1966 (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) {
1968 * Check the DD bit of a rx descriptor of each 4 in a group,
1969 * to avoid checking too frequently and downgrading performance
1972 desc += I40E_RXQ_SCAN_INTERVAL;
1973 rxdp += I40E_RXQ_SCAN_INTERVAL;
1974 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
1975 rxdp = &(rxq->rx_ring[rxq->rx_tail +
1976 desc - rxq->nb_rx_desc]);
1983 i40e_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
1985 volatile union i40e_rx_desc *rxdp;
1986 struct i40e_rx_queue *rxq = rx_queue;
1990 if (unlikely(offset >= rxq->nb_rx_desc)) {
1991 PMD_DRV_LOG(ERR, "Invalid RX queue id %u", offset);
1995 desc = rxq->rx_tail + offset;
1996 if (desc >= rxq->nb_rx_desc)
1997 desc -= rxq->nb_rx_desc;
1999 rxdp = &(rxq->rx_ring[desc]);
2001 ret = !!(((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
2002 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
2003 (1 << I40E_RX_DESC_STATUS_DD_SHIFT));
2009 i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
2012 unsigned int socket_id,
2013 const struct rte_eth_txconf *tx_conf)
2015 struct i40e_vsi *vsi;
2016 struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2017 struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
2018 struct i40e_tx_queue *txq;
2019 const struct rte_memzone *tz;
2021 uint16_t tx_rs_thresh, tx_free_thresh;
2023 if (hw->mac.type == I40E_MAC_VF) {
2024 struct i40e_vf *vf =
2025 I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2028 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
2031 PMD_DRV_LOG(ERR, "VSI is NULL, or queue index (%u) "
2032 "exceeds the maximum", queue_idx);
2033 return I40E_ERR_PARAM;
2036 if (((nb_desc * sizeof(struct i40e_tx_desc)) % I40E_ALIGN) != 0 ||
2037 (nb_desc > I40E_MAX_RING_DESC) ||
2038 (nb_desc < I40E_MIN_RING_DESC)) {
2039 PMD_DRV_LOG(ERR, "Number (%u) of transmit descriptors is "
2040 "invalid", nb_desc);
2041 return I40E_ERR_PARAM;
2045 * The following two parameters control the setting of the RS bit on
2046 * transmit descriptors. TX descriptors will have their RS bit set
2047 * after txq->tx_rs_thresh descriptors have been used. The TX
2048 * descriptor ring will be cleaned after txq->tx_free_thresh
2049 * descriptors are used or if the number of descriptors required to
2050 * transmit a packet is greater than the number of free TX descriptors.
2052 * The following constraints must be satisfied:
2053 * - tx_rs_thresh must be greater than 0.
2054 * - tx_rs_thresh must be less than the size of the ring minus 2.
2055 * - tx_rs_thresh must be less than or equal to tx_free_thresh.
2056 * - tx_rs_thresh must be a divisor of the ring size.
2057 * - tx_free_thresh must be greater than 0.
2058 * - tx_free_thresh must be less than the size of the ring minus 3.
2060 * One descriptor in the TX ring is used as a sentinel to avoid a H/W
2061 * race condition, hence the maximum threshold constraints. When set
2062 * to zero use default values.
2064 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2065 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2066 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2067 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2068 if (tx_rs_thresh >= (nb_desc - 2)) {
2069 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2070 "number of TX descriptors minus 2. "
2071 "(tx_rs_thresh=%u port=%d queue=%d)",
2072 (unsigned int)tx_rs_thresh,
2073 (int)dev->data->port_id,
2075 return I40E_ERR_PARAM;
2077 if (tx_free_thresh >= (nb_desc - 3)) {
2078 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2079 "tx_free_thresh must be less than the "
2080 "number of TX descriptors minus 3. "
2081 "(tx_free_thresh=%u port=%d queue=%d)",
2082 (unsigned int)tx_free_thresh,
2083 (int)dev->data->port_id,
2085 return I40E_ERR_PARAM;
2087 if (tx_rs_thresh > tx_free_thresh) {
2088 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or "
2089 "equal to tx_free_thresh. (tx_free_thresh=%u"
2090 " tx_rs_thresh=%u port=%d queue=%d)",
2091 (unsigned int)tx_free_thresh,
2092 (unsigned int)tx_rs_thresh,
2093 (int)dev->data->port_id,
2095 return I40E_ERR_PARAM;
2097 if ((nb_desc % tx_rs_thresh) != 0) {
2098 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2099 "number of TX descriptors. (tx_rs_thresh=%u"
2100 " port=%d queue=%d)",
2101 (unsigned int)tx_rs_thresh,
2102 (int)dev->data->port_id,
2104 return I40E_ERR_PARAM;
2106 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2107 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2108 "tx_rs_thresh is greater than 1. "
2109 "(tx_rs_thresh=%u port=%d queue=%d)",
2110 (unsigned int)tx_rs_thresh,
2111 (int)dev->data->port_id,
2113 return I40E_ERR_PARAM;
2116 /* Free memory if needed. */
2117 if (dev->data->tx_queues[queue_idx]) {
2118 i40e_dev_tx_queue_release(dev->data->tx_queues[queue_idx]);
2119 dev->data->tx_queues[queue_idx] = NULL;
2122 /* Allocate the TX queue data structure. */
2123 txq = rte_zmalloc_socket("i40e tx queue",
2124 sizeof(struct i40e_tx_queue),
2125 RTE_CACHE_LINE_SIZE,
2128 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2129 "tx queue structure");
2133 /* Allocate TX hardware ring descriptors. */
2134 ring_size = sizeof(struct i40e_tx_desc) * I40E_MAX_RING_DESC;
2135 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2136 tz = i40e_ring_dma_zone_reserve(dev,
2142 i40e_dev_tx_queue_release(txq);
2143 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX");
2147 txq->nb_tx_desc = nb_desc;
2148 txq->tx_rs_thresh = tx_rs_thresh;
2149 txq->tx_free_thresh = tx_free_thresh;
2150 txq->pthresh = tx_conf->tx_thresh.pthresh;
2151 txq->hthresh = tx_conf->tx_thresh.hthresh;
2152 txq->wthresh = tx_conf->tx_thresh.wthresh;
2153 txq->queue_id = queue_idx;
2154 if (hw->mac.type == I40E_MAC_VF)
2155 txq->reg_idx = queue_idx;
2156 else /* PF device */
2157 txq->reg_idx = vsi->base_queue +
2158 i40e_get_queue_offset_by_qindex(pf, queue_idx);
2160 txq->port_id = dev->data->port_id;
2161 txq->txq_flags = tx_conf->txq_flags;
2163 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2165 #ifdef RTE_LIBRTE_XEN_DOM0
2166 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2168 txq->tx_ring_phys_addr = (uint64_t)tz->phys_addr;
2170 txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2172 /* Allocate software ring */
2174 rte_zmalloc_socket("i40e tx sw ring",
2175 sizeof(struct i40e_tx_entry) * nb_desc,
2176 RTE_CACHE_LINE_SIZE,
2178 if (!txq->sw_ring) {
2179 i40e_dev_tx_queue_release(txq);
2180 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW TX ring");
2184 i40e_reset_tx_queue(txq);
2186 dev->data->tx_queues[queue_idx] = txq;
2188 /* Use a simple TX queue without offloads or multi segs if possible */
2189 if (((txq->txq_flags & I40E_SIMPLE_FLAGS) == I40E_SIMPLE_FLAGS) &&
2190 (txq->tx_rs_thresh >= I40E_TX_MAX_BURST)) {
2191 PMD_INIT_LOG(INFO, "Using simple tx path");
2192 dev->tx_pkt_burst = i40e_xmit_pkts_simple;
2194 PMD_INIT_LOG(INFO, "Using full-featured tx path");
2195 dev->tx_pkt_burst = i40e_xmit_pkts;
2202 i40e_dev_tx_queue_release(void *txq)
2204 struct i40e_tx_queue *q = (struct i40e_tx_queue *)txq;
2207 PMD_DRV_LOG(DEBUG, "Pointer to TX queue is NULL");
2211 i40e_tx_queue_release_mbufs(q);
2212 rte_free(q->sw_ring);
2216 static const struct rte_memzone *
2217 i40e_ring_dma_zone_reserve(struct rte_eth_dev *dev,
2218 const char *ring_name,
2223 char z_name[RTE_MEMZONE_NAMESIZE];
2224 const struct rte_memzone *mz;
2226 snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
2227 dev->driver->pci_drv.name, ring_name,
2228 dev->data->port_id, queue_id);
2229 mz = rte_memzone_lookup(z_name);
2233 #ifdef RTE_LIBRTE_XEN_DOM0
2234 return rte_memzone_reserve_bounded(z_name, ring_size,
2235 socket_id, 0, I40E_ALIGN, RTE_PGSIZE_2M);
2237 return rte_memzone_reserve_aligned(z_name, ring_size,
2238 socket_id, 0, I40E_ALIGN);
2242 const struct rte_memzone *
2243 i40e_memzone_reserve(const char *name, uint32_t len, int socket_id)
2245 const struct rte_memzone *mz = NULL;
2247 mz = rte_memzone_lookup(name);
2250 #ifdef RTE_LIBRTE_XEN_DOM0
2251 mz = rte_memzone_reserve_bounded(name, len,
2252 socket_id, 0, I40E_ALIGN, RTE_PGSIZE_2M);
2254 mz = rte_memzone_reserve_aligned(name, len,
2255 socket_id, 0, I40E_ALIGN);
2261 i40e_rx_queue_release_mbufs(struct i40e_rx_queue *rxq)
2265 if (!rxq || !rxq->sw_ring) {
2266 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL");
2270 for (i = 0; i < rxq->nb_rx_desc; i++) {
2271 if (rxq->sw_ring[i].mbuf) {
2272 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2273 rxq->sw_ring[i].mbuf = NULL;
2276 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2277 if (rxq->rx_nb_avail == 0)
2279 for (i = 0; i < rxq->rx_nb_avail; i++) {
2280 struct rte_mbuf *mbuf;
2282 mbuf = rxq->rx_stage[rxq->rx_next_avail + i];
2283 rte_pktmbuf_free_seg(mbuf);
2285 rxq->rx_nb_avail = 0;
2286 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2290 i40e_reset_rx_queue(struct i40e_rx_queue *rxq)
2295 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2296 if (check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
2297 len = (uint16_t)(rxq->nb_rx_desc + RTE_PMD_I40E_RX_MAX_BURST);
2299 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2300 len = rxq->nb_rx_desc;
2302 for (i = 0; i < len * sizeof(union i40e_rx_desc); i++)
2303 ((volatile char *)rxq->rx_ring)[i] = 0;
2305 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2306 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2307 for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; ++i)
2308 rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
2310 rxq->rx_nb_avail = 0;
2311 rxq->rx_next_avail = 0;
2312 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2313 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2315 rxq->nb_rx_hold = 0;
2316 rxq->pkt_first_seg = NULL;
2317 rxq->pkt_last_seg = NULL;
2321 i40e_tx_queue_release_mbufs(struct i40e_tx_queue *txq)
2325 if (!txq || !txq->sw_ring) {
2326 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL");
2330 for (i = 0; i < txq->nb_tx_desc; i++) {
2331 if (txq->sw_ring[i].mbuf) {
2332 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2333 txq->sw_ring[i].mbuf = NULL;
2339 i40e_reset_tx_queue(struct i40e_tx_queue *txq)
2341 struct i40e_tx_entry *txe;
2342 uint16_t i, prev, size;
2345 PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
2350 size = sizeof(struct i40e_tx_desc) * txq->nb_tx_desc;
2351 for (i = 0; i < size; i++)
2352 ((volatile char *)txq->tx_ring)[i] = 0;
2354 prev = (uint16_t)(txq->nb_tx_desc - 1);
2355 for (i = 0; i < txq->nb_tx_desc; i++) {
2356 volatile struct i40e_tx_desc *txd = &txq->tx_ring[i];
2358 txd->cmd_type_offset_bsz =
2359 rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE);
2362 txe[prev].next_id = i;
2366 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2367 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2370 txq->nb_tx_used = 0;
2372 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2373 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2376 /* Init the TX queue in hardware */
2378 i40e_tx_queue_init(struct i40e_tx_queue *txq)
2380 enum i40e_status_code err = I40E_SUCCESS;
2381 struct i40e_vsi *vsi = txq->vsi;
2382 struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2383 uint16_t pf_q = txq->reg_idx;
2384 struct i40e_hmc_obj_txq tx_ctx;
2387 /* clear the context structure first */
2388 memset(&tx_ctx, 0, sizeof(tx_ctx));
2389 tx_ctx.new_context = 1;
2390 tx_ctx.base = txq->tx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2391 tx_ctx.qlen = txq->nb_tx_desc;
2392 tx_ctx.rdylist = rte_le_to_cpu_16(vsi->info.qs_handle[0]);
2393 if (vsi->type == I40E_VSI_FDIR)
2394 tx_ctx.fd_ena = TRUE;
2396 err = i40e_clear_lan_tx_queue_context(hw, pf_q);
2397 if (err != I40E_SUCCESS) {
2398 PMD_DRV_LOG(ERR, "Failure of clean lan tx queue context");
2402 err = i40e_set_lan_tx_queue_context(hw, pf_q, &tx_ctx);
2403 if (err != I40E_SUCCESS) {
2404 PMD_DRV_LOG(ERR, "Failure of set lan tx queue context");
2408 /* Now associate this queue with this PCI function */
2409 qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
2410 qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
2411 I40E_QTX_CTL_PF_INDX_MASK);
2412 I40E_WRITE_REG(hw, I40E_QTX_CTL(pf_q), qtx_ctl);
2413 I40E_WRITE_FLUSH(hw);
2415 txq->qtx_tail = hw->hw_addr + I40E_QTX_TAIL(pf_q);
2421 i40e_alloc_rx_queue_mbufs(struct i40e_rx_queue *rxq)
2423 struct i40e_rx_entry *rxe = rxq->sw_ring;
2427 for (i = 0; i < rxq->nb_rx_desc; i++) {
2428 volatile union i40e_rx_desc *rxd;
2429 struct rte_mbuf *mbuf = rte_rxmbuf_alloc(rxq->mp);
2431 if (unlikely(!mbuf)) {
2432 PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
2436 rte_mbuf_refcnt_set(mbuf, 1);
2438 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
2440 mbuf->port = rxq->port_id;
2443 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf));
2445 rxd = &rxq->rx_ring[i];
2446 rxd->read.pkt_addr = dma_addr;
2447 rxd->read.hdr_addr = dma_addr;
2448 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2449 rxd->read.rsvd1 = 0;
2450 rxd->read.rsvd2 = 0;
2451 #endif /* RTE_LIBRTE_I40E_16BYTE_RX_DESC */
2460 * Calculate the buffer length, and check the jumbo frame
2461 * and maximum packet length.
2464 i40e_rx_queue_config(struct i40e_rx_queue *rxq)
2466 struct i40e_pf *pf = I40E_VSI_TO_PF(rxq->vsi);
2467 struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2468 struct rte_eth_dev_data *data = pf->dev_data;
2469 uint16_t buf_size, len;
2471 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
2472 RTE_PKTMBUF_HEADROOM);
2474 switch (pf->flags & (I40E_FLAG_HEADER_SPLIT_DISABLED |
2475 I40E_FLAG_HEADER_SPLIT_ENABLED)) {
2476 case I40E_FLAG_HEADER_SPLIT_ENABLED: /* Not supported */
2477 rxq->rx_hdr_len = RTE_ALIGN(I40E_RXBUF_SZ_1024,
2478 (1 << I40E_RXQ_CTX_HBUFF_SHIFT));
2479 rxq->rx_buf_len = RTE_ALIGN(I40E_RXBUF_SZ_2048,
2480 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2481 rxq->hs_mode = i40e_header_split_enabled;
2483 case I40E_FLAG_HEADER_SPLIT_DISABLED:
2485 rxq->rx_hdr_len = 0;
2486 rxq->rx_buf_len = RTE_ALIGN(buf_size,
2487 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2488 rxq->hs_mode = i40e_header_split_none;
2492 len = hw->func_caps.rx_buf_chain_len * rxq->rx_buf_len;
2493 rxq->max_pkt_len = RTE_MIN(len, data->dev_conf.rxmode.max_rx_pkt_len);
2494 if (data->dev_conf.rxmode.jumbo_frame == 1) {
2495 if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
2496 rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
2497 PMD_DRV_LOG(ERR, "maximum packet length must "
2498 "be larger than %u and smaller than %u,"
2499 "as jumbo frame is enabled",
2500 (uint32_t)ETHER_MAX_LEN,
2501 (uint32_t)I40E_FRAME_SIZE_MAX);
2502 return I40E_ERR_CONFIG;
2505 if (rxq->max_pkt_len < ETHER_MIN_LEN ||
2506 rxq->max_pkt_len > ETHER_MAX_LEN) {
2507 PMD_DRV_LOG(ERR, "maximum packet length must be "
2508 "larger than %u and smaller than %u, "
2509 "as jumbo frame is disabled",
2510 (uint32_t)ETHER_MIN_LEN,
2511 (uint32_t)ETHER_MAX_LEN);
2512 return I40E_ERR_CONFIG;
2519 /* Init the RX queue in hardware */
2521 i40e_rx_queue_init(struct i40e_rx_queue *rxq)
2523 int err = I40E_SUCCESS;
2524 struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2525 struct rte_eth_dev_data *dev_data = I40E_VSI_TO_DEV_DATA(rxq->vsi);
2526 struct rte_eth_dev *dev = I40E_VSI_TO_ETH_DEV(rxq->vsi);
2527 uint16_t pf_q = rxq->reg_idx;
2529 struct i40e_hmc_obj_rxq rx_ctx;
2531 err = i40e_rx_queue_config(rxq);
2533 PMD_DRV_LOG(ERR, "Failed to config RX queue");
2537 /* Clear the context structure first */
2538 memset(&rx_ctx, 0, sizeof(struct i40e_hmc_obj_rxq));
2539 rx_ctx.dbuff = rxq->rx_buf_len >> I40E_RXQ_CTX_DBUFF_SHIFT;
2540 rx_ctx.hbuff = rxq->rx_hdr_len >> I40E_RXQ_CTX_HBUFF_SHIFT;
2542 rx_ctx.base = rxq->rx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2543 rx_ctx.qlen = rxq->nb_rx_desc;
2544 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2547 rx_ctx.dtype = rxq->hs_mode;
2549 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_ALL;
2551 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_NONE;
2552 rx_ctx.rxmax = rxq->max_pkt_len;
2553 rx_ctx.tphrdesc_ena = 1;
2554 rx_ctx.tphwdesc_ena = 1;
2555 rx_ctx.tphdata_ena = 1;
2556 rx_ctx.tphhead_ena = 1;
2557 rx_ctx.lrxqthresh = 2;
2558 rx_ctx.crcstrip = (rxq->crc_len == 0) ? 1 : 0;
2563 err = i40e_clear_lan_rx_queue_context(hw, pf_q);
2564 if (err != I40E_SUCCESS) {
2565 PMD_DRV_LOG(ERR, "Failed to clear LAN RX queue context");
2568 err = i40e_set_lan_rx_queue_context(hw, pf_q, &rx_ctx);
2569 if (err != I40E_SUCCESS) {
2570 PMD_DRV_LOG(ERR, "Failed to set LAN RX queue context");
2574 rxq->qrx_tail = hw->hw_addr + I40E_QRX_TAIL(pf_q);
2576 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
2577 RTE_PKTMBUF_HEADROOM);
2579 /* Check if scattered RX needs to be used. */
2580 if ((rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
2581 dev_data->scattered_rx = 1;
2582 dev->rx_pkt_burst = i40e_recv_scattered_pkts;
2585 /* Init the RX tail regieter. */
2586 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
2592 i40e_dev_clear_queues(struct rte_eth_dev *dev)
2596 PMD_INIT_FUNC_TRACE();
2598 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2599 i40e_tx_queue_release_mbufs(dev->data->tx_queues[i]);
2600 i40e_reset_tx_queue(dev->data->tx_queues[i]);
2603 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2604 i40e_rx_queue_release_mbufs(dev->data->rx_queues[i]);
2605 i40e_reset_rx_queue(dev->data->rx_queues[i]);
2609 #define I40E_FDIR_NUM_TX_DESC I40E_MIN_RING_DESC
2610 #define I40E_FDIR_NUM_RX_DESC I40E_MIN_RING_DESC
2612 enum i40e_status_code
2613 i40e_fdir_setup_tx_resources(struct i40e_pf *pf)
2615 struct i40e_tx_queue *txq;
2616 const struct rte_memzone *tz = NULL;
2618 struct rte_eth_dev *dev = pf->adapter->eth_dev;
2621 PMD_DRV_LOG(ERR, "PF is not available");
2622 return I40E_ERR_BAD_PTR;
2625 /* Allocate the TX queue data structure. */
2626 txq = rte_zmalloc_socket("i40e fdir tx queue",
2627 sizeof(struct i40e_tx_queue),
2628 RTE_CACHE_LINE_SIZE,
2631 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2632 "tx queue structure.");
2633 return I40E_ERR_NO_MEMORY;
2636 /* Allocate TX hardware ring descriptors. */
2637 ring_size = sizeof(struct i40e_tx_desc) * I40E_FDIR_NUM_TX_DESC;
2638 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2640 tz = i40e_ring_dma_zone_reserve(dev,
2646 i40e_dev_tx_queue_release(txq);
2647 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX.");
2648 return I40E_ERR_NO_MEMORY;
2651 txq->nb_tx_desc = I40E_FDIR_NUM_TX_DESC;
2652 txq->queue_id = I40E_FDIR_QUEUE_ID;
2653 txq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2654 txq->vsi = pf->fdir.fdir_vsi;
2656 #ifdef RTE_LIBRTE_XEN_DOM0
2657 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2659 txq->tx_ring_phys_addr = (uint64_t)tz->phys_addr;
2661 txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2663 * don't need to allocate software ring and reset for the fdir
2664 * program queue just set the queue has been configured.
2669 return I40E_SUCCESS;
2672 enum i40e_status_code
2673 i40e_fdir_setup_rx_resources(struct i40e_pf *pf)
2675 struct i40e_rx_queue *rxq;
2676 const struct rte_memzone *rz = NULL;
2678 struct rte_eth_dev *dev = pf->adapter->eth_dev;
2681 PMD_DRV_LOG(ERR, "PF is not available");
2682 return I40E_ERR_BAD_PTR;
2685 /* Allocate the RX queue data structure. */
2686 rxq = rte_zmalloc_socket("i40e fdir rx queue",
2687 sizeof(struct i40e_rx_queue),
2688 RTE_CACHE_LINE_SIZE,
2691 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2692 "rx queue structure.");
2693 return I40E_ERR_NO_MEMORY;
2696 /* Allocate RX hardware ring descriptors. */
2697 ring_size = sizeof(union i40e_rx_desc) * I40E_FDIR_NUM_RX_DESC;
2698 ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2700 rz = i40e_ring_dma_zone_reserve(dev,
2706 i40e_dev_rx_queue_release(rxq);
2707 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX.");
2708 return I40E_ERR_NO_MEMORY;
2711 rxq->nb_rx_desc = I40E_FDIR_NUM_RX_DESC;
2712 rxq->queue_id = I40E_FDIR_QUEUE_ID;
2713 rxq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2714 rxq->vsi = pf->fdir.fdir_vsi;
2716 #ifdef RTE_LIBRTE_XEN_DOM0
2717 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
2719 rxq->rx_ring_phys_addr = (uint64_t)rz->phys_addr;
2721 rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
2724 * Don't need to allocate software ring and reset for the fdir
2725 * rx queue, just set the queue has been configured.
2730 return I40E_SUCCESS;