net/sfc: prepare to add more xstats
[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 ethdev_qid,
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         struct sfc_txq_info *txq_info;
534         sfc_sw_index_t sw_index;
535         int rc;
536
537         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
538                      ethdev_qid, nb_tx_desc, socket_id);
539
540         sfc_adapter_lock(sa);
541
542         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
543         rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
544         if (rc != 0)
545                 goto fail_tx_qinit;
546
547         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
548         dev->data->tx_queues[ethdev_qid] = txq_info->dp;
549
550         sfc_adapter_unlock(sa);
551         return 0;
552
553 fail_tx_qinit:
554         sfc_adapter_unlock(sa);
555         SFC_ASSERT(rc > 0);
556         return -rc;
557 }
558
559 static void
560 sfc_tx_queue_release(void *queue)
561 {
562         struct sfc_dp_txq *dp_txq = queue;
563         struct sfc_txq *txq;
564         sfc_sw_index_t sw_index;
565         struct sfc_adapter *sa;
566
567         if (dp_txq == NULL)
568                 return;
569
570         txq = sfc_txq_by_dp_txq(dp_txq);
571         sw_index = dp_txq->dpq.queue_id;
572
573         SFC_ASSERT(txq->evq != NULL);
574         sa = txq->evq->sa;
575
576         sfc_log_init(sa, "TxQ = %u", sw_index);
577
578         sfc_adapter_lock(sa);
579
580         sfc_tx_qfini(sa, sw_index);
581
582         sfc_adapter_unlock(sa);
583 }
584
585 /*
586  * Some statistics are computed as A - B where A and B each increase
587  * monotonically with some hardware counter(s) and the counters are read
588  * asynchronously.
589  *
590  * If packet X is counted in A, but not counted in B yet, computed value is
591  * greater than real.
592  *
593  * If packet X is not counted in A at the moment of reading the counter,
594  * but counted in B at the moment of reading the counter, computed value
595  * is less than real.
596  *
597  * However, counter which grows backward is worse evil than slightly wrong
598  * value. So, let's try to guarantee that it never happens except may be
599  * the case when the MAC stats are zeroed as a result of a NIC reset.
600  */
601 static void
602 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
603 {
604         if ((int64_t)(newval - *stat) > 0 || newval == 0)
605                 *stat = newval;
606 }
607
608 static int
609 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
610 {
611         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
612         struct sfc_port *port = &sa->port;
613         uint64_t *mac_stats;
614         int ret;
615
616         sfc_adapter_lock(sa);
617
618         ret = sfc_port_update_mac_stats(sa, B_FALSE);
619         if (ret != 0)
620                 goto unlock;
621
622         mac_stats = port->mac_stats_buf;
623
624         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
625                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
626                 stats->ipackets =
627                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
628                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
629                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
630                 stats->opackets =
631                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
632                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
633                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
634                 stats->ibytes =
635                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
636                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
637                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
638                 stats->obytes =
639                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
640                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
641                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
642                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
643                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
644
645                 /* CRC is included in these stats, but shouldn't be */
646                 stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
647                 stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
648         } else {
649                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
650                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
651                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
652
653                 /* CRC is included in these stats, but shouldn't be */
654                 stats->ibytes -= mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
655                 stats->obytes -= mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
656
657                 /*
658                  * Take into account stats which are whenever supported
659                  * on EF10. If some stat is not supported by current
660                  * firmware variant or HW revision, it is guaranteed
661                  * to be zero in mac_stats.
662                  */
663                 stats->imissed =
664                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
665                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
666                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
667                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
668                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
669                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
670                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
671                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
672                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
673                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
674                 stats->ierrors =
675                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
676                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
677                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
678                 /* no oerrors counters supported on EF10 */
679
680                 /* Exclude missed, errors and pauses from Rx packets */
681                 sfc_update_diff_stat(&port->ipackets,
682                         mac_stats[EFX_MAC_RX_PKTS] -
683                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
684                         stats->imissed - stats->ierrors);
685                 stats->ipackets = port->ipackets;
686         }
687
688 unlock:
689         sfc_adapter_unlock(sa);
690         SFC_ASSERT(ret >= 0);
691         return -ret;
692 }
693
694 static int
695 sfc_stats_reset(struct rte_eth_dev *dev)
696 {
697         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
698         struct sfc_port *port = &sa->port;
699         int rc;
700
701         sfc_adapter_lock(sa);
702
703         if (sa->state != SFC_ADAPTER_STARTED) {
704                 /*
705                  * The operation cannot be done if port is not started; it
706                  * will be scheduled to be done during the next port start
707                  */
708                 port->mac_stats_reset_pending = B_TRUE;
709                 sfc_adapter_unlock(sa);
710                 return 0;
711         }
712
713         rc = sfc_port_reset_mac_stats(sa);
714         if (rc != 0)
715                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
716
717         sfc_adapter_unlock(sa);
718
719         SFC_ASSERT(rc >= 0);
720         return -rc;
721 }
722
723 static int
724 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
725                unsigned int xstats_count)
726 {
727         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
728         struct sfc_port *port = &sa->port;
729         unsigned int nb_written = 0;
730         unsigned int nb_supp;
731
732         if (unlikely(xstats == NULL)) {
733                 sfc_adapter_lock(sa);
734                 nb_supp = port->mac_stats_nb_supported;
735                 sfc_adapter_unlock(sa);
736                 return nb_supp;
737         }
738
739         return sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
740 }
741
742 static int
743 sfc_xstats_get_names(struct rte_eth_dev *dev,
744                      struct rte_eth_xstat_name *xstats_names,
745                      unsigned int xstats_count)
746 {
747         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
748         struct sfc_port *port = &sa->port;
749         unsigned int i;
750         unsigned int nstats = 0;
751
752         if (unlikely(xstats_names == NULL)) {
753                 sfc_adapter_lock(sa);
754                 nstats = port->mac_stats_nb_supported;
755                 sfc_adapter_unlock(sa);
756                 return nstats;
757         }
758
759         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
760                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
761                         if (nstats < xstats_count)
762                                 strlcpy(xstats_names[nstats].name,
763                                         efx_mac_stat_name(sa->nic, i),
764                                         sizeof(xstats_names[0].name));
765                         nstats++;
766                 }
767         }
768
769         return nstats;
770 }
771
772 static int
773 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
774                      uint64_t *values, unsigned int n)
775 {
776         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
777
778         if (unlikely(ids == NULL || values == NULL))
779                 return -EINVAL;
780
781         return sfc_port_get_mac_stats_by_id(sa, ids, values, n);
782 }
783
784 static int
785 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
786                            struct rte_eth_xstat_name *xstats_names,
787                            const uint64_t *ids, unsigned int size)
788 {
789         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
790         struct sfc_port *port = &sa->port;
791         unsigned int nb_supported;
792         unsigned int i;
793
794         if (unlikely(xstats_names == NULL && ids != NULL) ||
795             unlikely(xstats_names != NULL && ids == NULL))
796                 return -EINVAL;
797
798         sfc_adapter_lock(sa);
799
800         if (unlikely(xstats_names == NULL && ids == NULL)) {
801                 nb_supported = port->mac_stats_nb_supported;
802                 sfc_adapter_unlock(sa);
803                 return nb_supported;
804         }
805
806         SFC_ASSERT(port->mac_stats_nb_supported <=
807                    RTE_DIM(port->mac_stats_by_id));
808
809         for (i = 0; i < size; i++) {
810                 if (ids[i] < port->mac_stats_nb_supported) {
811                         strlcpy(xstats_names[i].name,
812                                 efx_mac_stat_name(sa->nic,
813                                                  port->mac_stats_by_id[ids[i]]),
814                                 sizeof(xstats_names[0].name));
815                 } else {
816                         sfc_adapter_unlock(sa);
817                         return i;
818                 }
819         }
820
821         sfc_adapter_unlock(sa);
822
823         return size;
824 }
825
826 static int
827 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
828 {
829         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
830         unsigned int wanted_fc, link_fc;
831
832         memset(fc_conf, 0, sizeof(*fc_conf));
833
834         sfc_adapter_lock(sa);
835
836         if (sa->state == SFC_ADAPTER_STARTED)
837                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
838         else
839                 link_fc = sa->port.flow_ctrl;
840
841         switch (link_fc) {
842         case 0:
843                 fc_conf->mode = RTE_FC_NONE;
844                 break;
845         case EFX_FCNTL_RESPOND:
846                 fc_conf->mode = RTE_FC_RX_PAUSE;
847                 break;
848         case EFX_FCNTL_GENERATE:
849                 fc_conf->mode = RTE_FC_TX_PAUSE;
850                 break;
851         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
852                 fc_conf->mode = RTE_FC_FULL;
853                 break;
854         default:
855                 sfc_err(sa, "%s: unexpected flow control value %#x",
856                         __func__, link_fc);
857         }
858
859         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
860
861         sfc_adapter_unlock(sa);
862
863         return 0;
864 }
865
866 static int
867 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
868 {
869         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
870         struct sfc_port *port = &sa->port;
871         unsigned int fcntl;
872         int rc;
873
874         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
875             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
876             fc_conf->mac_ctrl_frame_fwd != 0) {
877                 sfc_err(sa, "unsupported flow control settings specified");
878                 rc = EINVAL;
879                 goto fail_inval;
880         }
881
882         switch (fc_conf->mode) {
883         case RTE_FC_NONE:
884                 fcntl = 0;
885                 break;
886         case RTE_FC_RX_PAUSE:
887                 fcntl = EFX_FCNTL_RESPOND;
888                 break;
889         case RTE_FC_TX_PAUSE:
890                 fcntl = EFX_FCNTL_GENERATE;
891                 break;
892         case RTE_FC_FULL:
893                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
894                 break;
895         default:
896                 rc = EINVAL;
897                 goto fail_inval;
898         }
899
900         sfc_adapter_lock(sa);
901
902         if (sa->state == SFC_ADAPTER_STARTED) {
903                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
904                 if (rc != 0)
905                         goto fail_mac_fcntl_set;
906         }
907
908         port->flow_ctrl = fcntl;
909         port->flow_ctrl_autoneg = fc_conf->autoneg;
910
911         sfc_adapter_unlock(sa);
912
913         return 0;
914
915 fail_mac_fcntl_set:
916         sfc_adapter_unlock(sa);
917 fail_inval:
918         SFC_ASSERT(rc > 0);
919         return -rc;
920 }
921
922 static int
923 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
924 {
925         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
926         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
927         boolean_t scatter_enabled;
928         const char *error;
929         unsigned int i;
930
931         for (i = 0; i < sas->rxq_count; i++) {
932                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
933                         continue;
934
935                 scatter_enabled = (sas->rxq_info[i].type_flags &
936                                    EFX_RXQ_FLAG_SCATTER);
937
938                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
939                                           encp->enc_rx_prefix_size,
940                                           scatter_enabled,
941                                           encp->enc_rx_scatter_max, &error)) {
942                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
943                                 error);
944                         return EINVAL;
945                 }
946         }
947
948         return 0;
949 }
950
951 static int
952 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
953 {
954         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
955         size_t pdu = EFX_MAC_PDU(mtu);
956         size_t old_pdu;
957         int rc;
958
959         sfc_log_init(sa, "mtu=%u", mtu);
960
961         rc = EINVAL;
962         if (pdu < EFX_MAC_PDU_MIN) {
963                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
964                         (unsigned int)mtu, (unsigned int)pdu,
965                         EFX_MAC_PDU_MIN);
966                 goto fail_inval;
967         }
968         if (pdu > EFX_MAC_PDU_MAX) {
969                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
970                         (unsigned int)mtu, (unsigned int)pdu,
971                         (unsigned int)EFX_MAC_PDU_MAX);
972                 goto fail_inval;
973         }
974
975         sfc_adapter_lock(sa);
976
977         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
978         if (rc != 0)
979                 goto fail_check_scatter;
980
981         if (pdu != sa->port.pdu) {
982                 if (sa->state == SFC_ADAPTER_STARTED) {
983                         sfc_stop(sa);
984
985                         old_pdu = sa->port.pdu;
986                         sa->port.pdu = pdu;
987                         rc = sfc_start(sa);
988                         if (rc != 0)
989                                 goto fail_start;
990                 } else {
991                         sa->port.pdu = pdu;
992                 }
993         }
994
995         /*
996          * The driver does not use it, but other PMDs update jumbo frame
997          * flag and max_rx_pkt_len when MTU is set.
998          */
999         if (mtu > RTE_ETHER_MTU) {
1000                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1001                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1002         }
1003
1004         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1005
1006         sfc_adapter_unlock(sa);
1007
1008         sfc_log_init(sa, "done");
1009         return 0;
1010
1011 fail_start:
1012         sa->port.pdu = old_pdu;
1013         if (sfc_start(sa) != 0)
1014                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1015                         "PDU max size - port is stopped",
1016                         (unsigned int)pdu, (unsigned int)old_pdu);
1017
1018 fail_check_scatter:
1019         sfc_adapter_unlock(sa);
1020
1021 fail_inval:
1022         sfc_log_init(sa, "failed %d", rc);
1023         SFC_ASSERT(rc > 0);
1024         return -rc;
1025 }
1026 static int
1027 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1028 {
1029         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1030         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1031         struct sfc_port *port = &sa->port;
1032         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1033         int rc = 0;
1034
1035         sfc_adapter_lock(sa);
1036
1037         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1038                 goto unlock;
1039
1040         /*
1041          * Copy the address to the device private data so that
1042          * it could be recalled in the case of adapter restart.
1043          */
1044         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1045
1046         /*
1047          * Neither of the two following checks can return
1048          * an error. The new MAC address is preserved in
1049          * the device private data and can be activated
1050          * on the next port start if the user prevents
1051          * isolated mode from being enabled.
1052          */
1053         if (sfc_sa2shared(sa)->isolated) {
1054                 sfc_warn(sa, "isolated mode is active on the port");
1055                 sfc_warn(sa, "will not set MAC address");
1056                 goto unlock;
1057         }
1058
1059         if (sa->state != SFC_ADAPTER_STARTED) {
1060                 sfc_notice(sa, "the port is not started");
1061                 sfc_notice(sa, "the new MAC address will be set on port start");
1062
1063                 goto unlock;
1064         }
1065
1066         if (encp->enc_allow_set_mac_with_installed_filters) {
1067                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1068                 if (rc != 0) {
1069                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1070                         goto unlock;
1071                 }
1072
1073                 /*
1074                  * Changing the MAC address by means of MCDI request
1075                  * has no effect on received traffic, therefore
1076                  * we also need to update unicast filters
1077                  */
1078                 rc = sfc_set_rx_mode_unchecked(sa);
1079                 if (rc != 0) {
1080                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1081                         /* Rollback the old address */
1082                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1083                         (void)sfc_set_rx_mode_unchecked(sa);
1084                 }
1085         } else {
1086                 sfc_warn(sa, "cannot set MAC address with filters installed");
1087                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1088                 sfc_warn(sa, "(some traffic may be dropped)");
1089
1090                 /*
1091                  * Since setting MAC address with filters installed is not
1092                  * allowed on the adapter, the new MAC address will be set
1093                  * by means of adapter restart. sfc_start() shall retrieve
1094                  * the new address from the device private data and set it.
1095                  */
1096                 sfc_stop(sa);
1097                 rc = sfc_start(sa);
1098                 if (rc != 0)
1099                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1100         }
1101
1102 unlock:
1103         if (rc != 0)
1104                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1105
1106         sfc_adapter_unlock(sa);
1107
1108         SFC_ASSERT(rc >= 0);
1109         return -rc;
1110 }
1111
1112
1113 static int
1114 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1115                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1116 {
1117         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1118         struct sfc_port *port = &sa->port;
1119         uint8_t *mc_addrs = port->mcast_addrs;
1120         int rc;
1121         unsigned int i;
1122
1123         if (sfc_sa2shared(sa)->isolated) {
1124                 sfc_err(sa, "isolated mode is active on the port");
1125                 sfc_err(sa, "will not set multicast address list");
1126                 return -ENOTSUP;
1127         }
1128
1129         if (mc_addrs == NULL)
1130                 return -ENOBUFS;
1131
1132         if (nb_mc_addr > port->max_mcast_addrs) {
1133                 sfc_err(sa, "too many multicast addresses: %u > %u",
1134                          nb_mc_addr, port->max_mcast_addrs);
1135                 return -EINVAL;
1136         }
1137
1138         for (i = 0; i < nb_mc_addr; ++i) {
1139                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1140                                  EFX_MAC_ADDR_LEN);
1141                 mc_addrs += EFX_MAC_ADDR_LEN;
1142         }
1143
1144         port->nb_mcast_addrs = nb_mc_addr;
1145
1146         if (sa->state != SFC_ADAPTER_STARTED)
1147                 return 0;
1148
1149         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1150                                         port->nb_mcast_addrs);
1151         if (rc != 0)
1152                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1153
1154         SFC_ASSERT(rc >= 0);
1155         return -rc;
1156 }
1157
1158 /*
1159  * The function is used by the secondary process as well. It must not
1160  * use any process-local pointers from the adapter data.
1161  */
1162 static void
1163 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1164                       struct rte_eth_rxq_info *qinfo)
1165 {
1166         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1167         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1168         struct sfc_rxq_info *rxq_info;
1169
1170         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1171
1172         qinfo->mp = rxq_info->refill_mb_pool;
1173         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1174         qinfo->conf.rx_drop_en = 1;
1175         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1176         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1177         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1178                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1179                 qinfo->scattered_rx = 1;
1180         }
1181         qinfo->nb_desc = rxq_info->entries;
1182 }
1183
1184 /*
1185  * The function is used by the secondary process as well. It must not
1186  * use any process-local pointers from the adapter data.
1187  */
1188 static void
1189 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1190                       struct rte_eth_txq_info *qinfo)
1191 {
1192         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1193         struct sfc_txq_info *txq_info;
1194
1195         SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1196
1197         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1198
1199         memset(qinfo, 0, sizeof(*qinfo));
1200
1201         qinfo->conf.offloads = txq_info->offloads;
1202         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1203         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1204         qinfo->nb_desc = txq_info->entries;
1205 }
1206
1207 /*
1208  * The function is used by the secondary process as well. It must not
1209  * use any process-local pointers from the adapter data.
1210  */
1211 static uint32_t
1212 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1213 {
1214         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1215         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1216         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1217         struct sfc_rxq_info *rxq_info;
1218
1219         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1220
1221         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1222                 return 0;
1223
1224         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1225 }
1226
1227 /*
1228  * The function is used by the secondary process as well. It must not
1229  * use any process-local pointers from the adapter data.
1230  */
1231 static int
1232 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1233 {
1234         struct sfc_dp_rxq *dp_rxq = queue;
1235         const struct sfc_dp_rx *dp_rx;
1236
1237         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1238
1239         return offset < dp_rx->qdesc_npending(dp_rxq);
1240 }
1241
1242 /*
1243  * The function is used by the secondary process as well. It must not
1244  * use any process-local pointers from the adapter data.
1245  */
1246 static int
1247 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1248 {
1249         struct sfc_dp_rxq *dp_rxq = queue;
1250         const struct sfc_dp_rx *dp_rx;
1251
1252         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1253
1254         return dp_rx->qdesc_status(dp_rxq, offset);
1255 }
1256
1257 /*
1258  * The function is used by the secondary process as well. It must not
1259  * use any process-local pointers from the adapter data.
1260  */
1261 static int
1262 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1263 {
1264         struct sfc_dp_txq *dp_txq = queue;
1265         const struct sfc_dp_tx *dp_tx;
1266
1267         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1268
1269         return dp_tx->qdesc_status(dp_txq, offset);
1270 }
1271
1272 static int
1273 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1274 {
1275         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1276         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1277         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1278         struct sfc_rxq_info *rxq_info;
1279         sfc_sw_index_t sw_index;
1280         int rc;
1281
1282         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1283
1284         sfc_adapter_lock(sa);
1285
1286         rc = EINVAL;
1287         if (sa->state != SFC_ADAPTER_STARTED)
1288                 goto fail_not_started;
1289
1290         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1291         if (rxq_info->state != SFC_RXQ_INITIALIZED)
1292                 goto fail_not_setup;
1293
1294         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1295         rc = sfc_rx_qstart(sa, sw_index);
1296         if (rc != 0)
1297                 goto fail_rx_qstart;
1298
1299         rxq_info->deferred_started = B_TRUE;
1300
1301         sfc_adapter_unlock(sa);
1302
1303         return 0;
1304
1305 fail_rx_qstart:
1306 fail_not_setup:
1307 fail_not_started:
1308         sfc_adapter_unlock(sa);
1309         SFC_ASSERT(rc > 0);
1310         return -rc;
1311 }
1312
1313 static int
1314 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1315 {
1316         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1317         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1318         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1319         struct sfc_rxq_info *rxq_info;
1320         sfc_sw_index_t sw_index;
1321
1322         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1323
1324         sfc_adapter_lock(sa);
1325
1326         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1327         sfc_rx_qstop(sa, sw_index);
1328
1329         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1330         rxq_info->deferred_started = B_FALSE;
1331
1332         sfc_adapter_unlock(sa);
1333
1334         return 0;
1335 }
1336
1337 static int
1338 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1339 {
1340         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1341         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1342         struct sfc_txq_info *txq_info;
1343         sfc_sw_index_t sw_index;
1344         int rc;
1345
1346         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1347
1348         sfc_adapter_lock(sa);
1349
1350         rc = EINVAL;
1351         if (sa->state != SFC_ADAPTER_STARTED)
1352                 goto fail_not_started;
1353
1354         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1355         if (txq_info->state != SFC_TXQ_INITIALIZED)
1356                 goto fail_not_setup;
1357
1358         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1359         rc = sfc_tx_qstart(sa, sw_index);
1360         if (rc != 0)
1361                 goto fail_tx_qstart;
1362
1363         txq_info->deferred_started = B_TRUE;
1364
1365         sfc_adapter_unlock(sa);
1366         return 0;
1367
1368 fail_tx_qstart:
1369
1370 fail_not_setup:
1371 fail_not_started:
1372         sfc_adapter_unlock(sa);
1373         SFC_ASSERT(rc > 0);
1374         return -rc;
1375 }
1376
1377 static int
1378 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1379 {
1380         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1381         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1382         struct sfc_txq_info *txq_info;
1383         sfc_sw_index_t sw_index;
1384
1385         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1386
1387         sfc_adapter_lock(sa);
1388
1389         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1390         sfc_tx_qstop(sa, sw_index);
1391
1392         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1393         txq_info->deferred_started = B_FALSE;
1394
1395         sfc_adapter_unlock(sa);
1396         return 0;
1397 }
1398
1399 static efx_tunnel_protocol_t
1400 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1401 {
1402         switch (rte_type) {
1403         case RTE_TUNNEL_TYPE_VXLAN:
1404                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1405         case RTE_TUNNEL_TYPE_GENEVE:
1406                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1407         default:
1408                 return EFX_TUNNEL_NPROTOS;
1409         }
1410 }
1411
1412 enum sfc_udp_tunnel_op_e {
1413         SFC_UDP_TUNNEL_ADD_PORT,
1414         SFC_UDP_TUNNEL_DEL_PORT,
1415 };
1416
1417 static int
1418 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1419                       struct rte_eth_udp_tunnel *tunnel_udp,
1420                       enum sfc_udp_tunnel_op_e op)
1421 {
1422         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1423         efx_tunnel_protocol_t tunnel_proto;
1424         int rc;
1425
1426         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1427                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1428                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1429                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1430
1431         tunnel_proto =
1432                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1433         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1434                 rc = ENOTSUP;
1435                 goto fail_bad_proto;
1436         }
1437
1438         sfc_adapter_lock(sa);
1439
1440         switch (op) {
1441         case SFC_UDP_TUNNEL_ADD_PORT:
1442                 rc = efx_tunnel_config_udp_add(sa->nic,
1443                                                tunnel_udp->udp_port,
1444                                                tunnel_proto);
1445                 break;
1446         case SFC_UDP_TUNNEL_DEL_PORT:
1447                 rc = efx_tunnel_config_udp_remove(sa->nic,
1448                                                   tunnel_udp->udp_port,
1449                                                   tunnel_proto);
1450                 break;
1451         default:
1452                 rc = EINVAL;
1453                 goto fail_bad_op;
1454         }
1455
1456         if (rc != 0)
1457                 goto fail_op;
1458
1459         if (sa->state == SFC_ADAPTER_STARTED) {
1460                 rc = efx_tunnel_reconfigure(sa->nic);
1461                 if (rc == EAGAIN) {
1462                         /*
1463                          * Configuration is accepted by FW and MC reboot
1464                          * is initiated to apply the changes. MC reboot
1465                          * will be handled in a usual way (MC reboot
1466                          * event on management event queue and adapter
1467                          * restart).
1468                          */
1469                         rc = 0;
1470                 } else if (rc != 0) {
1471                         goto fail_reconfigure;
1472                 }
1473         }
1474
1475         sfc_adapter_unlock(sa);
1476         return 0;
1477
1478 fail_reconfigure:
1479         /* Remove/restore entry since the change makes the trouble */
1480         switch (op) {
1481         case SFC_UDP_TUNNEL_ADD_PORT:
1482                 (void)efx_tunnel_config_udp_remove(sa->nic,
1483                                                    tunnel_udp->udp_port,
1484                                                    tunnel_proto);
1485                 break;
1486         case SFC_UDP_TUNNEL_DEL_PORT:
1487                 (void)efx_tunnel_config_udp_add(sa->nic,
1488                                                 tunnel_udp->udp_port,
1489                                                 tunnel_proto);
1490                 break;
1491         }
1492
1493 fail_op:
1494 fail_bad_op:
1495         sfc_adapter_unlock(sa);
1496
1497 fail_bad_proto:
1498         SFC_ASSERT(rc > 0);
1499         return -rc;
1500 }
1501
1502 static int
1503 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1504                             struct rte_eth_udp_tunnel *tunnel_udp)
1505 {
1506         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1507 }
1508
1509 static int
1510 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1511                             struct rte_eth_udp_tunnel *tunnel_udp)
1512 {
1513         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1514 }
1515
1516 /*
1517  * The function is used by the secondary process as well. It must not
1518  * use any process-local pointers from the adapter data.
1519  */
1520 static int
1521 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1522                           struct rte_eth_rss_conf *rss_conf)
1523 {
1524         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1525         struct sfc_rss *rss = &sas->rss;
1526
1527         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1528                 return -ENOTSUP;
1529
1530         /*
1531          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1532          * hence, conversion is done here to derive a correct set of ETH_RSS
1533          * flags which corresponds to the active EFX configuration stored
1534          * locally in 'sfc_adapter' and kept up-to-date
1535          */
1536         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1537         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1538         if (rss_conf->rss_key != NULL)
1539                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1540
1541         return 0;
1542 }
1543
1544 static int
1545 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1546                         struct rte_eth_rss_conf *rss_conf)
1547 {
1548         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1549         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1550         unsigned int efx_hash_types;
1551         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1552         unsigned int n_contexts;
1553         unsigned int mode_i = 0;
1554         unsigned int key_i = 0;
1555         unsigned int i = 0;
1556         int rc = 0;
1557
1558         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1559
1560         if (sfc_sa2shared(sa)->isolated)
1561                 return -ENOTSUP;
1562
1563         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1564                 sfc_err(sa, "RSS is not available");
1565                 return -ENOTSUP;
1566         }
1567
1568         if (rss->channels == 0) {
1569                 sfc_err(sa, "RSS is not configured");
1570                 return -EINVAL;
1571         }
1572
1573         if ((rss_conf->rss_key != NULL) &&
1574             (rss_conf->rss_key_len != sizeof(rss->key))) {
1575                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1576                         sizeof(rss->key));
1577                 return -EINVAL;
1578         }
1579
1580         sfc_adapter_lock(sa);
1581
1582         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1583         if (rc != 0)
1584                 goto fail_rx_hf_rte_to_efx;
1585
1586         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1587                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1588                                            rss->hash_alg, efx_hash_types,
1589                                            B_TRUE);
1590                 if (rc != 0)
1591                         goto fail_scale_mode_set;
1592         }
1593
1594         if (rss_conf->rss_key != NULL) {
1595                 if (sa->state == SFC_ADAPTER_STARTED) {
1596                         for (key_i = 0; key_i < n_contexts; key_i++) {
1597                                 rc = efx_rx_scale_key_set(sa->nic,
1598                                                           contexts[key_i],
1599                                                           rss_conf->rss_key,
1600                                                           sizeof(rss->key));
1601                                 if (rc != 0)
1602                                         goto fail_scale_key_set;
1603                         }
1604                 }
1605
1606                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1607         }
1608
1609         rss->hash_types = efx_hash_types;
1610
1611         sfc_adapter_unlock(sa);
1612
1613         return 0;
1614
1615 fail_scale_key_set:
1616         for (i = 0; i < key_i; i++) {
1617                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1618                                          sizeof(rss->key)) != 0)
1619                         sfc_err(sa, "failed to restore RSS key");
1620         }
1621
1622 fail_scale_mode_set:
1623         for (i = 0; i < mode_i; i++) {
1624                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1625                                           EFX_RX_HASHALG_TOEPLITZ,
1626                                           rss->hash_types, B_TRUE) != 0)
1627                         sfc_err(sa, "failed to restore RSS mode");
1628         }
1629
1630 fail_rx_hf_rte_to_efx:
1631         sfc_adapter_unlock(sa);
1632         return -rc;
1633 }
1634
1635 /*
1636  * The function is used by the secondary process as well. It must not
1637  * use any process-local pointers from the adapter data.
1638  */
1639 static int
1640 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1641                        struct rte_eth_rss_reta_entry64 *reta_conf,
1642                        uint16_t reta_size)
1643 {
1644         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1645         struct sfc_rss *rss = &sas->rss;
1646         int entry;
1647
1648         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1649                 return -ENOTSUP;
1650
1651         if (rss->channels == 0)
1652                 return -EINVAL;
1653
1654         if (reta_size != EFX_RSS_TBL_SIZE)
1655                 return -EINVAL;
1656
1657         for (entry = 0; entry < reta_size; entry++) {
1658                 int grp = entry / RTE_RETA_GROUP_SIZE;
1659                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1660
1661                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1662                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1663         }
1664
1665         return 0;
1666 }
1667
1668 static int
1669 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1670                         struct rte_eth_rss_reta_entry64 *reta_conf,
1671                         uint16_t reta_size)
1672 {
1673         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1674         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1675         unsigned int *rss_tbl_new;
1676         uint16_t entry;
1677         int rc = 0;
1678
1679
1680         if (sfc_sa2shared(sa)->isolated)
1681                 return -ENOTSUP;
1682
1683         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1684                 sfc_err(sa, "RSS is not available");
1685                 return -ENOTSUP;
1686         }
1687
1688         if (rss->channels == 0) {
1689                 sfc_err(sa, "RSS is not configured");
1690                 return -EINVAL;
1691         }
1692
1693         if (reta_size != EFX_RSS_TBL_SIZE) {
1694                 sfc_err(sa, "RETA size is wrong (should be %u)",
1695                         EFX_RSS_TBL_SIZE);
1696                 return -EINVAL;
1697         }
1698
1699         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1700         if (rss_tbl_new == NULL)
1701                 return -ENOMEM;
1702
1703         sfc_adapter_lock(sa);
1704
1705         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1706
1707         for (entry = 0; entry < reta_size; entry++) {
1708                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1709                 struct rte_eth_rss_reta_entry64 *grp;
1710
1711                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1712
1713                 if (grp->mask & (1ull << grp_idx)) {
1714                         if (grp->reta[grp_idx] >= rss->channels) {
1715                                 rc = EINVAL;
1716                                 goto bad_reta_entry;
1717                         }
1718                         rss_tbl_new[entry] = grp->reta[grp_idx];
1719                 }
1720         }
1721
1722         if (sa->state == SFC_ADAPTER_STARTED) {
1723                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1724                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1725                 if (rc != 0)
1726                         goto fail_scale_tbl_set;
1727         }
1728
1729         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1730
1731 fail_scale_tbl_set:
1732 bad_reta_entry:
1733         sfc_adapter_unlock(sa);
1734
1735         rte_free(rss_tbl_new);
1736
1737         SFC_ASSERT(rc >= 0);
1738         return -rc;
1739 }
1740
1741 static int
1742 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1743                      const struct rte_flow_ops **ops)
1744 {
1745         *ops = &sfc_flow_ops;
1746         return 0;
1747 }
1748
1749 static int
1750 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1751 {
1752         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1753
1754         /*
1755          * If Rx datapath does not provide callback to check mempool,
1756          * all pools are supported.
1757          */
1758         if (sap->dp_rx->pool_ops_supported == NULL)
1759                 return 1;
1760
1761         return sap->dp_rx->pool_ops_supported(pool);
1762 }
1763
1764 static int
1765 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1766 {
1767         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1768         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1769         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1770         struct sfc_rxq_info *rxq_info;
1771
1772         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1773
1774         return sap->dp_rx->intr_enable(rxq_info->dp);
1775 }
1776
1777 static int
1778 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1779 {
1780         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1781         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1782         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1783         struct sfc_rxq_info *rxq_info;
1784
1785         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1786
1787         return sap->dp_rx->intr_disable(rxq_info->dp);
1788 }
1789
1790 static const struct eth_dev_ops sfc_eth_dev_ops = {
1791         .dev_configure                  = sfc_dev_configure,
1792         .dev_start                      = sfc_dev_start,
1793         .dev_stop                       = sfc_dev_stop,
1794         .dev_set_link_up                = sfc_dev_set_link_up,
1795         .dev_set_link_down              = sfc_dev_set_link_down,
1796         .dev_close                      = sfc_dev_close,
1797         .promiscuous_enable             = sfc_dev_promisc_enable,
1798         .promiscuous_disable            = sfc_dev_promisc_disable,
1799         .allmulticast_enable            = sfc_dev_allmulti_enable,
1800         .allmulticast_disable           = sfc_dev_allmulti_disable,
1801         .link_update                    = sfc_dev_link_update,
1802         .stats_get                      = sfc_stats_get,
1803         .stats_reset                    = sfc_stats_reset,
1804         .xstats_get                     = sfc_xstats_get,
1805         .xstats_reset                   = sfc_stats_reset,
1806         .xstats_get_names               = sfc_xstats_get_names,
1807         .dev_infos_get                  = sfc_dev_infos_get,
1808         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1809         .mtu_set                        = sfc_dev_set_mtu,
1810         .rx_queue_start                 = sfc_rx_queue_start,
1811         .rx_queue_stop                  = sfc_rx_queue_stop,
1812         .tx_queue_start                 = sfc_tx_queue_start,
1813         .tx_queue_stop                  = sfc_tx_queue_stop,
1814         .rx_queue_setup                 = sfc_rx_queue_setup,
1815         .rx_queue_release               = sfc_rx_queue_release,
1816         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1817         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1818         .tx_queue_setup                 = sfc_tx_queue_setup,
1819         .tx_queue_release               = sfc_tx_queue_release,
1820         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1821         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1822         .mac_addr_set                   = sfc_mac_addr_set,
1823         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1824         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1825         .reta_update                    = sfc_dev_rss_reta_update,
1826         .reta_query                     = sfc_dev_rss_reta_query,
1827         .rss_hash_update                = sfc_dev_rss_hash_update,
1828         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1829         .flow_ops_get                   = sfc_dev_flow_ops_get,
1830         .set_mc_addr_list               = sfc_set_mc_addr_list,
1831         .rxq_info_get                   = sfc_rx_queue_info_get,
1832         .txq_info_get                   = sfc_tx_queue_info_get,
1833         .fw_version_get                 = sfc_fw_version_get,
1834         .xstats_get_by_id               = sfc_xstats_get_by_id,
1835         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1836         .pool_ops_supported             = sfc_pool_ops_supported,
1837 };
1838
1839 /**
1840  * Duplicate a string in potentially shared memory required for
1841  * multi-process support.
1842  *
1843  * strdup() allocates from process-local heap/memory.
1844  */
1845 static char *
1846 sfc_strdup(const char *str)
1847 {
1848         size_t size;
1849         char *copy;
1850
1851         if (str == NULL)
1852                 return NULL;
1853
1854         size = strlen(str) + 1;
1855         copy = rte_malloc(__func__, size, 0);
1856         if (copy != NULL)
1857                 rte_memcpy(copy, str, size);
1858
1859         return copy;
1860 }
1861
1862 static int
1863 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1864 {
1865         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1866         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1867         const struct sfc_dp_rx *dp_rx;
1868         const struct sfc_dp_tx *dp_tx;
1869         const efx_nic_cfg_t *encp;
1870         unsigned int avail_caps = 0;
1871         const char *rx_name = NULL;
1872         const char *tx_name = NULL;
1873         int rc;
1874
1875         switch (sa->family) {
1876         case EFX_FAMILY_HUNTINGTON:
1877         case EFX_FAMILY_MEDFORD:
1878         case EFX_FAMILY_MEDFORD2:
1879                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1880                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
1881                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
1882                 break;
1883         case EFX_FAMILY_RIVERHEAD:
1884                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
1885                 break;
1886         default:
1887                 break;
1888         }
1889
1890         encp = efx_nic_cfg_get(sa->nic);
1891         if (encp->enc_rx_es_super_buffer_supported)
1892                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1893
1894         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1895                                 sfc_kvarg_string_handler, &rx_name);
1896         if (rc != 0)
1897                 goto fail_kvarg_rx_datapath;
1898
1899         if (rx_name != NULL) {
1900                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1901                 if (dp_rx == NULL) {
1902                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1903                         rc = ENOENT;
1904                         goto fail_dp_rx;
1905                 }
1906                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1907                         sfc_err(sa,
1908                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1909                                 rx_name);
1910                         rc = EINVAL;
1911                         goto fail_dp_rx_caps;
1912                 }
1913         } else {
1914                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1915                 if (dp_rx == NULL) {
1916                         sfc_err(sa, "Rx datapath by caps %#x not found",
1917                                 avail_caps);
1918                         rc = ENOENT;
1919                         goto fail_dp_rx;
1920                 }
1921         }
1922
1923         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1924         if (sas->dp_rx_name == NULL) {
1925                 rc = ENOMEM;
1926                 goto fail_dp_rx_name;
1927         }
1928
1929         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1930
1931         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1932                                 sfc_kvarg_string_handler, &tx_name);
1933         if (rc != 0)
1934                 goto fail_kvarg_tx_datapath;
1935
1936         if (tx_name != NULL) {
1937                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1938                 if (dp_tx == NULL) {
1939                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1940                         rc = ENOENT;
1941                         goto fail_dp_tx;
1942                 }
1943                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1944                         sfc_err(sa,
1945                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1946                                 tx_name);
1947                         rc = EINVAL;
1948                         goto fail_dp_tx_caps;
1949                 }
1950         } else {
1951                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1952                 if (dp_tx == NULL) {
1953                         sfc_err(sa, "Tx datapath by caps %#x not found",
1954                                 avail_caps);
1955                         rc = ENOENT;
1956                         goto fail_dp_tx;
1957                 }
1958         }
1959
1960         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1961         if (sas->dp_tx_name == NULL) {
1962                 rc = ENOMEM;
1963                 goto fail_dp_tx_name;
1964         }
1965
1966         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1967
1968         sa->priv.dp_rx = dp_rx;
1969         sa->priv.dp_tx = dp_tx;
1970
1971         dev->rx_pkt_burst = dp_rx->pkt_burst;
1972         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1973         dev->tx_pkt_burst = dp_tx->pkt_burst;
1974
1975         dev->rx_queue_count = sfc_rx_queue_count;
1976         dev->rx_descriptor_done = sfc_rx_descriptor_done;
1977         dev->rx_descriptor_status = sfc_rx_descriptor_status;
1978         dev->tx_descriptor_status = sfc_tx_descriptor_status;
1979         dev->dev_ops = &sfc_eth_dev_ops;
1980
1981         return 0;
1982
1983 fail_dp_tx_name:
1984 fail_dp_tx_caps:
1985 fail_dp_tx:
1986 fail_kvarg_tx_datapath:
1987         rte_free(sas->dp_rx_name);
1988         sas->dp_rx_name = NULL;
1989
1990 fail_dp_rx_name:
1991 fail_dp_rx_caps:
1992 fail_dp_rx:
1993 fail_kvarg_rx_datapath:
1994         return rc;
1995 }
1996
1997 static void
1998 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1999 {
2000         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2001         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2002
2003         dev->dev_ops = NULL;
2004         dev->tx_pkt_prepare = NULL;
2005         dev->rx_pkt_burst = NULL;
2006         dev->tx_pkt_burst = NULL;
2007
2008         rte_free(sas->dp_tx_name);
2009         sas->dp_tx_name = NULL;
2010         sa->priv.dp_tx = NULL;
2011
2012         rte_free(sas->dp_rx_name);
2013         sas->dp_rx_name = NULL;
2014         sa->priv.dp_rx = NULL;
2015 }
2016
2017 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2018         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2019         .reta_query                     = sfc_dev_rss_reta_query,
2020         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2021         .rxq_info_get                   = sfc_rx_queue_info_get,
2022         .txq_info_get                   = sfc_tx_queue_info_get,
2023 };
2024
2025 static int
2026 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2027 {
2028         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2029         struct sfc_adapter_priv *sap;
2030         const struct sfc_dp_rx *dp_rx;
2031         const struct sfc_dp_tx *dp_tx;
2032         int rc;
2033
2034         /*
2035          * Allocate process private data from heap, since it should not
2036          * be located in shared memory allocated using rte_malloc() API.
2037          */
2038         sap = calloc(1, sizeof(*sap));
2039         if (sap == NULL) {
2040                 rc = ENOMEM;
2041                 goto fail_alloc_priv;
2042         }
2043
2044         sap->logtype_main = logtype_main;
2045
2046         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2047         if (dp_rx == NULL) {
2048                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2049                         "cannot find %s Rx datapath", sas->dp_rx_name);
2050                 rc = ENOENT;
2051                 goto fail_dp_rx;
2052         }
2053         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2054                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2055                         "%s Rx datapath does not support multi-process",
2056                         sas->dp_rx_name);
2057                 rc = EINVAL;
2058                 goto fail_dp_rx_multi_process;
2059         }
2060
2061         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2062         if (dp_tx == NULL) {
2063                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2064                         "cannot find %s Tx datapath", sas->dp_tx_name);
2065                 rc = ENOENT;
2066                 goto fail_dp_tx;
2067         }
2068         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2069                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2070                         "%s Tx datapath does not support multi-process",
2071                         sas->dp_tx_name);
2072                 rc = EINVAL;
2073                 goto fail_dp_tx_multi_process;
2074         }
2075
2076         sap->dp_rx = dp_rx;
2077         sap->dp_tx = dp_tx;
2078
2079         dev->process_private = sap;
2080         dev->rx_pkt_burst = dp_rx->pkt_burst;
2081         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2082         dev->tx_pkt_burst = dp_tx->pkt_burst;
2083         dev->rx_queue_count = sfc_rx_queue_count;
2084         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2085         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2086         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2087         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2088
2089         return 0;
2090
2091 fail_dp_tx_multi_process:
2092 fail_dp_tx:
2093 fail_dp_rx_multi_process:
2094 fail_dp_rx:
2095         free(sap);
2096
2097 fail_alloc_priv:
2098         return rc;
2099 }
2100
2101 static void
2102 sfc_register_dp(void)
2103 {
2104         /* Register once */
2105         if (TAILQ_EMPTY(&sfc_dp_head)) {
2106                 /* Prefer EF10 datapath */
2107                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2108                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2109                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2110                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2111
2112                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2113                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2114                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2115                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2116         }
2117 }
2118
2119 static int
2120 sfc_eth_dev_init(struct rte_eth_dev *dev)
2121 {
2122         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2123         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2124         uint32_t logtype_main;
2125         struct sfc_adapter *sa;
2126         int rc;
2127         const efx_nic_cfg_t *encp;
2128         const struct rte_ether_addr *from;
2129         int ret;
2130
2131         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2132                         SFC_EFX_DEV_CLASS_NET) {
2133                 SFC_GENERIC_LOG(DEBUG,
2134                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2135                 return 1;
2136         }
2137
2138         sfc_register_dp();
2139
2140         logtype_main = sfc_register_logtype(&pci_dev->addr,
2141                                             SFC_LOGTYPE_MAIN_STR,
2142                                             RTE_LOG_NOTICE);
2143
2144         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2145                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2146
2147         /* Required for logging */
2148         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2149                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2150                         pci_dev->addr.domain, pci_dev->addr.bus,
2151                         pci_dev->addr.devid, pci_dev->addr.function,
2152                         dev->data->port_id);
2153         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2154                 SFC_GENERIC_LOG(ERR,
2155                         "reserved log prefix is too short for " PCI_PRI_FMT,
2156                         pci_dev->addr.domain, pci_dev->addr.bus,
2157                         pci_dev->addr.devid, pci_dev->addr.function);
2158                 return -EINVAL;
2159         }
2160         sas->pci_addr = pci_dev->addr;
2161         sas->port_id = dev->data->port_id;
2162
2163         /*
2164          * Allocate process private data from heap, since it should not
2165          * be located in shared memory allocated using rte_malloc() API.
2166          */
2167         sa = calloc(1, sizeof(*sa));
2168         if (sa == NULL) {
2169                 rc = ENOMEM;
2170                 goto fail_alloc_sa;
2171         }
2172
2173         dev->process_private = sa;
2174
2175         /* Required for logging */
2176         sa->priv.shared = sas;
2177         sa->priv.logtype_main = logtype_main;
2178
2179         sa->eth_dev = dev;
2180
2181         /* Copy PCI device info to the dev->data */
2182         rte_eth_copy_pci_info(dev, pci_dev);
2183         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2184         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2185
2186         rc = sfc_kvargs_parse(sa);
2187         if (rc != 0)
2188                 goto fail_kvargs_parse;
2189
2190         sfc_log_init(sa, "entry");
2191
2192         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2193         if (dev->data->mac_addrs == NULL) {
2194                 rc = ENOMEM;
2195                 goto fail_mac_addrs;
2196         }
2197
2198         sfc_adapter_lock_init(sa);
2199         sfc_adapter_lock(sa);
2200
2201         sfc_log_init(sa, "probing");
2202         rc = sfc_probe(sa);
2203         if (rc != 0)
2204                 goto fail_probe;
2205
2206         sfc_log_init(sa, "set device ops");
2207         rc = sfc_eth_dev_set_ops(dev);
2208         if (rc != 0)
2209                 goto fail_set_ops;
2210
2211         sfc_log_init(sa, "attaching");
2212         rc = sfc_attach(sa);
2213         if (rc != 0)
2214                 goto fail_attach;
2215
2216         encp = efx_nic_cfg_get(sa->nic);
2217
2218         /*
2219          * The arguments are really reverse order in comparison to
2220          * Linux kernel. Copy from NIC config to Ethernet device data.
2221          */
2222         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2223         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2224
2225         sfc_adapter_unlock(sa);
2226
2227         sfc_log_init(sa, "done");
2228         return 0;
2229
2230 fail_attach:
2231         sfc_eth_dev_clear_ops(dev);
2232
2233 fail_set_ops:
2234         sfc_unprobe(sa);
2235
2236 fail_probe:
2237         sfc_adapter_unlock(sa);
2238         sfc_adapter_lock_fini(sa);
2239         rte_free(dev->data->mac_addrs);
2240         dev->data->mac_addrs = NULL;
2241
2242 fail_mac_addrs:
2243         sfc_kvargs_cleanup(sa);
2244
2245 fail_kvargs_parse:
2246         sfc_log_init(sa, "failed %d", rc);
2247         dev->process_private = NULL;
2248         free(sa);
2249
2250 fail_alloc_sa:
2251         SFC_ASSERT(rc > 0);
2252         return -rc;
2253 }
2254
2255 static int
2256 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2257 {
2258         sfc_dev_close(dev);
2259
2260         return 0;
2261 }
2262
2263 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2264         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2265         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2266         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2267         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2268         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2269         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2270         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2271         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2272         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2273         { .vendor_id = 0 /* sentinel */ }
2274 };
2275
2276 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2277         struct rte_pci_device *pci_dev)
2278 {
2279         return rte_eth_dev_pci_generic_probe(pci_dev,
2280                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2281 }
2282
2283 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2284 {
2285         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2286 }
2287
2288 static struct rte_pci_driver sfc_efx_pmd = {
2289         .id_table = pci_id_sfc_efx_map,
2290         .drv_flags =
2291                 RTE_PCI_DRV_INTR_LSC |
2292                 RTE_PCI_DRV_NEED_MAPPING,
2293         .probe = sfc_eth_dev_pci_probe,
2294         .remove = sfc_eth_dev_pci_remove,
2295 };
2296
2297 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2298 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2299 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2300 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2301         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2302         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2303         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2304         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2305         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2306         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2307
2308 RTE_INIT(sfc_driver_register_logtype)
2309 {
2310         int ret;
2311
2312         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2313                                                    RTE_LOG_NOTICE);
2314         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2315 }