837fd5555e4223cd0c63900e6501cb6a6104e50c
[dpdk.git] / drivers / net / sfc / sfc_ethdev.c
1 /*-
2  *   BSD LICENSE
3  *
4  * Copyright (c) 2016-2017 Solarflare Communications Inc.
5  * All rights reserved.
6  *
7  * This software was jointly developed between OKTET Labs (under contract
8  * for Solarflare) and Solarflare Communications, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright notice,
14  *    this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright notice,
16  *    this list of conditions and the following disclaimer in the documentation
17  *    and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
21  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
23  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
24  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
25  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
26  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
28  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
29  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31
32 #include <rte_dev.h>
33 #include <rte_ethdev.h>
34 #include <rte_ethdev_pci.h>
35 #include <rte_pci.h>
36 #include <rte_bus_pci.h>
37 #include <rte_errno.h>
38
39 #include "efx.h"
40
41 #include "sfc.h"
42 #include "sfc_debug.h"
43 #include "sfc_log.h"
44 #include "sfc_kvargs.h"
45 #include "sfc_ev.h"
46 #include "sfc_rx.h"
47 #include "sfc_tx.h"
48 #include "sfc_flow.h"
49 #include "sfc_dp.h"
50 #include "sfc_dp_rx.h"
51
52 static struct sfc_dp_list sfc_dp_head =
53         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
54
55 static int
56 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
57 {
58         struct sfc_adapter *sa = dev->data->dev_private;
59         efx_nic_fw_info_t enfi;
60         int ret;
61         int rc;
62
63         /*
64          * Return value of the callback is likely supposed to be
65          * equal to or greater than 0, nevertheless, if an error
66          * occurs, it will be desirable to pass it to the caller
67          */
68         if ((fw_version == NULL) || (fw_size == 0))
69                 return -EINVAL;
70
71         rc = efx_nic_get_fw_version(sa->nic, &enfi);
72         if (rc != 0)
73                 return -rc;
74
75         ret = snprintf(fw_version, fw_size,
76                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
77                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
78                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
79         if (ret < 0)
80                 return ret;
81
82         if (enfi.enfi_dpcpu_fw_ids_valid) {
83                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
84                 int ret_extra;
85
86                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
87                                      fw_size - dpcpu_fw_ids_offset,
88                                      " rx%" PRIx16 " tx%" PRIx16,
89                                      enfi.enfi_rx_dpcpu_fw_id,
90                                      enfi.enfi_tx_dpcpu_fw_id);
91                 if (ret_extra < 0)
92                         return ret_extra;
93
94                 ret += ret_extra;
95         }
96
97         if (fw_size < (size_t)(++ret))
98                 return ret;
99         else
100                 return 0;
101 }
102
103 static void
104 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
105 {
106         struct sfc_adapter *sa = dev->data->dev_private;
107         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
108
109         sfc_log_init(sa, "entry");
110
111         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
112         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
113
114         /* Autonegotiation may be disabled */
115         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
116         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
117                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
118         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
119                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
120         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
121                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
122
123         dev_info->max_rx_queues = sa->rxq_max;
124         dev_info->max_tx_queues = sa->txq_max;
125
126         /* By default packets are dropped if no descriptors are available */
127         dev_info->default_rxconf.rx_drop_en = 1;
128
129         dev_info->rx_offload_capa =
130                 DEV_RX_OFFLOAD_IPV4_CKSUM |
131                 DEV_RX_OFFLOAD_UDP_CKSUM |
132                 DEV_RX_OFFLOAD_TCP_CKSUM;
133
134         dev_info->tx_offload_capa =
135                 DEV_TX_OFFLOAD_IPV4_CKSUM |
136                 DEV_TX_OFFLOAD_UDP_CKSUM |
137                 DEV_TX_OFFLOAD_TCP_CKSUM;
138
139         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
140         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
141             !encp->enc_hw_tx_insert_vlan_enabled)
142                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
143         else
144                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
145
146         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
147                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
148
149         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
150                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
151
152         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
153                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
154
155 #if EFSYS_OPT_RX_SCALE
156         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
157                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
158                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
159                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
160         }
161 #endif
162
163         if (sa->tso)
164                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
165
166         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
167         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
168         /* The RXQ hardware requires that the descriptor count is a power
169          * of 2, but rx_desc_lim cannot properly describe that constraint.
170          */
171         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
172
173         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
174         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
175         /*
176          * The TXQ hardware requires that the descriptor count is a power
177          * of 2, but tx_desc_lim cannot properly describe that constraint
178          */
179         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
180 }
181
182 static const uint32_t *
183 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
184 {
185         struct sfc_adapter *sa = dev->data->dev_private;
186
187         return sa->dp_rx->supported_ptypes_get();
188 }
189
190 static int
191 sfc_dev_configure(struct rte_eth_dev *dev)
192 {
193         struct rte_eth_dev_data *dev_data = dev->data;
194         struct sfc_adapter *sa = dev_data->dev_private;
195         int rc;
196
197         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
198                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
199
200         sfc_adapter_lock(sa);
201         switch (sa->state) {
202         case SFC_ADAPTER_CONFIGURED:
203                 /* FALLTHROUGH */
204         case SFC_ADAPTER_INITIALIZED:
205                 rc = sfc_configure(sa);
206                 break;
207         default:
208                 sfc_err(sa, "unexpected adapter state %u to configure",
209                         sa->state);
210                 rc = EINVAL;
211                 break;
212         }
213         sfc_adapter_unlock(sa);
214
215         sfc_log_init(sa, "done %d", rc);
216         SFC_ASSERT(rc >= 0);
217         return -rc;
218 }
219
220 static int
221 sfc_dev_start(struct rte_eth_dev *dev)
222 {
223         struct sfc_adapter *sa = dev->data->dev_private;
224         int rc;
225
226         sfc_log_init(sa, "entry");
227
228         sfc_adapter_lock(sa);
229         rc = sfc_start(sa);
230         sfc_adapter_unlock(sa);
231
232         sfc_log_init(sa, "done %d", rc);
233         SFC_ASSERT(rc >= 0);
234         return -rc;
235 }
236
237 static int
238 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
239 {
240         struct sfc_adapter *sa = dev->data->dev_private;
241         struct rte_eth_link *dev_link = &dev->data->dev_link;
242         struct rte_eth_link old_link;
243         struct rte_eth_link current_link;
244
245         sfc_log_init(sa, "entry");
246
247 retry:
248         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
249         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
250
251         if (sa->state != SFC_ADAPTER_STARTED) {
252                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
253                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
254                                          *(uint64_t *)&old_link,
255                                          *(uint64_t *)&current_link))
256                         goto retry;
257         } else if (wait_to_complete) {
258                 efx_link_mode_t link_mode;
259
260                 if (efx_port_poll(sa->nic, &link_mode) != 0)
261                         link_mode = EFX_LINK_UNKNOWN;
262                 sfc_port_link_mode_to_info(link_mode, &current_link);
263
264                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
265                                          *(uint64_t *)&old_link,
266                                          *(uint64_t *)&current_link))
267                         goto retry;
268         } else {
269                 sfc_ev_mgmt_qpoll(sa);
270                 *(int64_t *)&current_link =
271                         rte_atomic64_read((rte_atomic64_t *)dev_link);
272         }
273
274         if (old_link.link_status != current_link.link_status)
275                 sfc_info(sa, "Link status is %s",
276                          current_link.link_status ? "UP" : "DOWN");
277
278         return old_link.link_status == current_link.link_status ? 0 : -1;
279 }
280
281 static void
282 sfc_dev_stop(struct rte_eth_dev *dev)
283 {
284         struct sfc_adapter *sa = dev->data->dev_private;
285
286         sfc_log_init(sa, "entry");
287
288         sfc_adapter_lock(sa);
289         sfc_stop(sa);
290         sfc_adapter_unlock(sa);
291
292         sfc_log_init(sa, "done");
293 }
294
295 static int
296 sfc_dev_set_link_up(struct rte_eth_dev *dev)
297 {
298         struct sfc_adapter *sa = dev->data->dev_private;
299         int rc;
300
301         sfc_log_init(sa, "entry");
302
303         sfc_adapter_lock(sa);
304         rc = sfc_start(sa);
305         sfc_adapter_unlock(sa);
306
307         SFC_ASSERT(rc >= 0);
308         return -rc;
309 }
310
311 static int
312 sfc_dev_set_link_down(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         sfc_stop(sa);
320         sfc_adapter_unlock(sa);
321
322         return 0;
323 }
324
325 static void
326 sfc_dev_close(struct rte_eth_dev *dev)
327 {
328         struct sfc_adapter *sa = dev->data->dev_private;
329
330         sfc_log_init(sa, "entry");
331
332         sfc_adapter_lock(sa);
333         switch (sa->state) {
334         case SFC_ADAPTER_STARTED:
335                 sfc_stop(sa);
336                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
337                 /* FALLTHROUGH */
338         case SFC_ADAPTER_CONFIGURED:
339                 sfc_close(sa);
340                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
341                 /* FALLTHROUGH */
342         case SFC_ADAPTER_INITIALIZED:
343                 break;
344         default:
345                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
346                 break;
347         }
348         sfc_adapter_unlock(sa);
349
350         sfc_log_init(sa, "done");
351 }
352
353 static void
354 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
355                    boolean_t enabled)
356 {
357         struct sfc_port *port;
358         boolean_t *toggle;
359         struct sfc_adapter *sa = dev->data->dev_private;
360         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
361         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
362
363         sfc_adapter_lock(sa);
364
365         port = &sa->port;
366         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
367
368         if (*toggle != enabled) {
369                 *toggle = enabled;
370
371                 if (port->isolated) {
372                         sfc_warn(sa, "isolated mode is active on the port");
373                         sfc_warn(sa, "the change is to be applied on the next "
374                                      "start provided that isolated mode is "
375                                      "disabled prior the next start");
376                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
377                            (sfc_set_rx_mode(sa) != 0)) {
378                         *toggle = !(enabled);
379                         sfc_warn(sa, "Failed to %s %s mode",
380                                  ((enabled) ? "enable" : "disable"), desc);
381                 }
382         }
383
384         sfc_adapter_unlock(sa);
385 }
386
387 static void
388 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
389 {
390         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
391 }
392
393 static void
394 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
395 {
396         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
397 }
398
399 static void
400 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
401 {
402         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
403 }
404
405 static void
406 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
407 {
408         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
409 }
410
411 static int
412 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
413                    uint16_t nb_rx_desc, unsigned int socket_id,
414                    const struct rte_eth_rxconf *rx_conf,
415                    struct rte_mempool *mb_pool)
416 {
417         struct sfc_adapter *sa = dev->data->dev_private;
418         int rc;
419
420         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
421                      rx_queue_id, nb_rx_desc, socket_id);
422
423         sfc_adapter_lock(sa);
424
425         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
426                           rx_conf, mb_pool);
427         if (rc != 0)
428                 goto fail_rx_qinit;
429
430         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
431
432         sfc_adapter_unlock(sa);
433
434         return 0;
435
436 fail_rx_qinit:
437         sfc_adapter_unlock(sa);
438         SFC_ASSERT(rc > 0);
439         return -rc;
440 }
441
442 static void
443 sfc_rx_queue_release(void *queue)
444 {
445         struct sfc_dp_rxq *dp_rxq = queue;
446         struct sfc_rxq *rxq;
447         struct sfc_adapter *sa;
448         unsigned int sw_index;
449
450         if (dp_rxq == NULL)
451                 return;
452
453         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
454         sa = rxq->evq->sa;
455         sfc_adapter_lock(sa);
456
457         sw_index = sfc_rxq_sw_index(rxq);
458
459         sfc_log_init(sa, "RxQ=%u", sw_index);
460
461         sa->eth_dev->data->rx_queues[sw_index] = NULL;
462
463         sfc_rx_qfini(sa, sw_index);
464
465         sfc_adapter_unlock(sa);
466 }
467
468 static int
469 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
470                    uint16_t nb_tx_desc, unsigned int socket_id,
471                    const struct rte_eth_txconf *tx_conf)
472 {
473         struct sfc_adapter *sa = dev->data->dev_private;
474         int rc;
475
476         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
477                      tx_queue_id, nb_tx_desc, socket_id);
478
479         sfc_adapter_lock(sa);
480
481         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
482         if (rc != 0)
483                 goto fail_tx_qinit;
484
485         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
486
487         sfc_adapter_unlock(sa);
488         return 0;
489
490 fail_tx_qinit:
491         sfc_adapter_unlock(sa);
492         SFC_ASSERT(rc > 0);
493         return -rc;
494 }
495
496 static void
497 sfc_tx_queue_release(void *queue)
498 {
499         struct sfc_dp_txq *dp_txq = queue;
500         struct sfc_txq *txq;
501         unsigned int sw_index;
502         struct sfc_adapter *sa;
503
504         if (dp_txq == NULL)
505                 return;
506
507         txq = sfc_txq_by_dp_txq(dp_txq);
508         sw_index = sfc_txq_sw_index(txq);
509
510         SFC_ASSERT(txq->evq != NULL);
511         sa = txq->evq->sa;
512
513         sfc_log_init(sa, "TxQ = %u", sw_index);
514
515         sfc_adapter_lock(sa);
516
517         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
518         sa->eth_dev->data->tx_queues[sw_index] = NULL;
519
520         sfc_tx_qfini(sa, sw_index);
521
522         sfc_adapter_unlock(sa);
523 }
524
525 static int
526 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
527 {
528         struct sfc_adapter *sa = dev->data->dev_private;
529         struct sfc_port *port = &sa->port;
530         uint64_t *mac_stats;
531         int ret;
532
533         rte_spinlock_lock(&port->mac_stats_lock);
534
535         ret = sfc_port_update_mac_stats(sa);
536         if (ret != 0)
537                 goto unlock;
538
539         mac_stats = port->mac_stats_buf;
540
541         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
542                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
543                 stats->ipackets =
544                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
545                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
546                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
547                 stats->opackets =
548                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
549                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
550                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
551                 stats->ibytes =
552                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
553                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
554                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
555                 stats->obytes =
556                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
557                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
558                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
559                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
560                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
561                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
562         } else {
563                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
564                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
565                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
566                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
567                 /*
568                  * Take into account stats which are whenever supported
569                  * on EF10. If some stat is not supported by current
570                  * firmware variant or HW revision, it is guaranteed
571                  * to be zero in mac_stats.
572                  */
573                 stats->imissed =
574                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
575                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
576                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
577                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
578                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
579                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
580                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
581                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
582                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
583                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
584                 stats->ierrors =
585                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
586                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
587                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
588                 /* no oerrors counters supported on EF10 */
589         }
590
591 unlock:
592         rte_spinlock_unlock(&port->mac_stats_lock);
593         SFC_ASSERT(ret >= 0);
594         return -ret;
595 }
596
597 static void
598 sfc_stats_reset(struct rte_eth_dev *dev)
599 {
600         struct sfc_adapter *sa = dev->data->dev_private;
601         struct sfc_port *port = &sa->port;
602         int rc;
603
604         if (sa->state != SFC_ADAPTER_STARTED) {
605                 /*
606                  * The operation cannot be done if port is not started; it
607                  * will be scheduled to be done during the next port start
608                  */
609                 port->mac_stats_reset_pending = B_TRUE;
610                 return;
611         }
612
613         rc = sfc_port_reset_mac_stats(sa);
614         if (rc != 0)
615                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
616 }
617
618 static int
619 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
620                unsigned int xstats_count)
621 {
622         struct sfc_adapter *sa = dev->data->dev_private;
623         struct sfc_port *port = &sa->port;
624         uint64_t *mac_stats;
625         int rc;
626         unsigned int i;
627         int nstats = 0;
628
629         rte_spinlock_lock(&port->mac_stats_lock);
630
631         rc = sfc_port_update_mac_stats(sa);
632         if (rc != 0) {
633                 SFC_ASSERT(rc > 0);
634                 nstats = -rc;
635                 goto unlock;
636         }
637
638         mac_stats = port->mac_stats_buf;
639
640         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
641                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
642                         if (xstats != NULL && nstats < (int)xstats_count) {
643                                 xstats[nstats].id = nstats;
644                                 xstats[nstats].value = mac_stats[i];
645                         }
646                         nstats++;
647                 }
648         }
649
650 unlock:
651         rte_spinlock_unlock(&port->mac_stats_lock);
652
653         return nstats;
654 }
655
656 static int
657 sfc_xstats_get_names(struct rte_eth_dev *dev,
658                      struct rte_eth_xstat_name *xstats_names,
659                      unsigned int xstats_count)
660 {
661         struct sfc_adapter *sa = dev->data->dev_private;
662         struct sfc_port *port = &sa->port;
663         unsigned int i;
664         unsigned int nstats = 0;
665
666         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
667                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
668                         if (xstats_names != NULL && nstats < xstats_count)
669                                 strncpy(xstats_names[nstats].name,
670                                         efx_mac_stat_name(sa->nic, i),
671                                         sizeof(xstats_names[0].name));
672                         nstats++;
673                 }
674         }
675
676         return nstats;
677 }
678
679 static int
680 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
681                      uint64_t *values, unsigned int n)
682 {
683         struct sfc_adapter *sa = dev->data->dev_private;
684         struct sfc_port *port = &sa->port;
685         uint64_t *mac_stats;
686         unsigned int nb_supported = 0;
687         unsigned int nb_written = 0;
688         unsigned int i;
689         int ret;
690         int rc;
691
692         if (unlikely(values == NULL) ||
693             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
694                 return port->mac_stats_nb_supported;
695
696         rte_spinlock_lock(&port->mac_stats_lock);
697
698         rc = sfc_port_update_mac_stats(sa);
699         if (rc != 0) {
700                 SFC_ASSERT(rc > 0);
701                 ret = -rc;
702                 goto unlock;
703         }
704
705         mac_stats = port->mac_stats_buf;
706
707         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
708                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
709                         continue;
710
711                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
712                         values[nb_written++] = mac_stats[i];
713
714                 ++nb_supported;
715         }
716
717         ret = nb_written;
718
719 unlock:
720         rte_spinlock_unlock(&port->mac_stats_lock);
721
722         return ret;
723 }
724
725 static int
726 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
727                            struct rte_eth_xstat_name *xstats_names,
728                            const uint64_t *ids, unsigned int size)
729 {
730         struct sfc_adapter *sa = dev->data->dev_private;
731         struct sfc_port *port = &sa->port;
732         unsigned int nb_supported = 0;
733         unsigned int nb_written = 0;
734         unsigned int i;
735
736         if (unlikely(xstats_names == NULL) ||
737             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
738                 return port->mac_stats_nb_supported;
739
740         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
741                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
742                         continue;
743
744                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
745                         char *name = xstats_names[nb_written++].name;
746
747                         strncpy(name, efx_mac_stat_name(sa->nic, i),
748                                 sizeof(xstats_names[0].name));
749                         name[sizeof(xstats_names[0].name) - 1] = '\0';
750                 }
751
752                 ++nb_supported;
753         }
754
755         return nb_written;
756 }
757
758 static int
759 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
760 {
761         struct sfc_adapter *sa = dev->data->dev_private;
762         unsigned int wanted_fc, link_fc;
763
764         memset(fc_conf, 0, sizeof(*fc_conf));
765
766         sfc_adapter_lock(sa);
767
768         if (sa->state == SFC_ADAPTER_STARTED)
769                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
770         else
771                 link_fc = sa->port.flow_ctrl;
772
773         switch (link_fc) {
774         case 0:
775                 fc_conf->mode = RTE_FC_NONE;
776                 break;
777         case EFX_FCNTL_RESPOND:
778                 fc_conf->mode = RTE_FC_RX_PAUSE;
779                 break;
780         case EFX_FCNTL_GENERATE:
781                 fc_conf->mode = RTE_FC_TX_PAUSE;
782                 break;
783         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
784                 fc_conf->mode = RTE_FC_FULL;
785                 break;
786         default:
787                 sfc_err(sa, "%s: unexpected flow control value %#x",
788                         __func__, link_fc);
789         }
790
791         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
792
793         sfc_adapter_unlock(sa);
794
795         return 0;
796 }
797
798 static int
799 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
800 {
801         struct sfc_adapter *sa = dev->data->dev_private;
802         struct sfc_port *port = &sa->port;
803         unsigned int fcntl;
804         int rc;
805
806         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
807             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
808             fc_conf->mac_ctrl_frame_fwd != 0) {
809                 sfc_err(sa, "unsupported flow control settings specified");
810                 rc = EINVAL;
811                 goto fail_inval;
812         }
813
814         switch (fc_conf->mode) {
815         case RTE_FC_NONE:
816                 fcntl = 0;
817                 break;
818         case RTE_FC_RX_PAUSE:
819                 fcntl = EFX_FCNTL_RESPOND;
820                 break;
821         case RTE_FC_TX_PAUSE:
822                 fcntl = EFX_FCNTL_GENERATE;
823                 break;
824         case RTE_FC_FULL:
825                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
826                 break;
827         default:
828                 rc = EINVAL;
829                 goto fail_inval;
830         }
831
832         sfc_adapter_lock(sa);
833
834         if (sa->state == SFC_ADAPTER_STARTED) {
835                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
836                 if (rc != 0)
837                         goto fail_mac_fcntl_set;
838         }
839
840         port->flow_ctrl = fcntl;
841         port->flow_ctrl_autoneg = fc_conf->autoneg;
842
843         sfc_adapter_unlock(sa);
844
845         return 0;
846
847 fail_mac_fcntl_set:
848         sfc_adapter_unlock(sa);
849 fail_inval:
850         SFC_ASSERT(rc > 0);
851         return -rc;
852 }
853
854 static int
855 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
856 {
857         struct sfc_adapter *sa = dev->data->dev_private;
858         size_t pdu = EFX_MAC_PDU(mtu);
859         size_t old_pdu;
860         int rc;
861
862         sfc_log_init(sa, "mtu=%u", mtu);
863
864         rc = EINVAL;
865         if (pdu < EFX_MAC_PDU_MIN) {
866                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
867                         (unsigned int)mtu, (unsigned int)pdu,
868                         EFX_MAC_PDU_MIN);
869                 goto fail_inval;
870         }
871         if (pdu > EFX_MAC_PDU_MAX) {
872                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
873                         (unsigned int)mtu, (unsigned int)pdu,
874                         EFX_MAC_PDU_MAX);
875                 goto fail_inval;
876         }
877
878         sfc_adapter_lock(sa);
879
880         if (pdu != sa->port.pdu) {
881                 if (sa->state == SFC_ADAPTER_STARTED) {
882                         sfc_stop(sa);
883
884                         old_pdu = sa->port.pdu;
885                         sa->port.pdu = pdu;
886                         rc = sfc_start(sa);
887                         if (rc != 0)
888                                 goto fail_start;
889                 } else {
890                         sa->port.pdu = pdu;
891                 }
892         }
893
894         /*
895          * The driver does not use it, but other PMDs update jumbo_frame
896          * flag and max_rx_pkt_len when MTU is set.
897          */
898         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
899         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
900
901         sfc_adapter_unlock(sa);
902
903         sfc_log_init(sa, "done");
904         return 0;
905
906 fail_start:
907         sa->port.pdu = old_pdu;
908         if (sfc_start(sa) != 0)
909                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
910                         "PDU max size - port is stopped",
911                         (unsigned int)pdu, (unsigned int)old_pdu);
912         sfc_adapter_unlock(sa);
913
914 fail_inval:
915         sfc_log_init(sa, "failed %d", rc);
916         SFC_ASSERT(rc > 0);
917         return -rc;
918 }
919 static void
920 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
921 {
922         struct sfc_adapter *sa = dev->data->dev_private;
923         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
924         struct sfc_port *port = &sa->port;
925         int rc;
926
927         sfc_adapter_lock(sa);
928
929         /*
930          * Copy the address to the device private data so that
931          * it could be recalled in the case of adapter restart.
932          */
933         ether_addr_copy(mac_addr, &port->default_mac_addr);
934
935         if (port->isolated) {
936                 sfc_err(sa, "isolated mode is active on the port");
937                 sfc_err(sa, "will not set MAC address");
938                 goto unlock;
939         }
940
941         if (sa->state != SFC_ADAPTER_STARTED) {
942                 sfc_info(sa, "the port is not started");
943                 sfc_info(sa, "the new MAC address will be set on port start");
944
945                 goto unlock;
946         }
947
948         if (encp->enc_allow_set_mac_with_installed_filters) {
949                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
950                 if (rc != 0) {
951                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
952                         goto unlock;
953                 }
954
955                 /*
956                  * Changing the MAC address by means of MCDI request
957                  * has no effect on received traffic, therefore
958                  * we also need to update unicast filters
959                  */
960                 rc = sfc_set_rx_mode(sa);
961                 if (rc != 0)
962                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
963         } else {
964                 sfc_warn(sa, "cannot set MAC address with filters installed");
965                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
966                 sfc_warn(sa, "(some traffic may be dropped)");
967
968                 /*
969                  * Since setting MAC address with filters installed is not
970                  * allowed on the adapter, the new MAC address will be set
971                  * by means of adapter restart. sfc_start() shall retrieve
972                  * the new address from the device private data and set it.
973                  */
974                 sfc_stop(sa);
975                 rc = sfc_start(sa);
976                 if (rc != 0)
977                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
978         }
979
980 unlock:
981         /*
982          * In the case of failure sa->port->default_mac_addr does not
983          * need rollback since no error code is returned, and the upper
984          * API will anyway update the external MAC address storage.
985          * To be consistent with that new value it is better to keep
986          * the device private value the same.
987          */
988         sfc_adapter_unlock(sa);
989 }
990
991
992 static int
993 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
994                      uint32_t nb_mc_addr)
995 {
996         struct sfc_adapter *sa = dev->data->dev_private;
997         struct sfc_port *port = &sa->port;
998         uint8_t *mc_addrs = port->mcast_addrs;
999         int rc;
1000         unsigned int i;
1001
1002         if (port->isolated) {
1003                 sfc_err(sa, "isolated mode is active on the port");
1004                 sfc_err(sa, "will not set multicast address list");
1005                 return -ENOTSUP;
1006         }
1007
1008         if (mc_addrs == NULL)
1009                 return -ENOBUFS;
1010
1011         if (nb_mc_addr > port->max_mcast_addrs) {
1012                 sfc_err(sa, "too many multicast addresses: %u > %u",
1013                          nb_mc_addr, port->max_mcast_addrs);
1014                 return -EINVAL;
1015         }
1016
1017         for (i = 0; i < nb_mc_addr; ++i) {
1018                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1019                                  EFX_MAC_ADDR_LEN);
1020                 mc_addrs += EFX_MAC_ADDR_LEN;
1021         }
1022
1023         port->nb_mcast_addrs = nb_mc_addr;
1024
1025         if (sa->state != SFC_ADAPTER_STARTED)
1026                 return 0;
1027
1028         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1029                                         port->nb_mcast_addrs);
1030         if (rc != 0)
1031                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1032
1033         SFC_ASSERT(rc > 0);
1034         return -rc;
1035 }
1036
1037 /*
1038  * The function is used by the secondary process as well. It must not
1039  * use any process-local pointers from the adapter data.
1040  */
1041 static void
1042 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1043                       struct rte_eth_rxq_info *qinfo)
1044 {
1045         struct sfc_adapter *sa = dev->data->dev_private;
1046         struct sfc_rxq_info *rxq_info;
1047         struct sfc_rxq *rxq;
1048
1049         sfc_adapter_lock(sa);
1050
1051         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1052
1053         rxq_info = &sa->rxq_info[rx_queue_id];
1054         rxq = rxq_info->rxq;
1055         SFC_ASSERT(rxq != NULL);
1056
1057         qinfo->mp = rxq->refill_mb_pool;
1058         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1059         qinfo->conf.rx_drop_en = 1;
1060         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1061         qinfo->scattered_rx =
1062                 ((rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) != 0);
1063         qinfo->nb_desc = rxq_info->entries;
1064
1065         sfc_adapter_unlock(sa);
1066 }
1067
1068 /*
1069  * The function is used by the secondary process as well. It must not
1070  * use any process-local pointers from the adapter data.
1071  */
1072 static void
1073 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1074                       struct rte_eth_txq_info *qinfo)
1075 {
1076         struct sfc_adapter *sa = dev->data->dev_private;
1077         struct sfc_txq_info *txq_info;
1078
1079         sfc_adapter_lock(sa);
1080
1081         SFC_ASSERT(tx_queue_id < sa->txq_count);
1082
1083         txq_info = &sa->txq_info[tx_queue_id];
1084         SFC_ASSERT(txq_info->txq != NULL);
1085
1086         memset(qinfo, 0, sizeof(*qinfo));
1087
1088         qinfo->conf.txq_flags = txq_info->txq->flags;
1089         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1090         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1091         qinfo->nb_desc = txq_info->entries;
1092
1093         sfc_adapter_unlock(sa);
1094 }
1095
1096 static uint32_t
1097 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1098 {
1099         struct sfc_adapter *sa = dev->data->dev_private;
1100
1101         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1102
1103         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1104 }
1105
1106 static int
1107 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1108 {
1109         struct sfc_dp_rxq *dp_rxq = queue;
1110
1111         return sfc_rx_qdesc_done(dp_rxq, offset);
1112 }
1113
1114 static int
1115 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1116 {
1117         struct sfc_dp_rxq *dp_rxq = queue;
1118         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1119
1120         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1121 }
1122
1123 static int
1124 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1125 {
1126         struct sfc_dp_txq *dp_txq = queue;
1127         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1128
1129         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1130 }
1131
1132 static int
1133 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1134 {
1135         struct sfc_adapter *sa = dev->data->dev_private;
1136         int rc;
1137
1138         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1139
1140         sfc_adapter_lock(sa);
1141
1142         rc = EINVAL;
1143         if (sa->state != SFC_ADAPTER_STARTED)
1144                 goto fail_not_started;
1145
1146         rc = sfc_rx_qstart(sa, rx_queue_id);
1147         if (rc != 0)
1148                 goto fail_rx_qstart;
1149
1150         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1151
1152         sfc_adapter_unlock(sa);
1153
1154         return 0;
1155
1156 fail_rx_qstart:
1157 fail_not_started:
1158         sfc_adapter_unlock(sa);
1159         SFC_ASSERT(rc > 0);
1160         return -rc;
1161 }
1162
1163 static int
1164 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1165 {
1166         struct sfc_adapter *sa = dev->data->dev_private;
1167
1168         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1169
1170         sfc_adapter_lock(sa);
1171         sfc_rx_qstop(sa, rx_queue_id);
1172
1173         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1174
1175         sfc_adapter_unlock(sa);
1176
1177         return 0;
1178 }
1179
1180 static int
1181 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1182 {
1183         struct sfc_adapter *sa = dev->data->dev_private;
1184         int rc;
1185
1186         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1187
1188         sfc_adapter_lock(sa);
1189
1190         rc = EINVAL;
1191         if (sa->state != SFC_ADAPTER_STARTED)
1192                 goto fail_not_started;
1193
1194         rc = sfc_tx_qstart(sa, tx_queue_id);
1195         if (rc != 0)
1196                 goto fail_tx_qstart;
1197
1198         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1199
1200         sfc_adapter_unlock(sa);
1201         return 0;
1202
1203 fail_tx_qstart:
1204
1205 fail_not_started:
1206         sfc_adapter_unlock(sa);
1207         SFC_ASSERT(rc > 0);
1208         return -rc;
1209 }
1210
1211 static int
1212 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1213 {
1214         struct sfc_adapter *sa = dev->data->dev_private;
1215
1216         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1217
1218         sfc_adapter_lock(sa);
1219
1220         sfc_tx_qstop(sa, tx_queue_id);
1221
1222         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1223
1224         sfc_adapter_unlock(sa);
1225         return 0;
1226 }
1227
1228 static efx_tunnel_protocol_t
1229 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1230 {
1231         switch (rte_type) {
1232         case RTE_TUNNEL_TYPE_VXLAN:
1233                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1234         case RTE_TUNNEL_TYPE_GENEVE:
1235                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1236         default:
1237                 return EFX_TUNNEL_NPROTOS;
1238         }
1239 }
1240
1241 enum sfc_udp_tunnel_op_e {
1242         SFC_UDP_TUNNEL_ADD_PORT,
1243         SFC_UDP_TUNNEL_DEL_PORT,
1244 };
1245
1246 static int
1247 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1248                       struct rte_eth_udp_tunnel *tunnel_udp,
1249                       enum sfc_udp_tunnel_op_e op)
1250 {
1251         struct sfc_adapter *sa = dev->data->dev_private;
1252         efx_tunnel_protocol_t tunnel_proto;
1253         int rc;
1254
1255         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1256                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1257                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1258                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1259
1260         tunnel_proto =
1261                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1262         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1263                 rc = ENOTSUP;
1264                 goto fail_bad_proto;
1265         }
1266
1267         sfc_adapter_lock(sa);
1268
1269         switch (op) {
1270         case SFC_UDP_TUNNEL_ADD_PORT:
1271                 rc = efx_tunnel_config_udp_add(sa->nic,
1272                                                tunnel_udp->udp_port,
1273                                                tunnel_proto);
1274                 break;
1275         case SFC_UDP_TUNNEL_DEL_PORT:
1276                 rc = efx_tunnel_config_udp_remove(sa->nic,
1277                                                   tunnel_udp->udp_port,
1278                                                   tunnel_proto);
1279                 break;
1280         default:
1281                 rc = EINVAL;
1282                 goto fail_bad_op;
1283         }
1284
1285         if (rc != 0)
1286                 goto fail_op;
1287
1288         if (sa->state == SFC_ADAPTER_STARTED) {
1289                 rc = efx_tunnel_reconfigure(sa->nic);
1290                 if (rc == EAGAIN) {
1291                         /*
1292                          * Configuration is accepted by FW and MC reboot
1293                          * is initiated to apply the changes. MC reboot
1294                          * will be handled in a usual way (MC reboot
1295                          * event on management event queue and adapter
1296                          * restart).
1297                          */
1298                         rc = 0;
1299                 } else if (rc != 0) {
1300                         goto fail_reconfigure;
1301                 }
1302         }
1303
1304         sfc_adapter_unlock(sa);
1305         return 0;
1306
1307 fail_reconfigure:
1308         /* Remove/restore entry since the change makes the trouble */
1309         switch (op) {
1310         case SFC_UDP_TUNNEL_ADD_PORT:
1311                 (void)efx_tunnel_config_udp_remove(sa->nic,
1312                                                    tunnel_udp->udp_port,
1313                                                    tunnel_proto);
1314                 break;
1315         case SFC_UDP_TUNNEL_DEL_PORT:
1316                 (void)efx_tunnel_config_udp_add(sa->nic,
1317                                                 tunnel_udp->udp_port,
1318                                                 tunnel_proto);
1319                 break;
1320         }
1321
1322 fail_op:
1323 fail_bad_op:
1324         sfc_adapter_unlock(sa);
1325
1326 fail_bad_proto:
1327         SFC_ASSERT(rc > 0);
1328         return -rc;
1329 }
1330
1331 static int
1332 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1333                             struct rte_eth_udp_tunnel *tunnel_udp)
1334 {
1335         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1336 }
1337
1338 static int
1339 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1340                             struct rte_eth_udp_tunnel *tunnel_udp)
1341 {
1342         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1343 }
1344
1345 #if EFSYS_OPT_RX_SCALE
1346 static int
1347 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1348                           struct rte_eth_rss_conf *rss_conf)
1349 {
1350         struct sfc_adapter *sa = dev->data->dev_private;
1351         struct sfc_port *port = &sa->port;
1352
1353         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1354                 return -ENOTSUP;
1355
1356         if (sa->rss_channels == 0)
1357                 return -EINVAL;
1358
1359         sfc_adapter_lock(sa);
1360
1361         /*
1362          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1363          * hence, conversion is done here to derive a correct set of ETH_RSS
1364          * flags which corresponds to the active EFX configuration stored
1365          * locally in 'sfc_adapter' and kept up-to-date
1366          */
1367         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1368         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1369         if (rss_conf->rss_key != NULL)
1370                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1371
1372         sfc_adapter_unlock(sa);
1373
1374         return 0;
1375 }
1376
1377 static int
1378 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1379                         struct rte_eth_rss_conf *rss_conf)
1380 {
1381         struct sfc_adapter *sa = dev->data->dev_private;
1382         struct sfc_port *port = &sa->port;
1383         unsigned int efx_hash_types;
1384         int rc = 0;
1385
1386         if (port->isolated)
1387                 return -ENOTSUP;
1388
1389         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1390                 sfc_err(sa, "RSS is not available");
1391                 return -ENOTSUP;
1392         }
1393
1394         if (sa->rss_channels == 0) {
1395                 sfc_err(sa, "RSS is not configured");
1396                 return -EINVAL;
1397         }
1398
1399         if ((rss_conf->rss_key != NULL) &&
1400             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1401                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1402                         sizeof(sa->rss_key));
1403                 return -EINVAL;
1404         }
1405
1406         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1407                 sfc_err(sa, "unsupported hash functions requested");
1408                 return -EINVAL;
1409         }
1410
1411         sfc_adapter_lock(sa);
1412
1413         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1414
1415         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1416                                    EFX_RX_HASHALG_TOEPLITZ,
1417                                    efx_hash_types, B_TRUE);
1418         if (rc != 0)
1419                 goto fail_scale_mode_set;
1420
1421         if (rss_conf->rss_key != NULL) {
1422                 if (sa->state == SFC_ADAPTER_STARTED) {
1423                         rc = efx_rx_scale_key_set(sa->nic,
1424                                                   EFX_RSS_CONTEXT_DEFAULT,
1425                                                   rss_conf->rss_key,
1426                                                   sizeof(sa->rss_key));
1427                         if (rc != 0)
1428                                 goto fail_scale_key_set;
1429                 }
1430
1431                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1432         }
1433
1434         sa->rss_hash_types = efx_hash_types;
1435
1436         sfc_adapter_unlock(sa);
1437
1438         return 0;
1439
1440 fail_scale_key_set:
1441         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1442                                   EFX_RX_HASHALG_TOEPLITZ,
1443                                   sa->rss_hash_types, B_TRUE) != 0)
1444                 sfc_err(sa, "failed to restore RSS mode");
1445
1446 fail_scale_mode_set:
1447         sfc_adapter_unlock(sa);
1448         return -rc;
1449 }
1450
1451 static int
1452 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1453                        struct rte_eth_rss_reta_entry64 *reta_conf,
1454                        uint16_t reta_size)
1455 {
1456         struct sfc_adapter *sa = dev->data->dev_private;
1457         struct sfc_port *port = &sa->port;
1458         int entry;
1459
1460         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1461                 return -ENOTSUP;
1462
1463         if (sa->rss_channels == 0)
1464                 return -EINVAL;
1465
1466         if (reta_size != EFX_RSS_TBL_SIZE)
1467                 return -EINVAL;
1468
1469         sfc_adapter_lock(sa);
1470
1471         for (entry = 0; entry < reta_size; entry++) {
1472                 int grp = entry / RTE_RETA_GROUP_SIZE;
1473                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1474
1475                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1476                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1477         }
1478
1479         sfc_adapter_unlock(sa);
1480
1481         return 0;
1482 }
1483
1484 static int
1485 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1486                         struct rte_eth_rss_reta_entry64 *reta_conf,
1487                         uint16_t reta_size)
1488 {
1489         struct sfc_adapter *sa = dev->data->dev_private;
1490         struct sfc_port *port = &sa->port;
1491         unsigned int *rss_tbl_new;
1492         uint16_t entry;
1493         int rc = 0;
1494
1495
1496         if (port->isolated)
1497                 return -ENOTSUP;
1498
1499         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1500                 sfc_err(sa, "RSS is not available");
1501                 return -ENOTSUP;
1502         }
1503
1504         if (sa->rss_channels == 0) {
1505                 sfc_err(sa, "RSS is not configured");
1506                 return -EINVAL;
1507         }
1508
1509         if (reta_size != EFX_RSS_TBL_SIZE) {
1510                 sfc_err(sa, "RETA size is wrong (should be %u)",
1511                         EFX_RSS_TBL_SIZE);
1512                 return -EINVAL;
1513         }
1514
1515         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1516         if (rss_tbl_new == NULL)
1517                 return -ENOMEM;
1518
1519         sfc_adapter_lock(sa);
1520
1521         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1522
1523         for (entry = 0; entry < reta_size; entry++) {
1524                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1525                 struct rte_eth_rss_reta_entry64 *grp;
1526
1527                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1528
1529                 if (grp->mask & (1ull << grp_idx)) {
1530                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1531                                 rc = EINVAL;
1532                                 goto bad_reta_entry;
1533                         }
1534                         rss_tbl_new[entry] = grp->reta[grp_idx];
1535                 }
1536         }
1537
1538         if (sa->state == SFC_ADAPTER_STARTED) {
1539                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1540                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1541                 if (rc != 0)
1542                         goto fail_scale_tbl_set;
1543         }
1544
1545         rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1546
1547 fail_scale_tbl_set:
1548 bad_reta_entry:
1549         sfc_adapter_unlock(sa);
1550
1551         rte_free(rss_tbl_new);
1552
1553         SFC_ASSERT(rc >= 0);
1554         return -rc;
1555 }
1556 #endif
1557
1558 static int
1559 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1560                     enum rte_filter_op filter_op,
1561                     void *arg)
1562 {
1563         struct sfc_adapter *sa = dev->data->dev_private;
1564         int rc = ENOTSUP;
1565
1566         sfc_log_init(sa, "entry");
1567
1568         switch (filter_type) {
1569         case RTE_ETH_FILTER_NONE:
1570                 sfc_err(sa, "Global filters configuration not supported");
1571                 break;
1572         case RTE_ETH_FILTER_MACVLAN:
1573                 sfc_err(sa, "MACVLAN filters not supported");
1574                 break;
1575         case RTE_ETH_FILTER_ETHERTYPE:
1576                 sfc_err(sa, "EtherType filters not supported");
1577                 break;
1578         case RTE_ETH_FILTER_FLEXIBLE:
1579                 sfc_err(sa, "Flexible filters not supported");
1580                 break;
1581         case RTE_ETH_FILTER_SYN:
1582                 sfc_err(sa, "SYN filters not supported");
1583                 break;
1584         case RTE_ETH_FILTER_NTUPLE:
1585                 sfc_err(sa, "NTUPLE filters not supported");
1586                 break;
1587         case RTE_ETH_FILTER_TUNNEL:
1588                 sfc_err(sa, "Tunnel filters not supported");
1589                 break;
1590         case RTE_ETH_FILTER_FDIR:
1591                 sfc_err(sa, "Flow Director filters not supported");
1592                 break;
1593         case RTE_ETH_FILTER_HASH:
1594                 sfc_err(sa, "Hash filters not supported");
1595                 break;
1596         case RTE_ETH_FILTER_GENERIC:
1597                 if (filter_op != RTE_ETH_FILTER_GET) {
1598                         rc = EINVAL;
1599                 } else {
1600                         *(const void **)arg = &sfc_flow_ops;
1601                         rc = 0;
1602                 }
1603                 break;
1604         default:
1605                 sfc_err(sa, "Unknown filter type %u", filter_type);
1606                 break;
1607         }
1608
1609         sfc_log_init(sa, "exit: %d", -rc);
1610         SFC_ASSERT(rc >= 0);
1611         return -rc;
1612 }
1613
1614 static const struct eth_dev_ops sfc_eth_dev_ops = {
1615         .dev_configure                  = sfc_dev_configure,
1616         .dev_start                      = sfc_dev_start,
1617         .dev_stop                       = sfc_dev_stop,
1618         .dev_set_link_up                = sfc_dev_set_link_up,
1619         .dev_set_link_down              = sfc_dev_set_link_down,
1620         .dev_close                      = sfc_dev_close,
1621         .promiscuous_enable             = sfc_dev_promisc_enable,
1622         .promiscuous_disable            = sfc_dev_promisc_disable,
1623         .allmulticast_enable            = sfc_dev_allmulti_enable,
1624         .allmulticast_disable           = sfc_dev_allmulti_disable,
1625         .link_update                    = sfc_dev_link_update,
1626         .stats_get                      = sfc_stats_get,
1627         .stats_reset                    = sfc_stats_reset,
1628         .xstats_get                     = sfc_xstats_get,
1629         .xstats_reset                   = sfc_stats_reset,
1630         .xstats_get_names               = sfc_xstats_get_names,
1631         .dev_infos_get                  = sfc_dev_infos_get,
1632         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1633         .mtu_set                        = sfc_dev_set_mtu,
1634         .rx_queue_start                 = sfc_rx_queue_start,
1635         .rx_queue_stop                  = sfc_rx_queue_stop,
1636         .tx_queue_start                 = sfc_tx_queue_start,
1637         .tx_queue_stop                  = sfc_tx_queue_stop,
1638         .rx_queue_setup                 = sfc_rx_queue_setup,
1639         .rx_queue_release               = sfc_rx_queue_release,
1640         .rx_queue_count                 = sfc_rx_queue_count,
1641         .rx_descriptor_done             = sfc_rx_descriptor_done,
1642         .rx_descriptor_status           = sfc_rx_descriptor_status,
1643         .tx_descriptor_status           = sfc_tx_descriptor_status,
1644         .tx_queue_setup                 = sfc_tx_queue_setup,
1645         .tx_queue_release               = sfc_tx_queue_release,
1646         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1647         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1648         .mac_addr_set                   = sfc_mac_addr_set,
1649         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1650         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1651 #if EFSYS_OPT_RX_SCALE
1652         .reta_update                    = sfc_dev_rss_reta_update,
1653         .reta_query                     = sfc_dev_rss_reta_query,
1654         .rss_hash_update                = sfc_dev_rss_hash_update,
1655         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1656 #endif
1657         .filter_ctrl                    = sfc_dev_filter_ctrl,
1658         .set_mc_addr_list               = sfc_set_mc_addr_list,
1659         .rxq_info_get                   = sfc_rx_queue_info_get,
1660         .txq_info_get                   = sfc_tx_queue_info_get,
1661         .fw_version_get                 = sfc_fw_version_get,
1662         .xstats_get_by_id               = sfc_xstats_get_by_id,
1663         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1664 };
1665
1666 /**
1667  * Duplicate a string in potentially shared memory required for
1668  * multi-process support.
1669  *
1670  * strdup() allocates from process-local heap/memory.
1671  */
1672 static char *
1673 sfc_strdup(const char *str)
1674 {
1675         size_t size;
1676         char *copy;
1677
1678         if (str == NULL)
1679                 return NULL;
1680
1681         size = strlen(str) + 1;
1682         copy = rte_malloc(__func__, size, 0);
1683         if (copy != NULL)
1684                 rte_memcpy(copy, str, size);
1685
1686         return copy;
1687 }
1688
1689 static int
1690 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1691 {
1692         struct sfc_adapter *sa = dev->data->dev_private;
1693         unsigned int avail_caps = 0;
1694         const char *rx_name = NULL;
1695         const char *tx_name = NULL;
1696         int rc;
1697
1698         switch (sa->family) {
1699         case EFX_FAMILY_HUNTINGTON:
1700         case EFX_FAMILY_MEDFORD:
1701                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1702                 break;
1703         default:
1704                 break;
1705         }
1706
1707         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1708                                 sfc_kvarg_string_handler, &rx_name);
1709         if (rc != 0)
1710                 goto fail_kvarg_rx_datapath;
1711
1712         if (rx_name != NULL) {
1713                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1714                 if (sa->dp_rx == NULL) {
1715                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1716                         rc = ENOENT;
1717                         goto fail_dp_rx;
1718                 }
1719                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1720                         sfc_err(sa,
1721                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1722                                 rx_name);
1723                         rc = EINVAL;
1724                         goto fail_dp_rx_caps;
1725                 }
1726         } else {
1727                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1728                 if (sa->dp_rx == NULL) {
1729                         sfc_err(sa, "Rx datapath by caps %#x not found",
1730                                 avail_caps);
1731                         rc = ENOENT;
1732                         goto fail_dp_rx;
1733                 }
1734         }
1735
1736         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1737         if (sa->dp_rx_name == NULL) {
1738                 rc = ENOMEM;
1739                 goto fail_dp_rx_name;
1740         }
1741
1742         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1743
1744         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1745
1746         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1747                                 sfc_kvarg_string_handler, &tx_name);
1748         if (rc != 0)
1749                 goto fail_kvarg_tx_datapath;
1750
1751         if (tx_name != NULL) {
1752                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1753                 if (sa->dp_tx == NULL) {
1754                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1755                         rc = ENOENT;
1756                         goto fail_dp_tx;
1757                 }
1758                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1759                         sfc_err(sa,
1760                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1761                                 tx_name);
1762                         rc = EINVAL;
1763                         goto fail_dp_tx_caps;
1764                 }
1765         } else {
1766                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1767                 if (sa->dp_tx == NULL) {
1768                         sfc_err(sa, "Tx datapath by caps %#x not found",
1769                                 avail_caps);
1770                         rc = ENOENT;
1771                         goto fail_dp_tx;
1772                 }
1773         }
1774
1775         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1776         if (sa->dp_tx_name == NULL) {
1777                 rc = ENOMEM;
1778                 goto fail_dp_tx_name;
1779         }
1780
1781         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1782
1783         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1784
1785         dev->dev_ops = &sfc_eth_dev_ops;
1786
1787         return 0;
1788
1789 fail_dp_tx_name:
1790 fail_dp_tx_caps:
1791         sa->dp_tx = NULL;
1792
1793 fail_dp_tx:
1794 fail_kvarg_tx_datapath:
1795         rte_free(sa->dp_rx_name);
1796         sa->dp_rx_name = NULL;
1797
1798 fail_dp_rx_name:
1799 fail_dp_rx_caps:
1800         sa->dp_rx = NULL;
1801
1802 fail_dp_rx:
1803 fail_kvarg_rx_datapath:
1804         return rc;
1805 }
1806
1807 static void
1808 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1809 {
1810         struct sfc_adapter *sa = dev->data->dev_private;
1811
1812         dev->dev_ops = NULL;
1813         dev->rx_pkt_burst = NULL;
1814         dev->tx_pkt_burst = NULL;
1815
1816         rte_free(sa->dp_tx_name);
1817         sa->dp_tx_name = NULL;
1818         sa->dp_tx = NULL;
1819
1820         rte_free(sa->dp_rx_name);
1821         sa->dp_rx_name = NULL;
1822         sa->dp_rx = NULL;
1823 }
1824
1825 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1826         .rxq_info_get                   = sfc_rx_queue_info_get,
1827         .txq_info_get                   = sfc_tx_queue_info_get,
1828 };
1829
1830 static int
1831 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1832 {
1833         /*
1834          * Device private data has really many process-local pointers.
1835          * Below code should be extremely careful to use data located
1836          * in shared memory only.
1837          */
1838         struct sfc_adapter *sa = dev->data->dev_private;
1839         const struct sfc_dp_rx *dp_rx;
1840         const struct sfc_dp_tx *dp_tx;
1841         int rc;
1842
1843         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1844         if (dp_rx == NULL) {
1845                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1846                 rc = ENOENT;
1847                 goto fail_dp_rx;
1848         }
1849         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1850                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1851                         sa->dp_tx_name);
1852                 rc = EINVAL;
1853                 goto fail_dp_rx_multi_process;
1854         }
1855
1856         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1857         if (dp_tx == NULL) {
1858                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1859                 rc = ENOENT;
1860                 goto fail_dp_tx;
1861         }
1862         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1863                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1864                         sa->dp_tx_name);
1865                 rc = EINVAL;
1866                 goto fail_dp_tx_multi_process;
1867         }
1868
1869         dev->rx_pkt_burst = dp_rx->pkt_burst;
1870         dev->tx_pkt_burst = dp_tx->pkt_burst;
1871         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1872
1873         return 0;
1874
1875 fail_dp_tx_multi_process:
1876 fail_dp_tx:
1877 fail_dp_rx_multi_process:
1878 fail_dp_rx:
1879         return rc;
1880 }
1881
1882 static void
1883 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1884 {
1885         dev->dev_ops = NULL;
1886         dev->tx_pkt_burst = NULL;
1887         dev->rx_pkt_burst = NULL;
1888 }
1889
1890 static void
1891 sfc_register_dp(void)
1892 {
1893         /* Register once */
1894         if (TAILQ_EMPTY(&sfc_dp_head)) {
1895                 /* Prefer EF10 datapath */
1896                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1897                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1898
1899                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1900                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1901                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1902         }
1903 }
1904
1905 static int
1906 sfc_eth_dev_init(struct rte_eth_dev *dev)
1907 {
1908         struct sfc_adapter *sa = dev->data->dev_private;
1909         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1910         int rc;
1911         const efx_nic_cfg_t *encp;
1912         const struct ether_addr *from;
1913
1914         sfc_register_dp();
1915
1916         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1917                 return -sfc_eth_dev_secondary_set_ops(dev);
1918
1919         /* Required for logging */
1920         sa->pci_addr = pci_dev->addr;
1921         sa->port_id = dev->data->port_id;
1922
1923         sa->eth_dev = dev;
1924
1925         /* Copy PCI device info to the dev->data */
1926         rte_eth_copy_pci_info(dev, pci_dev);
1927
1928         rc = sfc_kvargs_parse(sa);
1929         if (rc != 0)
1930                 goto fail_kvargs_parse;
1931
1932         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1933                                 sfc_kvarg_bool_handler, &sa->debug_init);
1934         if (rc != 0)
1935                 goto fail_kvarg_debug_init;
1936
1937         sfc_log_init(sa, "entry");
1938
1939         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1940         if (dev->data->mac_addrs == NULL) {
1941                 rc = ENOMEM;
1942                 goto fail_mac_addrs;
1943         }
1944
1945         sfc_adapter_lock_init(sa);
1946         sfc_adapter_lock(sa);
1947
1948         sfc_log_init(sa, "probing");
1949         rc = sfc_probe(sa);
1950         if (rc != 0)
1951                 goto fail_probe;
1952
1953         sfc_log_init(sa, "set device ops");
1954         rc = sfc_eth_dev_set_ops(dev);
1955         if (rc != 0)
1956                 goto fail_set_ops;
1957
1958         sfc_log_init(sa, "attaching");
1959         rc = sfc_attach(sa);
1960         if (rc != 0)
1961                 goto fail_attach;
1962
1963         encp = efx_nic_cfg_get(sa->nic);
1964
1965         /*
1966          * The arguments are really reverse order in comparison to
1967          * Linux kernel. Copy from NIC config to Ethernet device data.
1968          */
1969         from = (const struct ether_addr *)(encp->enc_mac_addr);
1970         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1971
1972         sfc_adapter_unlock(sa);
1973
1974         sfc_log_init(sa, "done");
1975         return 0;
1976
1977 fail_attach:
1978         sfc_eth_dev_clear_ops(dev);
1979
1980 fail_set_ops:
1981         sfc_unprobe(sa);
1982
1983 fail_probe:
1984         sfc_adapter_unlock(sa);
1985         sfc_adapter_lock_fini(sa);
1986         rte_free(dev->data->mac_addrs);
1987         dev->data->mac_addrs = NULL;
1988
1989 fail_mac_addrs:
1990 fail_kvarg_debug_init:
1991         sfc_kvargs_cleanup(sa);
1992
1993 fail_kvargs_parse:
1994         sfc_log_init(sa, "failed %d", rc);
1995         SFC_ASSERT(rc > 0);
1996         return -rc;
1997 }
1998
1999 static int
2000 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2001 {
2002         struct sfc_adapter *sa;
2003
2004         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2005                 sfc_eth_dev_secondary_clear_ops(dev);
2006                 return 0;
2007         }
2008
2009         sa = dev->data->dev_private;
2010         sfc_log_init(sa, "entry");
2011
2012         sfc_adapter_lock(sa);
2013
2014         sfc_eth_dev_clear_ops(dev);
2015
2016         sfc_detach(sa);
2017         sfc_unprobe(sa);
2018
2019         rte_free(dev->data->mac_addrs);
2020         dev->data->mac_addrs = NULL;
2021
2022         sfc_kvargs_cleanup(sa);
2023
2024         sfc_adapter_unlock(sa);
2025         sfc_adapter_lock_fini(sa);
2026
2027         sfc_log_init(sa, "done");
2028
2029         /* Required for logging, so cleanup last */
2030         sa->eth_dev = NULL;
2031         return 0;
2032 }
2033
2034 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2035         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2036         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2037         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2038         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2039         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2040         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2041         { .vendor_id = 0 /* sentinel */ }
2042 };
2043
2044 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2045         struct rte_pci_device *pci_dev)
2046 {
2047         return rte_eth_dev_pci_generic_probe(pci_dev,
2048                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2049 }
2050
2051 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2052 {
2053         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2054 }
2055
2056 static struct rte_pci_driver sfc_efx_pmd = {
2057         .id_table = pci_id_sfc_efx_map,
2058         .drv_flags =
2059                 RTE_PCI_DRV_INTR_LSC |
2060                 RTE_PCI_DRV_NEED_MAPPING,
2061         .probe = sfc_eth_dev_pci_probe,
2062         .remove = sfc_eth_dev_pci_remove,
2063 };
2064
2065 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2066 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2067 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2068 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2069         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2070         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2071         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2072         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2073         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2074         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);