net/i40e: fix queue offload initialization
[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->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT) {
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_SEG ||
1443                             m->nb_segs > I40E_TX_MAX_MTU_SEG) {
1444                                 rte_errno = -EINVAL;
1445                                 return i;
1446                         }
1447                 } else if ((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 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1462                 ret = rte_validate_tx_offload(m);
1463                 if (ret != 0) {
1464                         rte_errno = ret;
1465                         return i;
1466                 }
1467 #endif
1468                 ret = rte_net_intel_cksum_prepare(m);
1469                 if (ret != 0) {
1470                         rte_errno = ret;
1471                         return i;
1472                 }
1473         }
1474         return i;
1475 }
1476
1477 /*
1478  * Find the VSI the queue belongs to. 'queue_idx' is the queue index
1479  * application used, which assume having sequential ones. But from driver's
1480  * perspective, it's different. For example, q0 belongs to FDIR VSI, q1-q64
1481  * to MAIN VSI, , q65-96 to SRIOV VSIs, q97-128 to VMDQ VSIs. For application
1482  * running on host, q1-64 and q97-128 can be used, total 96 queues. They can
1483  * use queue_idx from 0 to 95 to access queues, while real queue would be
1484  * different. This function will do a queue mapping to find VSI the queue
1485  * belongs to.
1486  */
1487 static struct i40e_vsi*
1488 i40e_pf_get_vsi_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1489 {
1490         /* the queue in MAIN VSI range */
1491         if (queue_idx < pf->main_vsi->nb_qps)
1492                 return pf->main_vsi;
1493
1494         queue_idx -= pf->main_vsi->nb_qps;
1495
1496         /* queue_idx is greater than VMDQ VSIs range */
1497         if (queue_idx > pf->nb_cfg_vmdq_vsi * pf->vmdq_nb_qps - 1) {
1498                 PMD_INIT_LOG(ERR, "queue_idx out of range. VMDQ configured?");
1499                 return NULL;
1500         }
1501
1502         return pf->vmdq[queue_idx / pf->vmdq_nb_qps].vsi;
1503 }
1504
1505 static uint16_t
1506 i40e_get_queue_offset_by_qindex(struct i40e_pf *pf, uint16_t queue_idx)
1507 {
1508         /* the queue in MAIN VSI range */
1509         if (queue_idx < pf->main_vsi->nb_qps)
1510                 return queue_idx;
1511
1512         /* It's VMDQ queues */
1513         queue_idx -= pf->main_vsi->nb_qps;
1514
1515         if (pf->nb_cfg_vmdq_vsi)
1516                 return queue_idx % pf->vmdq_nb_qps;
1517         else {
1518                 PMD_INIT_LOG(ERR, "Fail to get queue offset");
1519                 return (uint16_t)(-1);
1520         }
1521 }
1522
1523 int
1524 i40e_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1525 {
1526         struct i40e_rx_queue *rxq;
1527         int err = -1;
1528         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1529
1530         PMD_INIT_FUNC_TRACE();
1531
1532         if (rx_queue_id < dev->data->nb_rx_queues) {
1533                 rxq = dev->data->rx_queues[rx_queue_id];
1534
1535                 err = i40e_alloc_rx_queue_mbufs(rxq);
1536                 if (err) {
1537                         PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
1538                         return err;
1539                 }
1540
1541                 rte_wmb();
1542
1543                 /* Init the RX tail regieter. */
1544                 I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
1545
1546                 err = i40e_switch_rx_queue(hw, rxq->reg_idx, TRUE);
1547
1548                 if (err) {
1549                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
1550                                     rx_queue_id);
1551
1552                         i40e_rx_queue_release_mbufs(rxq);
1553                         i40e_reset_rx_queue(rxq);
1554                 } else
1555                         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1556         }
1557
1558         return err;
1559 }
1560
1561 int
1562 i40e_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1563 {
1564         struct i40e_rx_queue *rxq;
1565         int err;
1566         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1567
1568         if (rx_queue_id < dev->data->nb_rx_queues) {
1569                 rxq = dev->data->rx_queues[rx_queue_id];
1570
1571                 /*
1572                 * rx_queue_id is queue id application refers to, while
1573                 * rxq->reg_idx is the real queue index.
1574                 */
1575                 err = i40e_switch_rx_queue(hw, rxq->reg_idx, FALSE);
1576
1577                 if (err) {
1578                         PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
1579                                     rx_queue_id);
1580                         return err;
1581                 }
1582                 i40e_rx_queue_release_mbufs(rxq);
1583                 i40e_reset_rx_queue(rxq);
1584                 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1585         }
1586
1587         return 0;
1588 }
1589
1590 int
1591 i40e_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1592 {
1593         int err = -1;
1594         struct i40e_tx_queue *txq;
1595         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1596
1597         PMD_INIT_FUNC_TRACE();
1598
1599         if (tx_queue_id < dev->data->nb_tx_queues) {
1600                 txq = dev->data->tx_queues[tx_queue_id];
1601
1602                 /*
1603                 * tx_queue_id is queue id application refers to, while
1604                 * rxq->reg_idx is the real queue index.
1605                 */
1606                 err = i40e_switch_tx_queue(hw, txq->reg_idx, TRUE);
1607                 if (err)
1608                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
1609                                     tx_queue_id);
1610                 else
1611                         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1612         }
1613
1614         return err;
1615 }
1616
1617 int
1618 i40e_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1619 {
1620         struct i40e_tx_queue *txq;
1621         int err;
1622         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1623
1624         if (tx_queue_id < dev->data->nb_tx_queues) {
1625                 txq = dev->data->tx_queues[tx_queue_id];
1626
1627                 /*
1628                 * tx_queue_id is queue id application refers to, while
1629                 * txq->reg_idx is the real queue index.
1630                 */
1631                 err = i40e_switch_tx_queue(hw, txq->reg_idx, FALSE);
1632
1633                 if (err) {
1634                         PMD_DRV_LOG(ERR, "Failed to switch TX queue %u of",
1635                                     tx_queue_id);
1636                         return err;
1637                 }
1638
1639                 i40e_tx_queue_release_mbufs(txq);
1640                 i40e_reset_tx_queue(txq);
1641                 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1642         }
1643
1644         return 0;
1645 }
1646
1647 const uint32_t *
1648 i40e_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1649 {
1650         static const uint32_t ptypes[] = {
1651                 /* refers to i40e_rxd_pkt_type_mapping() */
1652                 RTE_PTYPE_L2_ETHER,
1653                 RTE_PTYPE_L2_ETHER_TIMESYNC,
1654                 RTE_PTYPE_L2_ETHER_LLDP,
1655                 RTE_PTYPE_L2_ETHER_ARP,
1656                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1657                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
1658                 RTE_PTYPE_L4_FRAG,
1659                 RTE_PTYPE_L4_ICMP,
1660                 RTE_PTYPE_L4_NONFRAG,
1661                 RTE_PTYPE_L4_SCTP,
1662                 RTE_PTYPE_L4_TCP,
1663                 RTE_PTYPE_L4_UDP,
1664                 RTE_PTYPE_TUNNEL_GRENAT,
1665                 RTE_PTYPE_TUNNEL_IP,
1666                 RTE_PTYPE_INNER_L2_ETHER,
1667                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1668                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
1669                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
1670                 RTE_PTYPE_INNER_L4_FRAG,
1671                 RTE_PTYPE_INNER_L4_ICMP,
1672                 RTE_PTYPE_INNER_L4_NONFRAG,
1673                 RTE_PTYPE_INNER_L4_SCTP,
1674                 RTE_PTYPE_INNER_L4_TCP,
1675                 RTE_PTYPE_INNER_L4_UDP,
1676                 RTE_PTYPE_UNKNOWN
1677         };
1678
1679         if (dev->rx_pkt_burst == i40e_recv_pkts ||
1680 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1681             dev->rx_pkt_burst == i40e_recv_pkts_bulk_alloc ||
1682 #endif
1683             dev->rx_pkt_burst == i40e_recv_scattered_pkts ||
1684             dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec ||
1685             dev->rx_pkt_burst == i40e_recv_pkts_vec ||
1686             dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec_avx2 ||
1687             dev->rx_pkt_burst == i40e_recv_pkts_vec_avx2)
1688                 return ptypes;
1689         return NULL;
1690 }
1691
1692 static int
1693 i40e_check_rx_queue_offloads(struct rte_eth_dev *dev, uint64_t requested)
1694 {
1695         struct rte_eth_dev_info dev_info;
1696         uint64_t mandatory = dev->data->dev_conf.rxmode.offloads;
1697         uint64_t supported; /* All per port offloads */
1698
1699         dev->dev_ops->dev_infos_get(dev, &dev_info);
1700         supported = dev_info.rx_offload_capa ^ dev_info.rx_queue_offload_capa;
1701         if ((requested & dev_info.rx_offload_capa) != requested)
1702                 return 0; /* requested range check */
1703         return !((mandatory ^ requested) & supported);
1704 }
1705
1706 static int
1707 i40e_dev_first_queue(uint16_t idx, void **queues, int num)
1708 {
1709         uint16_t i;
1710
1711         for (i = 0; i < num; i++) {
1712                 if (i != idx && queues[i])
1713                         return 0;
1714         }
1715
1716         return 1;
1717 }
1718
1719 static int
1720 i40e_dev_rx_queue_setup_runtime(struct rte_eth_dev *dev,
1721                                 struct i40e_rx_queue *rxq)
1722 {
1723         struct i40e_adapter *ad =
1724                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1725         int use_def_burst_func =
1726                 check_rx_burst_bulk_alloc_preconditions(rxq);
1727         uint16_t buf_size =
1728                 (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
1729                            RTE_PKTMBUF_HEADROOM);
1730         int use_scattered_rx =
1731                 ((rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size);
1732
1733         if (i40e_rx_queue_init(rxq) != I40E_SUCCESS) {
1734                 PMD_DRV_LOG(ERR,
1735                             "Failed to do RX queue initialization");
1736                 return -EINVAL;
1737         }
1738
1739         if (i40e_dev_first_queue(rxq->queue_id,
1740                                  dev->data->rx_queues,
1741                                  dev->data->nb_rx_queues)) {
1742                 /**
1743                  * If it is the first queue to setup,
1744                  * set all flags to default and call
1745                  * i40e_set_rx_function.
1746                  */
1747                 ad->rx_bulk_alloc_allowed = true;
1748                 ad->rx_vec_allowed = true;
1749                 dev->data->scattered_rx = use_scattered_rx;
1750                 if (use_def_burst_func)
1751                         ad->rx_bulk_alloc_allowed = false;
1752                 i40e_set_rx_function(dev);
1753                 return 0;
1754         }
1755
1756         /* check bulk alloc conflict */
1757         if (ad->rx_bulk_alloc_allowed && use_def_burst_func) {
1758                 PMD_DRV_LOG(ERR, "Can't use default burst.");
1759                 return -EINVAL;
1760         }
1761         /* check scatterred conflict */
1762         if (!dev->data->scattered_rx && use_scattered_rx) {
1763                 PMD_DRV_LOG(ERR, "Scattered rx is required.");
1764                 return -EINVAL;
1765         }
1766         /* check vector conflict */
1767         if (ad->rx_vec_allowed && i40e_rxq_vec_setup(rxq)) {
1768                 PMD_DRV_LOG(ERR, "Failed vector rx setup.");
1769                 return -EINVAL;
1770         }
1771
1772         return 0;
1773 }
1774
1775 int
1776 i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
1777                         uint16_t queue_idx,
1778                         uint16_t nb_desc,
1779                         unsigned int socket_id,
1780                         const struct rte_eth_rxconf *rx_conf,
1781                         struct rte_mempool *mp)
1782 {
1783         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1784         struct i40e_adapter *ad =
1785                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
1786         struct i40e_vsi *vsi;
1787         struct i40e_pf *pf = NULL;
1788         struct i40e_vf *vf = NULL;
1789         struct i40e_rx_queue *rxq;
1790         const struct rte_memzone *rz;
1791         uint32_t ring_size;
1792         uint16_t len, i;
1793         uint16_t reg_idx, base, bsf, tc_mapping;
1794         int q_offset, use_def_burst_func = 1;
1795         struct rte_eth_dev_info dev_info;
1796
1797         if (!i40e_check_rx_queue_offloads(dev, rx_conf->offloads)) {
1798                 dev->dev_ops->dev_infos_get(dev, &dev_info);
1799                 PMD_INIT_LOG(ERR, "%p: Rx queue offloads 0x%" PRIx64
1800                         " don't match port  offloads 0x%" PRIx64
1801                         " or supported offloads 0x%" PRIx64,
1802                         (void *)dev, rx_conf->offloads,
1803                         dev->data->dev_conf.rxmode.offloads,
1804                         dev_info.rx_offload_capa);
1805                 return -ENOTSUP;
1806         }
1807
1808         if (hw->mac.type == I40E_MAC_VF || hw->mac.type == I40E_MAC_X722_VF) {
1809                 vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
1810                 vsi = &vf->vsi;
1811                 if (!vsi)
1812                         return -EINVAL;
1813                 reg_idx = queue_idx;
1814         } else {
1815                 pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
1816                 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
1817                 if (!vsi)
1818                         return -EINVAL;
1819                 q_offset = i40e_get_queue_offset_by_qindex(pf, queue_idx);
1820                 if (q_offset < 0)
1821                         return -EINVAL;
1822                 reg_idx = vsi->base_queue + q_offset;
1823         }
1824
1825         if (nb_desc % I40E_ALIGN_RING_DESC != 0 ||
1826             (nb_desc > I40E_MAX_RING_DESC) ||
1827             (nb_desc < I40E_MIN_RING_DESC)) {
1828                 PMD_DRV_LOG(ERR, "Number (%u) of receive descriptors is "
1829                             "invalid", nb_desc);
1830                 return -EINVAL;
1831         }
1832
1833         /* Free memory if needed */
1834         if (dev->data->rx_queues[queue_idx]) {
1835                 i40e_dev_rx_queue_release(dev->data->rx_queues[queue_idx]);
1836                 dev->data->rx_queues[queue_idx] = NULL;
1837         }
1838
1839         /* Allocate the rx queue data structure */
1840         rxq = rte_zmalloc_socket("i40e rx queue",
1841                                  sizeof(struct i40e_rx_queue),
1842                                  RTE_CACHE_LINE_SIZE,
1843                                  socket_id);
1844         if (!rxq) {
1845                 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
1846                             "rx queue data structure");
1847                 return -ENOMEM;
1848         }
1849         rxq->mp = mp;
1850         rxq->nb_rx_desc = nb_desc;
1851         rxq->rx_free_thresh = rx_conf->rx_free_thresh;
1852         rxq->queue_id = queue_idx;
1853         rxq->reg_idx = reg_idx;
1854         rxq->port_id = dev->data->port_id;
1855         rxq->crc_len = (uint8_t)((dev->data->dev_conf.rxmode.offloads &
1856                         DEV_RX_OFFLOAD_CRC_STRIP) ? 0 : ETHER_CRC_LEN);
1857         rxq->drop_en = rx_conf->rx_drop_en;
1858         rxq->vsi = vsi;
1859         rxq->rx_deferred_start = rx_conf->rx_deferred_start;
1860         rxq->offloads = rx_conf->offloads;
1861
1862         /* Allocate the maximun number of RX ring hardware descriptor. */
1863         len = I40E_MAX_RING_DESC;
1864
1865         /**
1866          * Allocating a little more memory because vectorized/bulk_alloc Rx
1867          * functions doesn't check boundaries each time.
1868          */
1869         len += RTE_PMD_I40E_RX_MAX_BURST;
1870
1871         ring_size = RTE_ALIGN(len * sizeof(union i40e_rx_desc),
1872                               I40E_DMA_MEM_ALIGN);
1873
1874         rz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_idx,
1875                               ring_size, I40E_RING_BASE_ALIGN, socket_id);
1876         if (!rz) {
1877                 i40e_dev_rx_queue_release(rxq);
1878                 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX");
1879                 return -ENOMEM;
1880         }
1881
1882         /* Zero all the descriptors in the ring. */
1883         memset(rz->addr, 0, ring_size);
1884
1885         rxq->rx_ring_phys_addr = rz->iova;
1886         rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
1887
1888         len = (uint16_t)(nb_desc + RTE_PMD_I40E_RX_MAX_BURST);
1889
1890         /* Allocate the software ring. */
1891         rxq->sw_ring =
1892                 rte_zmalloc_socket("i40e rx sw ring",
1893                                    sizeof(struct i40e_rx_entry) * len,
1894                                    RTE_CACHE_LINE_SIZE,
1895                                    socket_id);
1896         if (!rxq->sw_ring) {
1897                 i40e_dev_rx_queue_release(rxq);
1898                 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW ring");
1899                 return -ENOMEM;
1900         }
1901
1902         i40e_reset_rx_queue(rxq);
1903         rxq->q_set = TRUE;
1904
1905         for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
1906                 if (!(vsi->enabled_tc & (1 << i)))
1907                         continue;
1908                 tc_mapping = rte_le_to_cpu_16(vsi->info.tc_mapping[i]);
1909                 base = (tc_mapping & I40E_AQ_VSI_TC_QUE_OFFSET_MASK) >>
1910                         I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT;
1911                 bsf = (tc_mapping & I40E_AQ_VSI_TC_QUE_NUMBER_MASK) >>
1912                         I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT;
1913
1914                 if (queue_idx >= base && queue_idx < (base + BIT(bsf)))
1915                         rxq->dcb_tc = i;
1916         }
1917
1918         if (dev->data->dev_started) {
1919                 if (i40e_dev_rx_queue_setup_runtime(dev, rxq)) {
1920                         i40e_dev_rx_queue_release(rxq);
1921                         return -EINVAL;
1922                 }
1923         } else {
1924                 use_def_burst_func =
1925                         check_rx_burst_bulk_alloc_preconditions(rxq);
1926                 if (!use_def_burst_func) {
1927 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
1928                         PMD_INIT_LOG(DEBUG,
1929                           "Rx Burst Bulk Alloc Preconditions are "
1930                           "satisfied. Rx Burst Bulk Alloc function will be "
1931                           "used on port=%d, queue=%d.",
1932                           rxq->port_id, rxq->queue_id);
1933 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
1934                 } else {
1935                         PMD_INIT_LOG(DEBUG,
1936                           "Rx Burst Bulk Alloc Preconditions are "
1937                           "not satisfied, Scattered Rx is requested, "
1938                           "or RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC is "
1939                           "not enabled on port=%d, queue=%d.",
1940                           rxq->port_id, rxq->queue_id);
1941                         ad->rx_bulk_alloc_allowed = false;
1942                 }
1943         }
1944
1945         dev->data->rx_queues[queue_idx] = rxq;
1946         return 0;
1947 }
1948
1949 void
1950 i40e_dev_rx_queue_release(void *rxq)
1951 {
1952         struct i40e_rx_queue *q = (struct i40e_rx_queue *)rxq;
1953
1954         if (!q) {
1955                 PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL");
1956                 return;
1957         }
1958
1959         i40e_rx_queue_release_mbufs(q);
1960         rte_free(q->sw_ring);
1961         rte_free(q);
1962 }
1963
1964 uint32_t
1965 i40e_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1966 {
1967 #define I40E_RXQ_SCAN_INTERVAL 4
1968         volatile union i40e_rx_desc *rxdp;
1969         struct i40e_rx_queue *rxq;
1970         uint16_t desc = 0;
1971
1972         rxq = dev->data->rx_queues[rx_queue_id];
1973         rxdp = &(rxq->rx_ring[rxq->rx_tail]);
1974         while ((desc < rxq->nb_rx_desc) &&
1975                 ((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
1976                 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
1977                                 (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) {
1978                 /**
1979                  * Check the DD bit of a rx descriptor of each 4 in a group,
1980                  * to avoid checking too frequently and downgrading performance
1981                  * too much.
1982                  */
1983                 desc += I40E_RXQ_SCAN_INTERVAL;
1984                 rxdp += I40E_RXQ_SCAN_INTERVAL;
1985                 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
1986                         rxdp = &(rxq->rx_ring[rxq->rx_tail +
1987                                         desc - rxq->nb_rx_desc]);
1988         }
1989
1990         return desc;
1991 }
1992
1993 int
1994 i40e_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
1995 {
1996         volatile union i40e_rx_desc *rxdp;
1997         struct i40e_rx_queue *rxq = rx_queue;
1998         uint16_t desc;
1999         int ret;
2000
2001         if (unlikely(offset >= rxq->nb_rx_desc)) {
2002                 PMD_DRV_LOG(ERR, "Invalid RX descriptor id %u", offset);
2003                 return 0;
2004         }
2005
2006         desc = rxq->rx_tail + offset;
2007         if (desc >= rxq->nb_rx_desc)
2008                 desc -= rxq->nb_rx_desc;
2009
2010         rxdp = &(rxq->rx_ring[desc]);
2011
2012         ret = !!(((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
2013                 I40E_RXD_QW1_STATUS_MASK) >> I40E_RXD_QW1_STATUS_SHIFT) &
2014                                 (1 << I40E_RX_DESC_STATUS_DD_SHIFT));
2015
2016         return ret;
2017 }
2018
2019 int
2020 i40e_dev_rx_descriptor_status(void *rx_queue, uint16_t offset)
2021 {
2022         struct i40e_rx_queue *rxq = rx_queue;
2023         volatile uint64_t *status;
2024         uint64_t mask;
2025         uint32_t desc;
2026
2027         if (unlikely(offset >= rxq->nb_rx_desc))
2028                 return -EINVAL;
2029
2030         if (offset >= rxq->nb_rx_desc - rxq->nb_rx_hold)
2031                 return RTE_ETH_RX_DESC_UNAVAIL;
2032
2033         desc = rxq->rx_tail + offset;
2034         if (desc >= rxq->nb_rx_desc)
2035                 desc -= rxq->nb_rx_desc;
2036
2037         status = &rxq->rx_ring[desc].wb.qword1.status_error_len;
2038         mask = rte_le_to_cpu_64((1ULL << I40E_RX_DESC_STATUS_DD_SHIFT)
2039                 << I40E_RXD_QW1_STATUS_SHIFT);
2040         if (*status & mask)
2041                 return RTE_ETH_RX_DESC_DONE;
2042
2043         return RTE_ETH_RX_DESC_AVAIL;
2044 }
2045
2046 int
2047 i40e_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
2048 {
2049         struct i40e_tx_queue *txq = tx_queue;
2050         volatile uint64_t *status;
2051         uint64_t mask, expect;
2052         uint32_t desc;
2053
2054         if (unlikely(offset >= txq->nb_tx_desc))
2055                 return -EINVAL;
2056
2057         desc = txq->tx_tail + offset;
2058         /* go to next desc that has the RS bit */
2059         desc = ((desc + txq->tx_rs_thresh - 1) / txq->tx_rs_thresh) *
2060                 txq->tx_rs_thresh;
2061         if (desc >= txq->nb_tx_desc) {
2062                 desc -= txq->nb_tx_desc;
2063                 if (desc >= txq->nb_tx_desc)
2064                         desc -= txq->nb_tx_desc;
2065         }
2066
2067         status = &txq->tx_ring[desc].cmd_type_offset_bsz;
2068         mask = rte_le_to_cpu_64(I40E_TXD_QW1_DTYPE_MASK);
2069         expect = rte_cpu_to_le_64(
2070                 I40E_TX_DESC_DTYPE_DESC_DONE << I40E_TXD_QW1_DTYPE_SHIFT);
2071         if ((*status & mask) == expect)
2072                 return RTE_ETH_TX_DESC_DONE;
2073
2074         return RTE_ETH_TX_DESC_FULL;
2075 }
2076
2077 static int
2078 i40e_check_tx_queue_offloads(struct rte_eth_dev *dev, uint64_t requested)
2079 {
2080         struct rte_eth_dev_info dev_info;
2081         uint64_t mandatory = dev->data->dev_conf.txmode.offloads;
2082         uint64_t supported; /* All per port offloads */
2083
2084         dev->dev_ops->dev_infos_get(dev, &dev_info);
2085         supported = dev_info.tx_offload_capa ^ dev_info.tx_queue_offload_capa;
2086         if ((requested & dev_info.tx_offload_capa) != requested)
2087                 return 0; /* requested range check */
2088         return !((mandatory ^ requested) & supported);
2089 }
2090
2091 static int
2092 i40e_dev_tx_queue_setup_runtime(struct rte_eth_dev *dev,
2093                                 struct i40e_tx_queue *txq)
2094 {
2095         struct i40e_adapter *ad =
2096                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
2097
2098         if (i40e_tx_queue_init(txq) != I40E_SUCCESS) {
2099                 PMD_DRV_LOG(ERR,
2100                             "Failed to do TX queue initialization");
2101                 return -EINVAL;
2102         }
2103
2104         if (i40e_dev_first_queue(txq->queue_id,
2105                                  dev->data->tx_queues,
2106                                  dev->data->nb_tx_queues)) {
2107                 /**
2108                  * If it is the first queue to setup,
2109                  * set all flags and call
2110                  * i40e_set_tx_function.
2111                  */
2112                 i40e_set_tx_function_flag(dev, txq);
2113                 i40e_set_tx_function(dev);
2114                 return 0;
2115         }
2116
2117         /* check vector conflict */
2118         if (ad->tx_vec_allowed) {
2119                 if (txq->tx_rs_thresh > RTE_I40E_TX_MAX_FREE_BUF_SZ ||
2120                     i40e_txq_vec_setup(txq)) {
2121                         PMD_DRV_LOG(ERR, "Failed vector tx setup.");
2122                         return -EINVAL;
2123                 }
2124         }
2125         /* check simple tx conflict */
2126         if (ad->tx_simple_allowed) {
2127                 if (txq->offloads != 0 ||
2128                                 txq->tx_rs_thresh < RTE_PMD_I40E_TX_MAX_BURST) {
2129                         PMD_DRV_LOG(ERR, "No-simple tx is required.");
2130                         return -EINVAL;
2131                 }
2132         }
2133
2134         return 0;
2135 }
2136
2137 int
2138 i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
2139                         uint16_t queue_idx,
2140                         uint16_t nb_desc,
2141                         unsigned int socket_id,
2142                         const struct rte_eth_txconf *tx_conf)
2143 {
2144         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2145         struct i40e_vsi *vsi;
2146         struct i40e_pf *pf = NULL;
2147         struct i40e_vf *vf = NULL;
2148         struct i40e_tx_queue *txq;
2149         const struct rte_memzone *tz;
2150         uint32_t ring_size;
2151         uint16_t tx_rs_thresh, tx_free_thresh;
2152         uint16_t reg_idx, i, base, bsf, tc_mapping;
2153         int q_offset;
2154         struct rte_eth_dev_info dev_info;
2155
2156         if (!i40e_check_tx_queue_offloads(dev, tx_conf->offloads)) {
2157                 dev->dev_ops->dev_infos_get(dev, &dev_info);
2158                 PMD_INIT_LOG(ERR, "%p: Tx queue offloads 0x%" PRIx64
2159                         " don't match port  offloads 0x%" PRIx64
2160                         " or supported offloads 0x%" PRIx64,
2161                         (void *)dev, tx_conf->offloads,
2162                         dev->data->dev_conf.txmode.offloads,
2163                         dev_info.tx_offload_capa);
2164                         return -ENOTSUP;
2165         }
2166
2167         if (hw->mac.type == I40E_MAC_VF || hw->mac.type == I40E_MAC_X722_VF) {
2168                 vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
2169                 vsi = &vf->vsi;
2170                 if (!vsi)
2171                         return -EINVAL;
2172                 reg_idx = queue_idx;
2173         } else {
2174                 pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
2175                 vsi = i40e_pf_get_vsi_by_qindex(pf, queue_idx);
2176                 if (!vsi)
2177                         return -EINVAL;
2178                 q_offset = i40e_get_queue_offset_by_qindex(pf, queue_idx);
2179                 if (q_offset < 0)
2180                         return -EINVAL;
2181                 reg_idx = vsi->base_queue + q_offset;
2182         }
2183
2184         if (nb_desc % I40E_ALIGN_RING_DESC != 0 ||
2185             (nb_desc > I40E_MAX_RING_DESC) ||
2186             (nb_desc < I40E_MIN_RING_DESC)) {
2187                 PMD_DRV_LOG(ERR, "Number (%u) of transmit descriptors is "
2188                             "invalid", nb_desc);
2189                 return -EINVAL;
2190         }
2191
2192         /**
2193          * The following two parameters control the setting of the RS bit on
2194          * transmit descriptors. TX descriptors will have their RS bit set
2195          * after txq->tx_rs_thresh descriptors have been used. The TX
2196          * descriptor ring will be cleaned after txq->tx_free_thresh
2197          * descriptors are used or if the number of descriptors required to
2198          * transmit a packet is greater than the number of free TX descriptors.
2199          *
2200          * The following constraints must be satisfied:
2201          *  - tx_rs_thresh must be greater than 0.
2202          *  - tx_rs_thresh must be less than the size of the ring minus 2.
2203          *  - tx_rs_thresh must be less than or equal to tx_free_thresh.
2204          *  - tx_rs_thresh must be a divisor of the ring size.
2205          *  - tx_free_thresh must be greater than 0.
2206          *  - tx_free_thresh must be less than the size of the ring minus 3.
2207          *
2208          * One descriptor in the TX ring is used as a sentinel to avoid a H/W
2209          * race condition, hence the maximum threshold constraints. When set
2210          * to zero use default values.
2211          */
2212         tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2213                 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2214         tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2215                 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2216         if (tx_rs_thresh >= (nb_desc - 2)) {
2217                 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2218                              "number of TX descriptors minus 2. "
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         if (tx_free_thresh >= (nb_desc - 3)) {
2226                 PMD_INIT_LOG(ERR, "tx_free_thresh must be less than the "
2227                              "number of TX descriptors minus 3. "
2228                              "(tx_free_thresh=%u port=%d queue=%d)",
2229                              (unsigned int)tx_free_thresh,
2230                              (int)dev->data->port_id,
2231                              (int)queue_idx);
2232                 return I40E_ERR_PARAM;
2233         }
2234         if (tx_rs_thresh > tx_free_thresh) {
2235                 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or "
2236                              "equal to tx_free_thresh. (tx_free_thresh=%u"
2237                              " tx_rs_thresh=%u port=%d queue=%d)",
2238                              (unsigned int)tx_free_thresh,
2239                              (unsigned int)tx_rs_thresh,
2240                              (int)dev->data->port_id,
2241                              (int)queue_idx);
2242                 return I40E_ERR_PARAM;
2243         }
2244         if ((nb_desc % tx_rs_thresh) != 0) {
2245                 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2246                              "number of TX descriptors. (tx_rs_thresh=%u"
2247                              " port=%d queue=%d)",
2248                              (unsigned int)tx_rs_thresh,
2249                              (int)dev->data->port_id,
2250                              (int)queue_idx);
2251                 return I40E_ERR_PARAM;
2252         }
2253         if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2254                 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2255                              "tx_rs_thresh is greater than 1. "
2256                              "(tx_rs_thresh=%u port=%d queue=%d)",
2257                              (unsigned int)tx_rs_thresh,
2258                              (int)dev->data->port_id,
2259                              (int)queue_idx);
2260                 return I40E_ERR_PARAM;
2261         }
2262
2263         /* Free memory if needed. */
2264         if (dev->data->tx_queues[queue_idx]) {
2265                 i40e_dev_tx_queue_release(dev->data->tx_queues[queue_idx]);
2266                 dev->data->tx_queues[queue_idx] = NULL;
2267         }
2268
2269         /* Allocate the TX queue data structure. */
2270         txq = rte_zmalloc_socket("i40e tx queue",
2271                                   sizeof(struct i40e_tx_queue),
2272                                   RTE_CACHE_LINE_SIZE,
2273                                   socket_id);
2274         if (!txq) {
2275                 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2276                             "tx queue structure");
2277                 return -ENOMEM;
2278         }
2279
2280         /* Allocate TX hardware ring descriptors. */
2281         ring_size = sizeof(struct i40e_tx_desc) * I40E_MAX_RING_DESC;
2282         ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2283         tz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_idx,
2284                               ring_size, I40E_RING_BASE_ALIGN, socket_id);
2285         if (!tz) {
2286                 i40e_dev_tx_queue_release(txq);
2287                 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX");
2288                 return -ENOMEM;
2289         }
2290
2291         txq->nb_tx_desc = nb_desc;
2292         txq->tx_rs_thresh = tx_rs_thresh;
2293         txq->tx_free_thresh = tx_free_thresh;
2294         txq->pthresh = tx_conf->tx_thresh.pthresh;
2295         txq->hthresh = tx_conf->tx_thresh.hthresh;
2296         txq->wthresh = tx_conf->tx_thresh.wthresh;
2297         txq->queue_id = queue_idx;
2298         txq->reg_idx = reg_idx;
2299         txq->port_id = dev->data->port_id;
2300         txq->txq_flags = tx_conf->txq_flags;
2301         txq->offloads = tx_conf->offloads;
2302         txq->vsi = vsi;
2303         txq->tx_deferred_start = tx_conf->tx_deferred_start;
2304
2305         txq->tx_ring_phys_addr = tz->iova;
2306         txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2307
2308         /* Allocate software ring */
2309         txq->sw_ring =
2310                 rte_zmalloc_socket("i40e tx sw ring",
2311                                    sizeof(struct i40e_tx_entry) * nb_desc,
2312                                    RTE_CACHE_LINE_SIZE,
2313                                    socket_id);
2314         if (!txq->sw_ring) {
2315                 i40e_dev_tx_queue_release(txq);
2316                 PMD_DRV_LOG(ERR, "Failed to allocate memory for SW TX ring");
2317                 return -ENOMEM;
2318         }
2319
2320         i40e_reset_tx_queue(txq);
2321         txq->q_set = TRUE;
2322
2323         for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
2324                 if (!(vsi->enabled_tc & (1 << i)))
2325                         continue;
2326                 tc_mapping = rte_le_to_cpu_16(vsi->info.tc_mapping[i]);
2327                 base = (tc_mapping & I40E_AQ_VSI_TC_QUE_OFFSET_MASK) >>
2328                         I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT;
2329                 bsf = (tc_mapping & I40E_AQ_VSI_TC_QUE_NUMBER_MASK) >>
2330                         I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT;
2331
2332                 if (queue_idx >= base && queue_idx < (base + BIT(bsf)))
2333                         txq->dcb_tc = i;
2334         }
2335
2336         if (dev->data->dev_started) {
2337                 if (i40e_dev_tx_queue_setup_runtime(dev, txq)) {
2338                         i40e_dev_tx_queue_release(txq);
2339                         return -EINVAL;
2340                 }
2341         } else {
2342                 /**
2343                  * Use a simple TX queue without offloads or
2344                  * multi segs if possible
2345                  */
2346                 i40e_set_tx_function_flag(dev, txq);
2347         }
2348         dev->data->tx_queues[queue_idx] = txq;
2349
2350         return 0;
2351 }
2352
2353 void
2354 i40e_dev_tx_queue_release(void *txq)
2355 {
2356         struct i40e_tx_queue *q = (struct i40e_tx_queue *)txq;
2357
2358         if (!q) {
2359                 PMD_DRV_LOG(DEBUG, "Pointer to TX queue is NULL");
2360                 return;
2361         }
2362
2363         i40e_tx_queue_release_mbufs(q);
2364         rte_free(q->sw_ring);
2365         rte_free(q);
2366 }
2367
2368 const struct rte_memzone *
2369 i40e_memzone_reserve(const char *name, uint32_t len, int socket_id)
2370 {
2371         const struct rte_memzone *mz;
2372
2373         mz = rte_memzone_lookup(name);
2374         if (mz)
2375                 return mz;
2376
2377         mz = rte_memzone_reserve_aligned(name, len, socket_id,
2378                         RTE_MEMZONE_IOVA_CONTIG, I40E_RING_BASE_ALIGN);
2379         return mz;
2380 }
2381
2382 void
2383 i40e_rx_queue_release_mbufs(struct i40e_rx_queue *rxq)
2384 {
2385         uint16_t i;
2386
2387         /* SSE Vector driver has a different way of releasing mbufs. */
2388         if (rxq->rx_using_sse) {
2389                 i40e_rx_queue_release_mbufs_vec(rxq);
2390                 return;
2391         }
2392
2393         if (!rxq->sw_ring) {
2394                 PMD_DRV_LOG(DEBUG, "Pointer to sw_ring is NULL");
2395                 return;
2396         }
2397
2398         for (i = 0; i < rxq->nb_rx_desc; i++) {
2399                 if (rxq->sw_ring[i].mbuf) {
2400                         rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2401                         rxq->sw_ring[i].mbuf = NULL;
2402                 }
2403         }
2404 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2405         if (rxq->rx_nb_avail == 0)
2406                 return;
2407         for (i = 0; i < rxq->rx_nb_avail; i++) {
2408                 struct rte_mbuf *mbuf;
2409
2410                 mbuf = rxq->rx_stage[rxq->rx_next_avail + i];
2411                 rte_pktmbuf_free_seg(mbuf);
2412         }
2413         rxq->rx_nb_avail = 0;
2414 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2415 }
2416
2417 void
2418 i40e_reset_rx_queue(struct i40e_rx_queue *rxq)
2419 {
2420         unsigned i;
2421         uint16_t len;
2422
2423         if (!rxq) {
2424                 PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL");
2425                 return;
2426         }
2427
2428 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2429         if (check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
2430                 len = (uint16_t)(rxq->nb_rx_desc + RTE_PMD_I40E_RX_MAX_BURST);
2431         else
2432 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2433                 len = rxq->nb_rx_desc;
2434
2435         for (i = 0; i < len * sizeof(union i40e_rx_desc); i++)
2436                 ((volatile char *)rxq->rx_ring)[i] = 0;
2437
2438         memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2439         for (i = 0; i < RTE_PMD_I40E_RX_MAX_BURST; ++i)
2440                 rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
2441
2442 #ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
2443         rxq->rx_nb_avail = 0;
2444         rxq->rx_next_avail = 0;
2445         rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2446 #endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
2447         rxq->rx_tail = 0;
2448         rxq->nb_rx_hold = 0;
2449         rxq->pkt_first_seg = NULL;
2450         rxq->pkt_last_seg = NULL;
2451
2452         rxq->rxrearm_start = 0;
2453         rxq->rxrearm_nb = 0;
2454 }
2455
2456 void
2457 i40e_tx_queue_release_mbufs(struct i40e_tx_queue *txq)
2458 {
2459         struct rte_eth_dev *dev;
2460         uint16_t i;
2461
2462         dev = &rte_eth_devices[txq->port_id];
2463
2464         if (!txq || !txq->sw_ring) {
2465                 PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL");
2466                 return;
2467         }
2468
2469         /**
2470          *  vPMD tx will not set sw_ring's mbuf to NULL after free,
2471          *  so need to free remains more carefully.
2472          */
2473         if (dev->tx_pkt_burst == i40e_xmit_pkts_vec_avx2 ||
2474                         dev->tx_pkt_burst == i40e_xmit_pkts_vec) {
2475                 i = txq->tx_next_dd - txq->tx_rs_thresh + 1;
2476                 if (txq->tx_tail < i) {
2477                         for (; i < txq->nb_tx_desc; i++) {
2478                                 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2479                                 txq->sw_ring[i].mbuf = NULL;
2480                         }
2481                         i = 0;
2482                 }
2483                 for (; i < txq->tx_tail; i++) {
2484                         rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2485                         txq->sw_ring[i].mbuf = NULL;
2486                 }
2487         } else {
2488                 for (i = 0; i < txq->nb_tx_desc; i++) {
2489                         if (txq->sw_ring[i].mbuf) {
2490                                 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2491                                 txq->sw_ring[i].mbuf = NULL;
2492                         }
2493                 }
2494         }
2495 }
2496
2497 void
2498 i40e_reset_tx_queue(struct i40e_tx_queue *txq)
2499 {
2500         struct i40e_tx_entry *txe;
2501         uint16_t i, prev, size;
2502
2503         if (!txq) {
2504                 PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
2505                 return;
2506         }
2507
2508         txe = txq->sw_ring;
2509         size = sizeof(struct i40e_tx_desc) * txq->nb_tx_desc;
2510         for (i = 0; i < size; i++)
2511                 ((volatile char *)txq->tx_ring)[i] = 0;
2512
2513         prev = (uint16_t)(txq->nb_tx_desc - 1);
2514         for (i = 0; i < txq->nb_tx_desc; i++) {
2515                 volatile struct i40e_tx_desc *txd = &txq->tx_ring[i];
2516
2517                 txd->cmd_type_offset_bsz =
2518                         rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE);
2519                 txe[i].mbuf =  NULL;
2520                 txe[i].last_id = i;
2521                 txe[prev].next_id = i;
2522                 prev = i;
2523         }
2524
2525         txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2526         txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2527
2528         txq->tx_tail = 0;
2529         txq->nb_tx_used = 0;
2530
2531         txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2532         txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2533 }
2534
2535 /* Init the TX queue in hardware */
2536 int
2537 i40e_tx_queue_init(struct i40e_tx_queue *txq)
2538 {
2539         enum i40e_status_code err = I40E_SUCCESS;
2540         struct i40e_vsi *vsi = txq->vsi;
2541         struct i40e_hw *hw = I40E_VSI_TO_HW(vsi);
2542         uint16_t pf_q = txq->reg_idx;
2543         struct i40e_hmc_obj_txq tx_ctx;
2544         uint32_t qtx_ctl;
2545
2546         /* clear the context structure first */
2547         memset(&tx_ctx, 0, sizeof(tx_ctx));
2548         tx_ctx.new_context = 1;
2549         tx_ctx.base = txq->tx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2550         tx_ctx.qlen = txq->nb_tx_desc;
2551
2552 #ifdef RTE_LIBRTE_IEEE1588
2553         tx_ctx.timesync_ena = 1;
2554 #endif
2555         tx_ctx.rdylist = rte_le_to_cpu_16(vsi->info.qs_handle[txq->dcb_tc]);
2556         if (vsi->type == I40E_VSI_FDIR)
2557                 tx_ctx.fd_ena = TRUE;
2558
2559         err = i40e_clear_lan_tx_queue_context(hw, pf_q);
2560         if (err != I40E_SUCCESS) {
2561                 PMD_DRV_LOG(ERR, "Failure of clean lan tx queue context");
2562                 return err;
2563         }
2564
2565         err = i40e_set_lan_tx_queue_context(hw, pf_q, &tx_ctx);
2566         if (err != I40E_SUCCESS) {
2567                 PMD_DRV_LOG(ERR, "Failure of set lan tx queue context");
2568                 return err;
2569         }
2570
2571         /* Now associate this queue with this PCI function */
2572         qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
2573         qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
2574                                         I40E_QTX_CTL_PF_INDX_MASK);
2575         I40E_WRITE_REG(hw, I40E_QTX_CTL(pf_q), qtx_ctl);
2576         I40E_WRITE_FLUSH(hw);
2577
2578         txq->qtx_tail = hw->hw_addr + I40E_QTX_TAIL(pf_q);
2579
2580         return err;
2581 }
2582
2583 int
2584 i40e_alloc_rx_queue_mbufs(struct i40e_rx_queue *rxq)
2585 {
2586         struct i40e_rx_entry *rxe = rxq->sw_ring;
2587         uint64_t dma_addr;
2588         uint16_t i;
2589
2590         for (i = 0; i < rxq->nb_rx_desc; i++) {
2591                 volatile union i40e_rx_desc *rxd;
2592                 struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mp);
2593
2594                 if (unlikely(!mbuf)) {
2595                         PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
2596                         return -ENOMEM;
2597                 }
2598
2599                 rte_mbuf_refcnt_set(mbuf, 1);
2600                 mbuf->next = NULL;
2601                 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
2602                 mbuf->nb_segs = 1;
2603                 mbuf->port = rxq->port_id;
2604
2605                 dma_addr =
2606                         rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
2607
2608                 rxd = &rxq->rx_ring[i];
2609                 rxd->read.pkt_addr = dma_addr;
2610                 rxd->read.hdr_addr = 0;
2611 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2612                 rxd->read.rsvd1 = 0;
2613                 rxd->read.rsvd2 = 0;
2614 #endif /* RTE_LIBRTE_I40E_16BYTE_RX_DESC */
2615
2616                 rxe[i].mbuf = mbuf;
2617         }
2618
2619         return 0;
2620 }
2621
2622 /*
2623  * Calculate the buffer length, and check the jumbo frame
2624  * and maximum packet length.
2625  */
2626 static int
2627 i40e_rx_queue_config(struct i40e_rx_queue *rxq)
2628 {
2629         struct i40e_pf *pf = I40E_VSI_TO_PF(rxq->vsi);
2630         struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2631         struct rte_eth_dev_data *data = pf->dev_data;
2632         uint16_t buf_size, len;
2633
2634         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
2635                 RTE_PKTMBUF_HEADROOM);
2636
2637         switch (pf->flags & (I40E_FLAG_HEADER_SPLIT_DISABLED |
2638                         I40E_FLAG_HEADER_SPLIT_ENABLED)) {
2639         case I40E_FLAG_HEADER_SPLIT_ENABLED: /* Not supported */
2640                 rxq->rx_hdr_len = RTE_ALIGN(I40E_RXBUF_SZ_1024,
2641                                 (1 << I40E_RXQ_CTX_HBUFF_SHIFT));
2642                 rxq->rx_buf_len = RTE_ALIGN(I40E_RXBUF_SZ_2048,
2643                                 (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2644                 rxq->hs_mode = i40e_header_split_enabled;
2645                 break;
2646         case I40E_FLAG_HEADER_SPLIT_DISABLED:
2647         default:
2648                 rxq->rx_hdr_len = 0;
2649                 rxq->rx_buf_len = RTE_ALIGN_FLOOR(buf_size,
2650                         (1 << I40E_RXQ_CTX_DBUFF_SHIFT));
2651                 rxq->hs_mode = i40e_header_split_none;
2652                 break;
2653         }
2654
2655         len = hw->func_caps.rx_buf_chain_len * rxq->rx_buf_len;
2656         rxq->max_pkt_len = RTE_MIN(len, data->dev_conf.rxmode.max_rx_pkt_len);
2657         if (data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
2658                 if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
2659                         rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
2660                         PMD_DRV_LOG(ERR, "maximum packet length must "
2661                                     "be larger than %u and smaller than %u,"
2662                                     "as jumbo frame is enabled",
2663                                     (uint32_t)ETHER_MAX_LEN,
2664                                     (uint32_t)I40E_FRAME_SIZE_MAX);
2665                         return I40E_ERR_CONFIG;
2666                 }
2667         } else {
2668                 if (rxq->max_pkt_len < ETHER_MIN_LEN ||
2669                         rxq->max_pkt_len > ETHER_MAX_LEN) {
2670                         PMD_DRV_LOG(ERR, "maximum packet length must be "
2671                                     "larger than %u and smaller than %u, "
2672                                     "as jumbo frame is disabled",
2673                                     (uint32_t)ETHER_MIN_LEN,
2674                                     (uint32_t)ETHER_MAX_LEN);
2675                         return I40E_ERR_CONFIG;
2676                 }
2677         }
2678
2679         return 0;
2680 }
2681
2682 /* Init the RX queue in hardware */
2683 int
2684 i40e_rx_queue_init(struct i40e_rx_queue *rxq)
2685 {
2686         int err = I40E_SUCCESS;
2687         struct i40e_hw *hw = I40E_VSI_TO_HW(rxq->vsi);
2688         struct rte_eth_dev_data *dev_data = I40E_VSI_TO_DEV_DATA(rxq->vsi);
2689         uint16_t pf_q = rxq->reg_idx;
2690         uint16_t buf_size;
2691         struct i40e_hmc_obj_rxq rx_ctx;
2692
2693         err = i40e_rx_queue_config(rxq);
2694         if (err < 0) {
2695                 PMD_DRV_LOG(ERR, "Failed to config RX queue");
2696                 return err;
2697         }
2698
2699         /* Clear the context structure first */
2700         memset(&rx_ctx, 0, sizeof(struct i40e_hmc_obj_rxq));
2701         rx_ctx.dbuff = rxq->rx_buf_len >> I40E_RXQ_CTX_DBUFF_SHIFT;
2702         rx_ctx.hbuff = rxq->rx_hdr_len >> I40E_RXQ_CTX_HBUFF_SHIFT;
2703
2704         rx_ctx.base = rxq->rx_ring_phys_addr / I40E_QUEUE_BASE_ADDR_UNIT;
2705         rx_ctx.qlen = rxq->nb_rx_desc;
2706 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
2707         rx_ctx.dsize = 1;
2708 #endif
2709         rx_ctx.dtype = rxq->hs_mode;
2710         if (rxq->hs_mode)
2711                 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_ALL;
2712         else
2713                 rx_ctx.hsplit_0 = I40E_HEADER_SPLIT_NONE;
2714         rx_ctx.rxmax = rxq->max_pkt_len;
2715         rx_ctx.tphrdesc_ena = 1;
2716         rx_ctx.tphwdesc_ena = 1;
2717         rx_ctx.tphdata_ena = 1;
2718         rx_ctx.tphhead_ena = 1;
2719         rx_ctx.lrxqthresh = 2;
2720         rx_ctx.crcstrip = (rxq->crc_len == 0) ? 1 : 0;
2721         rx_ctx.l2tsel = 1;
2722         /* showiv indicates if inner VLAN is stripped inside of tunnel
2723          * packet. When set it to 1, vlan information is stripped from
2724          * the inner header, but the hardware does not put it in the
2725          * descriptor. So set it zero by default.
2726          */
2727         rx_ctx.showiv = 0;
2728         rx_ctx.prefena = 1;
2729
2730         err = i40e_clear_lan_rx_queue_context(hw, pf_q);
2731         if (err != I40E_SUCCESS) {
2732                 PMD_DRV_LOG(ERR, "Failed to clear LAN RX queue context");
2733                 return err;
2734         }
2735         err = i40e_set_lan_rx_queue_context(hw, pf_q, &rx_ctx);
2736         if (err != I40E_SUCCESS) {
2737                 PMD_DRV_LOG(ERR, "Failed to set LAN RX queue context");
2738                 return err;
2739         }
2740
2741         rxq->qrx_tail = hw->hw_addr + I40E_QRX_TAIL(pf_q);
2742
2743         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
2744                 RTE_PKTMBUF_HEADROOM);
2745
2746         /* Check if scattered RX needs to be used. */
2747         if ((rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
2748                 dev_data->scattered_rx = 1;
2749         }
2750
2751         /* Init the RX tail regieter. */
2752         I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
2753
2754         return 0;
2755 }
2756
2757 void
2758 i40e_dev_clear_queues(struct rte_eth_dev *dev)
2759 {
2760         uint16_t i;
2761
2762         PMD_INIT_FUNC_TRACE();
2763
2764         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2765                 if (!dev->data->tx_queues[i])
2766                         continue;
2767                 i40e_tx_queue_release_mbufs(dev->data->tx_queues[i]);
2768                 i40e_reset_tx_queue(dev->data->tx_queues[i]);
2769         }
2770
2771         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2772                 if (!dev->data->rx_queues[i])
2773                         continue;
2774                 i40e_rx_queue_release_mbufs(dev->data->rx_queues[i]);
2775                 i40e_reset_rx_queue(dev->data->rx_queues[i]);
2776         }
2777 }
2778
2779 void
2780 i40e_dev_free_queues(struct rte_eth_dev *dev)
2781 {
2782         uint16_t i;
2783
2784         PMD_INIT_FUNC_TRACE();
2785
2786         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2787                 if (!dev->data->rx_queues[i])
2788                         continue;
2789                 i40e_dev_rx_queue_release(dev->data->rx_queues[i]);
2790                 dev->data->rx_queues[i] = NULL;
2791         }
2792         dev->data->nb_rx_queues = 0;
2793
2794         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2795                 if (!dev->data->tx_queues[i])
2796                         continue;
2797                 i40e_dev_tx_queue_release(dev->data->tx_queues[i]);
2798                 dev->data->tx_queues[i] = NULL;
2799         }
2800         dev->data->nb_tx_queues = 0;
2801 }
2802
2803 #define I40E_FDIR_NUM_TX_DESC  I40E_MIN_RING_DESC
2804 #define I40E_FDIR_NUM_RX_DESC  I40E_MIN_RING_DESC
2805
2806 enum i40e_status_code
2807 i40e_fdir_setup_tx_resources(struct i40e_pf *pf)
2808 {
2809         struct i40e_tx_queue *txq;
2810         const struct rte_memzone *tz = NULL;
2811         uint32_t ring_size;
2812         struct rte_eth_dev *dev;
2813
2814         if (!pf) {
2815                 PMD_DRV_LOG(ERR, "PF is not available");
2816                 return I40E_ERR_BAD_PTR;
2817         }
2818
2819         dev = pf->adapter->eth_dev;
2820
2821         /* Allocate the TX queue data structure. */
2822         txq = rte_zmalloc_socket("i40e fdir tx queue",
2823                                   sizeof(struct i40e_tx_queue),
2824                                   RTE_CACHE_LINE_SIZE,
2825                                   SOCKET_ID_ANY);
2826         if (!txq) {
2827                 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2828                                         "tx queue structure.");
2829                 return I40E_ERR_NO_MEMORY;
2830         }
2831
2832         /* Allocate TX hardware ring descriptors. */
2833         ring_size = sizeof(struct i40e_tx_desc) * I40E_FDIR_NUM_TX_DESC;
2834         ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2835
2836         tz = rte_eth_dma_zone_reserve(dev, "fdir_tx_ring",
2837                                       I40E_FDIR_QUEUE_ID, ring_size,
2838                                       I40E_RING_BASE_ALIGN, SOCKET_ID_ANY);
2839         if (!tz) {
2840                 i40e_dev_tx_queue_release(txq);
2841                 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for TX.");
2842                 return I40E_ERR_NO_MEMORY;
2843         }
2844
2845         txq->nb_tx_desc = I40E_FDIR_NUM_TX_DESC;
2846         txq->queue_id = I40E_FDIR_QUEUE_ID;
2847         txq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2848         txq->vsi = pf->fdir.fdir_vsi;
2849
2850         txq->tx_ring_phys_addr = tz->iova;
2851         txq->tx_ring = (struct i40e_tx_desc *)tz->addr;
2852         /*
2853          * don't need to allocate software ring and reset for the fdir
2854          * program queue just set the queue has been configured.
2855          */
2856         txq->q_set = TRUE;
2857         pf->fdir.txq = txq;
2858
2859         return I40E_SUCCESS;
2860 }
2861
2862 enum i40e_status_code
2863 i40e_fdir_setup_rx_resources(struct i40e_pf *pf)
2864 {
2865         struct i40e_rx_queue *rxq;
2866         const struct rte_memzone *rz = NULL;
2867         uint32_t ring_size;
2868         struct rte_eth_dev *dev;
2869
2870         if (!pf) {
2871                 PMD_DRV_LOG(ERR, "PF is not available");
2872                 return I40E_ERR_BAD_PTR;
2873         }
2874
2875         dev = pf->adapter->eth_dev;
2876
2877         /* Allocate the RX queue data structure. */
2878         rxq = rte_zmalloc_socket("i40e fdir rx queue",
2879                                   sizeof(struct i40e_rx_queue),
2880                                   RTE_CACHE_LINE_SIZE,
2881                                   SOCKET_ID_ANY);
2882         if (!rxq) {
2883                 PMD_DRV_LOG(ERR, "Failed to allocate memory for "
2884                                         "rx queue structure.");
2885                 return I40E_ERR_NO_MEMORY;
2886         }
2887
2888         /* Allocate RX hardware ring descriptors. */
2889         ring_size = sizeof(union i40e_rx_desc) * I40E_FDIR_NUM_RX_DESC;
2890         ring_size = RTE_ALIGN(ring_size, I40E_DMA_MEM_ALIGN);
2891
2892         rz = rte_eth_dma_zone_reserve(dev, "fdir_rx_ring",
2893                                       I40E_FDIR_QUEUE_ID, ring_size,
2894                                       I40E_RING_BASE_ALIGN, SOCKET_ID_ANY);
2895         if (!rz) {
2896                 i40e_dev_rx_queue_release(rxq);
2897                 PMD_DRV_LOG(ERR, "Failed to reserve DMA memory for RX.");
2898                 return I40E_ERR_NO_MEMORY;
2899         }
2900
2901         rxq->nb_rx_desc = I40E_FDIR_NUM_RX_DESC;
2902         rxq->queue_id = I40E_FDIR_QUEUE_ID;
2903         rxq->reg_idx = pf->fdir.fdir_vsi->base_queue;
2904         rxq->vsi = pf->fdir.fdir_vsi;
2905
2906         rxq->rx_ring_phys_addr = rz->iova;
2907         memset(rz->addr, 0, I40E_FDIR_NUM_RX_DESC * sizeof(union i40e_rx_desc));
2908         rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
2909
2910         /*
2911          * Don't need to allocate software ring and reset for the fdir
2912          * rx queue, just set the queue has been configured.
2913          */
2914         rxq->q_set = TRUE;
2915         pf->fdir.rxq = rxq;
2916
2917         return I40E_SUCCESS;
2918 }
2919
2920 void
2921 i40e_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
2922         struct rte_eth_rxq_info *qinfo)
2923 {
2924         struct i40e_rx_queue *rxq;
2925
2926         rxq = dev->data->rx_queues[queue_id];
2927
2928         qinfo->mp = rxq->mp;
2929         qinfo->scattered_rx = dev->data->scattered_rx;
2930         qinfo->nb_desc = rxq->nb_rx_desc;
2931
2932         qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
2933         qinfo->conf.rx_drop_en = rxq->drop_en;
2934         qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
2935         qinfo->conf.offloads = rxq->offloads;
2936 }
2937
2938 void
2939 i40e_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
2940         struct rte_eth_txq_info *qinfo)
2941 {
2942         struct i40e_tx_queue *txq;
2943
2944         txq = dev->data->tx_queues[queue_id];
2945
2946         qinfo->nb_desc = txq->nb_tx_desc;
2947
2948         qinfo->conf.tx_thresh.pthresh = txq->pthresh;
2949         qinfo->conf.tx_thresh.hthresh = txq->hthresh;
2950         qinfo->conf.tx_thresh.wthresh = txq->wthresh;
2951
2952         qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
2953         qinfo->conf.tx_rs_thresh = txq->tx_rs_thresh;
2954         qinfo->conf.txq_flags = txq->txq_flags;
2955         qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
2956 }
2957
2958 void __attribute__((cold))
2959 i40e_set_rx_function(struct rte_eth_dev *dev)
2960 {
2961         struct i40e_adapter *ad =
2962                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
2963         uint16_t rx_using_sse, i;
2964         /* In order to allow Vector Rx there are a few configuration
2965          * conditions to be met and Rx Bulk Allocation should be allowed.
2966          */
2967         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
2968                 if (i40e_rx_vec_dev_conf_condition_check(dev) ||
2969                     !ad->rx_bulk_alloc_allowed) {
2970                         PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet"
2971                                      " Vector Rx preconditions",
2972                                      dev->data->port_id);
2973
2974                         ad->rx_vec_allowed = false;
2975                 }
2976                 if (ad->rx_vec_allowed) {
2977                         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2978                                 struct i40e_rx_queue *rxq =
2979                                         dev->data->rx_queues[i];
2980
2981                                 if (rxq && i40e_rxq_vec_setup(rxq)) {
2982                                         ad->rx_vec_allowed = false;
2983                                         break;
2984                                 }
2985                         }
2986                 }
2987         }
2988
2989         if (dev->data->scattered_rx) {
2990                 /* Set the non-LRO scattered callback: there are Vector and
2991                  * single allocation versions.
2992                  */
2993                 if (ad->rx_vec_allowed) {
2994                         PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
2995                                             "callback (port=%d).",
2996                                      dev->data->port_id);
2997
2998                         dev->rx_pkt_burst = i40e_recv_scattered_pkts_vec;
2999 #ifdef RTE_ARCH_X86
3000                         /*
3001                          * since AVX frequency can be different to base
3002                          * frequency, limit use of AVX2 version to later
3003                          * plaforms, not all those that could theoretically
3004                          * run it.
3005                          */
3006                         if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
3007                                 dev->rx_pkt_burst =
3008                                         i40e_recv_scattered_pkts_vec_avx2;
3009 #endif
3010                 } else {
3011                         PMD_INIT_LOG(DEBUG, "Using a Scattered with bulk "
3012                                            "allocation callback (port=%d).",
3013                                      dev->data->port_id);
3014                         dev->rx_pkt_burst = i40e_recv_scattered_pkts;
3015                 }
3016         /* If parameters allow we are going to choose between the following
3017          * callbacks:
3018          *    - Vector
3019          *    - Bulk Allocation
3020          *    - Single buffer allocation (the simplest one)
3021          */
3022         } else if (ad->rx_vec_allowed) {
3023                 PMD_INIT_LOG(DEBUG, "Vector rx enabled, please make sure RX "
3024                                     "burst size no less than %d (port=%d).",
3025                              RTE_I40E_DESCS_PER_LOOP,
3026                              dev->data->port_id);
3027
3028                 dev->rx_pkt_burst = i40e_recv_pkts_vec;
3029 #ifdef RTE_ARCH_X86
3030                 /*
3031                  * since AVX frequency can be different to base
3032                  * frequency, limit use of AVX2 version to later
3033                  * plaforms, not all those that could theoretically
3034                  * run it.
3035                  */
3036                 if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
3037                         dev->rx_pkt_burst = i40e_recv_pkts_vec_avx2;
3038 #endif
3039         } else if (ad->rx_bulk_alloc_allowed) {
3040                 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
3041                                     "satisfied. Rx Burst Bulk Alloc function "
3042                                     "will be used on port=%d.",
3043                              dev->data->port_id);
3044
3045                 dev->rx_pkt_burst = i40e_recv_pkts_bulk_alloc;
3046         } else {
3047                 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
3048                                     "satisfied, or Scattered Rx is requested "
3049                                     "(port=%d).",
3050                              dev->data->port_id);
3051
3052                 dev->rx_pkt_burst = i40e_recv_pkts;
3053         }
3054
3055         /* Propagate information about RX function choice through all queues. */
3056         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
3057                 rx_using_sse =
3058                         (dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec ||
3059                          dev->rx_pkt_burst == i40e_recv_pkts_vec ||
3060                          dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec_avx2 ||
3061                          dev->rx_pkt_burst == i40e_recv_pkts_vec_avx2);
3062
3063                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3064                         struct i40e_rx_queue *rxq = dev->data->rx_queues[i];
3065
3066                         if (rxq)
3067                                 rxq->rx_using_sse = rx_using_sse;
3068                 }
3069         }
3070 }
3071
3072 void __attribute__((cold))
3073 i40e_set_tx_function_flag(struct rte_eth_dev *dev, struct i40e_tx_queue *txq)
3074 {
3075         struct i40e_adapter *ad =
3076                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
3077
3078         /* Use a simple Tx queue (no offloads, no multi segs) if possible */
3079         ad->tx_simple_allowed = (txq->offloads == 0 &&
3080                         txq->tx_rs_thresh >= RTE_PMD_I40E_TX_MAX_BURST);
3081         ad->tx_vec_allowed = (ad->tx_simple_allowed &&
3082                         txq->tx_rs_thresh <= RTE_I40E_TX_MAX_FREE_BUF_SZ);
3083
3084         if (ad->tx_vec_allowed)
3085                 PMD_INIT_LOG(DEBUG, "Vector Tx can be enabled on Tx queue %u.",
3086                                 txq->queue_id);
3087         else if (ad->tx_simple_allowed)
3088                 PMD_INIT_LOG(DEBUG, "Simple Tx can be enabled on Tx queue %u.",
3089                                 txq->queue_id);
3090         else
3091                 PMD_INIT_LOG(DEBUG,
3092                                 "Neither simple nor vector Tx enabled on Tx queue %u\n",
3093                                 txq->queue_id);
3094 }
3095
3096 void __attribute__((cold))
3097 i40e_set_tx_function(struct rte_eth_dev *dev)
3098 {
3099         struct i40e_adapter *ad =
3100                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
3101         int i;
3102
3103         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
3104                 if (ad->tx_vec_allowed) {
3105                         for (i = 0; i < dev->data->nb_tx_queues; i++) {
3106                                 struct i40e_tx_queue *txq =
3107                                         dev->data->tx_queues[i];
3108
3109                                 if (txq && i40e_txq_vec_setup(txq)) {
3110                                         ad->tx_vec_allowed = false;
3111                                         break;
3112                                 }
3113                         }
3114                 }
3115         }
3116
3117         if (ad->tx_simple_allowed) {
3118                 if (ad->tx_vec_allowed) {
3119                         PMD_INIT_LOG(DEBUG, "Vector tx finally be used.");
3120                         dev->tx_pkt_burst = i40e_xmit_pkts_vec;
3121 #ifdef RTE_ARCH_X86
3122                         /*
3123                          * since AVX frequency can be different to base
3124                          * frequency, limit use of AVX2 version to later
3125                          * plaforms, not all those that could theoretically
3126                          * run it.
3127                          */
3128                         if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
3129                                 dev->tx_pkt_burst = i40e_xmit_pkts_vec_avx2;
3130 #endif
3131                 } else {
3132                         PMD_INIT_LOG(DEBUG, "Simple tx finally be used.");
3133                         dev->tx_pkt_burst = i40e_xmit_pkts_simple;
3134                 }
3135                 dev->tx_pkt_prepare = NULL;
3136         } else {
3137                 PMD_INIT_LOG(DEBUG, "Xmit tx finally be used.");
3138                 dev->tx_pkt_burst = i40e_xmit_pkts;
3139                 dev->tx_pkt_prepare = i40e_prep_pkts;
3140         }
3141 }
3142
3143 void __attribute__((cold))
3144 i40e_set_default_ptype_table(struct rte_eth_dev *dev)
3145 {
3146         struct i40e_adapter *ad =
3147                 I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
3148         int i;
3149
3150         for (i = 0; i < I40E_MAX_PKT_TYPE; i++)
3151                 ad->ptype_tbl[i] = i40e_get_default_pkt_type(i);
3152 }
3153
3154 void __attribute__((cold))
3155 i40e_set_default_pctype_table(struct rte_eth_dev *dev)
3156 {
3157         struct i40e_adapter *ad =
3158                         I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
3159         struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3160         int i;
3161
3162         for (i = 0; i < I40E_FLOW_TYPE_MAX; i++)
3163                 ad->pctypes_tbl[i] = 0ULL;
3164         ad->flow_types_mask = 0ULL;
3165         ad->pctypes_mask = 0ULL;
3166
3167         ad->pctypes_tbl[RTE_ETH_FLOW_FRAG_IPV4] =
3168                                 (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV4);
3169         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_UDP] =
3170                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_UDP);
3171         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_TCP] =
3172                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_TCP);
3173         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_SCTP] =
3174                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_SCTP);
3175         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_OTHER] =
3176                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_OTHER);
3177         ad->pctypes_tbl[RTE_ETH_FLOW_FRAG_IPV6] =
3178                                 (1ULL << I40E_FILTER_PCTYPE_FRAG_IPV6);
3179         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_UDP] =
3180                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_UDP);
3181         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_TCP] =
3182                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_TCP);
3183         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_SCTP] =
3184                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_SCTP);
3185         ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_OTHER] =
3186                                 (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_OTHER);
3187         ad->pctypes_tbl[RTE_ETH_FLOW_L2_PAYLOAD] =
3188                                 (1ULL << I40E_FILTER_PCTYPE_L2_PAYLOAD);
3189
3190         if (hw->mac.type == I40E_MAC_X722) {
3191                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_UDP] |=
3192                         (1ULL << I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP);
3193                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_UDP] |=
3194                         (1ULL << I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP);
3195                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV4_TCP] |=
3196                         (1ULL << I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK);
3197                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_UDP] |=
3198                         (1ULL << I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP);
3199                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_UDP] |=
3200                         (1ULL << I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP);
3201                 ad->pctypes_tbl[RTE_ETH_FLOW_NONFRAG_IPV6_TCP] |=
3202                         (1ULL << I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK);
3203         }
3204
3205         for (i = 0; i < I40E_FLOW_TYPE_MAX; i++) {
3206                 if (ad->pctypes_tbl[i])
3207                         ad->flow_types_mask |= (1ULL << i);
3208                 ad->pctypes_mask |= ad->pctypes_tbl[i];
3209         }
3210 }
3211
3212 /* Stubs needed for linkage when CONFIG_RTE_I40E_INC_VECTOR is set to 'n' */
3213 int __attribute__((weak))
3214 i40e_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
3215 {
3216         return -1;
3217 }
3218
3219 uint16_t __attribute__((weak))
3220 i40e_recv_pkts_vec(
3221         void __rte_unused *rx_queue,
3222         struct rte_mbuf __rte_unused **rx_pkts,
3223         uint16_t __rte_unused nb_pkts)
3224 {
3225         return 0;
3226 }
3227
3228 uint16_t __attribute__((weak))
3229 i40e_recv_scattered_pkts_vec(
3230         void __rte_unused *rx_queue,
3231         struct rte_mbuf __rte_unused **rx_pkts,
3232         uint16_t __rte_unused nb_pkts)
3233 {
3234         return 0;
3235 }
3236
3237 uint16_t __attribute__((weak))
3238 i40e_recv_pkts_vec_avx2(void __rte_unused *rx_queue,
3239                         struct rte_mbuf __rte_unused **rx_pkts,
3240                         uint16_t __rte_unused nb_pkts)
3241 {
3242         return 0;
3243 }
3244
3245 uint16_t __attribute__((weak))
3246 i40e_recv_scattered_pkts_vec_avx2(void __rte_unused *rx_queue,
3247                         struct rte_mbuf __rte_unused **rx_pkts,
3248                         uint16_t __rte_unused nb_pkts)
3249 {
3250         return 0;
3251 }
3252
3253 int __attribute__((weak))
3254 i40e_rxq_vec_setup(struct i40e_rx_queue __rte_unused *rxq)
3255 {
3256         return -1;
3257 }
3258
3259 int __attribute__((weak))
3260 i40e_txq_vec_setup(struct i40e_tx_queue __rte_unused *txq)
3261 {
3262         return -1;
3263 }
3264
3265 void __attribute__((weak))
3266 i40e_rx_queue_release_mbufs_vec(struct i40e_rx_queue __rte_unused*rxq)
3267 {
3268         return;
3269 }
3270
3271 uint16_t __attribute__((weak))
3272 i40e_xmit_fixed_burst_vec(void __rte_unused * tx_queue,
3273                           struct rte_mbuf __rte_unused **tx_pkts,
3274                           uint16_t __rte_unused nb_pkts)
3275 {
3276         return 0;
3277 }
3278
3279 uint16_t __attribute__((weak))
3280 i40e_xmit_pkts_vec_avx2(void __rte_unused * tx_queue,
3281                           struct rte_mbuf __rte_unused **tx_pkts,
3282                           uint16_t __rte_unused nb_pkts)
3283 {
3284         return 0;
3285 }