2651c412888940189a050061087d9be0565cf14a
[dpdk.git] / drivers / net / sfc / sfc_ethdev.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright(c) 2019-2021 Xilinx, Inc.
4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9
10 #include <rte_dev.h>
11 #include <ethdev_driver.h>
12 #include <ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
17 #include <rte_ether.h>
18
19 #include "efx.h"
20
21 #include "sfc.h"
22 #include "sfc_debug.h"
23 #include "sfc_log.h"
24 #include "sfc_kvargs.h"
25 #include "sfc_ev.h"
26 #include "sfc_rx.h"
27 #include "sfc_tx.h"
28 #include "sfc_flow.h"
29 #include "sfc_dp.h"
30 #include "sfc_dp_rx.h"
31
32 uint32_t sfc_logtype_driver;
33
34 static struct sfc_dp_list sfc_dp_head =
35         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
36
37
38 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
39
40
41 static int
42 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
43 {
44         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
45         efx_nic_fw_info_t enfi;
46         int ret;
47         int rc;
48
49         rc = efx_nic_get_fw_version(sa->nic, &enfi);
50         if (rc != 0)
51                 return -rc;
52
53         ret = snprintf(fw_version, fw_size,
54                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
55                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
56                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
57         if (ret < 0)
58                 return ret;
59
60         if (enfi.enfi_dpcpu_fw_ids_valid) {
61                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
62                 int ret_extra;
63
64                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
65                                      fw_size - dpcpu_fw_ids_offset,
66                                      " rx%" PRIx16 " tx%" PRIx16,
67                                      enfi.enfi_rx_dpcpu_fw_id,
68                                      enfi.enfi_tx_dpcpu_fw_id);
69                 if (ret_extra < 0)
70                         return ret_extra;
71
72                 ret += ret_extra;
73         }
74
75         if (fw_size < (size_t)(++ret))
76                 return ret;
77         else
78                 return 0;
79 }
80
81 static int
82 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
83 {
84         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
85         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
86         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
87         struct sfc_rss *rss = &sas->rss;
88         struct sfc_mae *mae = &sa->mae;
89         uint64_t txq_offloads_def = 0;
90
91         sfc_log_init(sa, "entry");
92
93         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
94         dev_info->max_mtu = EFX_MAC_SDU_MAX;
95
96         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
97
98         dev_info->max_vfs = sa->sriov.num_vfs;
99
100         /* Autonegotiation may be disabled */
101         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
102         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
103                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
104         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
105                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
106         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
107                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
108         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
109                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
110         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
111                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
112         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
113                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
114
115         dev_info->max_rx_queues = sa->rxq_max;
116         dev_info->max_tx_queues = sa->txq_max;
117
118         /* By default packets are dropped if no descriptors are available */
119         dev_info->default_rxconf.rx_drop_en = 1;
120
121         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
122
123         /*
124          * rx_offload_capa includes both device and queue offloads since
125          * the latter may be requested on a per device basis which makes
126          * sense when some offloads are needed to be set on all queues.
127          */
128         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
129                                     dev_info->rx_queue_offload_capa;
130
131         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
132
133         /*
134          * tx_offload_capa includes both device and queue offloads since
135          * the latter may be requested on a per device basis which makes
136          * sense when some offloads are needed to be set on all queues.
137          */
138         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
139                                     dev_info->tx_queue_offload_capa;
140
141         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
142                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
143
144         dev_info->default_txconf.offloads |= txq_offloads_def;
145
146         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
147                 uint64_t rte_hf = 0;
148                 unsigned int i;
149
150                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
151                         rte_hf |= rss->hf_map[i].rte;
152
153                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
154                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
155                 dev_info->flow_type_rss_offloads = rte_hf;
156         }
157
158         /* Initialize to hardware limits */
159         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
160         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
161         /* The RXQ hardware requires that the descriptor count is a power
162          * of 2, but rx_desc_lim cannot properly describe that constraint.
163          */
164         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
165
166         /* Initialize to hardware limits */
167         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
168         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
169         /*
170          * The TXQ hardware requires that the descriptor count is a power
171          * of 2, but tx_desc_lim cannot properly describe that constraint
172          */
173         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
174
175         if (sap->dp_rx->get_dev_info != NULL)
176                 sap->dp_rx->get_dev_info(dev_info);
177         if (sap->dp_tx->get_dev_info != NULL)
178                 sap->dp_tx->get_dev_info(dev_info);
179
180         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
181                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
182
183         if (mae->status == SFC_MAE_STATUS_SUPPORTED) {
184                 dev_info->switch_info.name = dev->device->driver->name;
185                 dev_info->switch_info.domain_id = mae->switch_domain_id;
186                 dev_info->switch_info.port_id = mae->switch_port_id;
187         }
188
189         return 0;
190 }
191
192 static const uint32_t *
193 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
194 {
195         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
196
197         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
198 }
199
200 static int
201 sfc_dev_configure(struct rte_eth_dev *dev)
202 {
203         struct rte_eth_dev_data *dev_data = dev->data;
204         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
205         int rc;
206
207         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
208                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
209
210         sfc_adapter_lock(sa);
211         switch (sa->state) {
212         case SFC_ADAPTER_CONFIGURED:
213                 /* FALLTHROUGH */
214         case SFC_ADAPTER_INITIALIZED:
215                 rc = sfc_configure(sa);
216                 break;
217         default:
218                 sfc_err(sa, "unexpected adapter state %u to configure",
219                         sa->state);
220                 rc = EINVAL;
221                 break;
222         }
223         sfc_adapter_unlock(sa);
224
225         sfc_log_init(sa, "done %d", rc);
226         SFC_ASSERT(rc >= 0);
227         return -rc;
228 }
229
230 static int
231 sfc_dev_start(struct rte_eth_dev *dev)
232 {
233         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
234         int rc;
235
236         sfc_log_init(sa, "entry");
237
238         sfc_adapter_lock(sa);
239         rc = sfc_start(sa);
240         sfc_adapter_unlock(sa);
241
242         sfc_log_init(sa, "done %d", rc);
243         SFC_ASSERT(rc >= 0);
244         return -rc;
245 }
246
247 static int
248 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
249 {
250         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
251         struct rte_eth_link current_link;
252         int ret;
253
254         sfc_log_init(sa, "entry");
255
256         if (sa->state != SFC_ADAPTER_STARTED) {
257                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
258         } else if (wait_to_complete) {
259                 efx_link_mode_t link_mode;
260
261                 if (efx_port_poll(sa->nic, &link_mode) != 0)
262                         link_mode = EFX_LINK_UNKNOWN;
263                 sfc_port_link_mode_to_info(link_mode, &current_link);
264
265         } else {
266                 sfc_ev_mgmt_qpoll(sa);
267                 rte_eth_linkstatus_get(dev, &current_link);
268         }
269
270         ret = rte_eth_linkstatus_set(dev, &current_link);
271         if (ret == 0)
272                 sfc_notice(sa, "Link status is %s",
273                            current_link.link_status ? "UP" : "DOWN");
274
275         return ret;
276 }
277
278 static int
279 sfc_dev_stop(struct rte_eth_dev *dev)
280 {
281         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
282
283         sfc_log_init(sa, "entry");
284
285         sfc_adapter_lock(sa);
286         sfc_stop(sa);
287         sfc_adapter_unlock(sa);
288
289         sfc_log_init(sa, "done");
290
291         return 0;
292 }
293
294 static int
295 sfc_dev_set_link_up(struct rte_eth_dev *dev)
296 {
297         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
298         int rc;
299
300         sfc_log_init(sa, "entry");
301
302         sfc_adapter_lock(sa);
303         rc = sfc_start(sa);
304         sfc_adapter_unlock(sa);
305
306         SFC_ASSERT(rc >= 0);
307         return -rc;
308 }
309
310 static int
311 sfc_dev_set_link_down(struct rte_eth_dev *dev)
312 {
313         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
314
315         sfc_log_init(sa, "entry");
316
317         sfc_adapter_lock(sa);
318         sfc_stop(sa);
319         sfc_adapter_unlock(sa);
320
321         return 0;
322 }
323
324 static void
325 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
326 {
327         free(dev->process_private);
328         rte_eth_dev_release_port(dev);
329 }
330
331 static int
332 sfc_dev_close(struct rte_eth_dev *dev)
333 {
334         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
335
336         sfc_log_init(sa, "entry");
337
338         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
339                 sfc_eth_dev_secondary_clear_ops(dev);
340                 return 0;
341         }
342
343         sfc_adapter_lock(sa);
344         switch (sa->state) {
345         case SFC_ADAPTER_STARTED:
346                 sfc_stop(sa);
347                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
348                 /* FALLTHROUGH */
349         case SFC_ADAPTER_CONFIGURED:
350                 sfc_close(sa);
351                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
352                 /* FALLTHROUGH */
353         case SFC_ADAPTER_INITIALIZED:
354                 break;
355         default:
356                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
357                 break;
358         }
359
360         /*
361          * Cleanup all resources.
362          * Rollback primary process sfc_eth_dev_init() below.
363          */
364
365         sfc_eth_dev_clear_ops(dev);
366
367         sfc_detach(sa);
368         sfc_unprobe(sa);
369
370         sfc_kvargs_cleanup(sa);
371
372         sfc_adapter_unlock(sa);
373         sfc_adapter_lock_fini(sa);
374
375         sfc_log_init(sa, "done");
376
377         /* Required for logging, so cleanup last */
378         sa->eth_dev = NULL;
379
380         free(sa);
381
382         return 0;
383 }
384
385 static int
386 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
387                    boolean_t enabled)
388 {
389         struct sfc_port *port;
390         boolean_t *toggle;
391         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
392         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
393         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
394         int rc = 0;
395
396         sfc_adapter_lock(sa);
397
398         port = &sa->port;
399         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
400
401         if (*toggle != enabled) {
402                 *toggle = enabled;
403
404                 if (sfc_sa2shared(sa)->isolated) {
405                         sfc_warn(sa, "isolated mode is active on the port");
406                         sfc_warn(sa, "the change is to be applied on the next "
407                                      "start provided that isolated mode is "
408                                      "disabled prior the next start");
409                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
410                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
411                         *toggle = !(enabled);
412                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
413                                  ((enabled) ? "enable" : "disable"), desc, rc);
414
415                         /*
416                          * For promiscuous and all-multicast filters a
417                          * permission failure should be reported as an
418                          * unsupported filter.
419                          */
420                         if (rc == EPERM)
421                                 rc = ENOTSUP;
422                 }
423         }
424
425         sfc_adapter_unlock(sa);
426         return rc;
427 }
428
429 static int
430 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
431 {
432         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
433
434         SFC_ASSERT(rc >= 0);
435         return -rc;
436 }
437
438 static int
439 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
440 {
441         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
442
443         SFC_ASSERT(rc >= 0);
444         return -rc;
445 }
446
447 static int
448 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
449 {
450         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
451
452         SFC_ASSERT(rc >= 0);
453         return -rc;
454 }
455
456 static int
457 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
458 {
459         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
460
461         SFC_ASSERT(rc >= 0);
462         return -rc;
463 }
464
465 static int
466 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
467                    uint16_t nb_rx_desc, unsigned int socket_id,
468                    const struct rte_eth_rxconf *rx_conf,
469                    struct rte_mempool *mb_pool)
470 {
471         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
472         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
473         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
474         struct sfc_rxq_info *rxq_info;
475         sfc_sw_index_t sw_index;
476         int rc;
477
478         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
479                      ethdev_qid, nb_rx_desc, socket_id);
480
481         sfc_adapter_lock(sa);
482
483         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
484         rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
485                           rx_conf, mb_pool);
486         if (rc != 0)
487                 goto fail_rx_qinit;
488
489         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
490         dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
491
492         sfc_adapter_unlock(sa);
493
494         return 0;
495
496 fail_rx_qinit:
497         sfc_adapter_unlock(sa);
498         SFC_ASSERT(rc > 0);
499         return -rc;
500 }
501
502 static void
503 sfc_rx_queue_release(void *queue)
504 {
505         struct sfc_dp_rxq *dp_rxq = queue;
506         struct sfc_rxq *rxq;
507         struct sfc_adapter *sa;
508         sfc_sw_index_t sw_index;
509
510         if (dp_rxq == NULL)
511                 return;
512
513         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
514         sa = rxq->evq->sa;
515         sfc_adapter_lock(sa);
516
517         sw_index = dp_rxq->dpq.queue_id;
518
519         sfc_log_init(sa, "RxQ=%u", sw_index);
520
521         sfc_rx_qfini(sa, sw_index);
522
523         sfc_adapter_unlock(sa);
524 }
525
526 static int
527 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
528                    uint16_t nb_tx_desc, unsigned int socket_id,
529                    const struct rte_eth_txconf *tx_conf)
530 {
531         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
532         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
533         int rc;
534
535         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
536                      tx_queue_id, nb_tx_desc, socket_id);
537
538         sfc_adapter_lock(sa);
539
540         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
541         if (rc != 0)
542                 goto fail_tx_qinit;
543
544         dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
545
546         sfc_adapter_unlock(sa);
547         return 0;
548
549 fail_tx_qinit:
550         sfc_adapter_unlock(sa);
551         SFC_ASSERT(rc > 0);
552         return -rc;
553 }
554
555 static void
556 sfc_tx_queue_release(void *queue)
557 {
558         struct sfc_dp_txq *dp_txq = queue;
559         struct sfc_txq *txq;
560         unsigned int sw_index;
561         struct sfc_adapter *sa;
562
563         if (dp_txq == NULL)
564                 return;
565
566         txq = sfc_txq_by_dp_txq(dp_txq);
567         sw_index = dp_txq->dpq.queue_id;
568
569         SFC_ASSERT(txq->evq != NULL);
570         sa = txq->evq->sa;
571
572         sfc_log_init(sa, "TxQ = %u", sw_index);
573
574         sfc_adapter_lock(sa);
575
576         sfc_tx_qfini(sa, sw_index);
577
578         sfc_adapter_unlock(sa);
579 }
580
581 /*
582  * Some statistics are computed as A - B where A and B each increase
583  * monotonically with some hardware counter(s) and the counters are read
584  * asynchronously.
585  *
586  * If packet X is counted in A, but not counted in B yet, computed value is
587  * greater than real.
588  *
589  * If packet X is not counted in A at the moment of reading the counter,
590  * but counted in B at the moment of reading the counter, computed value
591  * is less than real.
592  *
593  * However, counter which grows backward is worse evil than slightly wrong
594  * value. So, let's try to guarantee that it never happens except may be
595  * the case when the MAC stats are zeroed as a result of a NIC reset.
596  */
597 static void
598 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
599 {
600         if ((int64_t)(newval - *stat) > 0 || newval == 0)
601                 *stat = newval;
602 }
603
604 static int
605 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
606 {
607         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
608         struct sfc_port *port = &sa->port;
609         uint64_t *mac_stats;
610         int ret;
611
612         rte_spinlock_lock(&port->mac_stats_lock);
613
614         ret = sfc_port_update_mac_stats(sa);
615         if (ret != 0)
616                 goto unlock;
617
618         mac_stats = port->mac_stats_buf;
619
620         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
621                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
622                 stats->ipackets =
623                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
624                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
625                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
626                 stats->opackets =
627                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
628                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
629                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
630                 stats->ibytes =
631                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
632                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
633                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
634                 stats->obytes =
635                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
636                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
637                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
638                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
639                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
640
641                 /* CRC is included in these stats, but shouldn't be */
642                 stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
643                 stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
644         } else {
645                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
646                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
647                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
648
649                 /* CRC is included in these stats, but shouldn't be */
650                 stats->ibytes -= mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
651                 stats->obytes -= mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
652
653                 /*
654                  * Take into account stats which are whenever supported
655                  * on EF10. If some stat is not supported by current
656                  * firmware variant or HW revision, it is guaranteed
657                  * to be zero in mac_stats.
658                  */
659                 stats->imissed =
660                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
661                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
662                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
663                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
664                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
665                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
666                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
667                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
668                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
669                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
670                 stats->ierrors =
671                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
672                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
673                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
674                 /* no oerrors counters supported on EF10 */
675
676                 /* Exclude missed, errors and pauses from Rx packets */
677                 sfc_update_diff_stat(&port->ipackets,
678                         mac_stats[EFX_MAC_RX_PKTS] -
679                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
680                         stats->imissed - stats->ierrors);
681                 stats->ipackets = port->ipackets;
682         }
683
684 unlock:
685         rte_spinlock_unlock(&port->mac_stats_lock);
686         SFC_ASSERT(ret >= 0);
687         return -ret;
688 }
689
690 static int
691 sfc_stats_reset(struct rte_eth_dev *dev)
692 {
693         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
694         struct sfc_port *port = &sa->port;
695         int rc;
696
697         if (sa->state != SFC_ADAPTER_STARTED) {
698                 /*
699                  * The operation cannot be done if port is not started; it
700                  * will be scheduled to be done during the next port start
701                  */
702                 port->mac_stats_reset_pending = B_TRUE;
703                 return 0;
704         }
705
706         rc = sfc_port_reset_mac_stats(sa);
707         if (rc != 0)
708                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
709
710         SFC_ASSERT(rc >= 0);
711         return -rc;
712 }
713
714 static int
715 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
716                unsigned int xstats_count)
717 {
718         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
719         struct sfc_port *port = &sa->port;
720         uint64_t *mac_stats;
721         int rc;
722         unsigned int i;
723         int nstats = 0;
724
725         rte_spinlock_lock(&port->mac_stats_lock);
726
727         rc = sfc_port_update_mac_stats(sa);
728         if (rc != 0) {
729                 SFC_ASSERT(rc > 0);
730                 nstats = -rc;
731                 goto unlock;
732         }
733
734         mac_stats = port->mac_stats_buf;
735
736         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
737                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
738                         if (xstats != NULL && nstats < (int)xstats_count) {
739                                 xstats[nstats].id = nstats;
740                                 xstats[nstats].value = mac_stats[i];
741                         }
742                         nstats++;
743                 }
744         }
745
746 unlock:
747         rte_spinlock_unlock(&port->mac_stats_lock);
748
749         return nstats;
750 }
751
752 static int
753 sfc_xstats_get_names(struct rte_eth_dev *dev,
754                      struct rte_eth_xstat_name *xstats_names,
755                      unsigned int xstats_count)
756 {
757         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
758         struct sfc_port *port = &sa->port;
759         unsigned int i;
760         unsigned int nstats = 0;
761
762         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
763                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
764                         if (xstats_names != NULL && nstats < xstats_count)
765                                 strlcpy(xstats_names[nstats].name,
766                                         efx_mac_stat_name(sa->nic, i),
767                                         sizeof(xstats_names[0].name));
768                         nstats++;
769                 }
770         }
771
772         return nstats;
773 }
774
775 static int
776 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
777                      uint64_t *values, unsigned int n)
778 {
779         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
780         struct sfc_port *port = &sa->port;
781         uint64_t *mac_stats;
782         unsigned int nb_supported = 0;
783         unsigned int nb_written = 0;
784         unsigned int i;
785         int ret;
786         int rc;
787
788         if (unlikely(values == NULL) ||
789             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
790                 return port->mac_stats_nb_supported;
791
792         rte_spinlock_lock(&port->mac_stats_lock);
793
794         rc = sfc_port_update_mac_stats(sa);
795         if (rc != 0) {
796                 SFC_ASSERT(rc > 0);
797                 ret = -rc;
798                 goto unlock;
799         }
800
801         mac_stats = port->mac_stats_buf;
802
803         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
804                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
805                         continue;
806
807                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
808                         values[nb_written++] = mac_stats[i];
809
810                 ++nb_supported;
811         }
812
813         ret = nb_written;
814
815 unlock:
816         rte_spinlock_unlock(&port->mac_stats_lock);
817
818         return ret;
819 }
820
821 static int
822 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
823                            struct rte_eth_xstat_name *xstats_names,
824                            const uint64_t *ids, unsigned int size)
825 {
826         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
827         struct sfc_port *port = &sa->port;
828         unsigned int nb_supported = 0;
829         unsigned int nb_written = 0;
830         unsigned int i;
831
832         if (unlikely(xstats_names == NULL) ||
833             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
834                 return port->mac_stats_nb_supported;
835
836         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
837                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
838                         continue;
839
840                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
841                         char *name = xstats_names[nb_written++].name;
842
843                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
844                                 sizeof(xstats_names[0].name));
845                 }
846
847                 ++nb_supported;
848         }
849
850         return nb_written;
851 }
852
853 static int
854 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
855 {
856         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
857         unsigned int wanted_fc, link_fc;
858
859         memset(fc_conf, 0, sizeof(*fc_conf));
860
861         sfc_adapter_lock(sa);
862
863         if (sa->state == SFC_ADAPTER_STARTED)
864                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
865         else
866                 link_fc = sa->port.flow_ctrl;
867
868         switch (link_fc) {
869         case 0:
870                 fc_conf->mode = RTE_FC_NONE;
871                 break;
872         case EFX_FCNTL_RESPOND:
873                 fc_conf->mode = RTE_FC_RX_PAUSE;
874                 break;
875         case EFX_FCNTL_GENERATE:
876                 fc_conf->mode = RTE_FC_TX_PAUSE;
877                 break;
878         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
879                 fc_conf->mode = RTE_FC_FULL;
880                 break;
881         default:
882                 sfc_err(sa, "%s: unexpected flow control value %#x",
883                         __func__, link_fc);
884         }
885
886         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
887
888         sfc_adapter_unlock(sa);
889
890         return 0;
891 }
892
893 static int
894 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
895 {
896         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
897         struct sfc_port *port = &sa->port;
898         unsigned int fcntl;
899         int rc;
900
901         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
902             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
903             fc_conf->mac_ctrl_frame_fwd != 0) {
904                 sfc_err(sa, "unsupported flow control settings specified");
905                 rc = EINVAL;
906                 goto fail_inval;
907         }
908
909         switch (fc_conf->mode) {
910         case RTE_FC_NONE:
911                 fcntl = 0;
912                 break;
913         case RTE_FC_RX_PAUSE:
914                 fcntl = EFX_FCNTL_RESPOND;
915                 break;
916         case RTE_FC_TX_PAUSE:
917                 fcntl = EFX_FCNTL_GENERATE;
918                 break;
919         case RTE_FC_FULL:
920                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
921                 break;
922         default:
923                 rc = EINVAL;
924                 goto fail_inval;
925         }
926
927         sfc_adapter_lock(sa);
928
929         if (sa->state == SFC_ADAPTER_STARTED) {
930                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
931                 if (rc != 0)
932                         goto fail_mac_fcntl_set;
933         }
934
935         port->flow_ctrl = fcntl;
936         port->flow_ctrl_autoneg = fc_conf->autoneg;
937
938         sfc_adapter_unlock(sa);
939
940         return 0;
941
942 fail_mac_fcntl_set:
943         sfc_adapter_unlock(sa);
944 fail_inval:
945         SFC_ASSERT(rc > 0);
946         return -rc;
947 }
948
949 static int
950 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
951 {
952         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
953         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
954         boolean_t scatter_enabled;
955         const char *error;
956         unsigned int i;
957
958         for (i = 0; i < sas->rxq_count; i++) {
959                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
960                         continue;
961
962                 scatter_enabled = (sas->rxq_info[i].type_flags &
963                                    EFX_RXQ_FLAG_SCATTER);
964
965                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
966                                           encp->enc_rx_prefix_size,
967                                           scatter_enabled,
968                                           encp->enc_rx_scatter_max, &error)) {
969                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
970                                 error);
971                         return EINVAL;
972                 }
973         }
974
975         return 0;
976 }
977
978 static int
979 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
980 {
981         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
982         size_t pdu = EFX_MAC_PDU(mtu);
983         size_t old_pdu;
984         int rc;
985
986         sfc_log_init(sa, "mtu=%u", mtu);
987
988         rc = EINVAL;
989         if (pdu < EFX_MAC_PDU_MIN) {
990                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
991                         (unsigned int)mtu, (unsigned int)pdu,
992                         EFX_MAC_PDU_MIN);
993                 goto fail_inval;
994         }
995         if (pdu > EFX_MAC_PDU_MAX) {
996                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
997                         (unsigned int)mtu, (unsigned int)pdu,
998                         (unsigned int)EFX_MAC_PDU_MAX);
999                 goto fail_inval;
1000         }
1001
1002         sfc_adapter_lock(sa);
1003
1004         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1005         if (rc != 0)
1006                 goto fail_check_scatter;
1007
1008         if (pdu != sa->port.pdu) {
1009                 if (sa->state == SFC_ADAPTER_STARTED) {
1010                         sfc_stop(sa);
1011
1012                         old_pdu = sa->port.pdu;
1013                         sa->port.pdu = pdu;
1014                         rc = sfc_start(sa);
1015                         if (rc != 0)
1016                                 goto fail_start;
1017                 } else {
1018                         sa->port.pdu = pdu;
1019                 }
1020         }
1021
1022         /*
1023          * The driver does not use it, but other PMDs update jumbo frame
1024          * flag and max_rx_pkt_len when MTU is set.
1025          */
1026         if (mtu > RTE_ETHER_MTU) {
1027                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1028                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1029         }
1030
1031         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1032
1033         sfc_adapter_unlock(sa);
1034
1035         sfc_log_init(sa, "done");
1036         return 0;
1037
1038 fail_start:
1039         sa->port.pdu = old_pdu;
1040         if (sfc_start(sa) != 0)
1041                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1042                         "PDU max size - port is stopped",
1043                         (unsigned int)pdu, (unsigned int)old_pdu);
1044
1045 fail_check_scatter:
1046         sfc_adapter_unlock(sa);
1047
1048 fail_inval:
1049         sfc_log_init(sa, "failed %d", rc);
1050         SFC_ASSERT(rc > 0);
1051         return -rc;
1052 }
1053 static int
1054 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1055 {
1056         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1057         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1058         struct sfc_port *port = &sa->port;
1059         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1060         int rc = 0;
1061
1062         sfc_adapter_lock(sa);
1063
1064         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1065                 goto unlock;
1066
1067         /*
1068          * Copy the address to the device private data so that
1069          * it could be recalled in the case of adapter restart.
1070          */
1071         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1072
1073         /*
1074          * Neither of the two following checks can return
1075          * an error. The new MAC address is preserved in
1076          * the device private data and can be activated
1077          * on the next port start if the user prevents
1078          * isolated mode from being enabled.
1079          */
1080         if (sfc_sa2shared(sa)->isolated) {
1081                 sfc_warn(sa, "isolated mode is active on the port");
1082                 sfc_warn(sa, "will not set MAC address");
1083                 goto unlock;
1084         }
1085
1086         if (sa->state != SFC_ADAPTER_STARTED) {
1087                 sfc_notice(sa, "the port is not started");
1088                 sfc_notice(sa, "the new MAC address will be set on port start");
1089
1090                 goto unlock;
1091         }
1092
1093         if (encp->enc_allow_set_mac_with_installed_filters) {
1094                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1095                 if (rc != 0) {
1096                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1097                         goto unlock;
1098                 }
1099
1100                 /*
1101                  * Changing the MAC address by means of MCDI request
1102                  * has no effect on received traffic, therefore
1103                  * we also need to update unicast filters
1104                  */
1105                 rc = sfc_set_rx_mode_unchecked(sa);
1106                 if (rc != 0) {
1107                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1108                         /* Rollback the old address */
1109                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1110                         (void)sfc_set_rx_mode_unchecked(sa);
1111                 }
1112         } else {
1113                 sfc_warn(sa, "cannot set MAC address with filters installed");
1114                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1115                 sfc_warn(sa, "(some traffic may be dropped)");
1116
1117                 /*
1118                  * Since setting MAC address with filters installed is not
1119                  * allowed on the adapter, the new MAC address will be set
1120                  * by means of adapter restart. sfc_start() shall retrieve
1121                  * the new address from the device private data and set it.
1122                  */
1123                 sfc_stop(sa);
1124                 rc = sfc_start(sa);
1125                 if (rc != 0)
1126                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1127         }
1128
1129 unlock:
1130         if (rc != 0)
1131                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1132
1133         sfc_adapter_unlock(sa);
1134
1135         SFC_ASSERT(rc >= 0);
1136         return -rc;
1137 }
1138
1139
1140 static int
1141 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1142                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1143 {
1144         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1145         struct sfc_port *port = &sa->port;
1146         uint8_t *mc_addrs = port->mcast_addrs;
1147         int rc;
1148         unsigned int i;
1149
1150         if (sfc_sa2shared(sa)->isolated) {
1151                 sfc_err(sa, "isolated mode is active on the port");
1152                 sfc_err(sa, "will not set multicast address list");
1153                 return -ENOTSUP;
1154         }
1155
1156         if (mc_addrs == NULL)
1157                 return -ENOBUFS;
1158
1159         if (nb_mc_addr > port->max_mcast_addrs) {
1160                 sfc_err(sa, "too many multicast addresses: %u > %u",
1161                          nb_mc_addr, port->max_mcast_addrs);
1162                 return -EINVAL;
1163         }
1164
1165         for (i = 0; i < nb_mc_addr; ++i) {
1166                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1167                                  EFX_MAC_ADDR_LEN);
1168                 mc_addrs += EFX_MAC_ADDR_LEN;
1169         }
1170
1171         port->nb_mcast_addrs = nb_mc_addr;
1172
1173         if (sa->state != SFC_ADAPTER_STARTED)
1174                 return 0;
1175
1176         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1177                                         port->nb_mcast_addrs);
1178         if (rc != 0)
1179                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1180
1181         SFC_ASSERT(rc >= 0);
1182         return -rc;
1183 }
1184
1185 /*
1186  * The function is used by the secondary process as well. It must not
1187  * use any process-local pointers from the adapter data.
1188  */
1189 static void
1190 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1191                       struct rte_eth_rxq_info *qinfo)
1192 {
1193         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1194         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1195         struct sfc_rxq_info *rxq_info;
1196
1197         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1198
1199         qinfo->mp = rxq_info->refill_mb_pool;
1200         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1201         qinfo->conf.rx_drop_en = 1;
1202         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1203         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1204         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1205                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1206                 qinfo->scattered_rx = 1;
1207         }
1208         qinfo->nb_desc = rxq_info->entries;
1209 }
1210
1211 /*
1212  * The function is used by the secondary process as well. It must not
1213  * use any process-local pointers from the adapter data.
1214  */
1215 static void
1216 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1217                       struct rte_eth_txq_info *qinfo)
1218 {
1219         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1220         struct sfc_txq_info *txq_info;
1221
1222         SFC_ASSERT(tx_queue_id < sas->txq_count);
1223
1224         txq_info = &sas->txq_info[tx_queue_id];
1225
1226         memset(qinfo, 0, sizeof(*qinfo));
1227
1228         qinfo->conf.offloads = txq_info->offloads;
1229         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1230         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1231         qinfo->nb_desc = txq_info->entries;
1232 }
1233
1234 /*
1235  * The function is used by the secondary process as well. It must not
1236  * use any process-local pointers from the adapter data.
1237  */
1238 static uint32_t
1239 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1240 {
1241         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1242         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1243         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1244         struct sfc_rxq_info *rxq_info;
1245
1246         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1247
1248         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1249                 return 0;
1250
1251         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1252 }
1253
1254 /*
1255  * The function is used by the secondary process as well. It must not
1256  * use any process-local pointers from the adapter data.
1257  */
1258 static int
1259 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1260 {
1261         struct sfc_dp_rxq *dp_rxq = queue;
1262         const struct sfc_dp_rx *dp_rx;
1263
1264         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1265
1266         return offset < dp_rx->qdesc_npending(dp_rxq);
1267 }
1268
1269 /*
1270  * The function is used by the secondary process as well. It must not
1271  * use any process-local pointers from the adapter data.
1272  */
1273 static int
1274 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1275 {
1276         struct sfc_dp_rxq *dp_rxq = queue;
1277         const struct sfc_dp_rx *dp_rx;
1278
1279         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1280
1281         return dp_rx->qdesc_status(dp_rxq, offset);
1282 }
1283
1284 /*
1285  * The function is used by the secondary process as well. It must not
1286  * use any process-local pointers from the adapter data.
1287  */
1288 static int
1289 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1290 {
1291         struct sfc_dp_txq *dp_txq = queue;
1292         const struct sfc_dp_tx *dp_tx;
1293
1294         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1295
1296         return dp_tx->qdesc_status(dp_txq, offset);
1297 }
1298
1299 static int
1300 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1301 {
1302         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1303         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1304         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1305         struct sfc_rxq_info *rxq_info;
1306         sfc_sw_index_t sw_index;
1307         int rc;
1308
1309         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1310
1311         sfc_adapter_lock(sa);
1312
1313         rc = EINVAL;
1314         if (sa->state != SFC_ADAPTER_STARTED)
1315                 goto fail_not_started;
1316
1317         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1318         if (rxq_info->state != SFC_RXQ_INITIALIZED)
1319                 goto fail_not_setup;
1320
1321         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1322         rc = sfc_rx_qstart(sa, sw_index);
1323         if (rc != 0)
1324                 goto fail_rx_qstart;
1325
1326         rxq_info->deferred_started = B_TRUE;
1327
1328         sfc_adapter_unlock(sa);
1329
1330         return 0;
1331
1332 fail_rx_qstart:
1333 fail_not_setup:
1334 fail_not_started:
1335         sfc_adapter_unlock(sa);
1336         SFC_ASSERT(rc > 0);
1337         return -rc;
1338 }
1339
1340 static int
1341 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1342 {
1343         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1344         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1345         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1346         struct sfc_rxq_info *rxq_info;
1347         sfc_sw_index_t sw_index;
1348
1349         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1350
1351         sfc_adapter_lock(sa);
1352
1353         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1354         sfc_rx_qstop(sa, sw_index);
1355
1356         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1357         rxq_info->deferred_started = B_FALSE;
1358
1359         sfc_adapter_unlock(sa);
1360
1361         return 0;
1362 }
1363
1364 static int
1365 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1366 {
1367         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1368         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1369         int rc;
1370
1371         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1372
1373         sfc_adapter_lock(sa);
1374
1375         rc = EINVAL;
1376         if (sa->state != SFC_ADAPTER_STARTED)
1377                 goto fail_not_started;
1378
1379         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1380                 goto fail_not_setup;
1381
1382         rc = sfc_tx_qstart(sa, tx_queue_id);
1383         if (rc != 0)
1384                 goto fail_tx_qstart;
1385
1386         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1387
1388         sfc_adapter_unlock(sa);
1389         return 0;
1390
1391 fail_tx_qstart:
1392
1393 fail_not_setup:
1394 fail_not_started:
1395         sfc_adapter_unlock(sa);
1396         SFC_ASSERT(rc > 0);
1397         return -rc;
1398 }
1399
1400 static int
1401 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1402 {
1403         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1404         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1405
1406         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1407
1408         sfc_adapter_lock(sa);
1409
1410         sfc_tx_qstop(sa, tx_queue_id);
1411
1412         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1413
1414         sfc_adapter_unlock(sa);
1415         return 0;
1416 }
1417
1418 static efx_tunnel_protocol_t
1419 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1420 {
1421         switch (rte_type) {
1422         case RTE_TUNNEL_TYPE_VXLAN:
1423                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1424         case RTE_TUNNEL_TYPE_GENEVE:
1425                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1426         default:
1427                 return EFX_TUNNEL_NPROTOS;
1428         }
1429 }
1430
1431 enum sfc_udp_tunnel_op_e {
1432         SFC_UDP_TUNNEL_ADD_PORT,
1433         SFC_UDP_TUNNEL_DEL_PORT,
1434 };
1435
1436 static int
1437 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1438                       struct rte_eth_udp_tunnel *tunnel_udp,
1439                       enum sfc_udp_tunnel_op_e op)
1440 {
1441         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1442         efx_tunnel_protocol_t tunnel_proto;
1443         int rc;
1444
1445         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1446                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1447                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1448                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1449
1450         tunnel_proto =
1451                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1452         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1453                 rc = ENOTSUP;
1454                 goto fail_bad_proto;
1455         }
1456
1457         sfc_adapter_lock(sa);
1458
1459         switch (op) {
1460         case SFC_UDP_TUNNEL_ADD_PORT:
1461                 rc = efx_tunnel_config_udp_add(sa->nic,
1462                                                tunnel_udp->udp_port,
1463                                                tunnel_proto);
1464                 break;
1465         case SFC_UDP_TUNNEL_DEL_PORT:
1466                 rc = efx_tunnel_config_udp_remove(sa->nic,
1467                                                   tunnel_udp->udp_port,
1468                                                   tunnel_proto);
1469                 break;
1470         default:
1471                 rc = EINVAL;
1472                 goto fail_bad_op;
1473         }
1474
1475         if (rc != 0)
1476                 goto fail_op;
1477
1478         if (sa->state == SFC_ADAPTER_STARTED) {
1479                 rc = efx_tunnel_reconfigure(sa->nic);
1480                 if (rc == EAGAIN) {
1481                         /*
1482                          * Configuration is accepted by FW and MC reboot
1483                          * is initiated to apply the changes. MC reboot
1484                          * will be handled in a usual way (MC reboot
1485                          * event on management event queue and adapter
1486                          * restart).
1487                          */
1488                         rc = 0;
1489                 } else if (rc != 0) {
1490                         goto fail_reconfigure;
1491                 }
1492         }
1493
1494         sfc_adapter_unlock(sa);
1495         return 0;
1496
1497 fail_reconfigure:
1498         /* Remove/restore entry since the change makes the trouble */
1499         switch (op) {
1500         case SFC_UDP_TUNNEL_ADD_PORT:
1501                 (void)efx_tunnel_config_udp_remove(sa->nic,
1502                                                    tunnel_udp->udp_port,
1503                                                    tunnel_proto);
1504                 break;
1505         case SFC_UDP_TUNNEL_DEL_PORT:
1506                 (void)efx_tunnel_config_udp_add(sa->nic,
1507                                                 tunnel_udp->udp_port,
1508                                                 tunnel_proto);
1509                 break;
1510         }
1511
1512 fail_op:
1513 fail_bad_op:
1514         sfc_adapter_unlock(sa);
1515
1516 fail_bad_proto:
1517         SFC_ASSERT(rc > 0);
1518         return -rc;
1519 }
1520
1521 static int
1522 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1523                             struct rte_eth_udp_tunnel *tunnel_udp)
1524 {
1525         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1526 }
1527
1528 static int
1529 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1530                             struct rte_eth_udp_tunnel *tunnel_udp)
1531 {
1532         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1533 }
1534
1535 /*
1536  * The function is used by the secondary process as well. It must not
1537  * use any process-local pointers from the adapter data.
1538  */
1539 static int
1540 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1541                           struct rte_eth_rss_conf *rss_conf)
1542 {
1543         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1544         struct sfc_rss *rss = &sas->rss;
1545
1546         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1547                 return -ENOTSUP;
1548
1549         /*
1550          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1551          * hence, conversion is done here to derive a correct set of ETH_RSS
1552          * flags which corresponds to the active EFX configuration stored
1553          * locally in 'sfc_adapter' and kept up-to-date
1554          */
1555         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1556         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1557         if (rss_conf->rss_key != NULL)
1558                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1559
1560         return 0;
1561 }
1562
1563 static int
1564 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1565                         struct rte_eth_rss_conf *rss_conf)
1566 {
1567         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1568         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1569         unsigned int efx_hash_types;
1570         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1571         unsigned int n_contexts;
1572         unsigned int mode_i = 0;
1573         unsigned int key_i = 0;
1574         unsigned int i = 0;
1575         int rc = 0;
1576
1577         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1578
1579         if (sfc_sa2shared(sa)->isolated)
1580                 return -ENOTSUP;
1581
1582         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1583                 sfc_err(sa, "RSS is not available");
1584                 return -ENOTSUP;
1585         }
1586
1587         if (rss->channels == 0) {
1588                 sfc_err(sa, "RSS is not configured");
1589                 return -EINVAL;
1590         }
1591
1592         if ((rss_conf->rss_key != NULL) &&
1593             (rss_conf->rss_key_len != sizeof(rss->key))) {
1594                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1595                         sizeof(rss->key));
1596                 return -EINVAL;
1597         }
1598
1599         sfc_adapter_lock(sa);
1600
1601         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1602         if (rc != 0)
1603                 goto fail_rx_hf_rte_to_efx;
1604
1605         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1606                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1607                                            rss->hash_alg, efx_hash_types,
1608                                            B_TRUE);
1609                 if (rc != 0)
1610                         goto fail_scale_mode_set;
1611         }
1612
1613         if (rss_conf->rss_key != NULL) {
1614                 if (sa->state == SFC_ADAPTER_STARTED) {
1615                         for (key_i = 0; key_i < n_contexts; key_i++) {
1616                                 rc = efx_rx_scale_key_set(sa->nic,
1617                                                           contexts[key_i],
1618                                                           rss_conf->rss_key,
1619                                                           sizeof(rss->key));
1620                                 if (rc != 0)
1621                                         goto fail_scale_key_set;
1622                         }
1623                 }
1624
1625                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1626         }
1627
1628         rss->hash_types = efx_hash_types;
1629
1630         sfc_adapter_unlock(sa);
1631
1632         return 0;
1633
1634 fail_scale_key_set:
1635         for (i = 0; i < key_i; i++) {
1636                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1637                                          sizeof(rss->key)) != 0)
1638                         sfc_err(sa, "failed to restore RSS key");
1639         }
1640
1641 fail_scale_mode_set:
1642         for (i = 0; i < mode_i; i++) {
1643                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1644                                           EFX_RX_HASHALG_TOEPLITZ,
1645                                           rss->hash_types, B_TRUE) != 0)
1646                         sfc_err(sa, "failed to restore RSS mode");
1647         }
1648
1649 fail_rx_hf_rte_to_efx:
1650         sfc_adapter_unlock(sa);
1651         return -rc;
1652 }
1653
1654 /*
1655  * The function is used by the secondary process as well. It must not
1656  * use any process-local pointers from the adapter data.
1657  */
1658 static int
1659 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1660                        struct rte_eth_rss_reta_entry64 *reta_conf,
1661                        uint16_t reta_size)
1662 {
1663         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1664         struct sfc_rss *rss = &sas->rss;
1665         int entry;
1666
1667         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1668                 return -ENOTSUP;
1669
1670         if (rss->channels == 0)
1671                 return -EINVAL;
1672
1673         if (reta_size != EFX_RSS_TBL_SIZE)
1674                 return -EINVAL;
1675
1676         for (entry = 0; entry < reta_size; entry++) {
1677                 int grp = entry / RTE_RETA_GROUP_SIZE;
1678                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1679
1680                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1681                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1682         }
1683
1684         return 0;
1685 }
1686
1687 static int
1688 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1689                         struct rte_eth_rss_reta_entry64 *reta_conf,
1690                         uint16_t reta_size)
1691 {
1692         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1693         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1694         unsigned int *rss_tbl_new;
1695         uint16_t entry;
1696         int rc = 0;
1697
1698
1699         if (sfc_sa2shared(sa)->isolated)
1700                 return -ENOTSUP;
1701
1702         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1703                 sfc_err(sa, "RSS is not available");
1704                 return -ENOTSUP;
1705         }
1706
1707         if (rss->channels == 0) {
1708                 sfc_err(sa, "RSS is not configured");
1709                 return -EINVAL;
1710         }
1711
1712         if (reta_size != EFX_RSS_TBL_SIZE) {
1713                 sfc_err(sa, "RETA size is wrong (should be %u)",
1714                         EFX_RSS_TBL_SIZE);
1715                 return -EINVAL;
1716         }
1717
1718         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1719         if (rss_tbl_new == NULL)
1720                 return -ENOMEM;
1721
1722         sfc_adapter_lock(sa);
1723
1724         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1725
1726         for (entry = 0; entry < reta_size; entry++) {
1727                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1728                 struct rte_eth_rss_reta_entry64 *grp;
1729
1730                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1731
1732                 if (grp->mask & (1ull << grp_idx)) {
1733                         if (grp->reta[grp_idx] >= rss->channels) {
1734                                 rc = EINVAL;
1735                                 goto bad_reta_entry;
1736                         }
1737                         rss_tbl_new[entry] = grp->reta[grp_idx];
1738                 }
1739         }
1740
1741         if (sa->state == SFC_ADAPTER_STARTED) {
1742                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1743                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1744                 if (rc != 0)
1745                         goto fail_scale_tbl_set;
1746         }
1747
1748         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1749
1750 fail_scale_tbl_set:
1751 bad_reta_entry:
1752         sfc_adapter_unlock(sa);
1753
1754         rte_free(rss_tbl_new);
1755
1756         SFC_ASSERT(rc >= 0);
1757         return -rc;
1758 }
1759
1760 static int
1761 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1762                      const struct rte_flow_ops **ops)
1763 {
1764         *ops = &sfc_flow_ops;
1765         return 0;
1766 }
1767
1768 static int
1769 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1770 {
1771         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1772
1773         /*
1774          * If Rx datapath does not provide callback to check mempool,
1775          * all pools are supported.
1776          */
1777         if (sap->dp_rx->pool_ops_supported == NULL)
1778                 return 1;
1779
1780         return sap->dp_rx->pool_ops_supported(pool);
1781 }
1782
1783 static int
1784 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1785 {
1786         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1787         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1788         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1789         struct sfc_rxq_info *rxq_info;
1790
1791         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1792
1793         return sap->dp_rx->intr_enable(rxq_info->dp);
1794 }
1795
1796 static int
1797 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1798 {
1799         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1800         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1801         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1802         struct sfc_rxq_info *rxq_info;
1803
1804         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1805
1806         return sap->dp_rx->intr_disable(rxq_info->dp);
1807 }
1808
1809 static const struct eth_dev_ops sfc_eth_dev_ops = {
1810         .dev_configure                  = sfc_dev_configure,
1811         .dev_start                      = sfc_dev_start,
1812         .dev_stop                       = sfc_dev_stop,
1813         .dev_set_link_up                = sfc_dev_set_link_up,
1814         .dev_set_link_down              = sfc_dev_set_link_down,
1815         .dev_close                      = sfc_dev_close,
1816         .promiscuous_enable             = sfc_dev_promisc_enable,
1817         .promiscuous_disable            = sfc_dev_promisc_disable,
1818         .allmulticast_enable            = sfc_dev_allmulti_enable,
1819         .allmulticast_disable           = sfc_dev_allmulti_disable,
1820         .link_update                    = sfc_dev_link_update,
1821         .stats_get                      = sfc_stats_get,
1822         .stats_reset                    = sfc_stats_reset,
1823         .xstats_get                     = sfc_xstats_get,
1824         .xstats_reset                   = sfc_stats_reset,
1825         .xstats_get_names               = sfc_xstats_get_names,
1826         .dev_infos_get                  = sfc_dev_infos_get,
1827         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1828         .mtu_set                        = sfc_dev_set_mtu,
1829         .rx_queue_start                 = sfc_rx_queue_start,
1830         .rx_queue_stop                  = sfc_rx_queue_stop,
1831         .tx_queue_start                 = sfc_tx_queue_start,
1832         .tx_queue_stop                  = sfc_tx_queue_stop,
1833         .rx_queue_setup                 = sfc_rx_queue_setup,
1834         .rx_queue_release               = sfc_rx_queue_release,
1835         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1836         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1837         .tx_queue_setup                 = sfc_tx_queue_setup,
1838         .tx_queue_release               = sfc_tx_queue_release,
1839         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1840         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1841         .mac_addr_set                   = sfc_mac_addr_set,
1842         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1843         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1844         .reta_update                    = sfc_dev_rss_reta_update,
1845         .reta_query                     = sfc_dev_rss_reta_query,
1846         .rss_hash_update                = sfc_dev_rss_hash_update,
1847         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1848         .flow_ops_get                   = sfc_dev_flow_ops_get,
1849         .set_mc_addr_list               = sfc_set_mc_addr_list,
1850         .rxq_info_get                   = sfc_rx_queue_info_get,
1851         .txq_info_get                   = sfc_tx_queue_info_get,
1852         .fw_version_get                 = sfc_fw_version_get,
1853         .xstats_get_by_id               = sfc_xstats_get_by_id,
1854         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1855         .pool_ops_supported             = sfc_pool_ops_supported,
1856 };
1857
1858 /**
1859  * Duplicate a string in potentially shared memory required for
1860  * multi-process support.
1861  *
1862  * strdup() allocates from process-local heap/memory.
1863  */
1864 static char *
1865 sfc_strdup(const char *str)
1866 {
1867         size_t size;
1868         char *copy;
1869
1870         if (str == NULL)
1871                 return NULL;
1872
1873         size = strlen(str) + 1;
1874         copy = rte_malloc(__func__, size, 0);
1875         if (copy != NULL)
1876                 rte_memcpy(copy, str, size);
1877
1878         return copy;
1879 }
1880
1881 static int
1882 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1883 {
1884         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1885         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1886         const struct sfc_dp_rx *dp_rx;
1887         const struct sfc_dp_tx *dp_tx;
1888         const efx_nic_cfg_t *encp;
1889         unsigned int avail_caps = 0;
1890         const char *rx_name = NULL;
1891         const char *tx_name = NULL;
1892         int rc;
1893
1894         switch (sa->family) {
1895         case EFX_FAMILY_HUNTINGTON:
1896         case EFX_FAMILY_MEDFORD:
1897         case EFX_FAMILY_MEDFORD2:
1898                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1899                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
1900                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
1901                 break;
1902         case EFX_FAMILY_RIVERHEAD:
1903                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
1904                 break;
1905         default:
1906                 break;
1907         }
1908
1909         encp = efx_nic_cfg_get(sa->nic);
1910         if (encp->enc_rx_es_super_buffer_supported)
1911                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1912
1913         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1914                                 sfc_kvarg_string_handler, &rx_name);
1915         if (rc != 0)
1916                 goto fail_kvarg_rx_datapath;
1917
1918         if (rx_name != NULL) {
1919                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1920                 if (dp_rx == NULL) {
1921                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1922                         rc = ENOENT;
1923                         goto fail_dp_rx;
1924                 }
1925                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1926                         sfc_err(sa,
1927                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1928                                 rx_name);
1929                         rc = EINVAL;
1930                         goto fail_dp_rx_caps;
1931                 }
1932         } else {
1933                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1934                 if (dp_rx == NULL) {
1935                         sfc_err(sa, "Rx datapath by caps %#x not found",
1936                                 avail_caps);
1937                         rc = ENOENT;
1938                         goto fail_dp_rx;
1939                 }
1940         }
1941
1942         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1943         if (sas->dp_rx_name == NULL) {
1944                 rc = ENOMEM;
1945                 goto fail_dp_rx_name;
1946         }
1947
1948         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1949
1950         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1951                                 sfc_kvarg_string_handler, &tx_name);
1952         if (rc != 0)
1953                 goto fail_kvarg_tx_datapath;
1954
1955         if (tx_name != NULL) {
1956                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1957                 if (dp_tx == NULL) {
1958                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1959                         rc = ENOENT;
1960                         goto fail_dp_tx;
1961                 }
1962                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1963                         sfc_err(sa,
1964                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1965                                 tx_name);
1966                         rc = EINVAL;
1967                         goto fail_dp_tx_caps;
1968                 }
1969         } else {
1970                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1971                 if (dp_tx == NULL) {
1972                         sfc_err(sa, "Tx datapath by caps %#x not found",
1973                                 avail_caps);
1974                         rc = ENOENT;
1975                         goto fail_dp_tx;
1976                 }
1977         }
1978
1979         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1980         if (sas->dp_tx_name == NULL) {
1981                 rc = ENOMEM;
1982                 goto fail_dp_tx_name;
1983         }
1984
1985         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1986
1987         sa->priv.dp_rx = dp_rx;
1988         sa->priv.dp_tx = dp_tx;
1989
1990         dev->rx_pkt_burst = dp_rx->pkt_burst;
1991         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1992         dev->tx_pkt_burst = dp_tx->pkt_burst;
1993
1994         dev->rx_queue_count = sfc_rx_queue_count;
1995         dev->rx_descriptor_done = sfc_rx_descriptor_done;
1996         dev->rx_descriptor_status = sfc_rx_descriptor_status;
1997         dev->tx_descriptor_status = sfc_tx_descriptor_status;
1998         dev->dev_ops = &sfc_eth_dev_ops;
1999
2000         return 0;
2001
2002 fail_dp_tx_name:
2003 fail_dp_tx_caps:
2004 fail_dp_tx:
2005 fail_kvarg_tx_datapath:
2006         rte_free(sas->dp_rx_name);
2007         sas->dp_rx_name = NULL;
2008
2009 fail_dp_rx_name:
2010 fail_dp_rx_caps:
2011 fail_dp_rx:
2012 fail_kvarg_rx_datapath:
2013         return rc;
2014 }
2015
2016 static void
2017 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2018 {
2019         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2020         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2021
2022         dev->dev_ops = NULL;
2023         dev->tx_pkt_prepare = NULL;
2024         dev->rx_pkt_burst = NULL;
2025         dev->tx_pkt_burst = NULL;
2026
2027         rte_free(sas->dp_tx_name);
2028         sas->dp_tx_name = NULL;
2029         sa->priv.dp_tx = NULL;
2030
2031         rte_free(sas->dp_rx_name);
2032         sas->dp_rx_name = NULL;
2033         sa->priv.dp_rx = NULL;
2034 }
2035
2036 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2037         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2038         .reta_query                     = sfc_dev_rss_reta_query,
2039         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2040         .rxq_info_get                   = sfc_rx_queue_info_get,
2041         .txq_info_get                   = sfc_tx_queue_info_get,
2042 };
2043
2044 static int
2045 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2046 {
2047         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2048         struct sfc_adapter_priv *sap;
2049         const struct sfc_dp_rx *dp_rx;
2050         const struct sfc_dp_tx *dp_tx;
2051         int rc;
2052
2053         /*
2054          * Allocate process private data from heap, since it should not
2055          * be located in shared memory allocated using rte_malloc() API.
2056          */
2057         sap = calloc(1, sizeof(*sap));
2058         if (sap == NULL) {
2059                 rc = ENOMEM;
2060                 goto fail_alloc_priv;
2061         }
2062
2063         sap->logtype_main = logtype_main;
2064
2065         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2066         if (dp_rx == NULL) {
2067                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2068                         "cannot find %s Rx datapath", sas->dp_rx_name);
2069                 rc = ENOENT;
2070                 goto fail_dp_rx;
2071         }
2072         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2073                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2074                         "%s Rx datapath does not support multi-process",
2075                         sas->dp_rx_name);
2076                 rc = EINVAL;
2077                 goto fail_dp_rx_multi_process;
2078         }
2079
2080         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2081         if (dp_tx == NULL) {
2082                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2083                         "cannot find %s Tx datapath", sas->dp_tx_name);
2084                 rc = ENOENT;
2085                 goto fail_dp_tx;
2086         }
2087         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2088                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2089                         "%s Tx datapath does not support multi-process",
2090                         sas->dp_tx_name);
2091                 rc = EINVAL;
2092                 goto fail_dp_tx_multi_process;
2093         }
2094
2095         sap->dp_rx = dp_rx;
2096         sap->dp_tx = dp_tx;
2097
2098         dev->process_private = sap;
2099         dev->rx_pkt_burst = dp_rx->pkt_burst;
2100         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2101         dev->tx_pkt_burst = dp_tx->pkt_burst;
2102         dev->rx_queue_count = sfc_rx_queue_count;
2103         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2104         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2105         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2106         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2107
2108         return 0;
2109
2110 fail_dp_tx_multi_process:
2111 fail_dp_tx:
2112 fail_dp_rx_multi_process:
2113 fail_dp_rx:
2114         free(sap);
2115
2116 fail_alloc_priv:
2117         return rc;
2118 }
2119
2120 static void
2121 sfc_register_dp(void)
2122 {
2123         /* Register once */
2124         if (TAILQ_EMPTY(&sfc_dp_head)) {
2125                 /* Prefer EF10 datapath */
2126                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2127                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2128                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2129                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2130
2131                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2132                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2133                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2134                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2135         }
2136 }
2137
2138 static int
2139 sfc_eth_dev_init(struct rte_eth_dev *dev)
2140 {
2141         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2142         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2143         uint32_t logtype_main;
2144         struct sfc_adapter *sa;
2145         int rc;
2146         const efx_nic_cfg_t *encp;
2147         const struct rte_ether_addr *from;
2148         int ret;
2149
2150         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2151                         SFC_EFX_DEV_CLASS_NET) {
2152                 SFC_GENERIC_LOG(DEBUG,
2153                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2154                 return 1;
2155         }
2156
2157         sfc_register_dp();
2158
2159         logtype_main = sfc_register_logtype(&pci_dev->addr,
2160                                             SFC_LOGTYPE_MAIN_STR,
2161                                             RTE_LOG_NOTICE);
2162
2163         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2164                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2165
2166         /* Required for logging */
2167         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2168                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2169                         pci_dev->addr.domain, pci_dev->addr.bus,
2170                         pci_dev->addr.devid, pci_dev->addr.function,
2171                         dev->data->port_id);
2172         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2173                 SFC_GENERIC_LOG(ERR,
2174                         "reserved log prefix is too short for " PCI_PRI_FMT,
2175                         pci_dev->addr.domain, pci_dev->addr.bus,
2176                         pci_dev->addr.devid, pci_dev->addr.function);
2177                 return -EINVAL;
2178         }
2179         sas->pci_addr = pci_dev->addr;
2180         sas->port_id = dev->data->port_id;
2181
2182         /*
2183          * Allocate process private data from heap, since it should not
2184          * be located in shared memory allocated using rte_malloc() API.
2185          */
2186         sa = calloc(1, sizeof(*sa));
2187         if (sa == NULL) {
2188                 rc = ENOMEM;
2189                 goto fail_alloc_sa;
2190         }
2191
2192         dev->process_private = sa;
2193
2194         /* Required for logging */
2195         sa->priv.shared = sas;
2196         sa->priv.logtype_main = logtype_main;
2197
2198         sa->eth_dev = dev;
2199
2200         /* Copy PCI device info to the dev->data */
2201         rte_eth_copy_pci_info(dev, pci_dev);
2202         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2203         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2204
2205         rc = sfc_kvargs_parse(sa);
2206         if (rc != 0)
2207                 goto fail_kvargs_parse;
2208
2209         sfc_log_init(sa, "entry");
2210
2211         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2212         if (dev->data->mac_addrs == NULL) {
2213                 rc = ENOMEM;
2214                 goto fail_mac_addrs;
2215         }
2216
2217         sfc_adapter_lock_init(sa);
2218         sfc_adapter_lock(sa);
2219
2220         sfc_log_init(sa, "probing");
2221         rc = sfc_probe(sa);
2222         if (rc != 0)
2223                 goto fail_probe;
2224
2225         sfc_log_init(sa, "set device ops");
2226         rc = sfc_eth_dev_set_ops(dev);
2227         if (rc != 0)
2228                 goto fail_set_ops;
2229
2230         sfc_log_init(sa, "attaching");
2231         rc = sfc_attach(sa);
2232         if (rc != 0)
2233                 goto fail_attach;
2234
2235         encp = efx_nic_cfg_get(sa->nic);
2236
2237         /*
2238          * The arguments are really reverse order in comparison to
2239          * Linux kernel. Copy from NIC config to Ethernet device data.
2240          */
2241         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2242         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2243
2244         sfc_adapter_unlock(sa);
2245
2246         sfc_log_init(sa, "done");
2247         return 0;
2248
2249 fail_attach:
2250         sfc_eth_dev_clear_ops(dev);
2251
2252 fail_set_ops:
2253         sfc_unprobe(sa);
2254
2255 fail_probe:
2256         sfc_adapter_unlock(sa);
2257         sfc_adapter_lock_fini(sa);
2258         rte_free(dev->data->mac_addrs);
2259         dev->data->mac_addrs = NULL;
2260
2261 fail_mac_addrs:
2262         sfc_kvargs_cleanup(sa);
2263
2264 fail_kvargs_parse:
2265         sfc_log_init(sa, "failed %d", rc);
2266         dev->process_private = NULL;
2267         free(sa);
2268
2269 fail_alloc_sa:
2270         SFC_ASSERT(rc > 0);
2271         return -rc;
2272 }
2273
2274 static int
2275 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2276 {
2277         sfc_dev_close(dev);
2278
2279         return 0;
2280 }
2281
2282 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2283         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2284         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2285         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2286         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2287         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2288         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2289         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2290         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2291         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2292         { .vendor_id = 0 /* sentinel */ }
2293 };
2294
2295 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2296         struct rte_pci_device *pci_dev)
2297 {
2298         return rte_eth_dev_pci_generic_probe(pci_dev,
2299                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2300 }
2301
2302 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2303 {
2304         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2305 }
2306
2307 static struct rte_pci_driver sfc_efx_pmd = {
2308         .id_table = pci_id_sfc_efx_map,
2309         .drv_flags =
2310                 RTE_PCI_DRV_INTR_LSC |
2311                 RTE_PCI_DRV_NEED_MAPPING,
2312         .probe = sfc_eth_dev_pci_probe,
2313         .remove = sfc_eth_dev_pci_remove,
2314 };
2315
2316 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2317 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2318 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2319 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2320         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2321         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2322         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2323         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2324         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2325         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2326
2327 RTE_INIT(sfc_driver_register_logtype)
2328 {
2329         int ret;
2330
2331         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2332                                                    RTE_LOG_NOTICE);
2333         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2334 }