ethdev: add new offload flag to keep CRC
[dpdk.git] / drivers / net / i40e / i40e_rxtx.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4
5 #include <stdio.h>
6 #include <stdlib.h>
7 #include <string.h>
8 #include <errno.h>
9 #include <stdint.h>
10 #include <stdarg.h>
11 #include <unistd.h>
12 #include <inttypes.h>
13 #include <sys/queue.h>
14
15 #include <rte_string_fns.h>
16 #include <rte_memzone.h>
17 #include <rte_mbuf.h>
18 #include <rte_malloc.h>
19 #include <rte_ether.h>
20 #include <rte_ethdev_driver.h>
21 #include <rte_tcp.h>
22 #include <rte_sctp.h>
23 #include <rte_udp.h>
24 #include <rte_ip.h>
25 #include <rte_net.h>
26
27 #include "i40e_logs.h"
28 #include "base/i40e_prototype.h"
29 #include "base/i40e_type.h"
30 #include "i40e_ethdev.h"
31 #include "i40e_rxtx.h"
32
33 #define DEFAULT_TX_RS_THRESH   32
34 #define DEFAULT_TX_FREE_THRESH 32
35
36 #define I40E_TX_MAX_BURST  32
37
38 #define I40E_DMA_MEM_ALIGN 4096
39
40 /* Base address of the HW descriptor ring should be 128B aligned. */
41 #define I40E_RING_BASE_ALIGN    128
42
43 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
44
45 #ifdef RTE_LIBRTE_IEEE1588
46 #define I40E_TX_IEEE1588_TMST PKT_TX_IEEE1588_TMST
47 #else
48 #define I40E_TX_IEEE1588_TMST 0
49 #endif
50
51 #define I40E_TX_CKSUM_OFFLOAD_MASK (             \
52                 PKT_TX_IP_CKSUM |                \
53                 PKT_TX_L4_MASK |                 \
54                 PKT_TX_TCP_SEG |                 \
55                 PKT_TX_OUTER_IP_CKSUM)
56
57 #define I40E_TX_OFFLOAD_MASK (  \
58                 PKT_TX_IP_CKSUM |       \
59                 PKT_TX_L4_MASK |        \
60                 PKT_TX_OUTER_IP_CKSUM | \
61                 PKT_TX_TCP_SEG |        \
62                 PKT_TX_QINQ_PKT |       \
63                 PKT_TX_VLAN_PKT |       \
64                 PKT_TX_TUNNEL_MASK |    \
65                 I40E_TX_IEEE1588_TMST)
66
67 #define I40E_TX_OFFLOAD_NOTSUP_MASK \
68                 (PKT_TX_OFFLOAD_MASK ^ I40E_TX_OFFLOAD_MASK)
69
70 static inline void
71 i40e_rxd_to_vlan_tci(struct rte_mbuf *mb, volatile union i40e_rx_desc *rxdp)
72 {
73         if (rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
74                 (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) {
75                 mb->ol_flags |= PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED;
76                 mb->vlan_tci =
77                         rte_le_to_cpu_16(rxdp->wb.qword0.lo_dword.l2tag1);
78                 PMD_RX_LOG(DEBUG, "Descriptor l2tag1: %u",
79                            rte_le_to_cpu_16(rxdp->wb.qword0.lo_dword.l2tag1));
80         } else {
81                 mb->vlan_tci = 0;
82         }
83 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
84         if (rte_le_to_cpu_16(rxdp->wb.qword2.ext_status) &
85                 (1 << I40E_RX_DESC_EXT_STATUS_L2TAG2P_SHIFT)) {
86                 mb->ol_flags |= PKT_RX_QINQ_STRIPPED;
87                 mb->vlan_tci_outer = mb->vlan_tci;
88                 mb->vlan_tci = rte_le_to_cpu_16(rxdp->wb.qword2.l2tag2_2);
89                 PMD_RX_LOG(DEBUG, "Descriptor l2tag2_1: %u, l2tag2_2: %u",
90                            rte_le_to_cpu_16(rxdp->wb.qword2.l2tag2_1),
91                            rte_le_to_cpu_16(rxdp->wb.qword2.l2tag2_2));
92         } else {
93                 mb->vlan_tci_outer = 0;
94         }
95 #endif
96         PMD_RX_LOG(DEBUG, "Mbuf vlan_tci: %u, vlan_tci_outer: %u",
97                    mb->vlan_tci, mb->vlan_tci_outer);
98 }
99
100 /* Translate the rx descriptor status to pkt flags */
101 static inline uint64_t
102 i40e_rxd_status_to_pkt_flags(uint64_t qword)
103 {
104         uint64_t flags;
105
106         /* Check if RSS_HASH */
107         flags = (((qword >> I40E_RX_DESC_STATUS_FLTSTAT_SHIFT) &
108                                         I40E_RX_DESC_FLTSTAT_RSS_HASH) ==
109                         I40E_RX_DESC_FLTSTAT_RSS_HASH) ? PKT_RX_RSS_HASH : 0;
110
111         /* Check if FDIR Match */
112         flags |= (qword & (1 << I40E_RX_DESC_STATUS_FLM_SHIFT) ?
113                                                         PKT_RX_FDIR : 0);
114
115         return flags;
116 }
117
118 static inline uint64_t
119 i40e_rxd_error_to_pkt_flags(uint64_t qword)
120 {
121         uint64_t flags = 0;
122         uint64_t error_bits = (qword >> I40E_RXD_QW1_ERROR_SHIFT);
123
124 #define I40E_RX_ERR_BITS 0x3f
125         if (likely((error_bits & I40E_RX_ERR_BITS) == 0)) {
126                 flags |= (PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_GOOD);
127                 return flags;
128         }
129
130         if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_IPE_SHIFT)))
131                 flags |= PKT_RX_IP_CKSUM_BAD;
132         else
133                 flags |= PKT_RX_IP_CKSUM_GOOD;
134
135         if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_L4E_SHIFT)))
136                 flags |= PKT_RX_L4_CKSUM_BAD;
137         else
138                 flags |= PKT_RX_L4_CKSUM_GOOD;
139
140         if (unlikely(error_bits & (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT)))
141                 flags |= PKT_RX_EIP_CKSUM_BAD;
142
143         return flags;
144 }
145
146 /* Function to check and set the ieee1588 timesync index and get the
147  * appropriate flags.
148  */
149 #ifdef RTE_LIBRTE_IEEE1588
150 static inline uint64_t
151 i40e_get_iee15888_flags(struct rte_mbuf *mb, uint64_t qword)
152 {
153         uint64_t pkt_flags = 0;
154         uint16_t tsyn = (qword & (I40E_RXD_QW1_STATUS_TSYNVALID_MASK
155                                   | I40E_RXD_QW1_STATUS_TSYNINDX_MASK))
156                                     >> I40E_RX_DESC_STATUS_TSYNINDX_SHIFT;
157
158         if ((mb->packet_type & RTE_PTYPE_L2_MASK)
159                         == RTE_PTYPE_L2_ETHER_TIMESYNC)
160                 pkt_flags = PKT_RX_IEEE1588_PTP;
161         if (tsyn & 0x04) {
162                 pkt_flags |= PKT_RX_IEEE1588_TMST;
163                 mb->timesync = tsyn & 0x03;
164         }
165
166         return pkt_flags;
167 }
168 #endif
169
170 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_MASK   0x03
171 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_FD_ID  0x01
172 #define I40E_RX_DESC_EXT_STATUS_FLEXBH_FLEX   0x02
173 #define I40E_RX_DESC_EXT_STATUS_FLEXBL_MASK   0x03
174 #define I40E_RX_DESC_EXT_STATUS_FLEXBL_FLEX   0x01
175
176 static inline uint64_t
177 i40e_rxd_build_fdir(volatile union i40e_rx_desc *rxdp, struct rte_mbuf *mb)
178 {
179         uint64_t flags = 0;
180 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
181         uint16_t flexbh, flexbl;
182
183         flexbh = (rte_le_to_cpu_32(rxdp->wb.qword2.ext_status) >>
184                 I40E_RX_DESC_EXT_STATUS_FLEXBH_SHIFT) &
185                 I40E_RX_DESC_EXT_STATUS_FLEXBH_MASK;
186         flexbl = (rte_le_to_cpu_32(rxdp->wb.qword2.ext_status) >>
187                 I40E_RX_DESC_EXT_STATUS_FLEXBL_SHIFT) &
188                 I40E_RX_DESC_EXT_STATUS_FLEXBL_MASK;
189
190
191         if (flexbh == I40E_RX_DESC_EXT_STATUS_FLEXBH_FD_ID) {
192                 mb->hash.fdir.hi =
193                         rte_le_to_cpu_32(rxdp->wb.qword3.hi_dword.fd_id);
194                 flags |= PKT_RX_FDIR_ID;
195         } else if (flexbh == I40E_RX_DESC_EXT_STATUS_FLEXBH_FLEX) {
196                 mb->hash.fdir.hi =
197                         rte_le_to_cpu_32(rxdp->wb.qword3.hi_dword.flex_bytes_hi);
198                 flags |= PKT_RX_FDIR_FLX;
199         }
200         if (flexbl == I40E_RX_DESC_EXT_STATUS_FLEXBL_FLEX) {
201                 mb->hash.fdir.lo =
202                         rte_le_to_cpu_32(rxdp->wb.qword3.lo_dword.flex_bytes_lo);
203                 flags |= PKT_RX_FDIR_FLX;
204         }
205 #else
206         mb->hash.fdir.hi =
207                 rte_le_to_cpu_32(rxdp->wb.qword0.hi_dword.fd_id);
208         flags |= PKT_RX_FDIR_ID;
209 #endif
210         return flags;
211 }
212
213 static inline void
214 i40e_parse_tunneling_params(uint64_t ol_flags,
215                             union i40e_tx_offload tx_offload,
216                             uint32_t *cd_tunneling)
217 {
218         /* EIPT: External (outer) IP header type */
219         if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
220                 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
221         else if (ol_flags & PKT_TX_OUTER_IPV4)
222                 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
223         else if (ol_flags & PKT_TX_OUTER_IPV6)
224                 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
225
226         /* EIPLEN: External (outer) IP header length, in DWords */
227         *cd_tunneling |= (tx_offload.outer_l3_len >> 2) <<
228                 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
229
230         /* L4TUNT: L4 Tunneling Type */
231         switch (ol_flags & PKT_TX_TUNNEL_MASK) {
232         case PKT_TX_TUNNEL_IPIP:
233                 /* for non UDP / GRE tunneling, set to 00b */
234                 break;
235         case PKT_TX_TUNNEL_VXLAN:
236         case PKT_TX_TUNNEL_GENEVE:
237                 *cd_tunneling |= I40E_TXD_CTX_UDP_TUNNELING;
238                 break;
239         case PKT_TX_TUNNEL_GRE:
240                 *cd_tunneling |= I40E_TXD_CTX_GRE_TUNNELING;
241                 break;
242         default:
243                 PMD_TX_LOG(ERR, "Tunnel type not supported");
244                 return;
245         }
246
247         /* L4TUNLEN: L4 Tunneling Length, in Words
248          *
249          * We depend on app to set rte_mbuf.l2_len correctly.
250          * For IP in GRE it should be set to the length of the GRE
251          * header;
252          * for MAC in GRE or MAC in UDP it should be set to the length
253          * of the GRE or UDP headers plus the inner MAC up to including
254          * its last Ethertype.
255          */
256         *cd_tunneling |= (tx_offload.l2_len >> 1) <<
257                 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
258 }
259
260 static inline void
261 i40e_txd_enable_checksum(uint64_t ol_flags,
262                         uint32_t *td_cmd,
263                         uint32_t *td_offset,
264                         union i40e_tx_offload tx_offload)
265 {
266         /* Set MACLEN */
267         if (ol_flags & PKT_TX_TUNNEL_MASK)
268                 *td_offset |= (tx_offload.outer_l2_len >> 1)
269                                 << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
270         else
271                 *td_offset |= (tx_offload.l2_len >> 1)
272                         << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
273
274         /* Enable L3 checksum offloads */
275         if (ol_flags & PKT_TX_IP_CKSUM) {
276                 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
277                 *td_offset |= (tx_offload.l3_len >> 2)
278                                 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
279         } else if (ol_flags & PKT_TX_IPV4) {
280                 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
281                 *td_offset |= (tx_offload.l3_len >> 2)
282                                 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
283         } else if (ol_flags & PKT_TX_IPV6) {
284                 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
285                 *td_offset |= (tx_offload.l3_len >> 2)
286                                 << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
287         }
288
289         if (ol_flags & PKT_TX_TCP_SEG) {
290                 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
291                 *td_offset |= (tx_offload.l4_len >> 2)
292                         << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
293                 return;
294         }
295
296         /* Enable L4 checksum offloads */
297         switch (ol_flags & PKT_TX_L4_MASK) {
298         case PKT_TX_TCP_CKSUM:
299                 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
300                 *td_offset |= (sizeof(struct tcp_hdr) >> 2) <<
301                                 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
302                 break;
303         case PKT_TX_SCTP_CKSUM:
304                 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
305                 *td_offset |= (sizeof(struct sctp_hdr) >> 2) <<
306                                 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
307                 break;
308         case PKT_TX_UDP_CKSUM:
309                 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
310                 *td_offset |= (sizeof(struct udp_hdr) >> 2) <<
311                                 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
312                 break;
313         default:
314                 break;
315         }
316 }
317
318 /* Construct the tx flags */
319 static inline uint64_t
320 i40e_build_ctob(uint32_t td_cmd,
321                 uint32_t td_offset,
322                 unsigned int size,
323                 uint32_t td_tag)
324 {
325         return rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DATA |
326                         ((uint64_t)td_cmd  << I40E_TXD_QW1_CMD_SHIFT) |
327                         ((uint64_t)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
328                         ((uint64_t)size  << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
329                         ((uint64_t)td_tag  << I40E_TXD_QW1_L2TAG1_SHIFT));
330 }
331
332 static inline int
333 i40e_xmit_cleanup(struct i40e_tx_queue *txq)
334 {
335         struct i40e_tx_entry *sw_ring = txq->sw_ring;
336         volatile struct i40e_tx_desc *txd = txq->tx_ring;
337         uint16_t last_desc_cleaned = txq->last_desc_cleaned;
338         uint16_t nb_tx_desc = txq->nb_tx_desc;
339         uint16_t desc_to_clean_to;
340         uint16_t nb_tx_to_clean;
341
342         desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
343         if (desc_to_clean_to >= nb_tx_desc)
344                 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
345
346         desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
347         if ((txd[desc_to_clean_to].cmd_type_offset_bsz &
348                         rte_cpu_to_le_64(I40E_TXD_QW1_DTYPE_MASK)) !=
349                         rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE)) {
350                 PMD_TX_FREE_LOG(DEBUG, "TX descriptor %4u is not done "
351                         "(port=%d queue=%d)", desc_to_clean_to,
352                                 txq->port_id, txq->queue_id);
353                 return -1;
354         }
355
356         if (last_desc_cleaned > desc_to_clean_to)
357                 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
358                                                         desc_to_clean_to);
359         else
360                 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
361                                         last_desc_cleaned);
362
363         txd[desc_to_clean_to].cmd_type_offset_bsz = 0;
364
365         txq->last_desc_cleaned = desc_to_clean_to;
366         txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
367
368         return 0;
369 }
370
371 static inline int
372 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
373 check_rx_burst_bulk_alloc_preconditions(struct i40e_rx_queue *rxq)
374 #else
375 check_rx_burst_bulk_alloc_preconditions(__rte_unused struct i40e_rx_queue *rxq)
376 #endif
377 {
378         int ret = 0;
379
380 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
381         if (!(rxq->rx_free_thresh >= RTE_PMD_I40E_RX_MAX_BURST)) {
382                 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
383                              "rxq->rx_free_thresh=%d, "
384                              "RTE_PMD_I40E_RX_MAX_BURST=%d",
385                              rxq->rx_free_thresh, RTE_PMD_I40E_RX_MAX_BURST);
386                 ret = -EINVAL;
387         } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
388                 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
389                              "rxq->rx_free_thresh=%d, "
390                              "rxq->nb_rx_desc=%d",
391                              rxq->rx_free_thresh, rxq->nb_rx_desc);
392                 ret = -EINVAL;
393         } else if (rxq->nb_rx_desc % rxq->rx_free_thresh != 0) {
394                 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
395                              "rxq->nb_rx_desc=%d, "
396                              "rxq->rx_free_thresh=%d",
397                              rxq->nb_rx_desc, rxq->rx_free_thresh);
398                 ret = -EINVAL;
399         }
400 #else
401         ret = -EINVAL;
402 #endif
403
404         return ret;
405 }
406
407 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
408 #define I40E_LOOK_AHEAD 8
409 #if (I40E_LOOK_AHEAD != 8)
410 #error "PMD I40E: I40E_LOOK_AHEAD must be 8\n"
411 #endif
412 static inline int
413 i40e_rx_scan_hw_ring(struct i40e_rx_queue *rxq)
414 {
415         volatile union i40e_rx_desc *rxdp;
416         struct i40e_rx_entry *rxep;
417         struct rte_mbuf *mb;
418         uint16_t pkt_len;
419         uint64_t qword1;
420         uint32_t rx_status;
421         int32_t s[I40E_LOOK_AHEAD], nb_dd;
422         int32_t i, j, nb_rx = 0;
423         uint64_t pkt_flags;
424         uint32_t *ptype_tbl = rxq->vsi->adapter->ptype_tbl;
425
426         rxdp = &rxq->rx_ring[rxq->rx_tail];
427         rxep = &rxq->sw_ring[rxq->rx_tail];
428
429         qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
430         rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
431                                 I40E_RXD_QW1_STATUS_SHIFT;
432
433         /* Make sure there is at least 1 packet to receive */
434         if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
435                 return 0;
436
437         /**
438          * Scan LOOK_AHEAD descriptors at a time to determine which
439          * descriptors reference packets that are ready to be received.
440          */
441         for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; i+=I40E_LOOK_AHEAD,
442                         rxdp += I40E_LOOK_AHEAD, rxep += I40E_LOOK_AHEAD) {
443                 /* Read desc statuses backwards to avoid race condition */
444                 for (j = I40E_LOOK_AHEAD - 1; j >= 0; j--) {
445                         qword1 = rte_le_to_cpu_64(\
446                                 rxdp[j].wb.qword1.status_error_len);
447                         s[j] = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
448                                         I40E_RXD_QW1_STATUS_SHIFT;
449                 }
450
451                 rte_smp_rmb();
452
453                 /* Compute how many status bits were set */
454                 for (j = 0, nb_dd = 0; j < I40E_LOOK_AHEAD; j++)
455                         nb_dd += s[j] & (1 << I40E_RX_DESC_STATUS_DD_SHIFT);
456
457                 nb_rx += nb_dd;
458
459                 /* Translate descriptor info to mbuf parameters */
460                 for (j = 0; j < nb_dd; j++) {
461                         mb = rxep[j].mbuf;
462                         qword1 = rte_le_to_cpu_64(\
463                                 rxdp[j].wb.qword1.status_error_len);
464                         pkt_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
465                                 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
466                         mb->data_len = pkt_len;
467                         mb->pkt_len = pkt_len;
468                         mb->ol_flags = 0;
469                         i40e_rxd_to_vlan_tci(mb, &rxdp[j]);
470                         pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
471                         pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
472                         mb->packet_type =
473                                 ptype_tbl[(uint8_t)((qword1 &
474                                 I40E_RXD_QW1_PTYPE_MASK) >>
475                                 I40E_RXD_QW1_PTYPE_SHIFT)];
476                         if (pkt_flags & PKT_RX_RSS_HASH)
477                                 mb->hash.rss = rte_le_to_cpu_32(\
478                                         rxdp[j].wb.qword0.hi_dword.rss);
479                         if (pkt_flags & PKT_RX_FDIR)
480                                 pkt_flags |= i40e_rxd_build_fdir(&rxdp[j], mb);
481
482 #ifdef RTE_LIBRTE_IEEE1588
483                         pkt_flags |= i40e_get_iee15888_flags(mb, qword1);
484 #endif
485                         mb->ol_flags |= pkt_flags;
486
487                 }
488
489                 for (j = 0; j < I40E_LOOK_AHEAD; j++)
490                         rxq->rx_stage[i + j] = rxep[j].mbuf;
491
492                 if (nb_dd != I40E_LOOK_AHEAD)
493                         break;
494         }
495
496         /* Clear software ring entries */
497         for (i = 0; i < nb_rx; i++)
498                 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
499
500         return nb_rx;
501 }
502
503 static inline uint16_t
504 i40e_rx_fill_from_stage(struct i40e_rx_queue *rxq,
505                         struct rte_mbuf **rx_pkts,
506                         uint16_t nb_pkts)
507 {
508         uint16_t i;
509         struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
510
511         nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
512
513         for (i = 0; i < nb_pkts; i++)
514                 rx_pkts[i] = stage[i];
515
516         rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
517         rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
518
519         return nb_pkts;
520 }
521
522 static inline int
523 i40e_rx_alloc_bufs(struct i40e_rx_queue *rxq)
524 {
525         volatile union i40e_rx_desc *rxdp;
526         struct i40e_rx_entry *rxep;
527         struct rte_mbuf *mb;
528         uint16_t alloc_idx, i;
529         uint64_t dma_addr;
530         int diag;
531
532         /* Allocate buffers in bulk */
533         alloc_idx = (uint16_t)(rxq->rx_free_trigger -
534                                 (rxq->rx_free_thresh - 1));
535         rxep = &(rxq->sw_ring[alloc_idx]);
536         diag = rte_mempool_get_bulk(rxq->mp, (void *)rxep,
537                                         rxq->rx_free_thresh);
538         if (unlikely(diag != 0)) {
539                 PMD_DRV_LOG(ERR, "Failed to get mbufs in bulk");
540                 return -ENOMEM;
541         }
542
543         rxdp = &rxq->rx_ring[alloc_idx];
544         for (i = 0; i < rxq->rx_free_thresh; i++) {
545                 if (likely(i < (rxq->rx_free_thresh - 1)))
546                         /* Prefetch next mbuf */
547                         rte_prefetch0(rxep[i + 1].mbuf);
548
549                 mb = rxep[i].mbuf;
550                 rte_mbuf_refcnt_set(mb, 1);
551                 mb->next = NULL;
552                 mb->data_off = RTE_PKTMBUF_HEADROOM;
553                 mb->nb_segs = 1;
554                 mb->port = rxq->port_id;
555                 dma_addr = rte_cpu_to_le_64(\
556                         rte_mbuf_data_iova_default(mb));
557                 rxdp[i].read.hdr_addr = 0;
558                 rxdp[i].read.pkt_addr = dma_addr;
559         }
560
561         /* Update rx tail regsiter */
562         rte_wmb();
563         I40E_PCI_REG_WRITE_RELAXED(rxq->qrx_tail, rxq->rx_free_trigger);
564
565         rxq->rx_free_trigger =
566                 (uint16_t)(rxq->rx_free_trigger + rxq->rx_free_thresh);
567         if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
568                 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
569
570         return 0;
571 }
572
573 static inline uint16_t
574 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
575 {
576         struct i40e_rx_queue *rxq = (struct i40e_rx_queue *)rx_queue;
577         struct rte_eth_dev *dev;
578         uint16_t nb_rx = 0;
579
580         if (!nb_pkts)
581                 return 0;
582
583         if (rxq->rx_nb_avail)
584                 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
585
586         nb_rx = (uint16_t)i40e_rx_scan_hw_ring(rxq);
587         rxq->rx_next_avail = 0;
588         rxq->rx_nb_avail = nb_rx;
589         rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
590
591         if (rxq->rx_tail > rxq->rx_free_trigger) {
592                 if (i40e_rx_alloc_bufs(rxq) != 0) {
593                         uint16_t i, j;
594
595                         dev = I40E_VSI_TO_ETH_DEV(rxq->vsi);
596                         dev->data->rx_mbuf_alloc_failed +=
597                                 rxq->rx_free_thresh;
598
599                         rxq->rx_nb_avail = 0;
600                         rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
601                         for (i = 0, j = rxq->rx_tail; i < nb_rx; i++, j++)
602                                 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
603
604                         return 0;
605                 }
606         }
607
608         if (rxq->rx_tail >= rxq->nb_rx_desc)
609                 rxq->rx_tail = 0;
610
611         if (rxq->rx_nb_avail)
612                 return i40e_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
613
614         return 0;
615 }
616
617 static uint16_t
618 i40e_recv_pkts_bulk_alloc(void *rx_queue,
619                           struct rte_mbuf **rx_pkts,
620                           uint16_t nb_pkts)
621 {
622         uint16_t nb_rx = 0, n, count;
623
624         if (unlikely(nb_pkts == 0))
625                 return 0;
626
627         if (likely(nb_pkts <= RTE_PMD_I40E_RX_MAX_BURST))
628                 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
629
630         while (nb_pkts) {
631                 n = RTE_MIN(nb_pkts, RTE_PMD_I40E_RX_MAX_BURST);
632                 count = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
633                 nb_rx = (uint16_t)(nb_rx + count);
634                 nb_pkts = (uint16_t)(nb_pkts - count);
635                 if (count < n)
636                         break;
637         }
638
639         return nb_rx;
640 }
641 #else
642 static uint16_t
643 i40e_recv_pkts_bulk_alloc(void __rte_unused *rx_queue,
644                           struct rte_mbuf __rte_unused **rx_pkts,
645                           uint16_t __rte_unused nb_pkts)
646 {
647         return 0;
648 }
649 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
650
651 uint16_t
652 i40e_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
653 {
654         struct i40e_rx_queue *rxq;
655         volatile union i40e_rx_desc *rx_ring;
656         volatile union i40e_rx_desc *rxdp;
657         union i40e_rx_desc rxd;
658         struct i40e_rx_entry *sw_ring;
659         struct i40e_rx_entry *rxe;
660         struct rte_eth_dev *dev;
661         struct rte_mbuf *rxm;
662         struct rte_mbuf *nmb;
663         uint16_t nb_rx;
664         uint32_t rx_status;
665         uint64_t qword1;
666         uint16_t rx_packet_len;
667         uint16_t rx_id, nb_hold;
668         uint64_t dma_addr;
669         uint64_t pkt_flags;
670         uint32_t *ptype_tbl;
671
672         nb_rx = 0;
673         nb_hold = 0;
674         rxq = rx_queue;
675         rx_id = rxq->rx_tail;
676         rx_ring = rxq->rx_ring;
677         sw_ring = rxq->sw_ring;
678         ptype_tbl = rxq->vsi->adapter->ptype_tbl;
679
680         while (nb_rx < nb_pkts) {
681                 rxdp = &rx_ring[rx_id];
682                 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
683                 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK)
684                                 >> I40E_RXD_QW1_STATUS_SHIFT;
685
686                 /* Check the DD bit first */
687                 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
688                         break;
689
690                 nmb = rte_mbuf_raw_alloc(rxq->mp);
691                 if (unlikely(!nmb)) {
692                         dev = I40E_VSI_TO_ETH_DEV(rxq->vsi);
693                         dev->data->rx_mbuf_alloc_failed++;
694                         break;
695                 }
696
697                 rxd = *rxdp;
698                 nb_hold++;
699                 rxe = &sw_ring[rx_id];
700                 rx_id++;
701                 if (unlikely(rx_id == rxq->nb_rx_desc))
702                         rx_id = 0;
703
704                 /* Prefetch next mbuf */
705                 rte_prefetch0(sw_ring[rx_id].mbuf);
706
707                 /**
708                  * When next RX descriptor is on a cache line boundary,
709                  * prefetch the next 4 RX descriptors and next 8 pointers
710                  * to mbufs.
711                  */
712                 if ((rx_id & 0x3) == 0) {
713                         rte_prefetch0(&rx_ring[rx_id]);
714                         rte_prefetch0(&sw_ring[rx_id]);
715                 }
716                 rxm = rxe->mbuf;
717                 rxe->mbuf = nmb;
718                 dma_addr =
719                         rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb));
720                 rxdp->read.hdr_addr = 0;
721                 rxdp->read.pkt_addr = dma_addr;
722
723                 rx_packet_len = ((qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
724                                 I40E_RXD_QW1_LENGTH_PBUF_SHIFT) - rxq->crc_len;
725
726                 rxm->data_off = RTE_PKTMBUF_HEADROOM;
727                 rte_prefetch0(RTE_PTR_ADD(rxm->buf_addr, RTE_PKTMBUF_HEADROOM));
728                 rxm->nb_segs = 1;
729                 rxm->next = NULL;
730                 rxm->pkt_len = rx_packet_len;
731                 rxm->data_len = rx_packet_len;
732                 rxm->port = rxq->port_id;
733                 rxm->ol_flags = 0;
734                 i40e_rxd_to_vlan_tci(rxm, &rxd);
735                 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
736                 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
737                 rxm->packet_type =
738                         ptype_tbl[(uint8_t)((qword1 &
739                         I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT)];
740                 if (pkt_flags & PKT_RX_RSS_HASH)
741                         rxm->hash.rss =
742                                 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
743                 if (pkt_flags & PKT_RX_FDIR)
744                         pkt_flags |= i40e_rxd_build_fdir(&rxd, rxm);
745
746 #ifdef RTE_LIBRTE_IEEE1588
747                 pkt_flags |= i40e_get_iee15888_flags(rxm, qword1);
748 #endif
749                 rxm->ol_flags |= pkt_flags;
750
751                 rx_pkts[nb_rx++] = rxm;
752         }
753         rxq->rx_tail = rx_id;
754
755         /**
756          * If the number of free RX descriptors is greater than the RX free
757          * threshold of the queue, advance the receive tail register of queue.
758          * Update that register with the value of the last processed RX
759          * descriptor minus 1.
760          */
761         nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
762         if (nb_hold > rxq->rx_free_thresh) {
763                 rx_id = (uint16_t) ((rx_id == 0) ?
764                         (rxq->nb_rx_desc - 1) : (rx_id - 1));
765                 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
766                 nb_hold = 0;
767         }
768         rxq->nb_rx_hold = nb_hold;
769
770         return nb_rx;
771 }
772
773 uint16_t
774 i40e_recv_scattered_pkts(void *rx_queue,
775                          struct rte_mbuf **rx_pkts,
776                          uint16_t nb_pkts)
777 {
778         struct i40e_rx_queue *rxq = rx_queue;
779         volatile union i40e_rx_desc *rx_ring = rxq->rx_ring;
780         volatile union i40e_rx_desc *rxdp;
781         union i40e_rx_desc rxd;
782         struct i40e_rx_entry *sw_ring = rxq->sw_ring;
783         struct i40e_rx_entry *rxe;
784         struct rte_mbuf *first_seg = rxq->pkt_first_seg;
785         struct rte_mbuf *last_seg = rxq->pkt_last_seg;
786         struct rte_mbuf *nmb, *rxm;
787         uint16_t rx_id = rxq->rx_tail;
788         uint16_t nb_rx = 0, nb_hold = 0, rx_packet_len;
789         struct rte_eth_dev *dev;
790         uint32_t rx_status;
791         uint64_t qword1;
792         uint64_t dma_addr;
793         uint64_t pkt_flags;
794         uint32_t *ptype_tbl = rxq->vsi->adapter->ptype_tbl;
795
796         while (nb_rx < nb_pkts) {
797                 rxdp = &rx_ring[rx_id];
798                 qword1 = rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len);
799                 rx_status = (qword1 & I40E_RXD_QW1_STATUS_MASK) >>
800                                         I40E_RXD_QW1_STATUS_SHIFT;
801
802                 /* Check the DD bit */
803                 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
804                         break;
805
806                 nmb = rte_mbuf_raw_alloc(rxq->mp);
807                 if (unlikely(!nmb)) {
808                         dev = I40E_VSI_TO_ETH_DEV(rxq->vsi);
809                         dev->data->rx_mbuf_alloc_failed++;
810                         break;
811                 }
812
813                 rxd = *rxdp;
814                 nb_hold++;
815                 rxe = &sw_ring[rx_id];
816                 rx_id++;
817                 if (rx_id == rxq->nb_rx_desc)
818                         rx_id = 0;
819
820                 /* Prefetch next mbuf */
821                 rte_prefetch0(sw_ring[rx_id].mbuf);
822
823                 /**
824                  * When next RX descriptor is on a cache line boundary,
825                  * prefetch the next 4 RX descriptors and next 8 pointers
826                  * to mbufs.
827                  */
828                 if ((rx_id & 0x3) == 0) {
829                         rte_prefetch0(&rx_ring[rx_id]);
830                         rte_prefetch0(&sw_ring[rx_id]);
831                 }
832
833                 rxm = rxe->mbuf;
834                 rxe->mbuf = nmb;
835                 dma_addr =
836                         rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb));
837
838                 /* Set data buffer address and data length of the mbuf */
839                 rxdp->read.hdr_addr = 0;
840                 rxdp->read.pkt_addr = dma_addr;
841                 rx_packet_len = (qword1 & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
842                                         I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
843                 rxm->data_len = rx_packet_len;
844                 rxm->data_off = RTE_PKTMBUF_HEADROOM;
845
846                 /**
847                  * If this is the first buffer of the received packet, set the
848                  * pointer to the first mbuf of the packet and initialize its
849                  * context. Otherwise, update the total length and the number
850                  * of segments of the current scattered packet, and update the
851                  * pointer to the last mbuf of the current packet.
852                  */
853                 if (!first_seg) {
854                         first_seg = rxm;
855                         first_seg->nb_segs = 1;
856                         first_seg->pkt_len = rx_packet_len;
857                 } else {
858                         first_seg->pkt_len =
859                                 (uint16_t)(first_seg->pkt_len +
860                                                 rx_packet_len);
861                         first_seg->nb_segs++;
862                         last_seg->next = rxm;
863                 }
864
865                 /**
866                  * If this is not the last buffer of the received packet,
867                  * update the pointer to the last mbuf of the current scattered
868                  * packet and continue to parse the RX ring.
869                  */
870                 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT))) {
871                         last_seg = rxm;
872                         continue;
873                 }
874
875                 /**
876                  * This is the last buffer of the received packet. If the CRC
877                  * is not stripped by the hardware:
878                  *  - Subtract the CRC length from the total packet length.
879                  *  - If the last buffer only contains the whole CRC or a part
880                  *  of it, free the mbuf associated to the last buffer. If part
881                  *  of the CRC is also contained in the previous mbuf, subtract
882                  *  the length of that CRC part from the data length of the
883                  *  previous mbuf.
884                  */
885                 rxm->next = NULL;
886                 if (unlikely(rxq->crc_len > 0)) {
887                         first_seg->pkt_len -= ETHER_CRC_LEN;
888                         if (rx_packet_len <= ETHER_CRC_LEN) {
889                                 rte_pktmbuf_free_seg(rxm);
890                                 first_seg->nb_segs--;
891                                 last_seg->data_len =
892                                         (uint16_t)(last_seg->data_len -
893                                         (ETHER_CRC_LEN - rx_packet_len));
894                                 last_seg->next = NULL;
895                         } else
896                                 rxm->data_len = (uint16_t)(rx_packet_len -
897                                                                 ETHER_CRC_LEN);
898                 }
899
900                 first_seg->port = rxq->port_id;
901                 first_seg->ol_flags = 0;
902                 i40e_rxd_to_vlan_tci(first_seg, &rxd);
903                 pkt_flags = i40e_rxd_status_to_pkt_flags(qword1);
904                 pkt_flags |= i40e_rxd_error_to_pkt_flags(qword1);
905                 first_seg->packet_type =
906                         ptype_tbl[(uint8_t)((qword1 &
907                         I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT)];
908                 if (pkt_flags & PKT_RX_RSS_HASH)
909                         first_seg->hash.rss =
910                                 rte_le_to_cpu_32(rxd.wb.qword0.hi_dword.rss);
911                 if (pkt_flags & PKT_RX_FDIR)
912                         pkt_flags |= i40e_rxd_build_fdir(&rxd, first_seg);
913
914 #ifdef RTE_LIBRTE_IEEE1588
915                 pkt_flags |= i40e_get_iee15888_flags(first_seg, qword1);
916 #endif
917                 first_seg->ol_flags |= pkt_flags;
918
919                 /* Prefetch data of first segment, if configured to do so. */
920                 rte_prefetch0(RTE_PTR_ADD(first_seg->buf_addr,
921                         first_seg->data_off));
922                 rx_pkts[nb_rx++] = first_seg;
923                 first_seg = NULL;
924         }
925
926         /* Record index of the next RX descriptor to probe. */
927         rxq->rx_tail = rx_id;
928         rxq->pkt_first_seg = first_seg;
929         rxq->pkt_last_seg = last_seg;
930
931         /**
932          * If the number of free RX descriptors is greater than the RX free
933          * threshold of the queue, advance the Receive Descriptor Tail (RDT)
934          * register. Update the RDT with the value of the last processed RX
935          * descriptor minus 1, to guarantee that the RDT register is never
936          * equal to the RDH register, which creates a "full" ring situtation
937          * from the hardware point of view.
938          */
939         nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
940         if (nb_hold > rxq->rx_free_thresh) {
941                 rx_id = (uint16_t)(rx_id == 0 ?
942                         (rxq->nb_rx_desc - 1) : (rx_id - 1));
943                 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
944                 nb_hold = 0;
945         }
946         rxq->nb_rx_hold = nb_hold;
947
948         return nb_rx;
949 }
950
951 /* Check if the context descriptor is needed for TX offloading */
952 static inline uint16_t
953 i40e_calc_context_desc(uint64_t flags)
954 {
955         static uint64_t mask = PKT_TX_OUTER_IP_CKSUM |
956                 PKT_TX_TCP_SEG |
957                 PKT_TX_QINQ_PKT |
958                 PKT_TX_TUNNEL_MASK;
959
960 #ifdef RTE_LIBRTE_IEEE1588
961         mask |= PKT_TX_IEEE1588_TMST;
962 #endif
963
964         return (flags & mask) ? 1 : 0;
965 }
966
967 /* set i40e TSO context descriptor */
968 static inline uint64_t
969 i40e_set_tso_ctx(struct rte_mbuf *mbuf, union i40e_tx_offload tx_offload)
970 {
971         uint64_t ctx_desc = 0;
972         uint32_t cd_cmd, hdr_len, cd_tso_len;
973
974         if (!tx_offload.l4_len) {
975                 PMD_DRV_LOG(DEBUG, "L4 length set to 0");
976                 return ctx_desc;
977         }
978
979         /**
980          * in case of non tunneling packet, the outer_l2_len and
981          * outer_l3_len must be 0.
982          */
983         hdr_len = tx_offload.outer_l2_len +
984                 tx_offload.outer_l3_len +
985                 tx_offload.l2_len +
986                 tx_offload.l3_len +
987                 tx_offload.l4_len;
988
989         cd_cmd = I40E_TX_CTX_DESC_TSO;
990         cd_tso_len = mbuf->pkt_len - hdr_len;
991         ctx_desc |= ((uint64_t)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
992                 ((uint64_t)cd_tso_len <<
993                  I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
994                 ((uint64_t)mbuf->tso_segsz <<
995                  I40E_TXD_CTX_QW1_MSS_SHIFT);
996
997         return ctx_desc;
998 }
999
1000 uint16_t
1001 i40e_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1002 {
1003         struct i40e_tx_queue *txq;
1004         struct i40e_tx_entry *sw_ring;
1005         struct i40e_tx_entry *txe, *txn;
1006         volatile struct i40e_tx_desc *txd;
1007         volatile struct i40e_tx_desc *txr;
1008         struct rte_mbuf *tx_pkt;
1009         struct rte_mbuf *m_seg;
1010         uint32_t cd_tunneling_params;
1011         uint16_t tx_id;
1012         uint16_t nb_tx;
1013         uint32_t td_cmd;
1014         uint32_t td_offset;
1015         uint32_t td_tag;
1016         uint64_t ol_flags;
1017         uint16_t nb_used;
1018         uint16_t nb_ctx;
1019         uint16_t tx_last;
1020         uint16_t slen;
1021         uint64_t buf_dma_addr;
1022         union i40e_tx_offload tx_offload = {0};
1023
1024         txq = tx_queue;
1025         sw_ring = txq->sw_ring;
1026         txr = txq->tx_ring;
1027         tx_id = txq->tx_tail;
1028         txe = &sw_ring[tx_id];
1029
1030         /* Check if the descriptor ring needs to be cleaned. */
1031         if (txq->nb_tx_free < txq->tx_free_thresh)
1032                 i40e_xmit_cleanup(txq);
1033
1034         for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
1035                 td_cmd = 0;
1036                 td_tag = 0;
1037                 td_offset = 0;
1038
1039                 tx_pkt = *tx_pkts++;
1040                 RTE_MBUF_PREFETCH_TO_FREE(txe->mbuf);
1041
1042                 ol_flags = tx_pkt->ol_flags;
1043                 tx_offload.l2_len = tx_pkt->l2_len;
1044                 tx_offload.l3_len = tx_pkt->l3_len;
1045                 tx_offload.outer_l2_len = tx_pkt->outer_l2_len;
1046                 tx_offload.outer_l3_len = tx_pkt->outer_l3_len;
1047                 tx_offload.l4_len = tx_pkt->l4_len;
1048                 tx_offload.tso_segsz = tx_pkt->tso_segsz;
1049
1050                 /* Calculate the number of context descriptors needed. */
1051                 nb_ctx = i40e_calc_context_desc(ol_flags);
1052
1053                 /**
1054                  * The number of descriptors that must be allocated for
1055                  * a packet equals to the number of the segments of that
1056                  * packet plus 1 context descriptor if needed.
1057                  */
1058                 nb_used = (uint16_t)(tx_pkt->nb_segs + nb_ctx);
1059                 tx_last = (uint16_t)(tx_id + nb_used - 1);
1060
1061                 /* Circular ring */
1062                 if (tx_last >= txq->nb_tx_desc)
1063                         tx_last = (uint16_t)(tx_last - txq->nb_tx_desc);
1064
1065                 if (nb_used > txq->nb_tx_free) {
1066                         if (i40e_xmit_cleanup(txq) != 0) {
1067                                 if (nb_tx == 0)
1068                                         return 0;
1069                                 goto end_of_tx;
1070                         }
1071                         if (unlikely(nb_used > txq->tx_rs_thresh)) {
1072                                 while (nb_used > txq->nb_tx_free) {
1073                                         if (i40e_xmit_cleanup(txq) != 0) {
1074                                                 if (nb_tx == 0)
1075                                                         return 0;
1076                                                 goto end_of_tx;
1077                                         }
1078                                 }
1079                         }
1080                 }
1081
1082                 /* Descriptor based VLAN insertion */
1083                 if (ol_flags & (PKT_TX_VLAN_PKT | PKT_TX_QINQ_PKT)) {
1084                         td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1085                         td_tag = tx_pkt->vlan_tci;
1086                 }
1087
1088                 /* Always enable CRC offload insertion */
1089                 td_cmd |= I40E_TX_DESC_CMD_ICRC;
1090
1091                 /* Fill in tunneling parameters if necessary */
1092                 cd_tunneling_params = 0;
1093                 if (ol_flags & PKT_TX_TUNNEL_MASK)
1094                         i40e_parse_tunneling_params(ol_flags, tx_offload,
1095                                                     &cd_tunneling_params);
1096                 /* Enable checksum offloading */
1097                 if (ol_flags & I40E_TX_CKSUM_OFFLOAD_MASK)
1098                         i40e_txd_enable_checksum(ol_flags, &td_cmd,
1099                                                  &td_offset, tx_offload);
1100
1101                 if (nb_ctx) {
1102                         /* Setup TX context descriptor if required */
1103                         volatile struct i40e_tx_context_desc *ctx_txd =
1104                                 (volatile struct i40e_tx_context_desc *)\
1105                                                         &txr[tx_id];
1106                         uint16_t cd_l2tag2 = 0;
1107                         uint64_t cd_type_cmd_tso_mss =
1108                                 I40E_TX_DESC_DTYPE_CONTEXT;
1109
1110                         txn = &sw_ring[txe->next_id];
1111                         RTE_MBUF_PREFETCH_TO_FREE(txn->mbuf);
1112                         if (txe->mbuf != NULL) {
1113                                 rte_pktmbuf_free_seg(txe->mbuf);
1114                                 txe->mbuf = NULL;
1115                         }
1116
1117                         /* TSO enabled means no timestamp */
1118                         if (ol_flags & PKT_TX_TCP_SEG)
1119                                 cd_type_cmd_tso_mss |=
1120                                         i40e_set_tso_ctx(tx_pkt, tx_offload);
1121                         else {
1122 #ifdef RTE_LIBRTE_IEEE1588
1123                                 if (ol_flags & PKT_TX_IEEE1588_TMST)
1124                                         cd_type_cmd_tso_mss |=
1125                                                 ((uint64_t)I40E_TX_CTX_DESC_TSYN <<
1126                                                  I40E_TXD_CTX_QW1_CMD_SHIFT);
1127 #endif
1128                         }
1129
1130                         ctx_txd->tunneling_params =
1131                                 rte_cpu_to_le_32(cd_tunneling_params);
1132                         if (ol_flags & PKT_TX_QINQ_PKT) {
1133                                 cd_l2tag2 = tx_pkt->vlan_tci_outer;
1134                                 cd_type_cmd_tso_mss |=
1135                                         ((uint64_t)I40E_TX_CTX_DESC_IL2TAG2 <<
1136                                                 I40E_TXD_CTX_QW1_CMD_SHIFT);
1137                         }
1138                         ctx_txd->l2tag2 = rte_cpu_to_le_16(cd_l2tag2);
1139                         ctx_txd->type_cmd_tso_mss =
1140                                 rte_cpu_to_le_64(cd_type_cmd_tso_mss);
1141
1142                         PMD_TX_LOG(DEBUG, "mbuf: %p, TCD[%u]:\n"
1143                                 "tunneling_params: %#x;\n"
1144                                 "l2tag2: %#hx;\n"
1145                                 "rsvd: %#hx;\n"
1146                                 "type_cmd_tso_mss: %#"PRIx64";\n",
1147                                 tx_pkt, tx_id,
1148                                 ctx_txd->tunneling_params,
1149                                 ctx_txd->l2tag2,
1150                                 ctx_txd->rsvd,
1151                                 ctx_txd->type_cmd_tso_mss);
1152
1153                         txe->last_id = tx_last;
1154                         tx_id = txe->next_id;
1155                         txe = txn;
1156                 }
1157
1158                 m_seg = tx_pkt;
1159                 do {
1160                         txd = &txr[tx_id];
1161                         txn = &sw_ring[txe->next_id];
1162
1163                         if (txe->mbuf)
1164                                 rte_pktmbuf_free_seg(txe->mbuf);
1165                         txe->mbuf = m_seg;
1166
1167                         /* Setup TX Descriptor */
1168                         slen = m_seg->data_len;
1169                         buf_dma_addr = rte_mbuf_data_iova(m_seg);
1170
1171                         PMD_TX_LOG(DEBUG, "mbuf: %p, TDD[%u]:\n"
1172                                 "buf_dma_addr: %#"PRIx64";\n"
1173                                 "td_cmd: %#x;\n"
1174                                 "td_offset: %#x;\n"
1175                                 "td_len: %u;\n"
1176                                 "td_tag: %#x;\n",
1177                                 tx_pkt, tx_id, buf_dma_addr,
1178                                 td_cmd, td_offset, slen, td_tag);
1179
1180                         txd->buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
1181                         txd->cmd_type_offset_bsz = i40e_build_ctob(td_cmd,
1182                                                 td_offset, slen, td_tag);
1183                         txe->last_id = tx_last;
1184                         tx_id = txe->next_id;
1185                         txe = txn;
1186                         m_seg = m_seg->next;
1187                 } while (m_seg != NULL);
1188
1189                 /* The last packet data descriptor needs End Of Packet (EOP) */
1190                 td_cmd |= I40E_TX_DESC_CMD_EOP;
1191                 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
1192                 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
1193
1194                 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
1195                         PMD_TX_FREE_LOG(DEBUG,
1196                                         "Setting RS bit on TXD id="
1197                                         "%4u (port=%d queue=%d)",
1198                                         tx_last, txq->port_id, txq->queue_id);
1199
1200                         td_cmd |= I40E_TX_DESC_CMD_RS;
1201
1202                         /* Update txq RS bit counters */
1203                         txq->nb_tx_used = 0;
1204                 }
1205
1206                 txd->cmd_type_offset_bsz |=
1207                         rte_cpu_to_le_64(((uint64_t)td_cmd) <<
1208                                         I40E_TXD_QW1_CMD_SHIFT);
1209         }
1210
1211 end_of_tx:
1212         rte_wmb();
1213
1214         PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
1215                    (unsigned) txq->port_id, (unsigned) txq->queue_id,
1216                    (unsigned) tx_id, (unsigned) nb_tx);
1217
1218         I40E_PCI_REG_WRITE_RELAXED(txq->qtx_tail, tx_id);
1219         txq->tx_tail = tx_id;
1220
1221         return nb_tx;
1222 }
1223
1224 static __rte_always_inline int
1225 i40e_tx_free_bufs(struct i40e_tx_queue *txq)
1226 {
1227         struct i40e_tx_entry *txep;
1228         uint16_t i;
1229
1230         if ((txq->tx_ring[txq->tx_next_dd].cmd_type_offset_bsz &
1231                         rte_cpu_to_le_64(I40E_TXD_QW1_DTYPE_MASK)) !=
1232                         rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE))
1233                 return 0;
1234
1235         txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
1236
1237         for (i = 0; i < txq->tx_rs_thresh; i++)
1238                 rte_prefetch0((txep + i)->mbuf);
1239
1240         if (txq->offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE) {
1241                 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1242                         rte_mempool_put(txep->mbuf->pool, txep->mbuf);
1243                         txep->mbuf = NULL;
1244                 }
1245         } else {
1246                 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
1247                         rte_pktmbuf_free_seg(txep->mbuf);
1248                         txep->mbuf = NULL;
1249                 }
1250         }
1251
1252         txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
1253         txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
1254         if (txq->tx_next_dd >= txq->nb_tx_desc)
1255                 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
1256
1257         return txq->tx_rs_thresh;
1258 }
1259
1260 /* Populate 4 descriptors with data from 4 mbufs */
1261 static inline void
1262 tx4(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1263 {
1264         uint64_t dma_addr;
1265         uint32_t i;
1266
1267         for (i = 0; i < 4; i++, txdp++, pkts++) {
1268                 dma_addr = rte_mbuf_data_iova(*pkts);
1269                 txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1270                 txdp->cmd_type_offset_bsz =
1271                         i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1272                                         (*pkts)->data_len, 0);
1273         }
1274 }
1275
1276 /* Populate 1 descriptor with data from 1 mbuf */
1277 static inline void
1278 tx1(volatile struct i40e_tx_desc *txdp, struct rte_mbuf **pkts)
1279 {
1280         uint64_t dma_addr;
1281
1282         dma_addr = rte_mbuf_data_iova(*pkts);
1283         txdp->buffer_addr = rte_cpu_to_le_64(dma_addr);
1284         txdp->cmd_type_offset_bsz =
1285                 i40e_build_ctob((uint32_t)I40E_TD_CMD, 0,
1286                                 (*pkts)->data_len, 0);
1287 }
1288
1289 /* Fill hardware descriptor ring with mbuf data */
1290 static inline void
1291 i40e_tx_fill_hw_ring(struct i40e_tx_queue *txq,
1292                      struct rte_mbuf **pkts,
1293                      uint16_t nb_pkts)
1294 {
1295         volatile struct i40e_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
1296         struct i40e_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
1297         const int N_PER_LOOP = 4;
1298         const int N_PER_LOOP_MASK = N_PER_LOOP - 1;
1299         int mainpart, leftover;
1300         int i, j;
1301
1302         mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
1303         leftover = (nb_pkts & ((uint32_t)  N_PER_LOOP_MASK));
1304         for (i = 0; i < mainpart; i += N_PER_LOOP) {
1305                 for (j = 0; j < N_PER_LOOP; ++j) {
1306                         (txep + i + j)->mbuf = *(pkts + i + j);
1307                 }
1308                 tx4(txdp + i, pkts + i);
1309         }
1310         if (unlikely(leftover > 0)) {
1311                 for (i = 0; i < leftover; ++i) {
1312                         (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
1313                         tx1(txdp + mainpart + i, pkts + mainpart + i);
1314                 }
1315         }
1316 }
1317
1318 static inline uint16_t
1319 tx_xmit_pkts(struct i40e_tx_queue *txq,
1320              struct rte_mbuf **tx_pkts,
1321              uint16_t nb_pkts)
1322 {
1323         volatile struct i40e_tx_desc *txr = txq->tx_ring;
1324         uint16_t n = 0;
1325
1326         /**
1327          * Begin scanning the H/W ring for done descriptors when the number
1328          * of available descriptors drops below tx_free_thresh. For each done
1329          * descriptor, free the associated buffer.
1330          */
1331         if (txq->nb_tx_free < txq->tx_free_thresh)
1332                 i40e_tx_free_bufs(txq);
1333
1334         /* Use available descriptor only */
1335         nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
1336         if (unlikely(!nb_pkts))
1337                 return 0;
1338
1339         txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
1340         if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
1341                 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
1342                 i40e_tx_fill_hw_ring(txq, tx_pkts, n);
1343                 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1344                         rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1345                                                 I40E_TXD_QW1_CMD_SHIFT);
1346                 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1347                 txq->tx_tail = 0;
1348         }
1349
1350         /* Fill hardware descriptor ring with mbuf data */
1351         i40e_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
1352         txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
1353
1354         /* Determin if RS bit needs to be set */
1355         if (txq->tx_tail > txq->tx_next_rs) {
1356                 txr[txq->tx_next_rs].cmd_type_offset_bsz |=
1357                         rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
1358                                                 I40E_TXD_QW1_CMD_SHIFT);
1359                 txq->tx_next_rs =
1360                         (uint16_t)(txq->tx_next_rs + txq->tx_rs_thresh);
1361                 if (txq->tx_next_rs >= txq->nb_tx_desc)
1362                         txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1363         }
1364
1365         if (txq->tx_tail >= txq->nb_tx_desc)
1366                 txq->tx_tail = 0;
1367
1368         /* Update the tx tail register */
1369         rte_wmb();
1370         I40E_PCI_REG_WRITE_RELAXED(txq->qtx_tail, txq->tx_tail);
1371
1372         return nb_pkts;
1373 }
1374
1375 static uint16_t
1376 i40e_xmit_pkts_simple(void *tx_queue,
1377                       struct rte_mbuf **tx_pkts,
1378                       uint16_t nb_pkts)
1379 {
1380         uint16_t nb_tx = 0;
1381
1382         if (likely(nb_pkts <= I40E_TX_MAX_BURST))
1383                 return tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1384                                                 tx_pkts, nb_pkts);
1385
1386         while (nb_pkts) {
1387                 uint16_t ret, num = (uint16_t)RTE_MIN(nb_pkts,
1388                                                 I40E_TX_MAX_BURST);
1389
1390                 ret = tx_xmit_pkts((struct i40e_tx_queue *)tx_queue,
1391                                                 &tx_pkts[nb_tx], num);
1392                 nb_tx = (uint16_t)(nb_tx + ret);
1393                 nb_pkts = (uint16_t)(nb_pkts - ret);
1394                 if (ret < num)
1395                         break;
1396         }
1397
1398         return nb_tx;
1399 }
1400
1401 static uint16_t
1402 i40e_xmit_pkts_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
1403                    uint16_t nb_pkts)
1404 {
1405         uint16_t nb_tx = 0;
1406         struct i40e_tx_queue *txq = (struct i40e_tx_queue *)tx_queue;
1407
1408         while (nb_pkts) {
1409                 uint16_t ret, num;
1410
1411                 num = (uint16_t)RTE_MIN(nb_pkts, txq->tx_rs_thresh);
1412                 ret = i40e_xmit_fixed_burst_vec(tx_queue, &tx_pkts[nb_tx],
1413                                                 num);
1414                 nb_tx += ret;
1415                 nb_pkts -= ret;
1416                 if (ret < num)
1417                         break;
1418         }
1419
1420         return nb_tx;
1421 }
1422
1423 /*********************************************************************
1424  *
1425  *  TX prep functions
1426  *
1427  **********************************************************************/
1428 uint16_t
1429 i40e_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
1430                 uint16_t nb_pkts)
1431 {
1432         int i, ret;
1433         uint64_t ol_flags;
1434         struct rte_mbuf *m;
1435
1436         for (i = 0; i < nb_pkts; i++) {
1437                 m = tx_pkts[i];
1438                 ol_flags = m->ol_flags;
1439
1440                 /* Check for m->nb_segs to not exceed the limits. */
1441                 if (!(ol_flags & PKT_TX_TCP_SEG)) {
1442                         if (m->nb_segs > I40E_TX_MAX_MTU_SEG) {
1443                                 rte_errno = -EINVAL;
1444                                 return i;
1445                         }
1446                 } else if (m->nb_segs > I40E_TX_MAX_SEG ||
1447                            m->tso_segsz < I40E_MIN_TSO_MSS ||
1448                            m->tso_segsz > I40E_MAX_TSO_MSS) {
1449                         /* MSS outside the range (256B - 9674B) are considered
1450                          * malicious
1451                          */
1452                         rte_errno = -EINVAL;
1453                         return i;
1454                 }
1455
1456                 if (ol_flags & I40E_TX_OFFLOAD_NOTSUP_MASK) {
1457                         rte_errno = -ENOTSUP;
1458                         return i;
1459                 }
1460
1461                 /* check the size of packet */
1462                 if (m->pkt_len > I40E_FRAME_SIZE_MAX ||
1463                     m->pkt_len < I40E_TX_MIN_PKT_LEN) {
1464                         rte_errno = -EINVAL;
1465                         return i;
1466                 }
1467
1468 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1469                 ret = rte_validate_tx_offload(m);
1470                 if (ret != 0) {
1471                         rte_errno = ret;
1472                         return i;
1473                 }
1474 #endif
1475                 ret = rte_net_intel_cksum_prepare(m);
1476                 if (ret != 0) {
1477                         rte_errno = ret;
1478                         return i;
1479                 }
1480         }
1481         return i;
1482 }
1483
1484 /*
1485  * Find the VSI the queue belongs to. 'queue_idx' is the queue index
1486  * application used, which assume having sequential ones. But from driver's
1487  * perspective, it's different. For example, q0 belongs to FDIR VSI, q1-q64
1488  * to MAIN VSI, , q65-96 to SRIOV VSIs, q97-128 to VMDQ VSIs. For application
1489  * running on host, q1-64 and q97-128 can be used, total 96 queues. They can
1490  * use queue_idx from 0 to 95 to access queues, while real queue would be
1491  * different. This function will do a queue mapping to find VSI the queue
1492  * belongs to.
1493  */
1494 static struct i40e_vsi*
1495 i40e_pf_get_vsi_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1496 {
1497         /* the queue in MAIN VSI range */
1498         if (queue_idx < pf->main_vsi->nb_qps)
1499                 return pf->main_vsi;
1500
1501         queue_idx -= pf->main_vsi->nb_qps;
1502
1503         /* queue_idx is greater than VMDQ VSIs range */
1504         if (queue_idx > pf->nb_cfg_vmdq_vsi * pf->vmdq_nb_qps - 1) {
1505                 PMD_INIT_LOG(ERR, "queue_idx out of range. VMDQ configured?");
1506                 return NULL;
1507         }
1508
1509         return pf->vmdq[queue_idx / pf->vmdq_nb_qps].vsi;
1510 }
1511
1512 static uint16_t
1513 i40e_get_queue_offset_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1514 {
1515         /* the queue in MAIN VSI range */
1516         if (queue_idx < pf->main_vsi->nb_qps)
1517                 return queue_idx;
1518
1519         /* It's VMDQ queues */
1520         queue_idx -= pf->main_vsi->nb_qps;
1521
1522         if (pf->nb_cfg_vmdq_vsi)
1523                 return queue_idx % pf->vmdq_nb_qps;
1524         else {
1525                 PMD_INIT_LOG(ERR, "Fail to get queue offset");
1526                 return (uint16_t)(-1);
1527         }
1528 }
1529
1530 int
1531 i40e_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1532 {
1533         struct i40e_rx_queue *rxq;
1534         int err = -1;
1535         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1536
1537         PMD_INIT_FUNC_TRACE();
1538
1539         if (rx_queue_id < dev->data->nb_rx_queues) {
1540                 rxq = dev->data->rx_queues[rx_queue_id];
1541
1542                 err = i40e_alloc_rx_queue_mbufs(rxq);
1543                 if (err) {
1544                         PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
1545                         return err;
1546                 }
1547
1548                 rte_wmb();
1549
1550                 /* Init the RX tail regieter. */
1551                 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1552
1553                 err = i40e_switch_rx_queue(hw, rxq->reg_idx, TRUE);
1554
1555                 if (err) {
1556                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
1557                                     rx_queue_id);
1558
1559                         i40e_rx_queue_release_mbufs(rxq);
1560                         i40e_reset_rx_queue(rxq);
1561                 } else
1562                         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1563         }
1564
1565         return err;
1566 }
1567
1568 int
1569 i40e_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1570 {
1571         struct i40e_rx_queue *rxq;
1572         int err;
1573         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1574
1575         if (rx_queue_id < dev->data->nb_rx_queues) {
1576                 rxq = dev->data->rx_queues[rx_queue_id];
1577
1578                 /*
1579                 * rx_queue_id is queue id application refers to, while
1580                 * rxq->reg_idx is the real queue index.
1581                 */
1582                 err = i40e_switch_rx_queue(hw, rxq->reg_idx, FALSE);
1583
1584                 if (err) {
1585                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
1586                                     rx_queue_id);
1587                         return err;
1588                 }
1589                 i40e_rx_queue_release_mbufs(rxq);
1590                 i40e_reset_rx_queue(rxq);
1591                 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1592         }
1593
1594         return 0;
1595 }
1596
1597 int
1598 i40e_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1599 {
1600         int err = -1;
1601         struct i40e_tx_queue *txq;
1602         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1603
1604         PMD_INIT_FUNC_TRACE();
1605
1606         if (tx_queue_id < dev->data->nb_tx_queues) {
1607                 txq = dev->data->tx_queues[tx_queue_id];
1608
1609                 /*
1610                 * tx_queue_id is queue id application refers to, while
1611                 * rxq->reg_idx is the real queue index.
1612                 */
1613                 err = i40e_switch_tx_queue(hw, txq->reg_idx, TRUE);
1614                 if (err)
1615                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
1616                                     tx_queue_id);
1617                 else
1618                         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1619         }
1620
1621         return err;
1622 }
1623
1624 int
1625 i40e_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1626 {
1627         struct i40e_tx_queue *txq;
1628         int err;
1629         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1630
1631         if (tx_queue_id < dev->data->nb_tx_queues) {
1632                 txq = dev->data->tx_queues[tx_queue_id];
1633
1634                 /*
1635                 * tx_queue_id is queue id application refers to, while
1636                 * txq->reg_idx is the real queue index.
1637                 */
1638                 err = i40e_switch_tx_queue(hw, txq->reg_idx, FALSE);
1639
1640                 if (err) {
1641                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u of",
1642                                     tx_queue_id);
1643                         return err;
1644                 }
1645
1646                 i40e_tx_queue_release_mbufs(txq);
1647                 i40e_reset_tx_queue(txq);
1648                 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1649         }
1650
1651         return 0;
1652 }
1653
1654 const uint32_t *
1655 i40e_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1656 {
1657         static const uint32_t ptypes[] = {
1658                 /* refers to i40e_rxd_pkt_type_mapping() */
1659                 RTE_PTYPE_L2_ETHER,
1660                 RTE_PTYPE_L2_ETHER_TIMESYNC,
1661                 RTE_PTYPE_L2_ETHER_LLDP,
1662                 RTE_PTYPE_L2_ETHER_ARP,
1663                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1664                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
1665                 RTE_PTYPE_L4_FRAG,
1666                 RTE_PTYPE_L4_ICMP,
1667                 RTE_PTYPE_L4_NONFRAG,
1668                 RTE_PTYPE_L4_SCTP,
1669                 RTE_PTYPE_L4_TCP,
1670                 RTE_PTYPE_L4_UDP,
1671                 RTE_PTYPE_TUNNEL_GRENAT,
1672                 RTE_PTYPE_TUNNEL_IP,
1673                 RTE_PTYPE_INNER_L2_ETHER,
1674                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1675                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
1676                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
1677                 RTE_PTYPE_INNER_L4_FRAG,
1678                 RTE_PTYPE_INNER_L4_ICMP,
1679                 RTE_PTYPE_INNER_L4_NONFRAG,
1680                 RTE_PTYPE_INNER_L4_SCTP,
1681                 RTE_PTYPE_INNER_L4_TCP,
1682                 RTE_PTYPE_INNER_L4_UDP,
1683                 RTE_PTYPE_UNKNOWN
1684         };
1685
1686         if (dev->rx_pkt_burst == i40e_recv_pkts ||
1687 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1688             dev->rx_pkt_burst == i40e_recv_pkts_bulk_alloc ||
1689 #endif
1690             dev->rx_pkt_burst == i40e_recv_scattered_pkts ||
1691             dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec ||
1692             dev->rx_pkt_burst == i40e_recv_pkts_vec ||
1693             dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec_avx2 ||
1694             dev->rx_pkt_burst == i40e_recv_pkts_vec_avx2)
1695                 return ptypes;
1696         return NULL;
1697 }
1698
1699 static int
1700 i40e_dev_first_queue(uint16_t idx, void **queues, int num)
1701 {
1702         uint16_t i;
1703
1704         for (i = 0; i < num; i++) {
1705                 if (i != idx && queues[i])
1706                         return 0;
1707         }
1708
1709         return 1;
1710 }
1711
1712 static int
1713 i40e_dev_rx_queue_setup_runtime(struct rte_eth_dev *dev,
1714                                 struct i40e_rx_queue *rxq)
1715 {
1716         struct i40e_adapter *ad =
1717                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1718         int use_def_burst_func =
1719                 check_rx_burst_bulk_alloc_preconditions(rxq);
1720         uint16_t buf_size =
1721                 (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
1722                            RTE_PKTMBUF_HEADROOM);
1723         int use_scattered_rx =
1724                 ((rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size);
1725
1726         if (i40e_rx_queue_init(rxq) != I40E_SUCCESS) {
1727                 PMD_DRV_LOG(ERR,
1728                             "Failed to do RX queue initialization");
1729                 return -EINVAL;
1730         }
1731
1732         if (i40e_dev_first_queue(rxq->queue_id,
1733                                  dev->data->rx_queues,
1734                                  dev->data->nb_rx_queues)) {
1735                 /**
1736                  * If it is the first queue to setup,
1737                  * set all flags to default and call
1738                  * i40e_set_rx_function.
1739                  */
1740                 ad->rx_bulk_alloc_allowed = true;
1741                 ad->rx_vec_allowed = true;
1742                 dev->data->scattered_rx = use_scattered_rx;
1743                 if (use_def_burst_func)
1744                         ad->rx_bulk_alloc_allowed = false;
1745                 i40e_set_rx_function(dev);
1746                 return 0;
1747         }
1748
1749         /* check bulk alloc conflict */
1750         if (ad->rx_bulk_alloc_allowed && use_def_burst_func) {
1751                 PMD_DRV_LOG(ERR, "Can't use default burst.");
1752                 return -EINVAL;
1753         }
1754         /* check scatterred conflict */
1755         if (!dev->data->scattered_rx && use_scattered_rx) {
1756                 PMD_DRV_LOG(ERR, "Scattered rx is required.");
1757                 return -EINVAL;
1758         }
1759         /* check vector conflict */
1760         if (ad->rx_vec_allowed && i40e_rxq_vec_setup(rxq)) {
1761                 PMD_DRV_LOG(ERR, "Failed vector rx setup.");
1762                 return -EINVAL;
1763         }
1764
1765         return 0;
1766 }
1767
1768 int
1769 i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
1770                         uint16_t queue_idx,
1771                         uint16_t nb_desc,
1772                         unsigned int socket_id,
1773                         const struct rte_eth_rxconf *rx_conf,
1774                         struct rte_mempool *mp)
1775 {
1776         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1777         struct i40e_adapter *ad =
1778                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1779         struct i40e_vsi *vsi;
1780         struct i40e_pf *pf = NULL;
1781         struct i40e_vf *vf = NULL;
1782         struct i40e_rx_queue *rxq;
1783         const struct rte_memzone *rz;
1784         uint32_t ring_size;
1785         uint16_t len, i;
1786         uint16_t reg_idx, base, bsf, tc_mapping;
1787         int q_offset, use_def_burst_func = 1;
1788         uint64_t offloads;
1789
1790         offloads = rx_conf->offloads | dev->data->dev_conf.rxmode.offloads;
1791
1792         if (hw->mac.type == I40E_MAC_VF || hw->mac.type == I40E_MAC_X722_VF) {
1793                 vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1794                 vsi = &vf->vsi;
1795                 if (!vsi)
1796                         return -EINVAL;
1797                 reg_idx = queue_idx;
1798         } else {
1799                 pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
1800                 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
1801                 if (!vsi)
1802                         return -EINVAL;
1803                 q_offset = i40e_get_queue_offset_by_qindex(pf, queue_idx);
1804                 if (q_offset < 0)
1805                         return -EINVAL;
1806                 reg_idx = vsi->base_queue + q_offset;
1807         }
1808
1809         if (nb_desc % I40E_ALIGN_RING_DESC != 0 ||
1810             (nb_desc > I40E_MAX_RING_DESC) ||
1811             (nb_desc < I40E_MIN_RING_DESC)) {
1812                 PMD_DRV_LOG(ERR, "Number (%u) of receive descriptors is "
1813                             "invalid", nb_desc);
1814                 return -EINVAL;
1815         }
1816
1817         /* Free memory if needed */
1818         if (dev->data->rx_queues[queue_idx]) {
1819                 i40e_dev_rx_queue_release(dev->data->rx_queues[queue_idx]);
1820                 dev->data->rx_queues[queue_idx] = NULL;
1821         }
1822
1823         /* Allocate the rx queue data structure */
1824         rxq = rte_zmalloc_socket("i40e rx queue",
1825                                  sizeof(struct i40e_rx_queue),
1826                                  RTE_CACHE_LINE_SIZE,
1827                                  socket_id);
1828         if (!rxq) {
1829                 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
1830                             "rx queue data structure");
1831                 return -ENOMEM;
1832         }
1833         rxq->mp = mp;
1834         rxq->nb_rx_desc = nb_desc;
1835         rxq->rx_free_thresh = rx_conf->rx_free_thresh;
1836         rxq->queue_id = queue_idx;
1837         rxq->reg_idx = reg_idx;
1838         rxq->port_id = dev->data->port_id;
1839         if (rte_eth_dev_must_keep_crc(dev->data->dev_conf.rxmode.offloads))
1840                 rxq->crc_len = ETHER_CRC_LEN;
1841         else
1842                 rxq->crc_len = 0;
1843         rxq->drop_en = rx_conf->rx_drop_en;
1844         rxq->vsi = vsi;
1845         rxq->rx_deferred_start = rx_conf->rx_deferred_start;
1846         rxq->offloads = offloads;
1847
1848         /* Allocate the maximun number of RX ring hardware descriptor. */
1849         len = I40E_MAX_RING_DESC;
1850
1851         /**
1852          * Allocating a little more memory because vectorized/bulk_alloc Rx
1853          * functions doesn't check boundaries each time.
1854          */
1855         len += RTE_PMD_I40E_RX_MAX_BURST;
1856
1857         ring_size = RTE_ALIGN(len * sizeof(union i40e_rx_desc),
1858                               I40E_DMA_MEM_ALIGN);
1859
1860         rz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_idx,
1861                               ring_size, I40E_RING_BASE_ALIGN, socket_id);
1862         if (!rz) {
1863                 i40e_dev_rx_queue_release(rxq);
1864                 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX");
1865                 return -ENOMEM;
1866         }
1867
1868         /* Zero all the descriptors in the ring. */
1869         memset(rz->addr, 0, ring_size);
1870
1871         rxq->rx_ring_phys_addr = rz->iova;
1872         rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
1873
1874         len = (uint16_t)(nb_desc + RTE_PMD_I40E_RX_MAX_BURST);
1875
1876         /* Allocate the software ring. */
1877         rxq->sw_ring =
1878                 rte_zmalloc_socket("i40e rx sw ring",
1879                                    sizeof(struct i40e_rx_entry) * len,
1880                                    RTE_CACHE_LINE_SIZE,
1881                                    socket_id);
1882         if (!rxq->sw_ring) {
1883                 i40e_dev_rx_queue_release(rxq);
1884                 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW ring");
1885                 return -ENOMEM;
1886         }
1887
1888         i40e_reset_rx_queue(rxq);
1889         rxq->q_set = TRUE;
1890
1891         for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
1892                 if (!(vsi->enabled_tc & (1 << i)))
1893                         continue;
1894                 tc_mapping = rte_le_to_cpu_16(vsi->info.tc_mapping[i]);
1895                 base = (tc_mapping & I40E_AQ_VSI_TC_QUE_OFFSET_MASK) >>
1896                         I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT;
1897                 bsf = (tc_mapping & I40E_AQ_VSI_TC_QUE_NUMBER_MASK) >>
1898                         I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT;
1899
1900                 if (queue_idx >= base && queue_idx < (base + BIT(bsf)))
1901                         rxq->dcb_tc = i;
1902         }
1903
1904         if (dev->data->dev_started) {
1905                 if (i40e_dev_rx_queue_setup_runtime(dev, rxq)) {
1906                         i40e_dev_rx_queue_release(rxq);
1907                         return -EINVAL;
1908                 }
1909         } else {
1910                 use_def_burst_func =
1911                         check_rx_burst_bulk_alloc_preconditions(rxq);
1912                 if (!use_def_burst_func) {
1913 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1914                         PMD_INIT_LOG(DEBUG,
1915                           "Rx Burst Bulk Alloc Preconditions are "
1916                           "satisfied. Rx Burst Bulk Alloc function will be "
1917                           "used on port=%d, queue=%d.",
1918                           rxq->port_id, rxq->queue_id);
1919 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
1920                 } else {
1921                         PMD_INIT_LOG(DEBUG,
1922                           "Rx Burst Bulk Alloc Preconditions are "
1923                           "not satisfied, Scattered Rx is requested, "
1924                           "or RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC is "
1925                           "not enabled on port=%d, queue=%d.",
1926                           rxq->port_id, rxq->queue_id);
1927                         ad->rx_bulk_alloc_allowed = false;
1928                 }
1929         }
1930
1931         dev->data->rx_queues[queue_idx] = rxq;
1932         return 0;
1933 }
1934
1935 void
1936 i40e_dev_rx_queue_release(void *rxq)
1937 {
1938         struct i40e_rx_queue *q = (struct i40e_rx_queue *)rxq;
1939
1940         if (!q) {
1941                 PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL");
1942                 return;
1943         }
1944
1945         i40e_rx_queue_release_mbufs(q);
1946         rte_free(q->sw_ring);
1947         rte_free(q);
1948 }
1949
1950 uint32_t
1951 i40e_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1952 {
1953 #define I40E_RXQ_SCAN_INTERVAL 4
1954         volatile union i40e_rx_desc *rxdp;
1955         struct i40e_rx_queue *rxq;
1956         uint16_t desc = 0;
1957
1958         rxq = dev->data->rx_queues[rx_queue_id];
1959         rxdp = &(rxq->rx_ring[rxq->rx_tail]);
1960         while ((desc < rxq->nb_rx_desc) &&
1961                 ((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1962                 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
1963                                 (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) {
1964                 /**
1965                  * Check the DD bit of a rx descriptor of each 4 in a group,
1966                  * to avoid checking too frequently and downgrading performance
1967                  * too much.
1968                  */
1969                 desc += I40E_RXQ_SCAN_INTERVAL;
1970                 rxdp += I40E_RXQ_SCAN_INTERVAL;
1971                 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
1972                         rxdp = &(rxq->rx_ring[rxq->rx_tail +
1973                                         desc - rxq->nb_rx_desc]);
1974         }
1975
1976         return desc;
1977 }
1978
1979 int
1980 i40e_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
1981 {
1982         volatile union i40e_rx_desc *rxdp;
1983         struct i40e_rx_queue *rxq = rx_queue;
1984         uint16_t desc;
1985         int ret;
1986
1987         if (unlikely(offset >= rxq->nb_rx_desc)) {
1988                 PMD_DRV_LOG(ERR, "Invalid RX descriptor id %u", offset);
1989                 return 0;
1990         }
1991
1992         desc = rxq->rx_tail + offset;
1993         if (desc >= rxq->nb_rx_desc)
1994                 desc -= rxq->nb_rx_desc;
1995
1996         rxdp = &(rxq->rx_ring[desc]);
1997
1998         ret = !!(((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1999                 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
2000                                 (1 << I40E_RX_DESC_STATUS_DD_SHIFT));
2001
2002         return ret;
2003 }
2004
2005 int
2006 i40e_dev_rx_descriptor_status(void *rx_queue, uint16_t offset)
2007 {
2008         struct i40e_rx_queue *rxq = rx_queue;
2009         volatile uint64_t *status;
2010         uint64_t mask;
2011         uint32_t desc;
2012
2013         if (unlikely(offset >= rxq->nb_rx_desc))
2014                 return -EINVAL;
2015
2016         if (offset >= rxq->nb_rx_desc - rxq->nb_rx_hold)
2017                 return RTE_ETH_RX_DESC_UNAVAIL;
2018
2019         desc = rxq->rx_tail + offset;
2020         if (desc >= rxq->nb_rx_desc)
2021                 desc -= rxq->nb_rx_desc;
2022
2023         status = &rxq->rx_ring[desc].wb.qword1.status_error_len;
2024         mask = rte_le_to_cpu_64((1ULL << I40E_RX_DESC_STATUS_DD_SHIFT)
2025                 << I40E_RXD_QW1_STATUS_SHIFT);
2026         if (*status & mask)
2027                 return RTE_ETH_RX_DESC_DONE;
2028
2029         return RTE_ETH_RX_DESC_AVAIL;
2030 }
2031
2032 int
2033 i40e_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
2034 {
2035         struct i40e_tx_queue *txq = tx_queue;
2036         volatile uint64_t *status;
2037         uint64_t mask, expect;
2038         uint32_t desc;
2039
2040         if (unlikely(offset >= txq->nb_tx_desc))
2041                 return -EINVAL;
2042
2043         desc = txq->tx_tail + offset;
2044         /* go to next desc that has the RS bit */
2045         desc = ((desc + txq->tx_rs_thresh - 1) / txq->tx_rs_thresh) *
2046                 txq->tx_rs_thresh;
2047         if (desc >= txq->nb_tx_desc) {
2048                 desc -= txq->nb_tx_desc;
2049                 if (desc >= txq->nb_tx_desc)
2050                         desc -= txq->nb_tx_desc;
2051         }
2052
2053         status = &txq->tx_ring[desc].cmd_type_offset_bsz;
2054         mask = rte_le_to_cpu_64(I40E_TXD_QW1_DTYPE_MASK);
2055         expect = rte_cpu_to_le_64(
2056                 I40E_TX_DESC_DTYPE_DESC_DONE << I40E_TXD_QW1_DTYPE_SHIFT);
2057         if ((*status & mask) == expect)
2058                 return RTE_ETH_TX_DESC_DONE;
2059
2060         return RTE_ETH_TX_DESC_FULL;
2061 }
2062
2063 static int
2064 i40e_dev_tx_queue_setup_runtime(struct rte_eth_dev *dev,
2065                                 struct i40e_tx_queue *txq)
2066 {
2067         struct i40e_adapter *ad =
2068                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
2069
2070         if (i40e_tx_queue_init(txq) != I40E_SUCCESS) {
2071                 PMD_DRV_LOG(ERR,
2072                             "Failed to do TX queue initialization");
2073                 return -EINVAL;
2074         }
2075
2076         if (i40e_dev_first_queue(txq->queue_id,
2077                                  dev->data->tx_queues,
2078                                  dev->data->nb_tx_queues)) {
2079                 /**
2080                  * If it is the first queue to setup,
2081                  * set all flags and call
2082                  * i40e_set_tx_function.
2083                  */
2084                 i40e_set_tx_function_flag(dev, txq);
2085                 i40e_set_tx_function(dev);
2086                 return 0;
2087         }
2088
2089         /* check vector conflict */
2090         if (ad->tx_vec_allowed) {
2091                 if (txq->tx_rs_thresh > RTE_I40E_TX_MAX_FREE_BUF_SZ ||
2092                     i40e_txq_vec_setup(txq)) {
2093                         PMD_DRV_LOG(ERR, "Failed vector tx setup.");
2094                         return -EINVAL;
2095                 }
2096         }
2097         /* check simple tx conflict */
2098         if (ad->tx_simple_allowed) {
2099                 if (txq->offloads != 0 ||
2100                                 txq->tx_rs_thresh < RTE_PMD_I40E_TX_MAX_BURST) {
2101                         PMD_DRV_LOG(ERR, "No-simple tx is required.");
2102                         return -EINVAL;
2103                 }
2104         }
2105
2106         return 0;
2107 }
2108
2109 int
2110 i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
2111                         uint16_t queue_idx,
2112                         uint16_t nb_desc,
2113                         unsigned int socket_id,
2114                         const struct rte_eth_txconf *tx_conf)
2115 {
2116         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2117         struct i40e_vsi *vsi;
2118         struct i40e_pf *pf = NULL;
2119         struct i40e_vf *vf = NULL;
2120         struct i40e_tx_queue *txq;
2121         const struct rte_memzone *tz;
2122         uint32_t ring_size;
2123         uint16_t tx_rs_thresh, tx_free_thresh;
2124         uint16_t reg_idx, i, base, bsf, tc_mapping;
2125         int q_offset;
2126         uint64_t offloads;
2127
2128         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
2129
2130         if (hw->mac.type == I40E_MAC_VF || hw->mac.type == I40E_MAC_X722_VF) {
2131                 vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2132                 vsi = &vf->vsi;
2133                 if (!vsi)
2134                         return -EINVAL;
2135                 reg_idx = queue_idx;
2136         } else {
2137                 pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
2138                 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
2139                 if (!vsi)
2140                         return -EINVAL;
2141                 q_offset = i40e_get_queue_offset_by_qindex(pf, queue_idx);
2142                 if (q_offset < 0)
2143                         return -EINVAL;
2144                 reg_idx = vsi->base_queue + q_offset;
2145         }
2146
2147         if (nb_desc % I40E_ALIGN_RING_DESC != 0 ||
2148             (nb_desc > I40E_MAX_RING_DESC) ||
2149             (nb_desc < I40E_MIN_RING_DESC)) {
2150                 PMD_DRV_LOG(ERR, "Number (%u) of transmit descriptors is "
2151                             "invalid", nb_desc);
2152                 return -EINVAL;
2153         }
2154
2155         /**
2156          * The following two parameters control the setting of the RS bit on
2157          * transmit descriptors. TX descriptors will have their RS bit set
2158          * after txq->tx_rs_thresh descriptors have been used. The TX
2159          * descriptor ring will be cleaned after txq->tx_free_thresh
2160          * descriptors are used or if the number of descriptors required to
2161          * transmit a packet is greater than the number of free TX descriptors.
2162          *
2163          * The following constraints must be satisfied:
2164          *  - tx_rs_thresh must be greater than 0.
2165          *  - tx_rs_thresh must be less than the size of the ring minus 2.
2166          *  - tx_rs_thresh must be less than or equal to tx_free_thresh.
2167          *  - tx_rs_thresh must be a divisor of the ring size.
2168          *  - tx_free_thresh must be greater than 0.
2169          *  - tx_free_thresh must be less than the size of the ring minus 3.
2170          *
2171          * One descriptor in the TX ring is used as a sentinel to avoid a H/W
2172          * race condition, hence the maximum threshold constraints. When set
2173          * to zero use default values.
2174          */
2175         tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2176                 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2177         tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2178                 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2179         if (tx_rs_thresh >= (nb_desc - 2)) {
2180                 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2181                              "number of TX descriptors minus 2. "
2182                              "(tx_rs_thresh=%u port=%d queue=%d)",
2183                              (unsigned int)tx_rs_thresh,
2184                              (int)dev->data->port_id,
2185                              (int)queue_idx);
2186                 return I40E_ERR_PARAM;
2187         }
2188         if (tx_free_thresh >= (nb_desc - 3)) {
2189                 PMD_INIT_LOG(ERR, "tx_free_thresh must be less than the "
2190                              "number of TX descriptors minus 3. "
2191                              "(tx_free_thresh=%u port=%d queue=%d)",
2192                              (unsigned int)tx_free_thresh,
2193                              (int)dev->data->port_id,
2194                              (int)queue_idx);
2195                 return I40E_ERR_PARAM;
2196         }
2197         if (tx_rs_thresh > tx_free_thresh) {
2198                 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or "
2199                              "equal to tx_free_thresh. (tx_free_thresh=%u"
2200                              " tx_rs_thresh=%u port=%d queue=%d)",
2201                              (unsigned int)tx_free_thresh,
2202                              (unsigned int)tx_rs_thresh,
2203                              (int)dev->data->port_id,
2204                              (int)queue_idx);
2205                 return I40E_ERR_PARAM;
2206         }
2207         if ((nb_desc % tx_rs_thresh) != 0) {
2208                 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2209                              "number of TX descriptors. (tx_rs_thresh=%u"
2210                              " port=%d queue=%d)",
2211                              (unsigned int)tx_rs_thresh,
2212                              (int)dev->data->port_id,
2213                              (int)queue_idx);
2214                 return I40E_ERR_PARAM;
2215         }
2216         if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2217                 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2218                              "tx_rs_thresh is greater than 1. "
2219                              "(tx_rs_thresh=%u port=%d queue=%d)",
2220                              (unsigned int)tx_rs_thresh,
2221                              (int)dev->data->port_id,
2222                              (int)queue_idx);
2223                 return I40E_ERR_PARAM;
2224         }
2225
2226         /* Free memory if needed. */
2227         if (dev->data->tx_queues[queue_idx]) {
2228                 i40e_dev_tx_queue_release(dev->data->tx_queues[queue_idx]);
2229                 dev->data->tx_queues[queue_idx] = NULL;
2230         }
2231
2232         /* Allocate the TX queue data structure. */
2233         txq = rte_zmalloc_socket("i40e tx queue",
2234                                   sizeof(struct i40e_tx_queue),
2235                                   RTE_CACHE_LINE_SIZE,
2236                                   socket_id);
2237         if (!txq) {
2238                 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2239                             "tx queue structure");
2240                 return -ENOMEM;
2241         }
2242
2243         /* Allocate TX hardware ring descriptors. */
2244         ring_size = sizeof(struct i40e_tx_desc) * I40E_MAX_RING_DESC;
2245         ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2246         tz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_idx,
2247                               ring_size, I40E_RING_BASE_ALIGN, socket_id);
2248         if (!tz) {
2249                 i40e_dev_tx_queue_release(txq);
2250                 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX");
2251                 return -ENOMEM;
2252         }
2253
2254         txq->nb_tx_desc = nb_desc;
2255         txq->tx_rs_thresh = tx_rs_thresh;
2256         txq->tx_free_thresh = tx_free_thresh;
2257         txq->pthresh = tx_conf->tx_thresh.pthresh;
2258         txq->hthresh = tx_conf->tx_thresh.hthresh;
2259         txq->wthresh = tx_conf->tx_thresh.wthresh;
2260         txq->queue_id = queue_idx;
2261         txq->reg_idx = reg_idx;
2262         txq->port_id = dev->data->port_id;
2263         txq->offloads = offloads;
2264         txq->vsi = vsi;
2265         txq->tx_deferred_start = tx_conf->tx_deferred_start;
2266
2267         txq->tx_ring_phys_addr = tz->iova;
2268         txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2269
2270         /* Allocate software ring */
2271         txq->sw_ring =
2272                 rte_zmalloc_socket("i40e tx sw ring",
2273                                    sizeof(struct i40e_tx_entry) * nb_desc,
2274                                    RTE_CACHE_LINE_SIZE,
2275                                    socket_id);
2276         if (!txq->sw_ring) {
2277                 i40e_dev_tx_queue_release(txq);
2278                 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW TX ring");
2279                 return -ENOMEM;
2280         }
2281
2282         i40e_reset_tx_queue(txq);
2283         txq->q_set = TRUE;
2284
2285         for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
2286                 if (!(vsi->enabled_tc & (1 << i)))
2287                         continue;
2288                 tc_mapping = rte_le_to_cpu_16(vsi->info.tc_mapping[i]);
2289                 base = (tc_mapping & I40E_AQ_VSI_TC_QUE_OFFSET_MASK) >>
2290                         I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT;
2291                 bsf = (tc_mapping & I40E_AQ_VSI_TC_QUE_NUMBER_MASK) >>
2292                         I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT;
2293
2294                 if (queue_idx >= base && queue_idx < (base + BIT(bsf)))
2295                         txq->dcb_tc = i;
2296         }
2297
2298         if (dev->data->dev_started) {
2299                 if (i40e_dev_tx_queue_setup_runtime(dev, txq)) {
2300                         i40e_dev_tx_queue_release(txq);
2301                         return -EINVAL;
2302                 }
2303         } else {
2304                 /**
2305                  * Use a simple TX queue without offloads or
2306                  * multi segs if possible
2307                  */
2308                 i40e_set_tx_function_flag(dev, txq);
2309         }
2310         dev->data->tx_queues[queue_idx] = txq;
2311
2312         return 0;
2313 }
2314
2315 void
2316 i40e_dev_tx_queue_release(void *txq)
2317 {
2318         struct i40e_tx_queue *q = (struct i40e_tx_queue *)txq;
2319
2320         if (!q) {
2321                 PMD_DRV_LOG(DEBUG, "Pointer to TX queue is NULL");
2322                 return;
2323         }
2324
2325         i40e_tx_queue_release_mbufs(q);
2326         rte_free(q->sw_ring);
2327         rte_free(q);
2328 }
2329
2330 const struct rte_memzone *
2331 i40e_memzone_reserve(const char *name, uint32_t len, int socket_id)
2332 {
2333         const struct rte_memzone *mz;
2334
2335         mz = rte_memzone_lookup(name);
2336         if (mz)
2337                 return mz;
2338
2339         mz = rte_memzone_reserve_aligned(name, len, socket_id,
2340                         RTE_MEMZONE_IOVA_CONTIG, I40E_RING_BASE_ALIGN);
2341         return mz;
2342 }
2343
2344 void
2345 i40e_rx_queue_release_mbufs(struct i40e_rx_queue *rxq)
2346 {
2347         uint16_t i;
2348
2349         /* SSE Vector driver has a different way of releasing mbufs. */
2350         if (rxq->rx_using_sse) {
2351                 i40e_rx_queue_release_mbufs_vec(rxq);
2352                 return;
2353         }
2354
2355         if (!rxq->sw_ring) {
2356                 PMD_DRV_LOG(DEBUG, "Pointer to sw_ring is NULL");
2357                 return;
2358         }
2359
2360         for (i = 0; i < rxq->nb_rx_desc; i++) {
2361                 if (rxq->sw_ring[i].mbuf) {
2362                         rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2363                         rxq->sw_ring[i].mbuf = NULL;
2364                 }
2365         }
2366 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2367         if (rxq->rx_nb_avail == 0)
2368                 return;
2369         for (i = 0; i < rxq->rx_nb_avail; i++) {
2370                 struct rte_mbuf *mbuf;
2371
2372                 mbuf = rxq->rx_stage[rxq->rx_next_avail + i];
2373                 rte_pktmbuf_free_seg(mbuf);
2374         }
2375         rxq->rx_nb_avail = 0;
2376 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2377 }
2378
2379 void
2380 i40e_reset_rx_queue(struct i40e_rx_queue *rxq)
2381 {
2382         unsigned i;
2383         uint16_t len;
2384
2385         if (!rxq) {
2386                 PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL");
2387                 return;
2388         }
2389
2390 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2391         if (check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
2392                 len = (uint16_t)(rxq->nb_rx_desc + RTE_PMD_I40E_RX_MAX_BURST);
2393         else
2394 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2395                 len = rxq->nb_rx_desc;
2396
2397         for (i = 0; i < len * sizeof(union i40e_rx_desc); i++)
2398                 ((volatile char *)rxq->rx_ring)[i] = 0;
2399
2400         memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2401         for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; ++i)
2402                 rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
2403
2404 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2405         rxq->rx_nb_avail = 0;
2406         rxq->rx_next_avail = 0;
2407         rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2408 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2409         rxq->rx_tail = 0;
2410         rxq->nb_rx_hold = 0;
2411         rxq->pkt_first_seg = NULL;
2412         rxq->pkt_last_seg = NULL;
2413
2414         rxq->rxrearm_start = 0;
2415         rxq->rxrearm_nb = 0;
2416 }
2417
2418 void
2419 i40e_tx_queue_release_mbufs(struct i40e_tx_queue *txq)
2420 {
2421         struct rte_eth_dev *dev;
2422         uint16_t i;
2423
2424         dev = &rte_eth_devices[txq->port_id];
2425
2426         if (!txq || !txq->sw_ring) {
2427                 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL");
2428                 return;
2429         }
2430
2431         /**
2432          *  vPMD tx will not set sw_ring's mbuf to NULL after free,
2433          *  so need to free remains more carefully.
2434          */
2435         if (dev->tx_pkt_burst == i40e_xmit_pkts_vec_avx2 ||
2436                         dev->tx_pkt_burst == i40e_xmit_pkts_vec) {
2437                 i = txq->tx_next_dd - txq->tx_rs_thresh + 1;
2438                 if (txq->tx_tail < i) {
2439                         for (; i < txq->nb_tx_desc; i++) {
2440                                 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2441                                 txq->sw_ring[i].mbuf = NULL;
2442                         }
2443                         i = 0;
2444                 }
2445                 for (; i < txq->tx_tail; i++) {
2446                         rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2447                         txq->sw_ring[i].mbuf = NULL;
2448                 }
2449         } else {
2450                 for (i = 0; i < txq->nb_tx_desc; i++) {
2451                         if (txq->sw_ring[i].mbuf) {
2452                                 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2453                                 txq->sw_ring[i].mbuf = NULL;
2454                         }
2455                 }
2456         }
2457 }
2458
2459 void
2460 i40e_reset_tx_queue(struct i40e_tx_queue *txq)
2461 {
2462         struct i40e_tx_entry *txe;
2463         uint16_t i, prev, size;
2464
2465         if (!txq) {
2466                 PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
2467                 return;
2468         }
2469
2470         txe = txq->sw_ring;
2471         size = sizeof(struct i40e_tx_desc) * txq->nb_tx_desc;
2472         for (i = 0; i < size; i++)
2473                 ((volatile char *)txq->tx_ring)[i] = 0;
2474
2475         prev = (uint16_t)(txq->nb_tx_desc - 1);
2476         for (i = 0; i < txq->nb_tx_desc; i++) {
2477                 volatile struct i40e_tx_desc *txd = &txq->tx_ring[i];
2478
2479                 txd->cmd_type_offset_bsz =
2480                         rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE);
2481                 txe[i].mbuf =  NULL;
2482                 txe[i].last_id = i;
2483                 txe[prev].next_id = i;
2484                 prev = i;
2485         }
2486
2487         txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2488         txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2489
2490         txq->tx_tail = 0;
2491         txq->nb_tx_used = 0;
2492
2493         txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2494         txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2495 }
2496
2497 /* Init the TX queue in hardware */
2498 int
2499 i40e_tx_queue_init(struct i40e_tx_queue *txq)
2500 {
2501         enum i40e_status_code err = I40E_SUCCESS;
2502         struct i40e_vsi *vsi = txq->vsi;
2503         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2504         uint16_t pf_q = txq->reg_idx;
2505         struct i40e_hmc_obj_txq tx_ctx;
2506         uint32_t qtx_ctl;
2507
2508         /* clear the context structure first */
2509         memset(&tx_ctx, 0, sizeof(tx_ctx));
2510         tx_ctx.new_context = 1;
2511         tx_ctx.base = txq->tx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2512         tx_ctx.qlen = txq->nb_tx_desc;
2513
2514 #ifdef RTE_LIBRTE_IEEE1588
2515         tx_ctx.timesync_ena = 1;
2516 #endif
2517         tx_ctx.rdylist = rte_le_to_cpu_16(vsi->info.qs_handle[txq->dcb_tc]);
2518         if (vsi->type == I40E_VSI_FDIR)
2519                 tx_ctx.fd_ena = TRUE;
2520
2521         err = i40e_clear_lan_tx_queue_context(hw, pf_q);
2522         if (err != I40E_SUCCESS) {
2523                 PMD_DRV_LOG(ERR, "Failure of clean lan tx queue context");
2524                 return err;
2525         }
2526
2527         err = i40e_set_lan_tx_queue_context(hw, pf_q, &tx_ctx);
2528         if (err != I40E_SUCCESS) {
2529                 PMD_DRV_LOG(ERR, "Failure of set lan tx queue context");
2530                 return err;
2531         }
2532
2533         /* Now associate this queue with this PCI function */
2534         qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
2535         qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
2536                                         I40E_QTX_CTL_PF_INDX_MASK);
2537         I40E_WRITE_REG(hw, I40E_QTX_CTL(pf_q), qtx_ctl);
2538         I40E_WRITE_FLUSH(hw);
2539
2540         txq->qtx_tail = hw->hw_addr + I40E_QTX_TAIL(pf_q);
2541
2542         return err;
2543 }
2544
2545 int
2546 i40e_alloc_rx_queue_mbufs(struct i40e_rx_queue *rxq)
2547 {
2548         struct i40e_rx_entry *rxe = rxq->sw_ring;
2549         uint64_t dma_addr;
2550         uint16_t i;
2551
2552         for (i = 0; i < rxq->nb_rx_desc; i++) {
2553                 volatile union i40e_rx_desc *rxd;
2554                 struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mp);
2555
2556                 if (unlikely(!mbuf)) {
2557                         PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
2558                         return -ENOMEM;
2559                 }
2560
2561                 rte_mbuf_refcnt_set(mbuf, 1);
2562                 mbuf->next = NULL;
2563                 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
2564                 mbuf->nb_segs = 1;
2565                 mbuf->port = rxq->port_id;
2566
2567                 dma_addr =
2568                         rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
2569
2570                 rxd = &rxq->rx_ring[i];
2571                 rxd->read.pkt_addr = dma_addr;
2572                 rxd->read.hdr_addr = 0;
2573 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2574                 rxd->read.rsvd1 = 0;
2575                 rxd->read.rsvd2 = 0;
2576 #endif /* RTE_LIBRTE_I40E_16BYTE_RX_DESC */
2577
2578                 rxe[i].mbuf = mbuf;
2579         }
2580
2581         return 0;
2582 }
2583
2584 /*
2585  * Calculate the buffer length, and check the jumbo frame
2586  * and maximum packet length.
2587  */
2588 static int
2589 i40e_rx_queue_config(struct i40e_rx_queue *rxq)
2590 {
2591         struct i40e_pf *pf = I40E_VSI_TO_PF(rxq->vsi);
2592         struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2593         struct rte_eth_dev_data *data = pf->dev_data;
2594         uint16_t buf_size, len;
2595
2596         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
2597                 RTE_PKTMBUF_HEADROOM);
2598
2599         switch (pf->flags & (I40E_FLAG_HEADER_SPLIT_DISABLED |
2600                         I40E_FLAG_HEADER_SPLIT_ENABLED)) {
2601         case I40E_FLAG_HEADER_SPLIT_ENABLED: /* Not supported */
2602                 rxq->rx_hdr_len = RTE_ALIGN(I40E_RXBUF_SZ_1024,
2603                                 (1 << I40E_RXQ_CTX_HBUFF_SHIFT));
2604                 rxq->rx_buf_len = RTE_ALIGN(I40E_RXBUF_SZ_2048,
2605                                 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2606                 rxq->hs_mode = i40e_header_split_enabled;
2607                 break;
2608         case I40E_FLAG_HEADER_SPLIT_DISABLED:
2609         default:
2610                 rxq->rx_hdr_len = 0;
2611                 rxq->rx_buf_len = RTE_ALIGN_FLOOR(buf_size,
2612                         (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2613                 rxq->hs_mode = i40e_header_split_none;
2614                 break;
2615         }
2616
2617         len = hw->func_caps.rx_buf_chain_len * rxq->rx_buf_len;
2618         rxq->max_pkt_len = RTE_MIN(len, data->dev_conf.rxmode.max_rx_pkt_len);
2619         if (data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
2620                 if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
2621                         rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
2622                         PMD_DRV_LOG(ERR, "maximum packet length must "
2623                                     "be larger than %u and smaller than %u,"
2624                                     "as jumbo frame is enabled",
2625                                     (uint32_t)ETHER_MAX_LEN,
2626                                     (uint32_t)I40E_FRAME_SIZE_MAX);
2627                         return I40E_ERR_CONFIG;
2628                 }
2629         } else {
2630                 if (rxq->max_pkt_len < ETHER_MIN_LEN ||
2631                         rxq->max_pkt_len > ETHER_MAX_LEN) {
2632                         PMD_DRV_LOG(ERR, "maximum packet length must be "
2633                                     "larger than %u and smaller than %u, "
2634                                     "as jumbo frame is disabled",
2635                                     (uint32_t)ETHER_MIN_LEN,
2636                                     (uint32_t)ETHER_MAX_LEN);
2637                         return I40E_ERR_CONFIG;
2638                 }
2639         }
2640
2641         return 0;
2642 }
2643
2644 /* Init the RX queue in hardware */
2645 int
2646 i40e_rx_queue_init(struct i40e_rx_queue *rxq)
2647 {
2648         int err = I40E_SUCCESS;
2649         struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2650         struct rte_eth_dev_data *dev_data = I40E_VSI_TO_DEV_DATA(rxq->vsi);
2651         uint16_t pf_q = rxq->reg_idx;
2652         uint16_t buf_size;
2653         struct i40e_hmc_obj_rxq rx_ctx;
2654
2655         err = i40e_rx_queue_config(rxq);
2656         if (err < 0) {
2657                 PMD_DRV_LOG(ERR, "Failed to config RX queue");
2658                 return err;
2659         }
2660
2661         /* Clear the context structure first */
2662         memset(&rx_ctx, 0, sizeof(struct i40e_hmc_obj_rxq));
2663         rx_ctx.dbuff = rxq->rx_buf_len >> I40E_RXQ_CTX_DBUFF_SHIFT;
2664         rx_ctx.hbuff = rxq->rx_hdr_len >> I40E_RXQ_CTX_HBUFF_SHIFT;
2665
2666         rx_ctx.base = rxq->rx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2667         rx_ctx.qlen = rxq->nb_rx_desc;
2668 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2669         rx_ctx.dsize = 1;
2670 #endif
2671         rx_ctx.dtype = rxq->hs_mode;
2672         if (rxq->hs_mode)
2673                 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_ALL;
2674         else
2675                 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_NONE;
2676         rx_ctx.rxmax = rxq->max_pkt_len;
2677         rx_ctx.tphrdesc_ena = 1;
2678         rx_ctx.tphwdesc_ena = 1;
2679         rx_ctx.tphdata_ena = 1;
2680         rx_ctx.tphhead_ena = 1;
2681         rx_ctx.lrxqthresh = 2;
2682         rx_ctx.crcstrip = (rxq->crc_len == 0) ? 1 : 0;
2683         rx_ctx.l2tsel = 1;
2684         /* showiv indicates if inner VLAN is stripped inside of tunnel
2685          * packet. When set it to 1, vlan information is stripped from
2686          * the inner header, but the hardware does not put it in the
2687          * descriptor. So set it zero by default.
2688          */
2689         rx_ctx.showiv = 0;
2690         rx_ctx.prefena = 1;
2691
2692         err = i40e_clear_lan_rx_queue_context(hw, pf_q);
2693         if (err != I40E_SUCCESS) {
2694                 PMD_DRV_LOG(ERR, "Failed to clear LAN RX queue context");
2695                 return err;
2696         }
2697         err = i40e_set_lan_rx_queue_context(hw, pf_q, &rx_ctx);
2698         if (err != I40E_SUCCESS) {
2699                 PMD_DRV_LOG(ERR, "Failed to set LAN RX queue context");
2700                 return err;
2701         }
2702
2703         rxq->qrx_tail = hw->hw_addr + I40E_QRX_TAIL(pf_q);
2704
2705         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
2706                 RTE_PKTMBUF_HEADROOM);
2707
2708         /* Check if scattered RX needs to be used. */
2709         if ((rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
2710                 dev_data->scattered_rx = 1;
2711         }
2712
2713         /* Init the RX tail regieter. */
2714         I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
2715
2716         return 0;
2717 }
2718
2719 void
2720 i40e_dev_clear_queues(struct rte_eth_dev *dev)
2721 {
2722         uint16_t i;
2723
2724         PMD_INIT_FUNC_TRACE();
2725
2726         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2727                 if (!dev->data->tx_queues[i])
2728                         continue;
2729                 i40e_tx_queue_release_mbufs(dev->data->tx_queues[i]);
2730                 i40e_reset_tx_queue(dev->data->tx_queues[i]);
2731         }
2732
2733         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2734                 if (!dev->data->rx_queues[i])
2735                         continue;
2736                 i40e_rx_queue_release_mbufs(dev->data->rx_queues[i]);
2737                 i40e_reset_rx_queue(dev->data->rx_queues[i]);
2738         }
2739 }
2740
2741 void
2742 i40e_dev_free_queues(struct rte_eth_dev *dev)
2743 {
2744         uint16_t i;
2745
2746         PMD_INIT_FUNC_TRACE();
2747
2748         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2749                 if (!dev->data->rx_queues[i])
2750                         continue;
2751                 i40e_dev_rx_queue_release(dev->data->rx_queues[i]);
2752                 dev->data->rx_queues[i] = NULL;
2753         }
2754         dev->data->nb_rx_queues = 0;
2755
2756         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2757                 if (!dev->data->tx_queues[i])
2758                         continue;
2759                 i40e_dev_tx_queue_release(dev->data->tx_queues[i]);
2760                 dev->data->tx_queues[i] = NULL;
2761         }
2762         dev->data->nb_tx_queues = 0;
2763 }
2764
2765 #define I40E_FDIR_NUM_TX_DESC  I40E_MIN_RING_DESC
2766 #define I40E_FDIR_NUM_RX_DESC  I40E_MIN_RING_DESC
2767
2768 enum i40e_status_code
2769 i40e_fdir_setup_tx_resources(struct i40e_pf *pf)
2770 {
2771         struct i40e_tx_queue *txq;
2772         const struct rte_memzone *tz = NULL;
2773         uint32_t ring_size;
2774         struct rte_eth_dev *dev;
2775
2776         if (!pf) {
2777                 PMD_DRV_LOG(ERR, "PF is not available");
2778                 return I40E_ERR_BAD_PTR;
2779         }
2780
2781         dev = pf->adapter->eth_dev;
2782
2783         /* Allocate the TX queue data structure. */
2784         txq = rte_zmalloc_socket("i40e fdir tx queue",
2785                                   sizeof(struct i40e_tx_queue),
2786                                   RTE_CACHE_LINE_SIZE,
2787                                   SOCKET_ID_ANY);
2788         if (!txq) {
2789                 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2790                                         "tx queue structure.");
2791                 return I40E_ERR_NO_MEMORY;
2792         }
2793
2794         /* Allocate TX hardware ring descriptors. */
2795         ring_size = sizeof(struct i40e_tx_desc) * I40E_FDIR_NUM_TX_DESC;
2796         ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2797
2798         tz = rte_eth_dma_zone_reserve(dev, "fdir_tx_ring",
2799                                       I40E_FDIR_QUEUE_ID, ring_size,
2800                                       I40E_RING_BASE_ALIGN, SOCKET_ID_ANY);
2801         if (!tz) {
2802                 i40e_dev_tx_queue_release(txq);
2803                 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX.");
2804                 return I40E_ERR_NO_MEMORY;
2805         }
2806
2807         txq->nb_tx_desc = I40E_FDIR_NUM_TX_DESC;
2808         txq->queue_id = I40E_FDIR_QUEUE_ID;
2809         txq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2810         txq->vsi = pf->fdir.fdir_vsi;
2811
2812         txq->tx_ring_phys_addr = tz->iova;
2813         txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2814         /*
2815          * don't need to allocate software ring and reset for the fdir
2816          * program queue just set the queue has been configured.
2817          */
2818         txq->q_set = TRUE;
2819         pf->fdir.txq = txq;
2820
2821         return I40E_SUCCESS;
2822 }
2823
2824 enum i40e_status_code
2825 i40e_fdir_setup_rx_resources(struct i40e_pf *pf)
2826 {
2827         struct i40e_rx_queue *rxq;
2828         const struct rte_memzone *rz = NULL;
2829         uint32_t ring_size;
2830         struct rte_eth_dev *dev;
2831
2832         if (!pf) {
2833                 PMD_DRV_LOG(ERR, "PF is not available");
2834                 return I40E_ERR_BAD_PTR;
2835         }
2836
2837         dev = pf->adapter->eth_dev;
2838
2839         /* Allocate the RX queue data structure. */
2840         rxq = rte_zmalloc_socket("i40e fdir rx queue",
2841                                   sizeof(struct i40e_rx_queue),
2842                                   RTE_CACHE_LINE_SIZE,
2843                                   SOCKET_ID_ANY);
2844         if (!rxq) {
2845                 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2846                                         "rx queue structure.");
2847                 return I40E_ERR_NO_MEMORY;
2848         }
2849
2850         /* Allocate RX hardware ring descriptors. */
2851         ring_size = sizeof(union i40e_rx_desc) * I40E_FDIR_NUM_RX_DESC;
2852         ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2853
2854         rz = rte_eth_dma_zone_reserve(dev, "fdir_rx_ring",
2855                                       I40E_FDIR_QUEUE_ID, ring_size,
2856                                       I40E_RING_BASE_ALIGN, SOCKET_ID_ANY);
2857         if (!rz) {
2858                 i40e_dev_rx_queue_release(rxq);
2859                 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX.");
2860                 return I40E_ERR_NO_MEMORY;
2861         }
2862
2863         rxq->nb_rx_desc = I40E_FDIR_NUM_RX_DESC;
2864         rxq->queue_id = I40E_FDIR_QUEUE_ID;
2865         rxq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2866         rxq->vsi = pf->fdir.fdir_vsi;
2867
2868         rxq->rx_ring_phys_addr = rz->iova;
2869         memset(rz->addr, 0, I40E_FDIR_NUM_RX_DESC * sizeof(union i40e_rx_desc));
2870         rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
2871
2872         /*
2873          * Don't need to allocate software ring and reset for the fdir
2874          * rx queue, just set the queue has been configured.
2875          */
2876         rxq->q_set = TRUE;
2877         pf->fdir.rxq = rxq;
2878
2879         return I40E_SUCCESS;
2880 }
2881
2882 void
2883 i40e_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
2884         struct rte_eth_rxq_info *qinfo)
2885 {
2886         struct i40e_rx_queue *rxq;
2887
2888         rxq = dev->data->rx_queues[queue_id];
2889
2890         qinfo->mp = rxq->mp;
2891         qinfo->scattered_rx = dev->data->scattered_rx;
2892         qinfo->nb_desc = rxq->nb_rx_desc;
2893
2894         qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
2895         qinfo->conf.rx_drop_en = rxq->drop_en;
2896         qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
2897         qinfo->conf.offloads = rxq->offloads;
2898 }
2899
2900 void
2901 i40e_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
2902         struct rte_eth_txq_info *qinfo)
2903 {
2904         struct i40e_tx_queue *txq;
2905
2906         txq = dev->data->tx_queues[queue_id];
2907
2908         qinfo->nb_desc = txq->nb_tx_desc;
2909
2910         qinfo->conf.tx_thresh.pthresh = txq->pthresh;
2911         qinfo->conf.tx_thresh.hthresh = txq->hthresh;
2912         qinfo->conf.tx_thresh.wthresh = txq->wthresh;
2913
2914         qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
2915         qinfo->conf.tx_rs_thresh = txq->tx_rs_thresh;
2916         qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
2917         qinfo->conf.offloads = txq->offloads;
2918 }
2919
2920 void __attribute__((cold))
2921 i40e_set_rx_function(struct rte_eth_dev *dev)
2922 {
2923         struct i40e_adapter *ad =
2924                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
2925         uint16_t rx_using_sse, i;
2926         /* In order to allow Vector Rx there are a few configuration
2927          * conditions to be met and Rx Bulk Allocation should be allowed.
2928          */
2929         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
2930                 if (i40e_rx_vec_dev_conf_condition_check(dev) ||
2931                     !ad->rx_bulk_alloc_allowed) {
2932                         PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet"
2933                                      " Vector Rx preconditions",
2934                                      dev->data->port_id);
2935
2936                         ad->rx_vec_allowed = false;
2937                 }
2938                 if (ad->rx_vec_allowed) {
2939                         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2940                                 struct i40e_rx_queue *rxq =
2941                                         dev->data->rx_queues[i];
2942
2943                                 if (rxq && i40e_rxq_vec_setup(rxq)) {
2944                                         ad->rx_vec_allowed = false;
2945                                         break;
2946                                 }
2947                         }
2948                 }
2949         }
2950
2951         if (dev->data->scattered_rx) {
2952                 /* Set the non-LRO scattered callback: there are Vector and
2953                  * single allocation versions.
2954                  */
2955                 if (ad->rx_vec_allowed) {
2956                         PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
2957                                             "callback (port=%d).",
2958                                      dev->data->port_id);
2959
2960                         dev->rx_pkt_burst = i40e_recv_scattered_pkts_vec;
2961 #ifdef RTE_ARCH_X86
2962                         /*
2963                          * since AVX frequency can be different to base
2964                          * frequency, limit use of AVX2 version to later
2965                          * plaforms, not all those that could theoretically
2966                          * run it.
2967                          */
2968                         if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
2969                                 dev->rx_pkt_burst =
2970                                         i40e_recv_scattered_pkts_vec_avx2;
2971 #endif
2972                 } else {
2973                         PMD_INIT_LOG(DEBUG, "Using a Scattered with bulk "
2974                                            "allocation callback (port=%d).",
2975                                      dev->data->port_id);
2976                         dev->rx_pkt_burst = i40e_recv_scattered_pkts;
2977                 }
2978         /* If parameters allow we are going to choose between the following
2979          * callbacks:
2980          *    - Vector
2981          *    - Bulk Allocation
2982          *    - Single buffer allocation (the simplest one)
2983          */
2984         } else if (ad->rx_vec_allowed) {
2985                 PMD_INIT_LOG(DEBUG, "Vector rx enabled, please make sure RX "
2986                                     "burst size no less than %d (port=%d).",
2987                              RTE_I40E_DESCS_PER_LOOP,
2988                              dev->data->port_id);
2989
2990                 dev->rx_pkt_burst = i40e_recv_pkts_vec;
2991 #ifdef RTE_ARCH_X86
2992                 /*
2993                  * since AVX frequency can be different to base
2994                  * frequency, limit use of AVX2 version to later
2995                  * plaforms, not all those that could theoretically
2996                  * run it.
2997                  */
2998                 if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
2999                         dev->rx_pkt_burst = i40e_recv_pkts_vec_avx2;
3000 #endif
3001         } else if (ad->rx_bulk_alloc_allowed) {
3002                 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
3003                                     "satisfied. Rx Burst Bulk Alloc function "
3004                                     "will be used on port=%d.",
3005                              dev->data->port_id);
3006
3007                 dev->rx_pkt_burst = i40e_recv_pkts_bulk_alloc;
3008         } else {
3009                 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
3010                                     "satisfied, or Scattered Rx is requested "
3011                                     "(port=%d).",
3012                              dev->data->port_id);
3013
3014                 dev->rx_pkt_burst = i40e_recv_pkts;
3015         }
3016
3017         /* Propagate information about RX function choice through all queues. */
3018         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
3019                 rx_using_sse =
3020                         (dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec ||
3021                          dev->rx_pkt_burst == i40e_recv_pkts_vec ||
3022                          dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec_avx2 ||
3023                          dev->rx_pkt_burst == i40e_recv_pkts_vec_avx2);
3024
3025                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3026                         struct i40e_rx_queue *rxq = dev->data->rx_queues[i];
3027
3028                         if (rxq)
3029                                 rxq->rx_using_sse = rx_using_sse;
3030                 }
3031         }
3032 }
3033
3034 void __attribute__((cold))
3035 i40e_set_tx_function_flag(struct rte_eth_dev *dev, struct i40e_tx_queue *txq)
3036 {
3037         struct i40e_adapter *ad =
3038                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
3039
3040         /* Use a simple Tx queue if possible (only fast free is allowed) */
3041         ad->tx_simple_allowed =
3042                 (txq->offloads ==
3043                  (txq->offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE) &&
3044                  txq->tx_rs_thresh >= RTE_PMD_I40E_TX_MAX_BURST);
3045         ad->tx_vec_allowed = (ad->tx_simple_allowed &&
3046                         txq->tx_rs_thresh <= RTE_I40E_TX_MAX_FREE_BUF_SZ);
3047
3048         if (ad->tx_vec_allowed)
3049                 PMD_INIT_LOG(DEBUG, "Vector Tx can be enabled on Tx queue %u.",
3050                                 txq->queue_id);
3051         else if (ad->tx_simple_allowed)
3052                 PMD_INIT_LOG(DEBUG, "Simple Tx can be enabled on Tx queue %u.",
3053                                 txq->queue_id);
3054         else
3055                 PMD_INIT_LOG(DEBUG,
3056                                 "Neither simple nor vector Tx enabled on Tx queue %u\n",
3057                                 txq->queue_id);
3058 }
3059
3060 void __attribute__((cold))
3061 i40e_set_tx_function(struct rte_eth_dev *dev)
3062 {
3063         struct i40e_adapter *ad =
3064                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
3065         int i;
3066
3067         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
3068                 if (ad->tx_vec_allowed) {
3069                         for (i = 0; i < dev->data->nb_tx_queues; i++) {
3070                                 struct i40e_tx_queue *txq =
3071                                         dev->data->tx_queues[i];
3072
3073                                 if (txq && i40e_txq_vec_setup(txq)) {
3074                                         ad->tx_vec_allowed = false;
3075                                         break;
3076                                 }
3077                         }
3078                 }
3079         }
3080
3081         if (ad->tx_simple_allowed) {
3082                 if (ad->tx_vec_allowed) {
3083                         PMD_INIT_LOG(DEBUG, "Vector tx finally be used.");
3084                         dev->tx_pkt_burst = i40e_xmit_pkts_vec;
3085 #ifdef RTE_ARCH_X86
3086                         /*
3087                          * since AVX frequency can be different to base
3088                          * frequency, limit use of AVX2 version to later
3089                          * plaforms, not all those that could theoretically
3090                          * run it.
3091                          */
3092                         if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
3093                                 dev->tx_pkt_burst = i40e_xmit_pkts_vec_avx2;
3094 #endif
3095                 } else {
3096                         PMD_INIT_LOG(DEBUG, "Simple tx finally be used.");
3097                         dev->tx_pkt_burst = i40e_xmit_pkts_simple;
3098                 }
3099                 dev->tx_pkt_prepare = NULL;
3100         } else {
3101                 PMD_INIT_LOG(DEBUG, "Xmit tx finally be used.");
3102                 dev->tx_pkt_burst = i40e_xmit_pkts;
3103                 dev->tx_pkt_prepare = i40e_prep_pkts;
3104         }
3105 }
3106
3107 void __attribute__((cold))
3108 i40e_set_default_ptype_table(struct rte_eth_dev *dev)
3109 {
3110         struct i40e_adapter *ad =
3111                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
3112         int i;
3113
3114         for (i = 0; i < I40E_MAX_PKT_TYPE; i++)
3115                 ad->ptype_tbl[i] = i40e_get_default_pkt_type(i);
3116 }
3117
3118 void __attribute__((cold))
3119 i40e_set_default_pctype_table(struct rte_eth_dev *dev)
3120 {
3121         struct i40e_adapter *ad =
3122                         I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
3123         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3124         int i;
3125
3126         for (i = 0; i < I40E_FLOW_TYPE_MAX; i++)
3127                 ad->pctypes_tbl[i] = 0ULL;
3128         ad->flow_types_mask = 0ULL;
3129         ad->pctypes_mask = 0ULL;
3130
3131         ad->pctypes_tbl[RTE_ETH_FLOW_FRAG_IPV4] =
3132                                 (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV4);
3133         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_UDP] =
3134                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_UDP);
3135         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_TCP] =
3136                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_TCP);
3137         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_SCTP] =
3138                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_SCTP);
3139         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_OTHER] =
3140                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_OTHER);
3141         ad->pctypes_tbl[RTE_ETH_FLOW_FRAG_IPV6] =
3142                                 (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV6);
3143         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_UDP] =
3144                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_UDP);
3145         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_TCP] =
3146                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_TCP);
3147         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_SCTP] =
3148                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_SCTP);
3149         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_OTHER] =
3150                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_OTHER);
3151         ad->pctypes_tbl[RTE_ETH_FLOW_L2_PAYLOAD] =
3152                                 (1ULL << I40E_FILTER_PCTYPE_L2_PAYLOAD);
3153
3154         if (hw->mac.type == I40E_MAC_X722) {
3155                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_UDP] |=
3156                         (1ULL << I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP);
3157                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_UDP] |=
3158                         (1ULL << I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP);
3159                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_TCP] |=
3160                         (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK);
3161                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_UDP] |=
3162                         (1ULL << I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP);
3163                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_UDP] |=
3164                         (1ULL << I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP);
3165                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_TCP] |=
3166                         (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK);
3167         }
3168
3169         for (i = 0; i < I40E_FLOW_TYPE_MAX; i++) {
3170                 if (ad->pctypes_tbl[i])
3171                         ad->flow_types_mask |= (1ULL << i);
3172                 ad->pctypes_mask |= ad->pctypes_tbl[i];
3173         }
3174 }
3175
3176 /* Stubs needed for linkage when CONFIG_RTE_I40E_INC_VECTOR is set to 'n' */
3177 int __attribute__((weak))
3178 i40e_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
3179 {
3180         return -1;
3181 }
3182
3183 uint16_t __attribute__((weak))
3184 i40e_recv_pkts_vec(
3185         void __rte_unused *rx_queue,
3186         struct rte_mbuf __rte_unused **rx_pkts,
3187         uint16_t __rte_unused nb_pkts)
3188 {
3189         return 0;
3190 }
3191
3192 uint16_t __attribute__((weak))
3193 i40e_recv_scattered_pkts_vec(
3194         void __rte_unused *rx_queue,
3195         struct rte_mbuf __rte_unused **rx_pkts,
3196         uint16_t __rte_unused nb_pkts)
3197 {
3198         return 0;
3199 }
3200
3201 uint16_t __attribute__((weak))
3202 i40e_recv_pkts_vec_avx2(void __rte_unused *rx_queue,
3203                         struct rte_mbuf __rte_unused **rx_pkts,
3204                         uint16_t __rte_unused nb_pkts)
3205 {
3206         return 0;
3207 }
3208
3209 uint16_t __attribute__((weak))
3210 i40e_recv_scattered_pkts_vec_avx2(void __rte_unused *rx_queue,
3211                         struct rte_mbuf __rte_unused **rx_pkts,
3212                         uint16_t __rte_unused nb_pkts)
3213 {
3214         return 0;
3215 }
3216
3217 int __attribute__((weak))
3218 i40e_rxq_vec_setup(struct i40e_rx_queue __rte_unused *rxq)
3219 {
3220         return -1;
3221 }
3222
3223 int __attribute__((weak))
3224 i40e_txq_vec_setup(struct i40e_tx_queue __rte_unused *txq)
3225 {
3226         return -1;
3227 }
3228
3229 void __attribute__((weak))
3230 i40e_rx_queue_release_mbufs_vec(struct i40e_rx_queue __rte_unused*rxq)
3231 {
3232         return;
3233 }
3234
3235 uint16_t __attribute__((weak))
3236 i40e_xmit_fixed_burst_vec(void __rte_unused * tx_queue,
3237                           struct rte_mbuf __rte_unused **tx_pkts,
3238                           uint16_t __rte_unused nb_pkts)
3239 {
3240         return 0;
3241 }
3242
3243 uint16_t __attribute__((weak))
3244 i40e_xmit_pkts_vec_avx2(void __rte_unused * tx_queue,
3245                           struct rte_mbuf __rte_unused **tx_pkts,
3246                           uint16_t __rte_unused nb_pkts)
3247 {
3248         return 0;
3249 }