a7322a1eb3f8540f2d5a3a700360c0e84ee16b1e
[dpdk.git] / drivers / net / sfc / sfc_ethdev.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
4  * All rights reserved.
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 <rte_ethdev_driver.h>
12 #include <rte_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
18 #include "efx.h"
19
20 #include "sfc.h"
21 #include "sfc_debug.h"
22 #include "sfc_log.h"
23 #include "sfc_kvargs.h"
24 #include "sfc_ev.h"
25 #include "sfc_rx.h"
26 #include "sfc_tx.h"
27 #include "sfc_flow.h"
28 #include "sfc_dp.h"
29 #include "sfc_dp_rx.h"
30
31 uint32_t sfc_logtype_driver;
32
33 static struct sfc_dp_list sfc_dp_head =
34         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
35
36 static int
37 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
38 {
39         struct sfc_adapter *sa = dev->data->dev_private;
40         efx_nic_fw_info_t enfi;
41         int ret;
42         int rc;
43
44         /*
45          * Return value of the callback is likely supposed to be
46          * equal to or greater than 0, nevertheless, if an error
47          * occurs, it will be desirable to pass it to the caller
48          */
49         if ((fw_version == NULL) || (fw_size == 0))
50                 return -EINVAL;
51
52         rc = efx_nic_get_fw_version(sa->nic, &enfi);
53         if (rc != 0)
54                 return -rc;
55
56         ret = snprintf(fw_version, fw_size,
57                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
58                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
59                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
60         if (ret < 0)
61                 return ret;
62
63         if (enfi.enfi_dpcpu_fw_ids_valid) {
64                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
65                 int ret_extra;
66
67                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
68                                      fw_size - dpcpu_fw_ids_offset,
69                                      " rx%" PRIx16 " tx%" PRIx16,
70                                      enfi.enfi_rx_dpcpu_fw_id,
71                                      enfi.enfi_tx_dpcpu_fw_id);
72                 if (ret_extra < 0)
73                         return ret_extra;
74
75                 ret += ret_extra;
76         }
77
78         if (fw_size < (size_t)(++ret))
79                 return ret;
80         else
81                 return 0;
82 }
83
84 static void
85 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
86 {
87         struct sfc_adapter *sa = dev->data->dev_private;
88         struct sfc_rss *rss = &sa->rss;
89         uint64_t txq_offloads_def = 0;
90
91         sfc_log_init(sa, "entry");
92
93         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
94
95         /* Autonegotiation may be disabled */
96         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
97         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
98                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
99         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
100                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
101         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
102                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
103         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
104                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
105         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
106                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
107         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
108                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
109
110         dev_info->max_rx_queues = sa->rxq_max;
111         dev_info->max_tx_queues = sa->txq_max;
112
113         /* By default packets are dropped if no descriptors are available */
114         dev_info->default_rxconf.rx_drop_en = 1;
115
116         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
117
118         /*
119          * rx_offload_capa includes both device and queue offloads since
120          * the latter may be requested on a per device basis which makes
121          * sense when some offloads are needed to be set on all queues.
122          */
123         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
124                                     dev_info->rx_queue_offload_capa;
125
126         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
127
128         /*
129          * tx_offload_capa includes both device and queue offloads since
130          * the latter may be requested on a per device basis which makes
131          * sense when some offloads are needed to be set on all queues.
132          */
133         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
134                                     dev_info->tx_queue_offload_capa;
135
136         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
137                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
138
139         dev_info->default_txconf.offloads |= txq_offloads_def;
140
141         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
142                 uint64_t rte_hf = 0;
143                 unsigned int i;
144
145                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
146                         rte_hf |= rss->hf_map[i].rte;
147
148                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
149                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
150                 dev_info->flow_type_rss_offloads = rte_hf;
151         }
152
153         /* Initialize to hardware limits */
154         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
155         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
156         /* The RXQ hardware requires that the descriptor count is a power
157          * of 2, but rx_desc_lim cannot properly describe that constraint.
158          */
159         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
160
161         /* Initialize to hardware limits */
162         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
163         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
164         /*
165          * The TXQ hardware requires that the descriptor count is a power
166          * of 2, but tx_desc_lim cannot properly describe that constraint
167          */
168         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
169
170         if (sa->dp_rx->get_dev_info != NULL)
171                 sa->dp_rx->get_dev_info(dev_info);
172         if (sa->dp_tx->get_dev_info != NULL)
173                 sa->dp_tx->get_dev_info(dev_info);
174
175         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
176                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
177 }
178
179 static const uint32_t *
180 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
181 {
182         struct sfc_adapter *sa = dev->data->dev_private;
183         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
184         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
185
186         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
187 }
188
189 static int
190 sfc_dev_configure(struct rte_eth_dev *dev)
191 {
192         struct rte_eth_dev_data *dev_data = dev->data;
193         struct sfc_adapter *sa = dev_data->dev_private;
194         int rc;
195
196         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
197                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
198
199         sfc_adapter_lock(sa);
200         switch (sa->state) {
201         case SFC_ADAPTER_CONFIGURED:
202                 /* FALLTHROUGH */
203         case SFC_ADAPTER_INITIALIZED:
204                 rc = sfc_configure(sa);
205                 break;
206         default:
207                 sfc_err(sa, "unexpected adapter state %u to configure",
208                         sa->state);
209                 rc = EINVAL;
210                 break;
211         }
212         sfc_adapter_unlock(sa);
213
214         sfc_log_init(sa, "done %d", rc);
215         SFC_ASSERT(rc >= 0);
216         return -rc;
217 }
218
219 static int
220 sfc_dev_start(struct rte_eth_dev *dev)
221 {
222         struct sfc_adapter *sa = dev->data->dev_private;
223         int rc;
224
225         sfc_log_init(sa, "entry");
226
227         sfc_adapter_lock(sa);
228         rc = sfc_start(sa);
229         sfc_adapter_unlock(sa);
230
231         sfc_log_init(sa, "done %d", rc);
232         SFC_ASSERT(rc >= 0);
233         return -rc;
234 }
235
236 static int
237 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
238 {
239         struct sfc_adapter *sa = dev->data->dev_private;
240         struct rte_eth_link current_link;
241         int ret;
242
243         sfc_log_init(sa, "entry");
244
245         if (sa->state != SFC_ADAPTER_STARTED) {
246                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
247         } else if (wait_to_complete) {
248                 efx_link_mode_t link_mode;
249
250                 if (efx_port_poll(sa->nic, &link_mode) != 0)
251                         link_mode = EFX_LINK_UNKNOWN;
252                 sfc_port_link_mode_to_info(link_mode, &current_link);
253
254         } else {
255                 sfc_ev_mgmt_qpoll(sa);
256                 rte_eth_linkstatus_get(dev, &current_link);
257         }
258
259         ret = rte_eth_linkstatus_set(dev, &current_link);
260         if (ret == 0)
261                 sfc_notice(sa, "Link status is %s",
262                            current_link.link_status ? "UP" : "DOWN");
263
264         return ret;
265 }
266
267 static void
268 sfc_dev_stop(struct rte_eth_dev *dev)
269 {
270         struct sfc_adapter *sa = dev->data->dev_private;
271
272         sfc_log_init(sa, "entry");
273
274         sfc_adapter_lock(sa);
275         sfc_stop(sa);
276         sfc_adapter_unlock(sa);
277
278         sfc_log_init(sa, "done");
279 }
280
281 static int
282 sfc_dev_set_link_up(struct rte_eth_dev *dev)
283 {
284         struct sfc_adapter *sa = dev->data->dev_private;
285         int rc;
286
287         sfc_log_init(sa, "entry");
288
289         sfc_adapter_lock(sa);
290         rc = sfc_start(sa);
291         sfc_adapter_unlock(sa);
292
293         SFC_ASSERT(rc >= 0);
294         return -rc;
295 }
296
297 static int
298 sfc_dev_set_link_down(struct rte_eth_dev *dev)
299 {
300         struct sfc_adapter *sa = dev->data->dev_private;
301
302         sfc_log_init(sa, "entry");
303
304         sfc_adapter_lock(sa);
305         sfc_stop(sa);
306         sfc_adapter_unlock(sa);
307
308         return 0;
309 }
310
311 static void
312 sfc_dev_close(struct rte_eth_dev *dev)
313 {
314         struct sfc_adapter *sa = dev->data->dev_private;
315
316         sfc_log_init(sa, "entry");
317
318         sfc_adapter_lock(sa);
319         switch (sa->state) {
320         case SFC_ADAPTER_STARTED:
321                 sfc_stop(sa);
322                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
323                 /* FALLTHROUGH */
324         case SFC_ADAPTER_CONFIGURED:
325                 sfc_close(sa);
326                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
327                 /* FALLTHROUGH */
328         case SFC_ADAPTER_INITIALIZED:
329                 break;
330         default:
331                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
332                 break;
333         }
334         sfc_adapter_unlock(sa);
335
336         sfc_log_init(sa, "done");
337 }
338
339 static void
340 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
341                    boolean_t enabled)
342 {
343         struct sfc_port *port;
344         boolean_t *toggle;
345         struct sfc_adapter *sa = dev->data->dev_private;
346         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
347         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
348
349         sfc_adapter_lock(sa);
350
351         port = &sa->port;
352         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
353
354         if (*toggle != enabled) {
355                 *toggle = enabled;
356
357                 if (port->isolated) {
358                         sfc_warn(sa, "isolated mode is active on the port");
359                         sfc_warn(sa, "the change is to be applied on the next "
360                                      "start provided that isolated mode is "
361                                      "disabled prior the next start");
362                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
363                            (sfc_set_rx_mode(sa) != 0)) {
364                         *toggle = !(enabled);
365                         sfc_warn(sa, "Failed to %s %s mode",
366                                  ((enabled) ? "enable" : "disable"), desc);
367                 }
368         }
369
370         sfc_adapter_unlock(sa);
371 }
372
373 static void
374 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
375 {
376         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
377 }
378
379 static void
380 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
381 {
382         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
383 }
384
385 static void
386 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
387 {
388         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
389 }
390
391 static void
392 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
393 {
394         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
395 }
396
397 static int
398 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
399                    uint16_t nb_rx_desc, unsigned int socket_id,
400                    const struct rte_eth_rxconf *rx_conf,
401                    struct rte_mempool *mb_pool)
402 {
403         struct sfc_adapter *sa = dev->data->dev_private;
404         int rc;
405
406         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
407                      rx_queue_id, nb_rx_desc, socket_id);
408
409         sfc_adapter_lock(sa);
410
411         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
412                           rx_conf, mb_pool);
413         if (rc != 0)
414                 goto fail_rx_qinit;
415
416         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
417
418         sfc_adapter_unlock(sa);
419
420         return 0;
421
422 fail_rx_qinit:
423         sfc_adapter_unlock(sa);
424         SFC_ASSERT(rc > 0);
425         return -rc;
426 }
427
428 static void
429 sfc_rx_queue_release(void *queue)
430 {
431         struct sfc_dp_rxq *dp_rxq = queue;
432         struct sfc_rxq *rxq;
433         struct sfc_adapter *sa;
434         unsigned int sw_index;
435
436         if (dp_rxq == NULL)
437                 return;
438
439         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
440         sa = rxq->evq->sa;
441         sfc_adapter_lock(sa);
442
443         sw_index = sfc_rxq_sw_index(rxq);
444
445         sfc_log_init(sa, "RxQ=%u", sw_index);
446
447         sfc_rx_qfini(sa, sw_index);
448
449         sfc_adapter_unlock(sa);
450 }
451
452 static int
453 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
454                    uint16_t nb_tx_desc, unsigned int socket_id,
455                    const struct rte_eth_txconf *tx_conf)
456 {
457         struct sfc_adapter *sa = dev->data->dev_private;
458         int rc;
459
460         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
461                      tx_queue_id, nb_tx_desc, socket_id);
462
463         sfc_adapter_lock(sa);
464
465         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
466         if (rc != 0)
467                 goto fail_tx_qinit;
468
469         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
470
471         sfc_adapter_unlock(sa);
472         return 0;
473
474 fail_tx_qinit:
475         sfc_adapter_unlock(sa);
476         SFC_ASSERT(rc > 0);
477         return -rc;
478 }
479
480 static void
481 sfc_tx_queue_release(void *queue)
482 {
483         struct sfc_dp_txq *dp_txq = queue;
484         struct sfc_txq *txq;
485         unsigned int sw_index;
486         struct sfc_adapter *sa;
487
488         if (dp_txq == NULL)
489                 return;
490
491         txq = sfc_txq_by_dp_txq(dp_txq);
492         sw_index = sfc_txq_sw_index(txq);
493
494         SFC_ASSERT(txq->evq != NULL);
495         sa = txq->evq->sa;
496
497         sfc_log_init(sa, "TxQ = %u", sw_index);
498
499         sfc_adapter_lock(sa);
500
501         sfc_tx_qfini(sa, sw_index);
502
503         sfc_adapter_unlock(sa);
504 }
505
506 /*
507  * Some statistics are computed as A - B where A and B each increase
508  * monotonically with some hardware counter(s) and the counters are read
509  * asynchronously.
510  *
511  * If packet X is counted in A, but not counted in B yet, computed value is
512  * greater than real.
513  *
514  * If packet X is not counted in A at the moment of reading the counter,
515  * but counted in B at the moment of reading the counter, computed value
516  * is less than real.
517  *
518  * However, counter which grows backward is worse evil than slightly wrong
519  * value. So, let's try to guarantee that it never happens except may be
520  * the case when the MAC stats are zeroed as a result of a NIC reset.
521  */
522 static void
523 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
524 {
525         if ((int64_t)(newval - *stat) > 0 || newval == 0)
526                 *stat = newval;
527 }
528
529 static int
530 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
531 {
532         struct sfc_adapter *sa = dev->data->dev_private;
533         struct sfc_port *port = &sa->port;
534         uint64_t *mac_stats;
535         int ret;
536
537         rte_spinlock_lock(&port->mac_stats_lock);
538
539         ret = sfc_port_update_mac_stats(sa);
540         if (ret != 0)
541                 goto unlock;
542
543         mac_stats = port->mac_stats_buf;
544
545         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
546                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
547                 stats->ipackets =
548                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
549                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
550                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
551                 stats->opackets =
552                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
553                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
554                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
555                 stats->ibytes =
556                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
557                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
558                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
559                 stats->obytes =
560                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
561                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
562                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
563                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
564                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
565         } else {
566                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
567                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
568                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
569                 /*
570                  * Take into account stats which are whenever supported
571                  * on EF10. If some stat is not supported by current
572                  * firmware variant or HW revision, it is guaranteed
573                  * to be zero in mac_stats.
574                  */
575                 stats->imissed =
576                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
577                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
578                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
579                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
580                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
581                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
582                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
583                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
584                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
585                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
586                 stats->ierrors =
587                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
588                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
589                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
590                 /* no oerrors counters supported on EF10 */
591
592                 /* Exclude missed, errors and pauses from Rx packets */
593                 sfc_update_diff_stat(&port->ipackets,
594                         mac_stats[EFX_MAC_RX_PKTS] -
595                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
596                         stats->imissed - stats->ierrors);
597                 stats->ipackets = port->ipackets;
598         }
599
600 unlock:
601         rte_spinlock_unlock(&port->mac_stats_lock);
602         SFC_ASSERT(ret >= 0);
603         return -ret;
604 }
605
606 static void
607 sfc_stats_reset(struct rte_eth_dev *dev)
608 {
609         struct sfc_adapter *sa = dev->data->dev_private;
610         struct sfc_port *port = &sa->port;
611         int rc;
612
613         if (sa->state != SFC_ADAPTER_STARTED) {
614                 /*
615                  * The operation cannot be done if port is not started; it
616                  * will be scheduled to be done during the next port start
617                  */
618                 port->mac_stats_reset_pending = B_TRUE;
619                 return;
620         }
621
622         rc = sfc_port_reset_mac_stats(sa);
623         if (rc != 0)
624                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
625 }
626
627 static int
628 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
629                unsigned int xstats_count)
630 {
631         struct sfc_adapter *sa = dev->data->dev_private;
632         struct sfc_port *port = &sa->port;
633         uint64_t *mac_stats;
634         int rc;
635         unsigned int i;
636         int nstats = 0;
637
638         rte_spinlock_lock(&port->mac_stats_lock);
639
640         rc = sfc_port_update_mac_stats(sa);
641         if (rc != 0) {
642                 SFC_ASSERT(rc > 0);
643                 nstats = -rc;
644                 goto unlock;
645         }
646
647         mac_stats = port->mac_stats_buf;
648
649         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
650                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
651                         if (xstats != NULL && nstats < (int)xstats_count) {
652                                 xstats[nstats].id = nstats;
653                                 xstats[nstats].value = mac_stats[i];
654                         }
655                         nstats++;
656                 }
657         }
658
659 unlock:
660         rte_spinlock_unlock(&port->mac_stats_lock);
661
662         return nstats;
663 }
664
665 static int
666 sfc_xstats_get_names(struct rte_eth_dev *dev,
667                      struct rte_eth_xstat_name *xstats_names,
668                      unsigned int xstats_count)
669 {
670         struct sfc_adapter *sa = dev->data->dev_private;
671         struct sfc_port *port = &sa->port;
672         unsigned int i;
673         unsigned int nstats = 0;
674
675         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
676                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
677                         if (xstats_names != NULL && nstats < xstats_count)
678                                 strlcpy(xstats_names[nstats].name,
679                                         efx_mac_stat_name(sa->nic, i),
680                                         sizeof(xstats_names[0].name));
681                         nstats++;
682                 }
683         }
684
685         return nstats;
686 }
687
688 static int
689 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
690                      uint64_t *values, unsigned int n)
691 {
692         struct sfc_adapter *sa = dev->data->dev_private;
693         struct sfc_port *port = &sa->port;
694         uint64_t *mac_stats;
695         unsigned int nb_supported = 0;
696         unsigned int nb_written = 0;
697         unsigned int i;
698         int ret;
699         int rc;
700
701         if (unlikely(values == NULL) ||
702             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
703                 return port->mac_stats_nb_supported;
704
705         rte_spinlock_lock(&port->mac_stats_lock);
706
707         rc = sfc_port_update_mac_stats(sa);
708         if (rc != 0) {
709                 SFC_ASSERT(rc > 0);
710                 ret = -rc;
711                 goto unlock;
712         }
713
714         mac_stats = port->mac_stats_buf;
715
716         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
717                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
718                         continue;
719
720                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
721                         values[nb_written++] = mac_stats[i];
722
723                 ++nb_supported;
724         }
725
726         ret = nb_written;
727
728 unlock:
729         rte_spinlock_unlock(&port->mac_stats_lock);
730
731         return ret;
732 }
733
734 static int
735 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
736                            struct rte_eth_xstat_name *xstats_names,
737                            const uint64_t *ids, unsigned int size)
738 {
739         struct sfc_adapter *sa = dev->data->dev_private;
740         struct sfc_port *port = &sa->port;
741         unsigned int nb_supported = 0;
742         unsigned int nb_written = 0;
743         unsigned int i;
744
745         if (unlikely(xstats_names == NULL) ||
746             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
747                 return port->mac_stats_nb_supported;
748
749         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
750                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
751                         continue;
752
753                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
754                         char *name = xstats_names[nb_written++].name;
755
756                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
757                                 sizeof(xstats_names[0].name));
758                 }
759
760                 ++nb_supported;
761         }
762
763         return nb_written;
764 }
765
766 static int
767 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
768 {
769         struct sfc_adapter *sa = dev->data->dev_private;
770         unsigned int wanted_fc, link_fc;
771
772         memset(fc_conf, 0, sizeof(*fc_conf));
773
774         sfc_adapter_lock(sa);
775
776         if (sa->state == SFC_ADAPTER_STARTED)
777                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
778         else
779                 link_fc = sa->port.flow_ctrl;
780
781         switch (link_fc) {
782         case 0:
783                 fc_conf->mode = RTE_FC_NONE;
784                 break;
785         case EFX_FCNTL_RESPOND:
786                 fc_conf->mode = RTE_FC_RX_PAUSE;
787                 break;
788         case EFX_FCNTL_GENERATE:
789                 fc_conf->mode = RTE_FC_TX_PAUSE;
790                 break;
791         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
792                 fc_conf->mode = RTE_FC_FULL;
793                 break;
794         default:
795                 sfc_err(sa, "%s: unexpected flow control value %#x",
796                         __func__, link_fc);
797         }
798
799         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
800
801         sfc_adapter_unlock(sa);
802
803         return 0;
804 }
805
806 static int
807 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
808 {
809         struct sfc_adapter *sa = dev->data->dev_private;
810         struct sfc_port *port = &sa->port;
811         unsigned int fcntl;
812         int rc;
813
814         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
815             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
816             fc_conf->mac_ctrl_frame_fwd != 0) {
817                 sfc_err(sa, "unsupported flow control settings specified");
818                 rc = EINVAL;
819                 goto fail_inval;
820         }
821
822         switch (fc_conf->mode) {
823         case RTE_FC_NONE:
824                 fcntl = 0;
825                 break;
826         case RTE_FC_RX_PAUSE:
827                 fcntl = EFX_FCNTL_RESPOND;
828                 break;
829         case RTE_FC_TX_PAUSE:
830                 fcntl = EFX_FCNTL_GENERATE;
831                 break;
832         case RTE_FC_FULL:
833                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
834                 break;
835         default:
836                 rc = EINVAL;
837                 goto fail_inval;
838         }
839
840         sfc_adapter_lock(sa);
841
842         if (sa->state == SFC_ADAPTER_STARTED) {
843                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
844                 if (rc != 0)
845                         goto fail_mac_fcntl_set;
846         }
847
848         port->flow_ctrl = fcntl;
849         port->flow_ctrl_autoneg = fc_conf->autoneg;
850
851         sfc_adapter_unlock(sa);
852
853         return 0;
854
855 fail_mac_fcntl_set:
856         sfc_adapter_unlock(sa);
857 fail_inval:
858         SFC_ASSERT(rc > 0);
859         return -rc;
860 }
861
862 static int
863 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
864 {
865         struct sfc_adapter *sa = dev->data->dev_private;
866         size_t pdu = EFX_MAC_PDU(mtu);
867         size_t old_pdu;
868         int rc;
869
870         sfc_log_init(sa, "mtu=%u", mtu);
871
872         rc = EINVAL;
873         if (pdu < EFX_MAC_PDU_MIN) {
874                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
875                         (unsigned int)mtu, (unsigned int)pdu,
876                         EFX_MAC_PDU_MIN);
877                 goto fail_inval;
878         }
879         if (pdu > EFX_MAC_PDU_MAX) {
880                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
881                         (unsigned int)mtu, (unsigned int)pdu,
882                         EFX_MAC_PDU_MAX);
883                 goto fail_inval;
884         }
885
886         sfc_adapter_lock(sa);
887
888         if (pdu != sa->port.pdu) {
889                 if (sa->state == SFC_ADAPTER_STARTED) {
890                         sfc_stop(sa);
891
892                         old_pdu = sa->port.pdu;
893                         sa->port.pdu = pdu;
894                         rc = sfc_start(sa);
895                         if (rc != 0)
896                                 goto fail_start;
897                 } else {
898                         sa->port.pdu = pdu;
899                 }
900         }
901
902         /*
903          * The driver does not use it, but other PMDs update jumbo frame
904          * flag and max_rx_pkt_len when MTU is set.
905          */
906         if (mtu > ETHER_MAX_LEN) {
907                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
908                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
909         }
910
911         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
912
913         sfc_adapter_unlock(sa);
914
915         sfc_log_init(sa, "done");
916         return 0;
917
918 fail_start:
919         sa->port.pdu = old_pdu;
920         if (sfc_start(sa) != 0)
921                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
922                         "PDU max size - port is stopped",
923                         (unsigned int)pdu, (unsigned int)old_pdu);
924         sfc_adapter_unlock(sa);
925
926 fail_inval:
927         sfc_log_init(sa, "failed %d", rc);
928         SFC_ASSERT(rc > 0);
929         return -rc;
930 }
931 static int
932 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
933 {
934         struct sfc_adapter *sa = dev->data->dev_private;
935         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
936         struct sfc_port *port = &sa->port;
937         struct ether_addr *old_addr = &dev->data->mac_addrs[0];
938         int rc = 0;
939
940         sfc_adapter_lock(sa);
941
942         /*
943          * Copy the address to the device private data so that
944          * it could be recalled in the case of adapter restart.
945          */
946         ether_addr_copy(mac_addr, &port->default_mac_addr);
947
948         /*
949          * Neither of the two following checks can return
950          * an error. The new MAC address is preserved in
951          * the device private data and can be activated
952          * on the next port start if the user prevents
953          * isolated mode from being enabled.
954          */
955         if (port->isolated) {
956                 sfc_warn(sa, "isolated mode is active on the port");
957                 sfc_warn(sa, "will not set MAC address");
958                 goto unlock;
959         }
960
961         if (sa->state != SFC_ADAPTER_STARTED) {
962                 sfc_notice(sa, "the port is not started");
963                 sfc_notice(sa, "the new MAC address will be set on port start");
964
965                 goto unlock;
966         }
967
968         if (encp->enc_allow_set_mac_with_installed_filters) {
969                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
970                 if (rc != 0) {
971                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
972                         goto unlock;
973                 }
974
975                 /*
976                  * Changing the MAC address by means of MCDI request
977                  * has no effect on received traffic, therefore
978                  * we also need to update unicast filters
979                  */
980                 rc = sfc_set_rx_mode(sa);
981                 if (rc != 0) {
982                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
983                         /* Rollback the old address */
984                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
985                         (void)sfc_set_rx_mode(sa);
986                 }
987         } else {
988                 sfc_warn(sa, "cannot set MAC address with filters installed");
989                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
990                 sfc_warn(sa, "(some traffic may be dropped)");
991
992                 /*
993                  * Since setting MAC address with filters installed is not
994                  * allowed on the adapter, the new MAC address will be set
995                  * by means of adapter restart. sfc_start() shall retrieve
996                  * the new address from the device private data and set it.
997                  */
998                 sfc_stop(sa);
999                 rc = sfc_start(sa);
1000                 if (rc != 0)
1001                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1002         }
1003
1004 unlock:
1005         if (rc != 0)
1006                 ether_addr_copy(old_addr, &port->default_mac_addr);
1007
1008         sfc_adapter_unlock(sa);
1009
1010         SFC_ASSERT(rc >= 0);
1011         return -rc;
1012 }
1013
1014
1015 static int
1016 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1017                      uint32_t nb_mc_addr)
1018 {
1019         struct sfc_adapter *sa = dev->data->dev_private;
1020         struct sfc_port *port = &sa->port;
1021         uint8_t *mc_addrs = port->mcast_addrs;
1022         int rc;
1023         unsigned int i;
1024
1025         if (port->isolated) {
1026                 sfc_err(sa, "isolated mode is active on the port");
1027                 sfc_err(sa, "will not set multicast address list");
1028                 return -ENOTSUP;
1029         }
1030
1031         if (mc_addrs == NULL)
1032                 return -ENOBUFS;
1033
1034         if (nb_mc_addr > port->max_mcast_addrs) {
1035                 sfc_err(sa, "too many multicast addresses: %u > %u",
1036                          nb_mc_addr, port->max_mcast_addrs);
1037                 return -EINVAL;
1038         }
1039
1040         for (i = 0; i < nb_mc_addr; ++i) {
1041                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1042                                  EFX_MAC_ADDR_LEN);
1043                 mc_addrs += EFX_MAC_ADDR_LEN;
1044         }
1045
1046         port->nb_mcast_addrs = nb_mc_addr;
1047
1048         if (sa->state != SFC_ADAPTER_STARTED)
1049                 return 0;
1050
1051         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1052                                         port->nb_mcast_addrs);
1053         if (rc != 0)
1054                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1055
1056         SFC_ASSERT(rc >= 0);
1057         return -rc;
1058 }
1059
1060 /*
1061  * The function is used by the secondary process as well. It must not
1062  * use any process-local pointers from the adapter data.
1063  */
1064 static void
1065 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1066                       struct rte_eth_rxq_info *qinfo)
1067 {
1068         struct sfc_adapter *sa = dev->data->dev_private;
1069         struct sfc_rxq_info *rxq_info;
1070         struct sfc_rxq *rxq;
1071
1072         sfc_adapter_lock(sa);
1073
1074         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1075
1076         rxq_info = &sa->rxq_info[rx_queue_id];
1077         rxq = rxq_info->rxq;
1078         SFC_ASSERT(rxq != NULL);
1079
1080         qinfo->mp = rxq->refill_mb_pool;
1081         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1082         qinfo->conf.rx_drop_en = 1;
1083         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1084         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1085         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1086                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1087                 qinfo->scattered_rx = 1;
1088         }
1089         qinfo->nb_desc = rxq_info->entries;
1090
1091         sfc_adapter_unlock(sa);
1092 }
1093
1094 /*
1095  * The function is used by the secondary process as well. It must not
1096  * use any process-local pointers from the adapter data.
1097  */
1098 static void
1099 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1100                       struct rte_eth_txq_info *qinfo)
1101 {
1102         struct sfc_adapter *sa = dev->data->dev_private;
1103         struct sfc_txq_info *txq_info;
1104
1105         sfc_adapter_lock(sa);
1106
1107         SFC_ASSERT(tx_queue_id < sa->txq_count);
1108
1109         txq_info = &sa->txq_info[tx_queue_id];
1110         SFC_ASSERT(txq_info->txq != NULL);
1111
1112         memset(qinfo, 0, sizeof(*qinfo));
1113
1114         qinfo->conf.offloads = txq_info->txq->offloads;
1115         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1116         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1117         qinfo->nb_desc = txq_info->entries;
1118
1119         sfc_adapter_unlock(sa);
1120 }
1121
1122 static uint32_t
1123 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1124 {
1125         struct sfc_adapter *sa = dev->data->dev_private;
1126
1127         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1128
1129         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1130 }
1131
1132 static int
1133 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1134 {
1135         struct sfc_dp_rxq *dp_rxq = queue;
1136
1137         return sfc_rx_qdesc_done(dp_rxq, offset);
1138 }
1139
1140 static int
1141 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1142 {
1143         struct sfc_dp_rxq *dp_rxq = queue;
1144         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1145
1146         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1147 }
1148
1149 static int
1150 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1151 {
1152         struct sfc_dp_txq *dp_txq = queue;
1153         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1154
1155         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1156 }
1157
1158 static int
1159 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1160 {
1161         struct sfc_adapter *sa = dev->data->dev_private;
1162         int rc;
1163
1164         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1165
1166         sfc_adapter_lock(sa);
1167
1168         rc = EINVAL;
1169         if (sa->state != SFC_ADAPTER_STARTED)
1170                 goto fail_not_started;
1171
1172         if (sa->rxq_info[rx_queue_id].rxq == NULL)
1173                 goto fail_not_setup;
1174
1175         rc = sfc_rx_qstart(sa, rx_queue_id);
1176         if (rc != 0)
1177                 goto fail_rx_qstart;
1178
1179         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1180
1181         sfc_adapter_unlock(sa);
1182
1183         return 0;
1184
1185 fail_rx_qstart:
1186 fail_not_setup:
1187 fail_not_started:
1188         sfc_adapter_unlock(sa);
1189         SFC_ASSERT(rc > 0);
1190         return -rc;
1191 }
1192
1193 static int
1194 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1195 {
1196         struct sfc_adapter *sa = dev->data->dev_private;
1197
1198         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1199
1200         sfc_adapter_lock(sa);
1201         sfc_rx_qstop(sa, rx_queue_id);
1202
1203         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1204
1205         sfc_adapter_unlock(sa);
1206
1207         return 0;
1208 }
1209
1210 static int
1211 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1212 {
1213         struct sfc_adapter *sa = dev->data->dev_private;
1214         int rc;
1215
1216         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1217
1218         sfc_adapter_lock(sa);
1219
1220         rc = EINVAL;
1221         if (sa->state != SFC_ADAPTER_STARTED)
1222                 goto fail_not_started;
1223
1224         if (sa->txq_info[tx_queue_id].txq == NULL)
1225                 goto fail_not_setup;
1226
1227         rc = sfc_tx_qstart(sa, tx_queue_id);
1228         if (rc != 0)
1229                 goto fail_tx_qstart;
1230
1231         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1232
1233         sfc_adapter_unlock(sa);
1234         return 0;
1235
1236 fail_tx_qstart:
1237
1238 fail_not_setup:
1239 fail_not_started:
1240         sfc_adapter_unlock(sa);
1241         SFC_ASSERT(rc > 0);
1242         return -rc;
1243 }
1244
1245 static int
1246 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1247 {
1248         struct sfc_adapter *sa = dev->data->dev_private;
1249
1250         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1251
1252         sfc_adapter_lock(sa);
1253
1254         sfc_tx_qstop(sa, tx_queue_id);
1255
1256         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1257
1258         sfc_adapter_unlock(sa);
1259         return 0;
1260 }
1261
1262 static efx_tunnel_protocol_t
1263 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1264 {
1265         switch (rte_type) {
1266         case RTE_TUNNEL_TYPE_VXLAN:
1267                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1268         case RTE_TUNNEL_TYPE_GENEVE:
1269                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1270         default:
1271                 return EFX_TUNNEL_NPROTOS;
1272         }
1273 }
1274
1275 enum sfc_udp_tunnel_op_e {
1276         SFC_UDP_TUNNEL_ADD_PORT,
1277         SFC_UDP_TUNNEL_DEL_PORT,
1278 };
1279
1280 static int
1281 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1282                       struct rte_eth_udp_tunnel *tunnel_udp,
1283                       enum sfc_udp_tunnel_op_e op)
1284 {
1285         struct sfc_adapter *sa = dev->data->dev_private;
1286         efx_tunnel_protocol_t tunnel_proto;
1287         int rc;
1288
1289         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1290                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1291                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1292                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1293
1294         tunnel_proto =
1295                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1296         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1297                 rc = ENOTSUP;
1298                 goto fail_bad_proto;
1299         }
1300
1301         sfc_adapter_lock(sa);
1302
1303         switch (op) {
1304         case SFC_UDP_TUNNEL_ADD_PORT:
1305                 rc = efx_tunnel_config_udp_add(sa->nic,
1306                                                tunnel_udp->udp_port,
1307                                                tunnel_proto);
1308                 break;
1309         case SFC_UDP_TUNNEL_DEL_PORT:
1310                 rc = efx_tunnel_config_udp_remove(sa->nic,
1311                                                   tunnel_udp->udp_port,
1312                                                   tunnel_proto);
1313                 break;
1314         default:
1315                 rc = EINVAL;
1316                 goto fail_bad_op;
1317         }
1318
1319         if (rc != 0)
1320                 goto fail_op;
1321
1322         if (sa->state == SFC_ADAPTER_STARTED) {
1323                 rc = efx_tunnel_reconfigure(sa->nic);
1324                 if (rc == EAGAIN) {
1325                         /*
1326                          * Configuration is accepted by FW and MC reboot
1327                          * is initiated to apply the changes. MC reboot
1328                          * will be handled in a usual way (MC reboot
1329                          * event on management event queue and adapter
1330                          * restart).
1331                          */
1332                         rc = 0;
1333                 } else if (rc != 0) {
1334                         goto fail_reconfigure;
1335                 }
1336         }
1337
1338         sfc_adapter_unlock(sa);
1339         return 0;
1340
1341 fail_reconfigure:
1342         /* Remove/restore entry since the change makes the trouble */
1343         switch (op) {
1344         case SFC_UDP_TUNNEL_ADD_PORT:
1345                 (void)efx_tunnel_config_udp_remove(sa->nic,
1346                                                    tunnel_udp->udp_port,
1347                                                    tunnel_proto);
1348                 break;
1349         case SFC_UDP_TUNNEL_DEL_PORT:
1350                 (void)efx_tunnel_config_udp_add(sa->nic,
1351                                                 tunnel_udp->udp_port,
1352                                                 tunnel_proto);
1353                 break;
1354         }
1355
1356 fail_op:
1357 fail_bad_op:
1358         sfc_adapter_unlock(sa);
1359
1360 fail_bad_proto:
1361         SFC_ASSERT(rc > 0);
1362         return -rc;
1363 }
1364
1365 static int
1366 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1367                             struct rte_eth_udp_tunnel *tunnel_udp)
1368 {
1369         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1370 }
1371
1372 static int
1373 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1374                             struct rte_eth_udp_tunnel *tunnel_udp)
1375 {
1376         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1377 }
1378
1379 static int
1380 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1381                           struct rte_eth_rss_conf *rss_conf)
1382 {
1383         struct sfc_adapter *sa = dev->data->dev_private;
1384         struct sfc_rss *rss = &sa->rss;
1385
1386         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1387                 return -ENOTSUP;
1388
1389         sfc_adapter_lock(sa);
1390
1391         /*
1392          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1393          * hence, conversion is done here to derive a correct set of ETH_RSS
1394          * flags which corresponds to the active EFX configuration stored
1395          * locally in 'sfc_adapter' and kept up-to-date
1396          */
1397         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1398         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1399         if (rss_conf->rss_key != NULL)
1400                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1401
1402         sfc_adapter_unlock(sa);
1403
1404         return 0;
1405 }
1406
1407 static int
1408 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1409                         struct rte_eth_rss_conf *rss_conf)
1410 {
1411         struct sfc_adapter *sa = dev->data->dev_private;
1412         struct sfc_rss *rss = &sa->rss;
1413         struct sfc_port *port = &sa->port;
1414         unsigned int efx_hash_types;
1415         int rc = 0;
1416
1417         if (port->isolated)
1418                 return -ENOTSUP;
1419
1420         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1421                 sfc_err(sa, "RSS is not available");
1422                 return -ENOTSUP;
1423         }
1424
1425         if (rss->channels == 0) {
1426                 sfc_err(sa, "RSS is not configured");
1427                 return -EINVAL;
1428         }
1429
1430         if ((rss_conf->rss_key != NULL) &&
1431             (rss_conf->rss_key_len != sizeof(rss->key))) {
1432                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1433                         sizeof(rss->key));
1434                 return -EINVAL;
1435         }
1436
1437         sfc_adapter_lock(sa);
1438
1439         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1440         if (rc != 0)
1441                 goto fail_rx_hf_rte_to_efx;
1442
1443         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1444                                    rss->hash_alg, efx_hash_types, B_TRUE);
1445         if (rc != 0)
1446                 goto fail_scale_mode_set;
1447
1448         if (rss_conf->rss_key != NULL) {
1449                 if (sa->state == SFC_ADAPTER_STARTED) {
1450                         rc = efx_rx_scale_key_set(sa->nic,
1451                                                   EFX_RSS_CONTEXT_DEFAULT,
1452                                                   rss_conf->rss_key,
1453                                                   sizeof(rss->key));
1454                         if (rc != 0)
1455                                 goto fail_scale_key_set;
1456                 }
1457
1458                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1459         }
1460
1461         rss->hash_types = efx_hash_types;
1462
1463         sfc_adapter_unlock(sa);
1464
1465         return 0;
1466
1467 fail_scale_key_set:
1468         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1469                                   EFX_RX_HASHALG_TOEPLITZ,
1470                                   rss->hash_types, B_TRUE) != 0)
1471                 sfc_err(sa, "failed to restore RSS mode");
1472
1473 fail_scale_mode_set:
1474 fail_rx_hf_rte_to_efx:
1475         sfc_adapter_unlock(sa);
1476         return -rc;
1477 }
1478
1479 static int
1480 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1481                        struct rte_eth_rss_reta_entry64 *reta_conf,
1482                        uint16_t reta_size)
1483 {
1484         struct sfc_adapter *sa = dev->data->dev_private;
1485         struct sfc_rss *rss = &sa->rss;
1486         struct sfc_port *port = &sa->port;
1487         int entry;
1488
1489         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1490                 return -ENOTSUP;
1491
1492         if (rss->channels == 0)
1493                 return -EINVAL;
1494
1495         if (reta_size != EFX_RSS_TBL_SIZE)
1496                 return -EINVAL;
1497
1498         sfc_adapter_lock(sa);
1499
1500         for (entry = 0; entry < reta_size; entry++) {
1501                 int grp = entry / RTE_RETA_GROUP_SIZE;
1502                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1503
1504                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1505                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1506         }
1507
1508         sfc_adapter_unlock(sa);
1509
1510         return 0;
1511 }
1512
1513 static int
1514 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1515                         struct rte_eth_rss_reta_entry64 *reta_conf,
1516                         uint16_t reta_size)
1517 {
1518         struct sfc_adapter *sa = dev->data->dev_private;
1519         struct sfc_rss *rss = &sa->rss;
1520         struct sfc_port *port = &sa->port;
1521         unsigned int *rss_tbl_new;
1522         uint16_t entry;
1523         int rc = 0;
1524
1525
1526         if (port->isolated)
1527                 return -ENOTSUP;
1528
1529         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1530                 sfc_err(sa, "RSS is not available");
1531                 return -ENOTSUP;
1532         }
1533
1534         if (rss->channels == 0) {
1535                 sfc_err(sa, "RSS is not configured");
1536                 return -EINVAL;
1537         }
1538
1539         if (reta_size != EFX_RSS_TBL_SIZE) {
1540                 sfc_err(sa, "RETA size is wrong (should be %u)",
1541                         EFX_RSS_TBL_SIZE);
1542                 return -EINVAL;
1543         }
1544
1545         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1546         if (rss_tbl_new == NULL)
1547                 return -ENOMEM;
1548
1549         sfc_adapter_lock(sa);
1550
1551         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1552
1553         for (entry = 0; entry < reta_size; entry++) {
1554                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1555                 struct rte_eth_rss_reta_entry64 *grp;
1556
1557                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1558
1559                 if (grp->mask & (1ull << grp_idx)) {
1560                         if (grp->reta[grp_idx] >= rss->channels) {
1561                                 rc = EINVAL;
1562                                 goto bad_reta_entry;
1563                         }
1564                         rss_tbl_new[entry] = grp->reta[grp_idx];
1565                 }
1566         }
1567
1568         if (sa->state == SFC_ADAPTER_STARTED) {
1569                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1570                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1571                 if (rc != 0)
1572                         goto fail_scale_tbl_set;
1573         }
1574
1575         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1576
1577 fail_scale_tbl_set:
1578 bad_reta_entry:
1579         sfc_adapter_unlock(sa);
1580
1581         rte_free(rss_tbl_new);
1582
1583         SFC_ASSERT(rc >= 0);
1584         return -rc;
1585 }
1586
1587 static int
1588 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1589                     enum rte_filter_op filter_op,
1590                     void *arg)
1591 {
1592         struct sfc_adapter *sa = dev->data->dev_private;
1593         int rc = ENOTSUP;
1594
1595         sfc_log_init(sa, "entry");
1596
1597         switch (filter_type) {
1598         case RTE_ETH_FILTER_NONE:
1599                 sfc_err(sa, "Global filters configuration not supported");
1600                 break;
1601         case RTE_ETH_FILTER_MACVLAN:
1602                 sfc_err(sa, "MACVLAN filters not supported");
1603                 break;
1604         case RTE_ETH_FILTER_ETHERTYPE:
1605                 sfc_err(sa, "EtherType filters not supported");
1606                 break;
1607         case RTE_ETH_FILTER_FLEXIBLE:
1608                 sfc_err(sa, "Flexible filters not supported");
1609                 break;
1610         case RTE_ETH_FILTER_SYN:
1611                 sfc_err(sa, "SYN filters not supported");
1612                 break;
1613         case RTE_ETH_FILTER_NTUPLE:
1614                 sfc_err(sa, "NTUPLE filters not supported");
1615                 break;
1616         case RTE_ETH_FILTER_TUNNEL:
1617                 sfc_err(sa, "Tunnel filters not supported");
1618                 break;
1619         case RTE_ETH_FILTER_FDIR:
1620                 sfc_err(sa, "Flow Director filters not supported");
1621                 break;
1622         case RTE_ETH_FILTER_HASH:
1623                 sfc_err(sa, "Hash filters not supported");
1624                 break;
1625         case RTE_ETH_FILTER_GENERIC:
1626                 if (filter_op != RTE_ETH_FILTER_GET) {
1627                         rc = EINVAL;
1628                 } else {
1629                         *(const void **)arg = &sfc_flow_ops;
1630                         rc = 0;
1631                 }
1632                 break;
1633         default:
1634                 sfc_err(sa, "Unknown filter type %u", filter_type);
1635                 break;
1636         }
1637
1638         sfc_log_init(sa, "exit: %d", -rc);
1639         SFC_ASSERT(rc >= 0);
1640         return -rc;
1641 }
1642
1643 static int
1644 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1645 {
1646         struct sfc_adapter *sa = dev->data->dev_private;
1647
1648         /*
1649          * If Rx datapath does not provide callback to check mempool,
1650          * all pools are supported.
1651          */
1652         if (sa->dp_rx->pool_ops_supported == NULL)
1653                 return 1;
1654
1655         return sa->dp_rx->pool_ops_supported(pool);
1656 }
1657
1658 static const struct eth_dev_ops sfc_eth_dev_ops = {
1659         .dev_configure                  = sfc_dev_configure,
1660         .dev_start                      = sfc_dev_start,
1661         .dev_stop                       = sfc_dev_stop,
1662         .dev_set_link_up                = sfc_dev_set_link_up,
1663         .dev_set_link_down              = sfc_dev_set_link_down,
1664         .dev_close                      = sfc_dev_close,
1665         .promiscuous_enable             = sfc_dev_promisc_enable,
1666         .promiscuous_disable            = sfc_dev_promisc_disable,
1667         .allmulticast_enable            = sfc_dev_allmulti_enable,
1668         .allmulticast_disable           = sfc_dev_allmulti_disable,
1669         .link_update                    = sfc_dev_link_update,
1670         .stats_get                      = sfc_stats_get,
1671         .stats_reset                    = sfc_stats_reset,
1672         .xstats_get                     = sfc_xstats_get,
1673         .xstats_reset                   = sfc_stats_reset,
1674         .xstats_get_names               = sfc_xstats_get_names,
1675         .dev_infos_get                  = sfc_dev_infos_get,
1676         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1677         .mtu_set                        = sfc_dev_set_mtu,
1678         .rx_queue_start                 = sfc_rx_queue_start,
1679         .rx_queue_stop                  = sfc_rx_queue_stop,
1680         .tx_queue_start                 = sfc_tx_queue_start,
1681         .tx_queue_stop                  = sfc_tx_queue_stop,
1682         .rx_queue_setup                 = sfc_rx_queue_setup,
1683         .rx_queue_release               = sfc_rx_queue_release,
1684         .rx_queue_count                 = sfc_rx_queue_count,
1685         .rx_descriptor_done             = sfc_rx_descriptor_done,
1686         .rx_descriptor_status           = sfc_rx_descriptor_status,
1687         .tx_descriptor_status           = sfc_tx_descriptor_status,
1688         .tx_queue_setup                 = sfc_tx_queue_setup,
1689         .tx_queue_release               = sfc_tx_queue_release,
1690         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1691         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1692         .mac_addr_set                   = sfc_mac_addr_set,
1693         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1694         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1695         .reta_update                    = sfc_dev_rss_reta_update,
1696         .reta_query                     = sfc_dev_rss_reta_query,
1697         .rss_hash_update                = sfc_dev_rss_hash_update,
1698         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1699         .filter_ctrl                    = sfc_dev_filter_ctrl,
1700         .set_mc_addr_list               = sfc_set_mc_addr_list,
1701         .rxq_info_get                   = sfc_rx_queue_info_get,
1702         .txq_info_get                   = sfc_tx_queue_info_get,
1703         .fw_version_get                 = sfc_fw_version_get,
1704         .xstats_get_by_id               = sfc_xstats_get_by_id,
1705         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1706         .pool_ops_supported             = sfc_pool_ops_supported,
1707 };
1708
1709 /**
1710  * Duplicate a string in potentially shared memory required for
1711  * multi-process support.
1712  *
1713  * strdup() allocates from process-local heap/memory.
1714  */
1715 static char *
1716 sfc_strdup(const char *str)
1717 {
1718         size_t size;
1719         char *copy;
1720
1721         if (str == NULL)
1722                 return NULL;
1723
1724         size = strlen(str) + 1;
1725         copy = rte_malloc(__func__, size, 0);
1726         if (copy != NULL)
1727                 rte_memcpy(copy, str, size);
1728
1729         return copy;
1730 }
1731
1732 static int
1733 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1734 {
1735         struct sfc_adapter *sa = dev->data->dev_private;
1736         const efx_nic_cfg_t *encp;
1737         unsigned int avail_caps = 0;
1738         const char *rx_name = NULL;
1739         const char *tx_name = NULL;
1740         int rc;
1741
1742         switch (sa->family) {
1743         case EFX_FAMILY_HUNTINGTON:
1744         case EFX_FAMILY_MEDFORD:
1745         case EFX_FAMILY_MEDFORD2:
1746                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1747                 break;
1748         default:
1749                 break;
1750         }
1751
1752         encp = efx_nic_cfg_get(sa->nic);
1753         if (encp->enc_rx_es_super_buffer_supported)
1754                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1755
1756         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1757                                 sfc_kvarg_string_handler, &rx_name);
1758         if (rc != 0)
1759                 goto fail_kvarg_rx_datapath;
1760
1761         if (rx_name != NULL) {
1762                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1763                 if (sa->dp_rx == NULL) {
1764                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1765                         rc = ENOENT;
1766                         goto fail_dp_rx;
1767                 }
1768                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1769                         sfc_err(sa,
1770                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1771                                 rx_name);
1772                         rc = EINVAL;
1773                         goto fail_dp_rx_caps;
1774                 }
1775         } else {
1776                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1777                 if (sa->dp_rx == NULL) {
1778                         sfc_err(sa, "Rx datapath by caps %#x not found",
1779                                 avail_caps);
1780                         rc = ENOENT;
1781                         goto fail_dp_rx;
1782                 }
1783         }
1784
1785         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1786         if (sa->dp_rx_name == NULL) {
1787                 rc = ENOMEM;
1788                 goto fail_dp_rx_name;
1789         }
1790
1791         sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1792
1793         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1794
1795         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1796                                 sfc_kvarg_string_handler, &tx_name);
1797         if (rc != 0)
1798                 goto fail_kvarg_tx_datapath;
1799
1800         if (tx_name != NULL) {
1801                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1802                 if (sa->dp_tx == NULL) {
1803                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1804                         rc = ENOENT;
1805                         goto fail_dp_tx;
1806                 }
1807                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1808                         sfc_err(sa,
1809                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1810                                 tx_name);
1811                         rc = EINVAL;
1812                         goto fail_dp_tx_caps;
1813                 }
1814         } else {
1815                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1816                 if (sa->dp_tx == NULL) {
1817                         sfc_err(sa, "Tx datapath by caps %#x not found",
1818                                 avail_caps);
1819                         rc = ENOENT;
1820                         goto fail_dp_tx;
1821                 }
1822         }
1823
1824         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1825         if (sa->dp_tx_name == NULL) {
1826                 rc = ENOMEM;
1827                 goto fail_dp_tx_name;
1828         }
1829
1830         sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1831
1832         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1833
1834         dev->dev_ops = &sfc_eth_dev_ops;
1835
1836         return 0;
1837
1838 fail_dp_tx_name:
1839 fail_dp_tx_caps:
1840         sa->dp_tx = NULL;
1841
1842 fail_dp_tx:
1843 fail_kvarg_tx_datapath:
1844         rte_free(sa->dp_rx_name);
1845         sa->dp_rx_name = NULL;
1846
1847 fail_dp_rx_name:
1848 fail_dp_rx_caps:
1849         sa->dp_rx = NULL;
1850
1851 fail_dp_rx:
1852 fail_kvarg_rx_datapath:
1853         return rc;
1854 }
1855
1856 static void
1857 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1858 {
1859         struct sfc_adapter *sa = dev->data->dev_private;
1860
1861         dev->dev_ops = NULL;
1862         dev->rx_pkt_burst = NULL;
1863         dev->tx_pkt_burst = NULL;
1864
1865         rte_free(sa->dp_tx_name);
1866         sa->dp_tx_name = NULL;
1867         sa->dp_tx = NULL;
1868
1869         rte_free(sa->dp_rx_name);
1870         sa->dp_rx_name = NULL;
1871         sa->dp_rx = NULL;
1872 }
1873
1874 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1875         .rxq_info_get                   = sfc_rx_queue_info_get,
1876         .txq_info_get                   = sfc_tx_queue_info_get,
1877 };
1878
1879 static int
1880 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1881 {
1882         /*
1883          * Device private data has really many process-local pointers.
1884          * Below code should be extremely careful to use data located
1885          * in shared memory only.
1886          */
1887         struct sfc_adapter *sa = dev->data->dev_private;
1888         const struct sfc_dp_rx *dp_rx;
1889         const struct sfc_dp_tx *dp_tx;
1890         int rc;
1891
1892         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1893         if (dp_rx == NULL) {
1894                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_rx_name);
1895                 rc = ENOENT;
1896                 goto fail_dp_rx;
1897         }
1898         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1899                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1900                         sa->dp_rx_name);
1901                 rc = EINVAL;
1902                 goto fail_dp_rx_multi_process;
1903         }
1904
1905         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1906         if (dp_tx == NULL) {
1907                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1908                 rc = ENOENT;
1909                 goto fail_dp_tx;
1910         }
1911         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1912                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1913                         sa->dp_tx_name);
1914                 rc = EINVAL;
1915                 goto fail_dp_tx_multi_process;
1916         }
1917
1918         dev->rx_pkt_burst = dp_rx->pkt_burst;
1919         dev->tx_pkt_burst = dp_tx->pkt_burst;
1920         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1921
1922         return 0;
1923
1924 fail_dp_tx_multi_process:
1925 fail_dp_tx:
1926 fail_dp_rx_multi_process:
1927 fail_dp_rx:
1928         return rc;
1929 }
1930
1931 static void
1932 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1933 {
1934         dev->dev_ops = NULL;
1935         dev->tx_pkt_burst = NULL;
1936         dev->rx_pkt_burst = NULL;
1937 }
1938
1939 static void
1940 sfc_register_dp(void)
1941 {
1942         /* Register once */
1943         if (TAILQ_EMPTY(&sfc_dp_head)) {
1944                 /* Prefer EF10 datapath */
1945                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1946                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1947                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1948
1949                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1950                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1951                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1952         }
1953 }
1954
1955 static int
1956 sfc_eth_dev_init(struct rte_eth_dev *dev)
1957 {
1958         struct sfc_adapter *sa = dev->data->dev_private;
1959         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1960         int rc;
1961         const efx_nic_cfg_t *encp;
1962         const struct ether_addr *from;
1963
1964         sfc_register_dp();
1965
1966         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1967                 return -sfc_eth_dev_secondary_set_ops(dev);
1968
1969         /* Required for logging */
1970         sa->pci_addr = pci_dev->addr;
1971         sa->port_id = dev->data->port_id;
1972
1973         sa->eth_dev = dev;
1974
1975         /* Copy PCI device info to the dev->data */
1976         rte_eth_copy_pci_info(dev, pci_dev);
1977
1978         sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1979                                                 RTE_LOG_NOTICE);
1980
1981         rc = sfc_kvargs_parse(sa);
1982         if (rc != 0)
1983                 goto fail_kvargs_parse;
1984
1985         sfc_log_init(sa, "entry");
1986
1987         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1988         if (dev->data->mac_addrs == NULL) {
1989                 rc = ENOMEM;
1990                 goto fail_mac_addrs;
1991         }
1992
1993         sfc_adapter_lock_init(sa);
1994         sfc_adapter_lock(sa);
1995
1996         sfc_log_init(sa, "probing");
1997         rc = sfc_probe(sa);
1998         if (rc != 0)
1999                 goto fail_probe;
2000
2001         sfc_log_init(sa, "set device ops");
2002         rc = sfc_eth_dev_set_ops(dev);
2003         if (rc != 0)
2004                 goto fail_set_ops;
2005
2006         sfc_log_init(sa, "attaching");
2007         rc = sfc_attach(sa);
2008         if (rc != 0)
2009                 goto fail_attach;
2010
2011         encp = efx_nic_cfg_get(sa->nic);
2012
2013         /*
2014          * The arguments are really reverse order in comparison to
2015          * Linux kernel. Copy from NIC config to Ethernet device data.
2016          */
2017         from = (const struct ether_addr *)(encp->enc_mac_addr);
2018         ether_addr_copy(from, &dev->data->mac_addrs[0]);
2019
2020         sfc_adapter_unlock(sa);
2021
2022         sfc_log_init(sa, "done");
2023         return 0;
2024
2025 fail_attach:
2026         sfc_eth_dev_clear_ops(dev);
2027
2028 fail_set_ops:
2029         sfc_unprobe(sa);
2030
2031 fail_probe:
2032         sfc_adapter_unlock(sa);
2033         sfc_adapter_lock_fini(sa);
2034         rte_free(dev->data->mac_addrs);
2035         dev->data->mac_addrs = NULL;
2036
2037 fail_mac_addrs:
2038         sfc_kvargs_cleanup(sa);
2039
2040 fail_kvargs_parse:
2041         sfc_log_init(sa, "failed %d", rc);
2042         SFC_ASSERT(rc > 0);
2043         return -rc;
2044 }
2045
2046 static int
2047 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2048 {
2049         struct sfc_adapter *sa;
2050
2051         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2052                 sfc_eth_dev_secondary_clear_ops(dev);
2053                 return 0;
2054         }
2055
2056         sa = dev->data->dev_private;
2057         sfc_log_init(sa, "entry");
2058
2059         sfc_adapter_lock(sa);
2060
2061         sfc_eth_dev_clear_ops(dev);
2062
2063         sfc_detach(sa);
2064         sfc_unprobe(sa);
2065
2066         sfc_kvargs_cleanup(sa);
2067
2068         sfc_adapter_unlock(sa);
2069         sfc_adapter_lock_fini(sa);
2070
2071         sfc_log_init(sa, "done");
2072
2073         /* Required for logging, so cleanup last */
2074         sa->eth_dev = NULL;
2075         return 0;
2076 }
2077
2078 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2079         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2080         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2081         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2082         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2083         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2084         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2085         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2086         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2087         { .vendor_id = 0 /* sentinel */ }
2088 };
2089
2090 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2091         struct rte_pci_device *pci_dev)
2092 {
2093         return rte_eth_dev_pci_generic_probe(pci_dev,
2094                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2095 }
2096
2097 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2098 {
2099         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2100 }
2101
2102 static struct rte_pci_driver sfc_efx_pmd = {
2103         .id_table = pci_id_sfc_efx_map,
2104         .drv_flags =
2105                 RTE_PCI_DRV_INTR_LSC |
2106                 RTE_PCI_DRV_NEED_MAPPING,
2107         .probe = sfc_eth_dev_pci_probe,
2108         .remove = sfc_eth_dev_pci_remove,
2109 };
2110
2111 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2112 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2113 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2114 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2115         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2116         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2117         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2118         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2119         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2120         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2121
2122 RTE_INIT(sfc_driver_register_logtype)
2123 {
2124         int ret;
2125
2126         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2127                                                    RTE_LOG_NOTICE);
2128         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2129 }