ethdev: fix TPID handling in flow API
[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
17 #include "efx.h"
18
19 #include "sfc.h"
20 #include "sfc_debug.h"
21 #include "sfc_log.h"
22 #include "sfc_kvargs.h"
23 #include "sfc_ev.h"
24 #include "sfc_rx.h"
25 #include "sfc_tx.h"
26 #include "sfc_flow.h"
27 #include "sfc_dp.h"
28 #include "sfc_dp_rx.h"
29
30 uint32_t sfc_logtype_driver;
31
32 static struct sfc_dp_list sfc_dp_head =
33         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
34
35 static int
36 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
37 {
38         struct sfc_adapter *sa = dev->data->dev_private;
39         efx_nic_fw_info_t enfi;
40         int ret;
41         int rc;
42
43         /*
44          * Return value of the callback is likely supposed to be
45          * equal to or greater than 0, nevertheless, if an error
46          * occurs, it will be desirable to pass it to the caller
47          */
48         if ((fw_version == NULL) || (fw_size == 0))
49                 return -EINVAL;
50
51         rc = efx_nic_get_fw_version(sa->nic, &enfi);
52         if (rc != 0)
53                 return -rc;
54
55         ret = snprintf(fw_version, fw_size,
56                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
57                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
58                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
59         if (ret < 0)
60                 return ret;
61
62         if (enfi.enfi_dpcpu_fw_ids_valid) {
63                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
64                 int ret_extra;
65
66                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
67                                      fw_size - dpcpu_fw_ids_offset,
68                                      " rx%" PRIx16 " tx%" PRIx16,
69                                      enfi.enfi_rx_dpcpu_fw_id,
70                                      enfi.enfi_tx_dpcpu_fw_id);
71                 if (ret_extra < 0)
72                         return ret_extra;
73
74                 ret += ret_extra;
75         }
76
77         if (fw_size < (size_t)(++ret))
78                 return ret;
79         else
80                 return 0;
81 }
82
83 static void
84 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
85 {
86         struct sfc_adapter *sa = dev->data->dev_private;
87         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
88         uint64_t txq_offloads_def = 0;
89
90         sfc_log_init(sa, "entry");
91
92         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
93
94         /* Autonegotiation may be disabled */
95         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
96         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
97                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
98         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
99                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
100         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
101                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
102         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
103                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
104         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
105                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
106         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
107                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
108
109         dev_info->max_rx_queues = sa->rxq_max;
110         dev_info->max_tx_queues = sa->txq_max;
111
112         /* By default packets are dropped if no descriptors are available */
113         dev_info->default_rxconf.rx_drop_en = 1;
114
115         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
116
117         /*
118          * rx_offload_capa includes both device and queue offloads since
119          * the latter may be requested on a per device basis which makes
120          * sense when some offloads are needed to be set on all queues.
121          */
122         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
123                                     dev_info->rx_queue_offload_capa;
124
125         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
126
127         /*
128          * tx_offload_capa includes both device and queue offloads since
129          * the latter may be requested on a per device basis which makes
130          * sense when some offloads are needed to be set on all queues.
131          */
132         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
133                                     dev_info->tx_queue_offload_capa;
134
135         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
136                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
137
138         dev_info->default_txconf.offloads |= txq_offloads_def;
139
140         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
141         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
142             !encp->enc_hw_tx_insert_vlan_enabled)
143                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
144
145         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
146                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
147
148         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
149                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
150
151         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
152                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
153
154 #if EFSYS_OPT_RX_SCALE
155         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
156                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
157                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
158                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
159         }
160 #endif
161
162         /* Initialize to hardware limits */
163         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
164         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
165         /* The RXQ hardware requires that the descriptor count is a power
166          * of 2, but rx_desc_lim cannot properly describe that constraint.
167          */
168         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
169
170         /* Initialize to hardware limits */
171         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
172         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
173         /*
174          * The TXQ hardware requires that the descriptor count is a power
175          * of 2, but tx_desc_lim cannot properly describe that constraint
176          */
177         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
178
179         if (sa->dp_rx->get_dev_info != NULL)
180                 sa->dp_rx->get_dev_info(dev_info);
181         if (sa->dp_tx->get_dev_info != NULL)
182                 sa->dp_tx->get_dev_info(dev_info);
183 }
184
185 static const uint32_t *
186 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
187 {
188         struct sfc_adapter *sa = dev->data->dev_private;
189         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
190         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
191
192         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
193 }
194
195 static int
196 sfc_dev_configure(struct rte_eth_dev *dev)
197 {
198         struct rte_eth_dev_data *dev_data = dev->data;
199         struct sfc_adapter *sa = dev_data->dev_private;
200         int rc;
201
202         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
203                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
204
205         sfc_adapter_lock(sa);
206         switch (sa->state) {
207         case SFC_ADAPTER_CONFIGURED:
208                 /* FALLTHROUGH */
209         case SFC_ADAPTER_INITIALIZED:
210                 rc = sfc_configure(sa);
211                 break;
212         default:
213                 sfc_err(sa, "unexpected adapter state %u to configure",
214                         sa->state);
215                 rc = EINVAL;
216                 break;
217         }
218         sfc_adapter_unlock(sa);
219
220         sfc_log_init(sa, "done %d", rc);
221         SFC_ASSERT(rc >= 0);
222         return -rc;
223 }
224
225 static int
226 sfc_dev_start(struct rte_eth_dev *dev)
227 {
228         struct sfc_adapter *sa = dev->data->dev_private;
229         int rc;
230
231         sfc_log_init(sa, "entry");
232
233         sfc_adapter_lock(sa);
234         rc = sfc_start(sa);
235         sfc_adapter_unlock(sa);
236
237         sfc_log_init(sa, "done %d", rc);
238         SFC_ASSERT(rc >= 0);
239         return -rc;
240 }
241
242 static int
243 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
244 {
245         struct sfc_adapter *sa = dev->data->dev_private;
246         struct rte_eth_link current_link;
247         int ret;
248
249         sfc_log_init(sa, "entry");
250
251         if (sa->state != SFC_ADAPTER_STARTED) {
252                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
253         } else if (wait_to_complete) {
254                 efx_link_mode_t link_mode;
255
256                 if (efx_port_poll(sa->nic, &link_mode) != 0)
257                         link_mode = EFX_LINK_UNKNOWN;
258                 sfc_port_link_mode_to_info(link_mode, &current_link);
259
260         } else {
261                 sfc_ev_mgmt_qpoll(sa);
262                 rte_eth_linkstatus_get(dev, &current_link);
263         }
264
265         ret = rte_eth_linkstatus_set(dev, &current_link);
266         if (ret == 0)
267                 sfc_notice(sa, "Link status is %s",
268                            current_link.link_status ? "UP" : "DOWN");
269
270         return ret;
271 }
272
273 static void
274 sfc_dev_stop(struct rte_eth_dev *dev)
275 {
276         struct sfc_adapter *sa = dev->data->dev_private;
277
278         sfc_log_init(sa, "entry");
279
280         sfc_adapter_lock(sa);
281         sfc_stop(sa);
282         sfc_adapter_unlock(sa);
283
284         sfc_log_init(sa, "done");
285 }
286
287 static int
288 sfc_dev_set_link_up(struct rte_eth_dev *dev)
289 {
290         struct sfc_adapter *sa = dev->data->dev_private;
291         int rc;
292
293         sfc_log_init(sa, "entry");
294
295         sfc_adapter_lock(sa);
296         rc = sfc_start(sa);
297         sfc_adapter_unlock(sa);
298
299         SFC_ASSERT(rc >= 0);
300         return -rc;
301 }
302
303 static int
304 sfc_dev_set_link_down(struct rte_eth_dev *dev)
305 {
306         struct sfc_adapter *sa = dev->data->dev_private;
307
308         sfc_log_init(sa, "entry");
309
310         sfc_adapter_lock(sa);
311         sfc_stop(sa);
312         sfc_adapter_unlock(sa);
313
314         return 0;
315 }
316
317 static void
318 sfc_dev_close(struct rte_eth_dev *dev)
319 {
320         struct sfc_adapter *sa = dev->data->dev_private;
321
322         sfc_log_init(sa, "entry");
323
324         sfc_adapter_lock(sa);
325         switch (sa->state) {
326         case SFC_ADAPTER_STARTED:
327                 sfc_stop(sa);
328                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
329                 /* FALLTHROUGH */
330         case SFC_ADAPTER_CONFIGURED:
331                 sfc_close(sa);
332                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
333                 /* FALLTHROUGH */
334         case SFC_ADAPTER_INITIALIZED:
335                 break;
336         default:
337                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
338                 break;
339         }
340         sfc_adapter_unlock(sa);
341
342         sfc_log_init(sa, "done");
343 }
344
345 static void
346 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
347                    boolean_t enabled)
348 {
349         struct sfc_port *port;
350         boolean_t *toggle;
351         struct sfc_adapter *sa = dev->data->dev_private;
352         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
353         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
354
355         sfc_adapter_lock(sa);
356
357         port = &sa->port;
358         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
359
360         if (*toggle != enabled) {
361                 *toggle = enabled;
362
363                 if (port->isolated) {
364                         sfc_warn(sa, "isolated mode is active on the port");
365                         sfc_warn(sa, "the change is to be applied on the next "
366                                      "start provided that isolated mode is "
367                                      "disabled prior the next start");
368                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
369                            (sfc_set_rx_mode(sa) != 0)) {
370                         *toggle = !(enabled);
371                         sfc_warn(sa, "Failed to %s %s mode",
372                                  ((enabled) ? "enable" : "disable"), desc);
373                 }
374         }
375
376         sfc_adapter_unlock(sa);
377 }
378
379 static void
380 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
381 {
382         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
383 }
384
385 static void
386 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
387 {
388         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
389 }
390
391 static void
392 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
393 {
394         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
395 }
396
397 static void
398 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
399 {
400         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
401 }
402
403 static int
404 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
405                    uint16_t nb_rx_desc, unsigned int socket_id,
406                    const struct rte_eth_rxconf *rx_conf,
407                    struct rte_mempool *mb_pool)
408 {
409         struct sfc_adapter *sa = dev->data->dev_private;
410         int rc;
411
412         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
413                      rx_queue_id, nb_rx_desc, socket_id);
414
415         sfc_adapter_lock(sa);
416
417         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
418                           rx_conf, mb_pool);
419         if (rc != 0)
420                 goto fail_rx_qinit;
421
422         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
423
424         sfc_adapter_unlock(sa);
425
426         return 0;
427
428 fail_rx_qinit:
429         sfc_adapter_unlock(sa);
430         SFC_ASSERT(rc > 0);
431         return -rc;
432 }
433
434 static void
435 sfc_rx_queue_release(void *queue)
436 {
437         struct sfc_dp_rxq *dp_rxq = queue;
438         struct sfc_rxq *rxq;
439         struct sfc_adapter *sa;
440         unsigned int sw_index;
441
442         if (dp_rxq == NULL)
443                 return;
444
445         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
446         sa = rxq->evq->sa;
447         sfc_adapter_lock(sa);
448
449         sw_index = sfc_rxq_sw_index(rxq);
450
451         sfc_log_init(sa, "RxQ=%u", sw_index);
452
453         sa->eth_dev->data->rx_queues[sw_index] = NULL;
454
455         sfc_rx_qfini(sa, sw_index);
456
457         sfc_adapter_unlock(sa);
458 }
459
460 static int
461 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
462                    uint16_t nb_tx_desc, unsigned int socket_id,
463                    const struct rte_eth_txconf *tx_conf)
464 {
465         struct sfc_adapter *sa = dev->data->dev_private;
466         int rc;
467
468         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
469                      tx_queue_id, nb_tx_desc, socket_id);
470
471         sfc_adapter_lock(sa);
472
473         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
474         if (rc != 0)
475                 goto fail_tx_qinit;
476
477         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
478
479         sfc_adapter_unlock(sa);
480         return 0;
481
482 fail_tx_qinit:
483         sfc_adapter_unlock(sa);
484         SFC_ASSERT(rc > 0);
485         return -rc;
486 }
487
488 static void
489 sfc_tx_queue_release(void *queue)
490 {
491         struct sfc_dp_txq *dp_txq = queue;
492         struct sfc_txq *txq;
493         unsigned int sw_index;
494         struct sfc_adapter *sa;
495
496         if (dp_txq == NULL)
497                 return;
498
499         txq = sfc_txq_by_dp_txq(dp_txq);
500         sw_index = sfc_txq_sw_index(txq);
501
502         SFC_ASSERT(txq->evq != NULL);
503         sa = txq->evq->sa;
504
505         sfc_log_init(sa, "TxQ = %u", sw_index);
506
507         sfc_adapter_lock(sa);
508
509         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
510         sa->eth_dev->data->tx_queues[sw_index] = NULL;
511
512         sfc_tx_qfini(sa, sw_index);
513
514         sfc_adapter_unlock(sa);
515 }
516
517 static int
518 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
519 {
520         struct sfc_adapter *sa = dev->data->dev_private;
521         struct sfc_port *port = &sa->port;
522         uint64_t *mac_stats;
523         int ret;
524
525         rte_spinlock_lock(&port->mac_stats_lock);
526
527         ret = sfc_port_update_mac_stats(sa);
528         if (ret != 0)
529                 goto unlock;
530
531         mac_stats = port->mac_stats_buf;
532
533         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
534                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
535                 stats->ipackets =
536                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
537                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
538                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
539                 stats->opackets =
540                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
541                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
542                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
543                 stats->ibytes =
544                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
545                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
546                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
547                 stats->obytes =
548                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
549                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
550                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
551                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
552                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
553                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
554         } else {
555                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
556                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
557                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
558                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
559                 /*
560                  * Take into account stats which are whenever supported
561                  * on EF10. If some stat is not supported by current
562                  * firmware variant or HW revision, it is guaranteed
563                  * to be zero in mac_stats.
564                  */
565                 stats->imissed =
566                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
567                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
568                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
569                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
570                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
571                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
572                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
573                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
574                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
575                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
576                 stats->ierrors =
577                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
578                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
579                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
580                 /* no oerrors counters supported on EF10 */
581         }
582
583 unlock:
584         rte_spinlock_unlock(&port->mac_stats_lock);
585         SFC_ASSERT(ret >= 0);
586         return -ret;
587 }
588
589 static void
590 sfc_stats_reset(struct rte_eth_dev *dev)
591 {
592         struct sfc_adapter *sa = dev->data->dev_private;
593         struct sfc_port *port = &sa->port;
594         int rc;
595
596         if (sa->state != SFC_ADAPTER_STARTED) {
597                 /*
598                  * The operation cannot be done if port is not started; it
599                  * will be scheduled to be done during the next port start
600                  */
601                 port->mac_stats_reset_pending = B_TRUE;
602                 return;
603         }
604
605         rc = sfc_port_reset_mac_stats(sa);
606         if (rc != 0)
607                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
608 }
609
610 static int
611 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
612                unsigned int xstats_count)
613 {
614         struct sfc_adapter *sa = dev->data->dev_private;
615         struct sfc_port *port = &sa->port;
616         uint64_t *mac_stats;
617         int rc;
618         unsigned int i;
619         int nstats = 0;
620
621         rte_spinlock_lock(&port->mac_stats_lock);
622
623         rc = sfc_port_update_mac_stats(sa);
624         if (rc != 0) {
625                 SFC_ASSERT(rc > 0);
626                 nstats = -rc;
627                 goto unlock;
628         }
629
630         mac_stats = port->mac_stats_buf;
631
632         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
633                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
634                         if (xstats != NULL && nstats < (int)xstats_count) {
635                                 xstats[nstats].id = nstats;
636                                 xstats[nstats].value = mac_stats[i];
637                         }
638                         nstats++;
639                 }
640         }
641
642 unlock:
643         rte_spinlock_unlock(&port->mac_stats_lock);
644
645         return nstats;
646 }
647
648 static int
649 sfc_xstats_get_names(struct rte_eth_dev *dev,
650                      struct rte_eth_xstat_name *xstats_names,
651                      unsigned int xstats_count)
652 {
653         struct sfc_adapter *sa = dev->data->dev_private;
654         struct sfc_port *port = &sa->port;
655         unsigned int i;
656         unsigned int nstats = 0;
657
658         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
659                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
660                         if (xstats_names != NULL && nstats < xstats_count)
661                                 strncpy(xstats_names[nstats].name,
662                                         efx_mac_stat_name(sa->nic, i),
663                                         sizeof(xstats_names[0].name));
664                         nstats++;
665                 }
666         }
667
668         return nstats;
669 }
670
671 static int
672 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
673                      uint64_t *values, unsigned int n)
674 {
675         struct sfc_adapter *sa = dev->data->dev_private;
676         struct sfc_port *port = &sa->port;
677         uint64_t *mac_stats;
678         unsigned int nb_supported = 0;
679         unsigned int nb_written = 0;
680         unsigned int i;
681         int ret;
682         int rc;
683
684         if (unlikely(values == NULL) ||
685             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
686                 return port->mac_stats_nb_supported;
687
688         rte_spinlock_lock(&port->mac_stats_lock);
689
690         rc = sfc_port_update_mac_stats(sa);
691         if (rc != 0) {
692                 SFC_ASSERT(rc > 0);
693                 ret = -rc;
694                 goto unlock;
695         }
696
697         mac_stats = port->mac_stats_buf;
698
699         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
700                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
701                         continue;
702
703                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
704                         values[nb_written++] = mac_stats[i];
705
706                 ++nb_supported;
707         }
708
709         ret = nb_written;
710
711 unlock:
712         rte_spinlock_unlock(&port->mac_stats_lock);
713
714         return ret;
715 }
716
717 static int
718 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
719                            struct rte_eth_xstat_name *xstats_names,
720                            const uint64_t *ids, unsigned int size)
721 {
722         struct sfc_adapter *sa = dev->data->dev_private;
723         struct sfc_port *port = &sa->port;
724         unsigned int nb_supported = 0;
725         unsigned int nb_written = 0;
726         unsigned int i;
727
728         if (unlikely(xstats_names == NULL) ||
729             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
730                 return port->mac_stats_nb_supported;
731
732         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
733                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
734                         continue;
735
736                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
737                         char *name = xstats_names[nb_written++].name;
738
739                         strncpy(name, efx_mac_stat_name(sa->nic, i),
740                                 sizeof(xstats_names[0].name));
741                         name[sizeof(xstats_names[0].name) - 1] = '\0';
742                 }
743
744                 ++nb_supported;
745         }
746
747         return nb_written;
748 }
749
750 static int
751 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
752 {
753         struct sfc_adapter *sa = dev->data->dev_private;
754         unsigned int wanted_fc, link_fc;
755
756         memset(fc_conf, 0, sizeof(*fc_conf));
757
758         sfc_adapter_lock(sa);
759
760         if (sa->state == SFC_ADAPTER_STARTED)
761                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
762         else
763                 link_fc = sa->port.flow_ctrl;
764
765         switch (link_fc) {
766         case 0:
767                 fc_conf->mode = RTE_FC_NONE;
768                 break;
769         case EFX_FCNTL_RESPOND:
770                 fc_conf->mode = RTE_FC_RX_PAUSE;
771                 break;
772         case EFX_FCNTL_GENERATE:
773                 fc_conf->mode = RTE_FC_TX_PAUSE;
774                 break;
775         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
776                 fc_conf->mode = RTE_FC_FULL;
777                 break;
778         default:
779                 sfc_err(sa, "%s: unexpected flow control value %#x",
780                         __func__, link_fc);
781         }
782
783         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
784
785         sfc_adapter_unlock(sa);
786
787         return 0;
788 }
789
790 static int
791 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
792 {
793         struct sfc_adapter *sa = dev->data->dev_private;
794         struct sfc_port *port = &sa->port;
795         unsigned int fcntl;
796         int rc;
797
798         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
799             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
800             fc_conf->mac_ctrl_frame_fwd != 0) {
801                 sfc_err(sa, "unsupported flow control settings specified");
802                 rc = EINVAL;
803                 goto fail_inval;
804         }
805
806         switch (fc_conf->mode) {
807         case RTE_FC_NONE:
808                 fcntl = 0;
809                 break;
810         case RTE_FC_RX_PAUSE:
811                 fcntl = EFX_FCNTL_RESPOND;
812                 break;
813         case RTE_FC_TX_PAUSE:
814                 fcntl = EFX_FCNTL_GENERATE;
815                 break;
816         case RTE_FC_FULL:
817                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
818                 break;
819         default:
820                 rc = EINVAL;
821                 goto fail_inval;
822         }
823
824         sfc_adapter_lock(sa);
825
826         if (sa->state == SFC_ADAPTER_STARTED) {
827                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
828                 if (rc != 0)
829                         goto fail_mac_fcntl_set;
830         }
831
832         port->flow_ctrl = fcntl;
833         port->flow_ctrl_autoneg = fc_conf->autoneg;
834
835         sfc_adapter_unlock(sa);
836
837         return 0;
838
839 fail_mac_fcntl_set:
840         sfc_adapter_unlock(sa);
841 fail_inval:
842         SFC_ASSERT(rc > 0);
843         return -rc;
844 }
845
846 static int
847 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
848 {
849         struct sfc_adapter *sa = dev->data->dev_private;
850         size_t pdu = EFX_MAC_PDU(mtu);
851         size_t old_pdu;
852         int rc;
853
854         sfc_log_init(sa, "mtu=%u", mtu);
855
856         rc = EINVAL;
857         if (pdu < EFX_MAC_PDU_MIN) {
858                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
859                         (unsigned int)mtu, (unsigned int)pdu,
860                         EFX_MAC_PDU_MIN);
861                 goto fail_inval;
862         }
863         if (pdu > EFX_MAC_PDU_MAX) {
864                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
865                         (unsigned int)mtu, (unsigned int)pdu,
866                         EFX_MAC_PDU_MAX);
867                 goto fail_inval;
868         }
869
870         sfc_adapter_lock(sa);
871
872         if (pdu != sa->port.pdu) {
873                 if (sa->state == SFC_ADAPTER_STARTED) {
874                         sfc_stop(sa);
875
876                         old_pdu = sa->port.pdu;
877                         sa->port.pdu = pdu;
878                         rc = sfc_start(sa);
879                         if (rc != 0)
880                                 goto fail_start;
881                 } else {
882                         sa->port.pdu = pdu;
883                 }
884         }
885
886         /*
887          * The driver does not use it, but other PMDs update jumbo_frame
888          * flag and max_rx_pkt_len when MTU is set.
889          */
890         if (mtu > ETHER_MAX_LEN) {
891                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
892
893                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
894                 rxmode->jumbo_frame = 1;
895         }
896
897         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
898
899         sfc_adapter_unlock(sa);
900
901         sfc_log_init(sa, "done");
902         return 0;
903
904 fail_start:
905         sa->port.pdu = old_pdu;
906         if (sfc_start(sa) != 0)
907                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
908                         "PDU max size - port is stopped",
909                         (unsigned int)pdu, (unsigned int)old_pdu);
910         sfc_adapter_unlock(sa);
911
912 fail_inval:
913         sfc_log_init(sa, "failed %d", rc);
914         SFC_ASSERT(rc > 0);
915         return -rc;
916 }
917 static int
918 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
919 {
920         struct sfc_adapter *sa = dev->data->dev_private;
921         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
922         struct sfc_port *port = &sa->port;
923         struct ether_addr *old_addr = &dev->data->mac_addrs[0];
924         int rc = 0;
925
926         sfc_adapter_lock(sa);
927
928         /*
929          * Copy the address to the device private data so that
930          * it could be recalled in the case of adapter restart.
931          */
932         ether_addr_copy(mac_addr, &port->default_mac_addr);
933
934         /*
935          * Neither of the two following checks can return
936          * an error. The new MAC address is preserved in
937          * the device private data and can be activated
938          * on the next port start if the user prevents
939          * isolated mode from being enabled.
940          */
941         if (port->isolated) {
942                 sfc_warn(sa, "isolated mode is active on the port");
943                 sfc_warn(sa, "will not set MAC address");
944                 goto unlock;
945         }
946
947         if (sa->state != SFC_ADAPTER_STARTED) {
948                 sfc_notice(sa, "the port is not started");
949                 sfc_notice(sa, "the new MAC address will be set on port start");
950
951                 goto unlock;
952         }
953
954         if (encp->enc_allow_set_mac_with_installed_filters) {
955                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
956                 if (rc != 0) {
957                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
958                         goto unlock;
959                 }
960
961                 /*
962                  * Changing the MAC address by means of MCDI request
963                  * has no effect on received traffic, therefore
964                  * we also need to update unicast filters
965                  */
966                 rc = sfc_set_rx_mode(sa);
967                 if (rc != 0) {
968                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
969                         /* Rollback the old address */
970                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
971                         (void)sfc_set_rx_mode(sa);
972                 }
973         } else {
974                 sfc_warn(sa, "cannot set MAC address with filters installed");
975                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
976                 sfc_warn(sa, "(some traffic may be dropped)");
977
978                 /*
979                  * Since setting MAC address with filters installed is not
980                  * allowed on the adapter, the new MAC address will be set
981                  * by means of adapter restart. sfc_start() shall retrieve
982                  * the new address from the device private data and set it.
983                  */
984                 sfc_stop(sa);
985                 rc = sfc_start(sa);
986                 if (rc != 0)
987                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
988         }
989
990 unlock:
991         if (rc != 0)
992                 ether_addr_copy(old_addr, &port->default_mac_addr);
993
994         sfc_adapter_unlock(sa);
995
996         SFC_ASSERT(rc >= 0);
997         return -rc;
998 }
999
1000
1001 static int
1002 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1003                      uint32_t nb_mc_addr)
1004 {
1005         struct sfc_adapter *sa = dev->data->dev_private;
1006         struct sfc_port *port = &sa->port;
1007         uint8_t *mc_addrs = port->mcast_addrs;
1008         int rc;
1009         unsigned int i;
1010
1011         if (port->isolated) {
1012                 sfc_err(sa, "isolated mode is active on the port");
1013                 sfc_err(sa, "will not set multicast address list");
1014                 return -ENOTSUP;
1015         }
1016
1017         if (mc_addrs == NULL)
1018                 return -ENOBUFS;
1019
1020         if (nb_mc_addr > port->max_mcast_addrs) {
1021                 sfc_err(sa, "too many multicast addresses: %u > %u",
1022                          nb_mc_addr, port->max_mcast_addrs);
1023                 return -EINVAL;
1024         }
1025
1026         for (i = 0; i < nb_mc_addr; ++i) {
1027                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1028                                  EFX_MAC_ADDR_LEN);
1029                 mc_addrs += EFX_MAC_ADDR_LEN;
1030         }
1031
1032         port->nb_mcast_addrs = nb_mc_addr;
1033
1034         if (sa->state != SFC_ADAPTER_STARTED)
1035                 return 0;
1036
1037         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1038                                         port->nb_mcast_addrs);
1039         if (rc != 0)
1040                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1041
1042         SFC_ASSERT(rc > 0);
1043         return -rc;
1044 }
1045
1046 /*
1047  * The function is used by the secondary process as well. It must not
1048  * use any process-local pointers from the adapter data.
1049  */
1050 static void
1051 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1052                       struct rte_eth_rxq_info *qinfo)
1053 {
1054         struct sfc_adapter *sa = dev->data->dev_private;
1055         struct sfc_rxq_info *rxq_info;
1056         struct sfc_rxq *rxq;
1057
1058         sfc_adapter_lock(sa);
1059
1060         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1061
1062         rxq_info = &sa->rxq_info[rx_queue_id];
1063         rxq = rxq_info->rxq;
1064         SFC_ASSERT(rxq != NULL);
1065
1066         qinfo->mp = rxq->refill_mb_pool;
1067         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1068         qinfo->conf.rx_drop_en = 1;
1069         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1070         qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1071                                DEV_RX_OFFLOAD_UDP_CKSUM |
1072                                DEV_RX_OFFLOAD_TCP_CKSUM;
1073         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1074                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1075                 qinfo->scattered_rx = 1;
1076         }
1077         qinfo->nb_desc = rxq_info->entries;
1078
1079         sfc_adapter_unlock(sa);
1080 }
1081
1082 /*
1083  * The function is used by the secondary process as well. It must not
1084  * use any process-local pointers from the adapter data.
1085  */
1086 static void
1087 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1088                       struct rte_eth_txq_info *qinfo)
1089 {
1090         struct sfc_adapter *sa = dev->data->dev_private;
1091         struct sfc_txq_info *txq_info;
1092
1093         sfc_adapter_lock(sa);
1094
1095         SFC_ASSERT(tx_queue_id < sa->txq_count);
1096
1097         txq_info = &sa->txq_info[tx_queue_id];
1098         SFC_ASSERT(txq_info->txq != NULL);
1099
1100         memset(qinfo, 0, sizeof(*qinfo));
1101
1102         qinfo->conf.txq_flags = txq_info->txq->flags;
1103         qinfo->conf.offloads = txq_info->txq->offloads;
1104         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1105         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1106         qinfo->nb_desc = txq_info->entries;
1107
1108         sfc_adapter_unlock(sa);
1109 }
1110
1111 static uint32_t
1112 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1113 {
1114         struct sfc_adapter *sa = dev->data->dev_private;
1115
1116         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1117
1118         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1119 }
1120
1121 static int
1122 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1123 {
1124         struct sfc_dp_rxq *dp_rxq = queue;
1125
1126         return sfc_rx_qdesc_done(dp_rxq, offset);
1127 }
1128
1129 static int
1130 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1131 {
1132         struct sfc_dp_rxq *dp_rxq = queue;
1133         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1134
1135         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1136 }
1137
1138 static int
1139 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1140 {
1141         struct sfc_dp_txq *dp_txq = queue;
1142         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1143
1144         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1145 }
1146
1147 static int
1148 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1149 {
1150         struct sfc_adapter *sa = dev->data->dev_private;
1151         int rc;
1152
1153         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1154
1155         sfc_adapter_lock(sa);
1156
1157         rc = EINVAL;
1158         if (sa->state != SFC_ADAPTER_STARTED)
1159                 goto fail_not_started;
1160
1161         rc = sfc_rx_qstart(sa, rx_queue_id);
1162         if (rc != 0)
1163                 goto fail_rx_qstart;
1164
1165         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1166
1167         sfc_adapter_unlock(sa);
1168
1169         return 0;
1170
1171 fail_rx_qstart:
1172 fail_not_started:
1173         sfc_adapter_unlock(sa);
1174         SFC_ASSERT(rc > 0);
1175         return -rc;
1176 }
1177
1178 static int
1179 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1180 {
1181         struct sfc_adapter *sa = dev->data->dev_private;
1182
1183         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1184
1185         sfc_adapter_lock(sa);
1186         sfc_rx_qstop(sa, rx_queue_id);
1187
1188         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1189
1190         sfc_adapter_unlock(sa);
1191
1192         return 0;
1193 }
1194
1195 static int
1196 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1197 {
1198         struct sfc_adapter *sa = dev->data->dev_private;
1199         int rc;
1200
1201         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1202
1203         sfc_adapter_lock(sa);
1204
1205         rc = EINVAL;
1206         if (sa->state != SFC_ADAPTER_STARTED)
1207                 goto fail_not_started;
1208
1209         rc = sfc_tx_qstart(sa, tx_queue_id);
1210         if (rc != 0)
1211                 goto fail_tx_qstart;
1212
1213         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1214
1215         sfc_adapter_unlock(sa);
1216         return 0;
1217
1218 fail_tx_qstart:
1219
1220 fail_not_started:
1221         sfc_adapter_unlock(sa);
1222         SFC_ASSERT(rc > 0);
1223         return -rc;
1224 }
1225
1226 static int
1227 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1228 {
1229         struct sfc_adapter *sa = dev->data->dev_private;
1230
1231         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1232
1233         sfc_adapter_lock(sa);
1234
1235         sfc_tx_qstop(sa, tx_queue_id);
1236
1237         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1238
1239         sfc_adapter_unlock(sa);
1240         return 0;
1241 }
1242
1243 static efx_tunnel_protocol_t
1244 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1245 {
1246         switch (rte_type) {
1247         case RTE_TUNNEL_TYPE_VXLAN:
1248                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1249         case RTE_TUNNEL_TYPE_GENEVE:
1250                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1251         default:
1252                 return EFX_TUNNEL_NPROTOS;
1253         }
1254 }
1255
1256 enum sfc_udp_tunnel_op_e {
1257         SFC_UDP_TUNNEL_ADD_PORT,
1258         SFC_UDP_TUNNEL_DEL_PORT,
1259 };
1260
1261 static int
1262 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1263                       struct rte_eth_udp_tunnel *tunnel_udp,
1264                       enum sfc_udp_tunnel_op_e op)
1265 {
1266         struct sfc_adapter *sa = dev->data->dev_private;
1267         efx_tunnel_protocol_t tunnel_proto;
1268         int rc;
1269
1270         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1271                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1272                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1273                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1274
1275         tunnel_proto =
1276                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1277         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1278                 rc = ENOTSUP;
1279                 goto fail_bad_proto;
1280         }
1281
1282         sfc_adapter_lock(sa);
1283
1284         switch (op) {
1285         case SFC_UDP_TUNNEL_ADD_PORT:
1286                 rc = efx_tunnel_config_udp_add(sa->nic,
1287                                                tunnel_udp->udp_port,
1288                                                tunnel_proto);
1289                 break;
1290         case SFC_UDP_TUNNEL_DEL_PORT:
1291                 rc = efx_tunnel_config_udp_remove(sa->nic,
1292                                                   tunnel_udp->udp_port,
1293                                                   tunnel_proto);
1294                 break;
1295         default:
1296                 rc = EINVAL;
1297                 goto fail_bad_op;
1298         }
1299
1300         if (rc != 0)
1301                 goto fail_op;
1302
1303         if (sa->state == SFC_ADAPTER_STARTED) {
1304                 rc = efx_tunnel_reconfigure(sa->nic);
1305                 if (rc == EAGAIN) {
1306                         /*
1307                          * Configuration is accepted by FW and MC reboot
1308                          * is initiated to apply the changes. MC reboot
1309                          * will be handled in a usual way (MC reboot
1310                          * event on management event queue and adapter
1311                          * restart).
1312                          */
1313                         rc = 0;
1314                 } else if (rc != 0) {
1315                         goto fail_reconfigure;
1316                 }
1317         }
1318
1319         sfc_adapter_unlock(sa);
1320         return 0;
1321
1322 fail_reconfigure:
1323         /* Remove/restore entry since the change makes the trouble */
1324         switch (op) {
1325         case SFC_UDP_TUNNEL_ADD_PORT:
1326                 (void)efx_tunnel_config_udp_remove(sa->nic,
1327                                                    tunnel_udp->udp_port,
1328                                                    tunnel_proto);
1329                 break;
1330         case SFC_UDP_TUNNEL_DEL_PORT:
1331                 (void)efx_tunnel_config_udp_add(sa->nic,
1332                                                 tunnel_udp->udp_port,
1333                                                 tunnel_proto);
1334                 break;
1335         }
1336
1337 fail_op:
1338 fail_bad_op:
1339         sfc_adapter_unlock(sa);
1340
1341 fail_bad_proto:
1342         SFC_ASSERT(rc > 0);
1343         return -rc;
1344 }
1345
1346 static int
1347 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1348                             struct rte_eth_udp_tunnel *tunnel_udp)
1349 {
1350         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1351 }
1352
1353 static int
1354 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1355                             struct rte_eth_udp_tunnel *tunnel_udp)
1356 {
1357         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1358 }
1359
1360 #if EFSYS_OPT_RX_SCALE
1361 static int
1362 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1363                           struct rte_eth_rss_conf *rss_conf)
1364 {
1365         struct sfc_adapter *sa = dev->data->dev_private;
1366         struct sfc_port *port = &sa->port;
1367
1368         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1369                 return -ENOTSUP;
1370
1371         if (sa->rss_channels == 0)
1372                 return -EINVAL;
1373
1374         sfc_adapter_lock(sa);
1375
1376         /*
1377          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1378          * hence, conversion is done here to derive a correct set of ETH_RSS
1379          * flags which corresponds to the active EFX configuration stored
1380          * locally in 'sfc_adapter' and kept up-to-date
1381          */
1382         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1383         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1384         if (rss_conf->rss_key != NULL)
1385                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1386
1387         sfc_adapter_unlock(sa);
1388
1389         return 0;
1390 }
1391
1392 static int
1393 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1394                         struct rte_eth_rss_conf *rss_conf)
1395 {
1396         struct sfc_adapter *sa = dev->data->dev_private;
1397         struct sfc_port *port = &sa->port;
1398         unsigned int efx_hash_types;
1399         int rc = 0;
1400
1401         if (port->isolated)
1402                 return -ENOTSUP;
1403
1404         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1405                 sfc_err(sa, "RSS is not available");
1406                 return -ENOTSUP;
1407         }
1408
1409         if (sa->rss_channels == 0) {
1410                 sfc_err(sa, "RSS is not configured");
1411                 return -EINVAL;
1412         }
1413
1414         if ((rss_conf->rss_key != NULL) &&
1415             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1416                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1417                         sizeof(sa->rss_key));
1418                 return -EINVAL;
1419         }
1420
1421         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1422                 sfc_err(sa, "unsupported hash functions requested");
1423                 return -EINVAL;
1424         }
1425
1426         sfc_adapter_lock(sa);
1427
1428         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1429
1430         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1431                                    EFX_RX_HASHALG_TOEPLITZ,
1432                                    efx_hash_types, B_TRUE);
1433         if (rc != 0)
1434                 goto fail_scale_mode_set;
1435
1436         if (rss_conf->rss_key != NULL) {
1437                 if (sa->state == SFC_ADAPTER_STARTED) {
1438                         rc = efx_rx_scale_key_set(sa->nic,
1439                                                   EFX_RSS_CONTEXT_DEFAULT,
1440                                                   rss_conf->rss_key,
1441                                                   sizeof(sa->rss_key));
1442                         if (rc != 0)
1443                                 goto fail_scale_key_set;
1444                 }
1445
1446                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1447         }
1448
1449         sa->rss_hash_types = efx_hash_types;
1450
1451         sfc_adapter_unlock(sa);
1452
1453         return 0;
1454
1455 fail_scale_key_set:
1456         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1457                                   EFX_RX_HASHALG_TOEPLITZ,
1458                                   sa->rss_hash_types, B_TRUE) != 0)
1459                 sfc_err(sa, "failed to restore RSS mode");
1460
1461 fail_scale_mode_set:
1462         sfc_adapter_unlock(sa);
1463         return -rc;
1464 }
1465
1466 static int
1467 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1468                        struct rte_eth_rss_reta_entry64 *reta_conf,
1469                        uint16_t reta_size)
1470 {
1471         struct sfc_adapter *sa = dev->data->dev_private;
1472         struct sfc_port *port = &sa->port;
1473         int entry;
1474
1475         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1476                 return -ENOTSUP;
1477
1478         if (sa->rss_channels == 0)
1479                 return -EINVAL;
1480
1481         if (reta_size != EFX_RSS_TBL_SIZE)
1482                 return -EINVAL;
1483
1484         sfc_adapter_lock(sa);
1485
1486         for (entry = 0; entry < reta_size; entry++) {
1487                 int grp = entry / RTE_RETA_GROUP_SIZE;
1488                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1489
1490                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1491                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1492         }
1493
1494         sfc_adapter_unlock(sa);
1495
1496         return 0;
1497 }
1498
1499 static int
1500 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1501                         struct rte_eth_rss_reta_entry64 *reta_conf,
1502                         uint16_t reta_size)
1503 {
1504         struct sfc_adapter *sa = dev->data->dev_private;
1505         struct sfc_port *port = &sa->port;
1506         unsigned int *rss_tbl_new;
1507         uint16_t entry;
1508         int rc = 0;
1509
1510
1511         if (port->isolated)
1512                 return -ENOTSUP;
1513
1514         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1515                 sfc_err(sa, "RSS is not available");
1516                 return -ENOTSUP;
1517         }
1518
1519         if (sa->rss_channels == 0) {
1520                 sfc_err(sa, "RSS is not configured");
1521                 return -EINVAL;
1522         }
1523
1524         if (reta_size != EFX_RSS_TBL_SIZE) {
1525                 sfc_err(sa, "RETA size is wrong (should be %u)",
1526                         EFX_RSS_TBL_SIZE);
1527                 return -EINVAL;
1528         }
1529
1530         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1531         if (rss_tbl_new == NULL)
1532                 return -ENOMEM;
1533
1534         sfc_adapter_lock(sa);
1535
1536         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1537
1538         for (entry = 0; entry < reta_size; entry++) {
1539                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1540                 struct rte_eth_rss_reta_entry64 *grp;
1541
1542                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1543
1544                 if (grp->mask & (1ull << grp_idx)) {
1545                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1546                                 rc = EINVAL;
1547                                 goto bad_reta_entry;
1548                         }
1549                         rss_tbl_new[entry] = grp->reta[grp_idx];
1550                 }
1551         }
1552
1553         if (sa->state == SFC_ADAPTER_STARTED) {
1554                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1555                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1556                 if (rc != 0)
1557                         goto fail_scale_tbl_set;
1558         }
1559
1560         rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1561
1562 fail_scale_tbl_set:
1563 bad_reta_entry:
1564         sfc_adapter_unlock(sa);
1565
1566         rte_free(rss_tbl_new);
1567
1568         SFC_ASSERT(rc >= 0);
1569         return -rc;
1570 }
1571 #endif
1572
1573 static int
1574 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1575                     enum rte_filter_op filter_op,
1576                     void *arg)
1577 {
1578         struct sfc_adapter *sa = dev->data->dev_private;
1579         int rc = ENOTSUP;
1580
1581         sfc_log_init(sa, "entry");
1582
1583         switch (filter_type) {
1584         case RTE_ETH_FILTER_NONE:
1585                 sfc_err(sa, "Global filters configuration not supported");
1586                 break;
1587         case RTE_ETH_FILTER_MACVLAN:
1588                 sfc_err(sa, "MACVLAN filters not supported");
1589                 break;
1590         case RTE_ETH_FILTER_ETHERTYPE:
1591                 sfc_err(sa, "EtherType filters not supported");
1592                 break;
1593         case RTE_ETH_FILTER_FLEXIBLE:
1594                 sfc_err(sa, "Flexible filters not supported");
1595                 break;
1596         case RTE_ETH_FILTER_SYN:
1597                 sfc_err(sa, "SYN filters not supported");
1598                 break;
1599         case RTE_ETH_FILTER_NTUPLE:
1600                 sfc_err(sa, "NTUPLE filters not supported");
1601                 break;
1602         case RTE_ETH_FILTER_TUNNEL:
1603                 sfc_err(sa, "Tunnel filters not supported");
1604                 break;
1605         case RTE_ETH_FILTER_FDIR:
1606                 sfc_err(sa, "Flow Director filters not supported");
1607                 break;
1608         case RTE_ETH_FILTER_HASH:
1609                 sfc_err(sa, "Hash filters not supported");
1610                 break;
1611         case RTE_ETH_FILTER_GENERIC:
1612                 if (filter_op != RTE_ETH_FILTER_GET) {
1613                         rc = EINVAL;
1614                 } else {
1615                         *(const void **)arg = &sfc_flow_ops;
1616                         rc = 0;
1617                 }
1618                 break;
1619         default:
1620                 sfc_err(sa, "Unknown filter type %u", filter_type);
1621                 break;
1622         }
1623
1624         sfc_log_init(sa, "exit: %d", -rc);
1625         SFC_ASSERT(rc >= 0);
1626         return -rc;
1627 }
1628
1629 static const struct eth_dev_ops sfc_eth_dev_ops = {
1630         .dev_configure                  = sfc_dev_configure,
1631         .dev_start                      = sfc_dev_start,
1632         .dev_stop                       = sfc_dev_stop,
1633         .dev_set_link_up                = sfc_dev_set_link_up,
1634         .dev_set_link_down              = sfc_dev_set_link_down,
1635         .dev_close                      = sfc_dev_close,
1636         .promiscuous_enable             = sfc_dev_promisc_enable,
1637         .promiscuous_disable            = sfc_dev_promisc_disable,
1638         .allmulticast_enable            = sfc_dev_allmulti_enable,
1639         .allmulticast_disable           = sfc_dev_allmulti_disable,
1640         .link_update                    = sfc_dev_link_update,
1641         .stats_get                      = sfc_stats_get,
1642         .stats_reset                    = sfc_stats_reset,
1643         .xstats_get                     = sfc_xstats_get,
1644         .xstats_reset                   = sfc_stats_reset,
1645         .xstats_get_names               = sfc_xstats_get_names,
1646         .dev_infos_get                  = sfc_dev_infos_get,
1647         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1648         .mtu_set                        = sfc_dev_set_mtu,
1649         .rx_queue_start                 = sfc_rx_queue_start,
1650         .rx_queue_stop                  = sfc_rx_queue_stop,
1651         .tx_queue_start                 = sfc_tx_queue_start,
1652         .tx_queue_stop                  = sfc_tx_queue_stop,
1653         .rx_queue_setup                 = sfc_rx_queue_setup,
1654         .rx_queue_release               = sfc_rx_queue_release,
1655         .rx_queue_count                 = sfc_rx_queue_count,
1656         .rx_descriptor_done             = sfc_rx_descriptor_done,
1657         .rx_descriptor_status           = sfc_rx_descriptor_status,
1658         .tx_descriptor_status           = sfc_tx_descriptor_status,
1659         .tx_queue_setup                 = sfc_tx_queue_setup,
1660         .tx_queue_release               = sfc_tx_queue_release,
1661         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1662         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1663         .mac_addr_set                   = sfc_mac_addr_set,
1664         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1665         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1666 #if EFSYS_OPT_RX_SCALE
1667         .reta_update                    = sfc_dev_rss_reta_update,
1668         .reta_query                     = sfc_dev_rss_reta_query,
1669         .rss_hash_update                = sfc_dev_rss_hash_update,
1670         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1671 #endif
1672         .filter_ctrl                    = sfc_dev_filter_ctrl,
1673         .set_mc_addr_list               = sfc_set_mc_addr_list,
1674         .rxq_info_get                   = sfc_rx_queue_info_get,
1675         .txq_info_get                   = sfc_tx_queue_info_get,
1676         .fw_version_get                 = sfc_fw_version_get,
1677         .xstats_get_by_id               = sfc_xstats_get_by_id,
1678         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1679 };
1680
1681 /**
1682  * Duplicate a string in potentially shared memory required for
1683  * multi-process support.
1684  *
1685  * strdup() allocates from process-local heap/memory.
1686  */
1687 static char *
1688 sfc_strdup(const char *str)
1689 {
1690         size_t size;
1691         char *copy;
1692
1693         if (str == NULL)
1694                 return NULL;
1695
1696         size = strlen(str) + 1;
1697         copy = rte_malloc(__func__, size, 0);
1698         if (copy != NULL)
1699                 rte_memcpy(copy, str, size);
1700
1701         return copy;
1702 }
1703
1704 static int
1705 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1706 {
1707         struct sfc_adapter *sa = dev->data->dev_private;
1708         unsigned int avail_caps = 0;
1709         const char *rx_name = NULL;
1710         const char *tx_name = NULL;
1711         int rc;
1712
1713         switch (sa->family) {
1714         case EFX_FAMILY_HUNTINGTON:
1715         case EFX_FAMILY_MEDFORD:
1716         case EFX_FAMILY_MEDFORD2:
1717                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1718                 break;
1719         default:
1720                 break;
1721         }
1722
1723         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1724                                 sfc_kvarg_string_handler, &rx_name);
1725         if (rc != 0)
1726                 goto fail_kvarg_rx_datapath;
1727
1728         if (rx_name != NULL) {
1729                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1730                 if (sa->dp_rx == NULL) {
1731                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1732                         rc = ENOENT;
1733                         goto fail_dp_rx;
1734                 }
1735                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1736                         sfc_err(sa,
1737                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1738                                 rx_name);
1739                         rc = EINVAL;
1740                         goto fail_dp_rx_caps;
1741                 }
1742         } else {
1743                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1744                 if (sa->dp_rx == NULL) {
1745                         sfc_err(sa, "Rx datapath by caps %#x not found",
1746                                 avail_caps);
1747                         rc = ENOENT;
1748                         goto fail_dp_rx;
1749                 }
1750         }
1751
1752         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1753         if (sa->dp_rx_name == NULL) {
1754                 rc = ENOMEM;
1755                 goto fail_dp_rx_name;
1756         }
1757
1758         sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1759
1760         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1761
1762         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1763                                 sfc_kvarg_string_handler, &tx_name);
1764         if (rc != 0)
1765                 goto fail_kvarg_tx_datapath;
1766
1767         if (tx_name != NULL) {
1768                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1769                 if (sa->dp_tx == NULL) {
1770                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1771                         rc = ENOENT;
1772                         goto fail_dp_tx;
1773                 }
1774                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1775                         sfc_err(sa,
1776                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1777                                 tx_name);
1778                         rc = EINVAL;
1779                         goto fail_dp_tx_caps;
1780                 }
1781         } else {
1782                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1783                 if (sa->dp_tx == NULL) {
1784                         sfc_err(sa, "Tx datapath by caps %#x not found",
1785                                 avail_caps);
1786                         rc = ENOENT;
1787                         goto fail_dp_tx;
1788                 }
1789         }
1790
1791         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1792         if (sa->dp_tx_name == NULL) {
1793                 rc = ENOMEM;
1794                 goto fail_dp_tx_name;
1795         }
1796
1797         sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1798
1799         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1800
1801         dev->dev_ops = &sfc_eth_dev_ops;
1802
1803         return 0;
1804
1805 fail_dp_tx_name:
1806 fail_dp_tx_caps:
1807         sa->dp_tx = NULL;
1808
1809 fail_dp_tx:
1810 fail_kvarg_tx_datapath:
1811         rte_free(sa->dp_rx_name);
1812         sa->dp_rx_name = NULL;
1813
1814 fail_dp_rx_name:
1815 fail_dp_rx_caps:
1816         sa->dp_rx = NULL;
1817
1818 fail_dp_rx:
1819 fail_kvarg_rx_datapath:
1820         return rc;
1821 }
1822
1823 static void
1824 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1825 {
1826         struct sfc_adapter *sa = dev->data->dev_private;
1827
1828         dev->dev_ops = NULL;
1829         dev->rx_pkt_burst = NULL;
1830         dev->tx_pkt_burst = NULL;
1831
1832         rte_free(sa->dp_tx_name);
1833         sa->dp_tx_name = NULL;
1834         sa->dp_tx = NULL;
1835
1836         rte_free(sa->dp_rx_name);
1837         sa->dp_rx_name = NULL;
1838         sa->dp_rx = NULL;
1839 }
1840
1841 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1842         .rxq_info_get                   = sfc_rx_queue_info_get,
1843         .txq_info_get                   = sfc_tx_queue_info_get,
1844 };
1845
1846 static int
1847 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1848 {
1849         /*
1850          * Device private data has really many process-local pointers.
1851          * Below code should be extremely careful to use data located
1852          * in shared memory only.
1853          */
1854         struct sfc_adapter *sa = dev->data->dev_private;
1855         const struct sfc_dp_rx *dp_rx;
1856         const struct sfc_dp_tx *dp_tx;
1857         int rc;
1858
1859         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1860         if (dp_rx == NULL) {
1861                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1862                 rc = ENOENT;
1863                 goto fail_dp_rx;
1864         }
1865         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1866                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1867                         sa->dp_tx_name);
1868                 rc = EINVAL;
1869                 goto fail_dp_rx_multi_process;
1870         }
1871
1872         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1873         if (dp_tx == NULL) {
1874                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1875                 rc = ENOENT;
1876                 goto fail_dp_tx;
1877         }
1878         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1879                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1880                         sa->dp_tx_name);
1881                 rc = EINVAL;
1882                 goto fail_dp_tx_multi_process;
1883         }
1884
1885         dev->rx_pkt_burst = dp_rx->pkt_burst;
1886         dev->tx_pkt_burst = dp_tx->pkt_burst;
1887         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1888
1889         return 0;
1890
1891 fail_dp_tx_multi_process:
1892 fail_dp_tx:
1893 fail_dp_rx_multi_process:
1894 fail_dp_rx:
1895         return rc;
1896 }
1897
1898 static void
1899 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1900 {
1901         dev->dev_ops = NULL;
1902         dev->tx_pkt_burst = NULL;
1903         dev->rx_pkt_burst = NULL;
1904 }
1905
1906 static void
1907 sfc_register_dp(void)
1908 {
1909         /* Register once */
1910         if (TAILQ_EMPTY(&sfc_dp_head)) {
1911                 /* Prefer EF10 datapath */
1912                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1913                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1914
1915                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1916                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1917                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1918         }
1919 }
1920
1921 static int
1922 sfc_eth_dev_init(struct rte_eth_dev *dev)
1923 {
1924         struct sfc_adapter *sa = dev->data->dev_private;
1925         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1926         int rc;
1927         const efx_nic_cfg_t *encp;
1928         const struct ether_addr *from;
1929
1930         sfc_register_dp();
1931
1932         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1933                 return -sfc_eth_dev_secondary_set_ops(dev);
1934
1935         /* Required for logging */
1936         sa->pci_addr = pci_dev->addr;
1937         sa->port_id = dev->data->port_id;
1938
1939         sa->eth_dev = dev;
1940
1941         /* Copy PCI device info to the dev->data */
1942         rte_eth_copy_pci_info(dev, pci_dev);
1943
1944         sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1945                                                 RTE_LOG_NOTICE);
1946
1947         rc = sfc_kvargs_parse(sa);
1948         if (rc != 0)
1949                 goto fail_kvargs_parse;
1950
1951         sfc_log_init(sa, "entry");
1952
1953         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1954         if (dev->data->mac_addrs == NULL) {
1955                 rc = ENOMEM;
1956                 goto fail_mac_addrs;
1957         }
1958
1959         sfc_adapter_lock_init(sa);
1960         sfc_adapter_lock(sa);
1961
1962         sfc_log_init(sa, "probing");
1963         rc = sfc_probe(sa);
1964         if (rc != 0)
1965                 goto fail_probe;
1966
1967         sfc_log_init(sa, "set device ops");
1968         rc = sfc_eth_dev_set_ops(dev);
1969         if (rc != 0)
1970                 goto fail_set_ops;
1971
1972         sfc_log_init(sa, "attaching");
1973         rc = sfc_attach(sa);
1974         if (rc != 0)
1975                 goto fail_attach;
1976
1977         encp = efx_nic_cfg_get(sa->nic);
1978
1979         /*
1980          * The arguments are really reverse order in comparison to
1981          * Linux kernel. Copy from NIC config to Ethernet device data.
1982          */
1983         from = (const struct ether_addr *)(encp->enc_mac_addr);
1984         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1985
1986         sfc_adapter_unlock(sa);
1987
1988         sfc_log_init(sa, "done");
1989         return 0;
1990
1991 fail_attach:
1992         sfc_eth_dev_clear_ops(dev);
1993
1994 fail_set_ops:
1995         sfc_unprobe(sa);
1996
1997 fail_probe:
1998         sfc_adapter_unlock(sa);
1999         sfc_adapter_lock_fini(sa);
2000         rte_free(dev->data->mac_addrs);
2001         dev->data->mac_addrs = NULL;
2002
2003 fail_mac_addrs:
2004         sfc_kvargs_cleanup(sa);
2005
2006 fail_kvargs_parse:
2007         sfc_log_init(sa, "failed %d", rc);
2008         SFC_ASSERT(rc > 0);
2009         return -rc;
2010 }
2011
2012 static int
2013 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2014 {
2015         struct sfc_adapter *sa;
2016
2017         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2018                 sfc_eth_dev_secondary_clear_ops(dev);
2019                 return 0;
2020         }
2021
2022         sa = dev->data->dev_private;
2023         sfc_log_init(sa, "entry");
2024
2025         sfc_adapter_lock(sa);
2026
2027         sfc_eth_dev_clear_ops(dev);
2028
2029         sfc_detach(sa);
2030         sfc_unprobe(sa);
2031
2032         rte_free(dev->data->mac_addrs);
2033         dev->data->mac_addrs = NULL;
2034
2035         sfc_kvargs_cleanup(sa);
2036
2037         sfc_adapter_unlock(sa);
2038         sfc_adapter_lock_fini(sa);
2039
2040         sfc_log_init(sa, "done");
2041
2042         /* Required for logging, so cleanup last */
2043         sa->eth_dev = NULL;
2044         return 0;
2045 }
2046
2047 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2048         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2049         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2050         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2051         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2052         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2053         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2054         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2055         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2056         { .vendor_id = 0 /* sentinel */ }
2057 };
2058
2059 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2060         struct rte_pci_device *pci_dev)
2061 {
2062         return rte_eth_dev_pci_generic_probe(pci_dev,
2063                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2064 }
2065
2066 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2067 {
2068         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2069 }
2070
2071 static struct rte_pci_driver sfc_efx_pmd = {
2072         .id_table = pci_id_sfc_efx_map,
2073         .drv_flags =
2074                 RTE_PCI_DRV_INTR_LSC |
2075                 RTE_PCI_DRV_NEED_MAPPING,
2076         .probe = sfc_eth_dev_pci_probe,
2077         .remove = sfc_eth_dev_pci_remove,
2078 };
2079
2080 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2081 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2082 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2083 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2084         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2085         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2086         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2087         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2088         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2089
2090 RTE_INIT(sfc_driver_register_logtype);
2091 static void
2092 sfc_driver_register_logtype(void)
2093 {
2094         int ret;
2095
2096         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2097                                                    RTE_LOG_NOTICE);
2098         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2099 }