net/sfc: support flow API isolated mode
[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_errno.h>
37
38 #include "efx.h"
39
40 #include "sfc.h"
41 #include "sfc_debug.h"
42 #include "sfc_log.h"
43 #include "sfc_kvargs.h"
44 #include "sfc_ev.h"
45 #include "sfc_rx.h"
46 #include "sfc_tx.h"
47 #include "sfc_flow.h"
48 #include "sfc_dp.h"
49 #include "sfc_dp_rx.h"
50
51 static struct sfc_dp_list sfc_dp_head =
52         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
53
54 static int
55 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
56 {
57         struct sfc_adapter *sa = dev->data->dev_private;
58         efx_nic_fw_info_t enfi;
59         int ret;
60         int rc;
61
62         /*
63          * Return value of the callback is likely supposed to be
64          * equal to or greater than 0, nevertheless, if an error
65          * occurs, it will be desirable to pass it to the caller
66          */
67         if ((fw_version == NULL) || (fw_size == 0))
68                 return -EINVAL;
69
70         rc = efx_nic_get_fw_version(sa->nic, &enfi);
71         if (rc != 0)
72                 return -rc;
73
74         ret = snprintf(fw_version, fw_size,
75                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
76                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
77                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
78         if (ret < 0)
79                 return ret;
80
81         if (enfi.enfi_dpcpu_fw_ids_valid) {
82                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
83                 int ret_extra;
84
85                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
86                                      fw_size - dpcpu_fw_ids_offset,
87                                      " rx%" PRIx16 " tx%" PRIx16,
88                                      enfi.enfi_rx_dpcpu_fw_id,
89                                      enfi.enfi_tx_dpcpu_fw_id);
90                 if (ret_extra < 0)
91                         return ret_extra;
92
93                 ret += ret_extra;
94         }
95
96         if (fw_size < (size_t)(++ret))
97                 return ret;
98         else
99                 return 0;
100 }
101
102 static void
103 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
104 {
105         struct sfc_adapter *sa = dev->data->dev_private;
106         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
107
108         sfc_log_init(sa, "entry");
109
110         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
111         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
112
113         /* Autonegotiation may be disabled */
114         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
115         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
116                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
117         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
118                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
119         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
120                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
121
122         dev_info->max_rx_queues = sa->rxq_max;
123         dev_info->max_tx_queues = sa->txq_max;
124
125         /* By default packets are dropped if no descriptors are available */
126         dev_info->default_rxconf.rx_drop_en = 1;
127
128         dev_info->rx_offload_capa =
129                 DEV_RX_OFFLOAD_IPV4_CKSUM |
130                 DEV_RX_OFFLOAD_UDP_CKSUM |
131                 DEV_RX_OFFLOAD_TCP_CKSUM;
132
133         dev_info->tx_offload_capa =
134                 DEV_TX_OFFLOAD_IPV4_CKSUM |
135                 DEV_TX_OFFLOAD_UDP_CKSUM |
136                 DEV_TX_OFFLOAD_TCP_CKSUM;
137
138         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
139         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
140             !encp->enc_hw_tx_insert_vlan_enabled)
141                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
142         else
143                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
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 EFSYS_OPT_RX_SCALE
149         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
150                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
151                 dev_info->hash_key_size = SFC_RSS_KEY_SIZE;
152                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
153         }
154 #endif
155
156         if (sa->tso)
157                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
158
159         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
160         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
161         /* The RXQ hardware requires that the descriptor count is a power
162          * of 2, but rx_desc_lim cannot properly describe that constraint.
163          */
164         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
165
166         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
167         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
168         /*
169          * The TXQ hardware requires that the descriptor count is a power
170          * of 2, but tx_desc_lim cannot properly describe that constraint
171          */
172         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
173 }
174
175 static const uint32_t *
176 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
177 {
178         struct sfc_adapter *sa = dev->data->dev_private;
179
180         return sa->dp_rx->supported_ptypes_get();
181 }
182
183 static int
184 sfc_dev_configure(struct rte_eth_dev *dev)
185 {
186         struct rte_eth_dev_data *dev_data = dev->data;
187         struct sfc_adapter *sa = dev_data->dev_private;
188         int rc;
189
190         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
191                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
192
193         sfc_adapter_lock(sa);
194         switch (sa->state) {
195         case SFC_ADAPTER_CONFIGURED:
196                 /* FALLTHROUGH */
197         case SFC_ADAPTER_INITIALIZED:
198                 rc = sfc_configure(sa);
199                 break;
200         default:
201                 sfc_err(sa, "unexpected adapter state %u to configure",
202                         sa->state);
203                 rc = EINVAL;
204                 break;
205         }
206         sfc_adapter_unlock(sa);
207
208         sfc_log_init(sa, "done %d", rc);
209         SFC_ASSERT(rc >= 0);
210         return -rc;
211 }
212
213 static int
214 sfc_dev_start(struct rte_eth_dev *dev)
215 {
216         struct sfc_adapter *sa = dev->data->dev_private;
217         int rc;
218
219         sfc_log_init(sa, "entry");
220
221         sfc_adapter_lock(sa);
222         rc = sfc_start(sa);
223         sfc_adapter_unlock(sa);
224
225         sfc_log_init(sa, "done %d", rc);
226         SFC_ASSERT(rc >= 0);
227         return -rc;
228 }
229
230 static int
231 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
232 {
233         struct sfc_adapter *sa = dev->data->dev_private;
234         struct rte_eth_link *dev_link = &dev->data->dev_link;
235         struct rte_eth_link old_link;
236         struct rte_eth_link current_link;
237
238         sfc_log_init(sa, "entry");
239
240 retry:
241         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
242         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
243
244         if (sa->state != SFC_ADAPTER_STARTED) {
245                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
246                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
247                                          *(uint64_t *)&old_link,
248                                          *(uint64_t *)&current_link))
249                         goto retry;
250         } else if (wait_to_complete) {
251                 efx_link_mode_t link_mode;
252
253                 if (efx_port_poll(sa->nic, &link_mode) != 0)
254                         link_mode = EFX_LINK_UNKNOWN;
255                 sfc_port_link_mode_to_info(link_mode, &current_link);
256
257                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
258                                          *(uint64_t *)&old_link,
259                                          *(uint64_t *)&current_link))
260                         goto retry;
261         } else {
262                 sfc_ev_mgmt_qpoll(sa);
263                 *(int64_t *)&current_link =
264                         rte_atomic64_read((rte_atomic64_t *)dev_link);
265         }
266
267         if (old_link.link_status != current_link.link_status)
268                 sfc_info(sa, "Link status is %s",
269                          current_link.link_status ? "UP" : "DOWN");
270
271         return old_link.link_status == current_link.link_status ? 0 : -1;
272 }
273
274 static void
275 sfc_dev_stop(struct rte_eth_dev *dev)
276 {
277         struct sfc_adapter *sa = dev->data->dev_private;
278
279         sfc_log_init(sa, "entry");
280
281         sfc_adapter_lock(sa);
282         sfc_stop(sa);
283         sfc_adapter_unlock(sa);
284
285         sfc_log_init(sa, "done");
286 }
287
288 static int
289 sfc_dev_set_link_up(struct rte_eth_dev *dev)
290 {
291         struct sfc_adapter *sa = dev->data->dev_private;
292         int rc;
293
294         sfc_log_init(sa, "entry");
295
296         sfc_adapter_lock(sa);
297         rc = sfc_start(sa);
298         sfc_adapter_unlock(sa);
299
300         SFC_ASSERT(rc >= 0);
301         return -rc;
302 }
303
304 static int
305 sfc_dev_set_link_down(struct rte_eth_dev *dev)
306 {
307         struct sfc_adapter *sa = dev->data->dev_private;
308
309         sfc_log_init(sa, "entry");
310
311         sfc_adapter_lock(sa);
312         sfc_stop(sa);
313         sfc_adapter_unlock(sa);
314
315         return 0;
316 }
317
318 static void
319 sfc_dev_close(struct rte_eth_dev *dev)
320 {
321         struct sfc_adapter *sa = dev->data->dev_private;
322
323         sfc_log_init(sa, "entry");
324
325         sfc_adapter_lock(sa);
326         switch (sa->state) {
327         case SFC_ADAPTER_STARTED:
328                 sfc_stop(sa);
329                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
330                 /* FALLTHROUGH */
331         case SFC_ADAPTER_CONFIGURED:
332                 sfc_close(sa);
333                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
334                 /* FALLTHROUGH */
335         case SFC_ADAPTER_INITIALIZED:
336                 break;
337         default:
338                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
339                 break;
340         }
341         sfc_adapter_unlock(sa);
342
343         sfc_log_init(sa, "done");
344 }
345
346 static void
347 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
348                    boolean_t enabled)
349 {
350         struct sfc_port *port;
351         boolean_t *toggle;
352         struct sfc_adapter *sa = dev->data->dev_private;
353         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
354         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
355
356         sfc_adapter_lock(sa);
357
358         port = &sa->port;
359         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
360
361         if (*toggle != enabled) {
362                 *toggle = enabled;
363
364                 if (port->isolated) {
365                         sfc_warn(sa, "isolated mode is active on the port");
366                         sfc_warn(sa, "the change is to be applied on the next "
367                                      "start provided that isolated mode is "
368                                      "disabled prior the next start");
369                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
370                            (sfc_set_rx_mode(sa) != 0)) {
371                         *toggle = !(enabled);
372                         sfc_warn(sa, "Failed to %s %s mode",
373                                  ((enabled) ? "enable" : "disable"), desc);
374                 }
375         }
376
377         sfc_adapter_unlock(sa);
378 }
379
380 static void
381 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
382 {
383         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
384 }
385
386 static void
387 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
388 {
389         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
390 }
391
392 static void
393 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
394 {
395         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
396 }
397
398 static void
399 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
400 {
401         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
402 }
403
404 static int
405 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
406                    uint16_t nb_rx_desc, unsigned int socket_id,
407                    const struct rte_eth_rxconf *rx_conf,
408                    struct rte_mempool *mb_pool)
409 {
410         struct sfc_adapter *sa = dev->data->dev_private;
411         int rc;
412
413         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
414                      rx_queue_id, nb_rx_desc, socket_id);
415
416         sfc_adapter_lock(sa);
417
418         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
419                           rx_conf, mb_pool);
420         if (rc != 0)
421                 goto fail_rx_qinit;
422
423         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
424
425         sfc_adapter_unlock(sa);
426
427         return 0;
428
429 fail_rx_qinit:
430         sfc_adapter_unlock(sa);
431         SFC_ASSERT(rc > 0);
432         return -rc;
433 }
434
435 static void
436 sfc_rx_queue_release(void *queue)
437 {
438         struct sfc_dp_rxq *dp_rxq = queue;
439         struct sfc_rxq *rxq;
440         struct sfc_adapter *sa;
441         unsigned int sw_index;
442
443         if (dp_rxq == NULL)
444                 return;
445
446         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
447         sa = rxq->evq->sa;
448         sfc_adapter_lock(sa);
449
450         sw_index = sfc_rxq_sw_index(rxq);
451
452         sfc_log_init(sa, "RxQ=%u", sw_index);
453
454         sa->eth_dev->data->rx_queues[sw_index] = NULL;
455
456         sfc_rx_qfini(sa, sw_index);
457
458         sfc_adapter_unlock(sa);
459 }
460
461 static int
462 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
463                    uint16_t nb_tx_desc, unsigned int socket_id,
464                    const struct rte_eth_txconf *tx_conf)
465 {
466         struct sfc_adapter *sa = dev->data->dev_private;
467         int rc;
468
469         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
470                      tx_queue_id, nb_tx_desc, socket_id);
471
472         sfc_adapter_lock(sa);
473
474         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
475         if (rc != 0)
476                 goto fail_tx_qinit;
477
478         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
479
480         sfc_adapter_unlock(sa);
481         return 0;
482
483 fail_tx_qinit:
484         sfc_adapter_unlock(sa);
485         SFC_ASSERT(rc > 0);
486         return -rc;
487 }
488
489 static void
490 sfc_tx_queue_release(void *queue)
491 {
492         struct sfc_dp_txq *dp_txq = queue;
493         struct sfc_txq *txq;
494         unsigned int sw_index;
495         struct sfc_adapter *sa;
496
497         if (dp_txq == NULL)
498                 return;
499
500         txq = sfc_txq_by_dp_txq(dp_txq);
501         sw_index = sfc_txq_sw_index(txq);
502
503         SFC_ASSERT(txq->evq != NULL);
504         sa = txq->evq->sa;
505
506         sfc_log_init(sa, "TxQ = %u", sw_index);
507
508         sfc_adapter_lock(sa);
509
510         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
511         sa->eth_dev->data->tx_queues[sw_index] = NULL;
512
513         sfc_tx_qfini(sa, sw_index);
514
515         sfc_adapter_unlock(sa);
516 }
517
518 static void
519 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
520 {
521         struct sfc_adapter *sa = dev->data->dev_private;
522         struct sfc_port *port = &sa->port;
523         uint64_t *mac_stats;
524
525         rte_spinlock_lock(&port->mac_stats_lock);
526
527         if (sfc_port_update_mac_stats(sa) != 0)
528                 goto unlock;
529
530         mac_stats = port->mac_stats_buf;
531
532         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
533                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
534                 stats->ipackets =
535                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
536                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
537                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
538                 stats->opackets =
539                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
540                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
541                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
542                 stats->ibytes =
543                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
544                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
545                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
546                 stats->obytes =
547                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
548                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
549                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
550                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
551                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
552                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
553         } else {
554                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
555                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
556                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
557                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
558                 /*
559                  * Take into account stats which are whenever supported
560                  * on EF10. If some stat is not supported by current
561                  * firmware variant or HW revision, it is guaranteed
562                  * to be zero in mac_stats.
563                  */
564                 stats->imissed =
565                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
566                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
567                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
568                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
569                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
570                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
571                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
572                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
573                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
574                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
575                 stats->ierrors =
576                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
577                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
578                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
579                 /* no oerrors counters supported on EF10 */
580         }
581
582 unlock:
583         rte_spinlock_unlock(&port->mac_stats_lock);
584 }
585
586 static void
587 sfc_stats_reset(struct rte_eth_dev *dev)
588 {
589         struct sfc_adapter *sa = dev->data->dev_private;
590         struct sfc_port *port = &sa->port;
591         int rc;
592
593         if (sa->state != SFC_ADAPTER_STARTED) {
594                 /*
595                  * The operation cannot be done if port is not started; it
596                  * will be scheduled to be done during the next port start
597                  */
598                 port->mac_stats_reset_pending = B_TRUE;
599                 return;
600         }
601
602         rc = sfc_port_reset_mac_stats(sa);
603         if (rc != 0)
604                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
605 }
606
607 static int
608 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
609                unsigned int xstats_count)
610 {
611         struct sfc_adapter *sa = dev->data->dev_private;
612         struct sfc_port *port = &sa->port;
613         uint64_t *mac_stats;
614         int rc;
615         unsigned int i;
616         int nstats = 0;
617
618         rte_spinlock_lock(&port->mac_stats_lock);
619
620         rc = sfc_port_update_mac_stats(sa);
621         if (rc != 0) {
622                 SFC_ASSERT(rc > 0);
623                 nstats = -rc;
624                 goto unlock;
625         }
626
627         mac_stats = port->mac_stats_buf;
628
629         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
630                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
631                         if (xstats != NULL && nstats < (int)xstats_count) {
632                                 xstats[nstats].id = nstats;
633                                 xstats[nstats].value = mac_stats[i];
634                         }
635                         nstats++;
636                 }
637         }
638
639 unlock:
640         rte_spinlock_unlock(&port->mac_stats_lock);
641
642         return nstats;
643 }
644
645 static int
646 sfc_xstats_get_names(struct rte_eth_dev *dev,
647                      struct rte_eth_xstat_name *xstats_names,
648                      unsigned int xstats_count)
649 {
650         struct sfc_adapter *sa = dev->data->dev_private;
651         struct sfc_port *port = &sa->port;
652         unsigned int i;
653         unsigned int nstats = 0;
654
655         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
656                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
657                         if (xstats_names != NULL && nstats < xstats_count)
658                                 strncpy(xstats_names[nstats].name,
659                                         efx_mac_stat_name(sa->nic, i),
660                                         sizeof(xstats_names[0].name));
661                         nstats++;
662                 }
663         }
664
665         return nstats;
666 }
667
668 static int
669 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
670 {
671         struct sfc_adapter *sa = dev->data->dev_private;
672         unsigned int wanted_fc, link_fc;
673
674         memset(fc_conf, 0, sizeof(*fc_conf));
675
676         sfc_adapter_lock(sa);
677
678         if (sa->state == SFC_ADAPTER_STARTED)
679                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
680         else
681                 link_fc = sa->port.flow_ctrl;
682
683         switch (link_fc) {
684         case 0:
685                 fc_conf->mode = RTE_FC_NONE;
686                 break;
687         case EFX_FCNTL_RESPOND:
688                 fc_conf->mode = RTE_FC_RX_PAUSE;
689                 break;
690         case EFX_FCNTL_GENERATE:
691                 fc_conf->mode = RTE_FC_TX_PAUSE;
692                 break;
693         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
694                 fc_conf->mode = RTE_FC_FULL;
695                 break;
696         default:
697                 sfc_err(sa, "%s: unexpected flow control value %#x",
698                         __func__, link_fc);
699         }
700
701         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
702
703         sfc_adapter_unlock(sa);
704
705         return 0;
706 }
707
708 static int
709 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
710 {
711         struct sfc_adapter *sa = dev->data->dev_private;
712         struct sfc_port *port = &sa->port;
713         unsigned int fcntl;
714         int rc;
715
716         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
717             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
718             fc_conf->mac_ctrl_frame_fwd != 0) {
719                 sfc_err(sa, "unsupported flow control settings specified");
720                 rc = EINVAL;
721                 goto fail_inval;
722         }
723
724         switch (fc_conf->mode) {
725         case RTE_FC_NONE:
726                 fcntl = 0;
727                 break;
728         case RTE_FC_RX_PAUSE:
729                 fcntl = EFX_FCNTL_RESPOND;
730                 break;
731         case RTE_FC_TX_PAUSE:
732                 fcntl = EFX_FCNTL_GENERATE;
733                 break;
734         case RTE_FC_FULL:
735                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
736                 break;
737         default:
738                 rc = EINVAL;
739                 goto fail_inval;
740         }
741
742         sfc_adapter_lock(sa);
743
744         if (sa->state == SFC_ADAPTER_STARTED) {
745                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
746                 if (rc != 0)
747                         goto fail_mac_fcntl_set;
748         }
749
750         port->flow_ctrl = fcntl;
751         port->flow_ctrl_autoneg = fc_conf->autoneg;
752
753         sfc_adapter_unlock(sa);
754
755         return 0;
756
757 fail_mac_fcntl_set:
758         sfc_adapter_unlock(sa);
759 fail_inval:
760         SFC_ASSERT(rc > 0);
761         return -rc;
762 }
763
764 static int
765 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
766 {
767         struct sfc_adapter *sa = dev->data->dev_private;
768         size_t pdu = EFX_MAC_PDU(mtu);
769         size_t old_pdu;
770         int rc;
771
772         sfc_log_init(sa, "mtu=%u", mtu);
773
774         rc = EINVAL;
775         if (pdu < EFX_MAC_PDU_MIN) {
776                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
777                         (unsigned int)mtu, (unsigned int)pdu,
778                         EFX_MAC_PDU_MIN);
779                 goto fail_inval;
780         }
781         if (pdu > EFX_MAC_PDU_MAX) {
782                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
783                         (unsigned int)mtu, (unsigned int)pdu,
784                         EFX_MAC_PDU_MAX);
785                 goto fail_inval;
786         }
787
788         sfc_adapter_lock(sa);
789
790         if (pdu != sa->port.pdu) {
791                 if (sa->state == SFC_ADAPTER_STARTED) {
792                         sfc_stop(sa);
793
794                         old_pdu = sa->port.pdu;
795                         sa->port.pdu = pdu;
796                         rc = sfc_start(sa);
797                         if (rc != 0)
798                                 goto fail_start;
799                 } else {
800                         sa->port.pdu = pdu;
801                 }
802         }
803
804         /*
805          * The driver does not use it, but other PMDs update jumbo_frame
806          * flag and max_rx_pkt_len when MTU is set.
807          */
808         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
809         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
810
811         sfc_adapter_unlock(sa);
812
813         sfc_log_init(sa, "done");
814         return 0;
815
816 fail_start:
817         sa->port.pdu = old_pdu;
818         if (sfc_start(sa) != 0)
819                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
820                         "PDU max size - port is stopped",
821                         (unsigned int)pdu, (unsigned int)old_pdu);
822         sfc_adapter_unlock(sa);
823
824 fail_inval:
825         sfc_log_init(sa, "failed %d", rc);
826         SFC_ASSERT(rc > 0);
827         return -rc;
828 }
829 static void
830 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
831 {
832         struct sfc_adapter *sa = dev->data->dev_private;
833         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
834         struct sfc_port *port = &sa->port;
835         int rc;
836
837         sfc_adapter_lock(sa);
838
839         if (port->isolated) {
840                 sfc_err(sa, "isolated mode is active on the port");
841                 sfc_err(sa, "will not set MAC address");
842                 goto unlock;
843         }
844
845         if (sa->state != SFC_ADAPTER_STARTED) {
846                 sfc_info(sa, "the port is not started");
847                 sfc_info(sa, "the new MAC address will be set on port start");
848
849                 goto unlock;
850         }
851
852         if (encp->enc_allow_set_mac_with_installed_filters) {
853                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
854                 if (rc != 0) {
855                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
856                         goto unlock;
857                 }
858
859                 /*
860                  * Changing the MAC address by means of MCDI request
861                  * has no effect on received traffic, therefore
862                  * we also need to update unicast filters
863                  */
864                 rc = sfc_set_rx_mode(sa);
865                 if (rc != 0)
866                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
867         } else {
868                 sfc_warn(sa, "cannot set MAC address with filters installed");
869                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
870                 sfc_warn(sa, "(some traffic may be dropped)");
871
872                 /*
873                  * Since setting MAC address with filters installed is not
874                  * allowed on the adapter, one needs to simply restart adapter
875                  * so that the new MAC address will be taken from an outer
876                  * storage and set flawlessly by means of sfc_start() call
877                  */
878                 sfc_stop(sa);
879                 rc = sfc_start(sa);
880                 if (rc != 0)
881                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
882         }
883
884 unlock:
885         sfc_adapter_unlock(sa);
886 }
887
888
889 static int
890 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
891                      uint32_t nb_mc_addr)
892 {
893         struct sfc_adapter *sa = dev->data->dev_private;
894         struct sfc_port *port = &sa->port;
895         uint8_t *mc_addrs = port->mcast_addrs;
896         int rc;
897         unsigned int i;
898
899         if (port->isolated) {
900                 sfc_err(sa, "isolated mode is active on the port");
901                 sfc_err(sa, "will not set multicast address list");
902                 return -ENOTSUP;
903         }
904
905         if (mc_addrs == NULL)
906                 return -ENOBUFS;
907
908         if (nb_mc_addr > port->max_mcast_addrs) {
909                 sfc_err(sa, "too many multicast addresses: %u > %u",
910                          nb_mc_addr, port->max_mcast_addrs);
911                 return -EINVAL;
912         }
913
914         for (i = 0; i < nb_mc_addr; ++i) {
915                 (void)rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
916                                  EFX_MAC_ADDR_LEN);
917                 mc_addrs += EFX_MAC_ADDR_LEN;
918         }
919
920         port->nb_mcast_addrs = nb_mc_addr;
921
922         if (sa->state != SFC_ADAPTER_STARTED)
923                 return 0;
924
925         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
926                                         port->nb_mcast_addrs);
927         if (rc != 0)
928                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
929
930         SFC_ASSERT(rc > 0);
931         return -rc;
932 }
933
934 /*
935  * The function is used by the secondary process as well. It must not
936  * use any process-local pointers from the adapter data.
937  */
938 static void
939 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
940                       struct rte_eth_rxq_info *qinfo)
941 {
942         struct sfc_adapter *sa = dev->data->dev_private;
943         struct sfc_rxq_info *rxq_info;
944         struct sfc_rxq *rxq;
945
946         sfc_adapter_lock(sa);
947
948         SFC_ASSERT(rx_queue_id < sa->rxq_count);
949
950         rxq_info = &sa->rxq_info[rx_queue_id];
951         rxq = rxq_info->rxq;
952         SFC_ASSERT(rxq != NULL);
953
954         qinfo->mp = rxq->refill_mb_pool;
955         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
956         qinfo->conf.rx_drop_en = 1;
957         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
958         qinfo->scattered_rx = (rxq_info->type == EFX_RXQ_TYPE_SCATTER);
959         qinfo->nb_desc = rxq_info->entries;
960
961         sfc_adapter_unlock(sa);
962 }
963
964 /*
965  * The function is used by the secondary process as well. It must not
966  * use any process-local pointers from the adapter data.
967  */
968 static void
969 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
970                       struct rte_eth_txq_info *qinfo)
971 {
972         struct sfc_adapter *sa = dev->data->dev_private;
973         struct sfc_txq_info *txq_info;
974
975         sfc_adapter_lock(sa);
976
977         SFC_ASSERT(tx_queue_id < sa->txq_count);
978
979         txq_info = &sa->txq_info[tx_queue_id];
980         SFC_ASSERT(txq_info->txq != NULL);
981
982         memset(qinfo, 0, sizeof(*qinfo));
983
984         qinfo->conf.txq_flags = txq_info->txq->flags;
985         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
986         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
987         qinfo->nb_desc = txq_info->entries;
988
989         sfc_adapter_unlock(sa);
990 }
991
992 static uint32_t
993 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
994 {
995         struct sfc_adapter *sa = dev->data->dev_private;
996
997         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
998
999         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1000 }
1001
1002 static int
1003 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1004 {
1005         struct sfc_dp_rxq *dp_rxq = queue;
1006
1007         return sfc_rx_qdesc_done(dp_rxq, offset);
1008 }
1009
1010 static int
1011 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1012 {
1013         struct sfc_adapter *sa = dev->data->dev_private;
1014         int rc;
1015
1016         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1017
1018         sfc_adapter_lock(sa);
1019
1020         rc = EINVAL;
1021         if (sa->state != SFC_ADAPTER_STARTED)
1022                 goto fail_not_started;
1023
1024         rc = sfc_rx_qstart(sa, rx_queue_id);
1025         if (rc != 0)
1026                 goto fail_rx_qstart;
1027
1028         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1029
1030         sfc_adapter_unlock(sa);
1031
1032         return 0;
1033
1034 fail_rx_qstart:
1035 fail_not_started:
1036         sfc_adapter_unlock(sa);
1037         SFC_ASSERT(rc > 0);
1038         return -rc;
1039 }
1040
1041 static int
1042 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1043 {
1044         struct sfc_adapter *sa = dev->data->dev_private;
1045
1046         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1047
1048         sfc_adapter_lock(sa);
1049         sfc_rx_qstop(sa, rx_queue_id);
1050
1051         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1052
1053         sfc_adapter_unlock(sa);
1054
1055         return 0;
1056 }
1057
1058 static int
1059 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1060 {
1061         struct sfc_adapter *sa = dev->data->dev_private;
1062         int rc;
1063
1064         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1065
1066         sfc_adapter_lock(sa);
1067
1068         rc = EINVAL;
1069         if (sa->state != SFC_ADAPTER_STARTED)
1070                 goto fail_not_started;
1071
1072         rc = sfc_tx_qstart(sa, tx_queue_id);
1073         if (rc != 0)
1074                 goto fail_tx_qstart;
1075
1076         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1077
1078         sfc_adapter_unlock(sa);
1079         return 0;
1080
1081 fail_tx_qstart:
1082
1083 fail_not_started:
1084         sfc_adapter_unlock(sa);
1085         SFC_ASSERT(rc > 0);
1086         return -rc;
1087 }
1088
1089 static int
1090 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1091 {
1092         struct sfc_adapter *sa = dev->data->dev_private;
1093
1094         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1095
1096         sfc_adapter_lock(sa);
1097
1098         sfc_tx_qstop(sa, tx_queue_id);
1099
1100         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1101
1102         sfc_adapter_unlock(sa);
1103         return 0;
1104 }
1105
1106 #if EFSYS_OPT_RX_SCALE
1107 static int
1108 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1109                           struct rte_eth_rss_conf *rss_conf)
1110 {
1111         struct sfc_adapter *sa = dev->data->dev_private;
1112         struct sfc_port *port = &sa->port;
1113
1114         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1115                 return -ENOTSUP;
1116
1117         if (sa->rss_channels == 0)
1118                 return -EINVAL;
1119
1120         sfc_adapter_lock(sa);
1121
1122         /*
1123          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1124          * hence, conversion is done here to derive a correct set of ETH_RSS
1125          * flags which corresponds to the active EFX configuration stored
1126          * locally in 'sfc_adapter' and kept up-to-date
1127          */
1128         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1129         rss_conf->rss_key_len = SFC_RSS_KEY_SIZE;
1130         if (rss_conf->rss_key != NULL)
1131                 rte_memcpy(rss_conf->rss_key, sa->rss_key, SFC_RSS_KEY_SIZE);
1132
1133         sfc_adapter_unlock(sa);
1134
1135         return 0;
1136 }
1137
1138 static int
1139 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1140                         struct rte_eth_rss_conf *rss_conf)
1141 {
1142         struct sfc_adapter *sa = dev->data->dev_private;
1143         struct sfc_port *port = &sa->port;
1144         unsigned int efx_hash_types;
1145         int rc = 0;
1146
1147         if (port->isolated)
1148                 return -ENOTSUP;
1149
1150         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1151                 sfc_err(sa, "RSS is not available");
1152                 return -ENOTSUP;
1153         }
1154
1155         if (sa->rss_channels == 0) {
1156                 sfc_err(sa, "RSS is not configured");
1157                 return -EINVAL;
1158         }
1159
1160         if ((rss_conf->rss_key != NULL) &&
1161             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1162                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1163                         sizeof(sa->rss_key));
1164                 return -EINVAL;
1165         }
1166
1167         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1168                 sfc_err(sa, "unsupported hash functions requested");
1169                 return -EINVAL;
1170         }
1171
1172         sfc_adapter_lock(sa);
1173
1174         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1175
1176         rc = efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1177                                    efx_hash_types, B_TRUE);
1178         if (rc != 0)
1179                 goto fail_scale_mode_set;
1180
1181         if (rss_conf->rss_key != NULL) {
1182                 if (sa->state == SFC_ADAPTER_STARTED) {
1183                         rc = efx_rx_scale_key_set(sa->nic, rss_conf->rss_key,
1184                                                   sizeof(sa->rss_key));
1185                         if (rc != 0)
1186                                 goto fail_scale_key_set;
1187                 }
1188
1189                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1190         }
1191
1192         sa->rss_hash_types = efx_hash_types;
1193
1194         sfc_adapter_unlock(sa);
1195
1196         return 0;
1197
1198 fail_scale_key_set:
1199         if (efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1200                                   sa->rss_hash_types, B_TRUE) != 0)
1201                 sfc_err(sa, "failed to restore RSS mode");
1202
1203 fail_scale_mode_set:
1204         sfc_adapter_unlock(sa);
1205         return -rc;
1206 }
1207
1208 static int
1209 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1210                        struct rte_eth_rss_reta_entry64 *reta_conf,
1211                        uint16_t reta_size)
1212 {
1213         struct sfc_adapter *sa = dev->data->dev_private;
1214         struct sfc_port *port = &sa->port;
1215         int entry;
1216
1217         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1218                 return -ENOTSUP;
1219
1220         if (sa->rss_channels == 0)
1221                 return -EINVAL;
1222
1223         if (reta_size != EFX_RSS_TBL_SIZE)
1224                 return -EINVAL;
1225
1226         sfc_adapter_lock(sa);
1227
1228         for (entry = 0; entry < reta_size; entry++) {
1229                 int grp = entry / RTE_RETA_GROUP_SIZE;
1230                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1231
1232                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1233                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1234         }
1235
1236         sfc_adapter_unlock(sa);
1237
1238         return 0;
1239 }
1240
1241 static int
1242 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1243                         struct rte_eth_rss_reta_entry64 *reta_conf,
1244                         uint16_t reta_size)
1245 {
1246         struct sfc_adapter *sa = dev->data->dev_private;
1247         struct sfc_port *port = &sa->port;
1248         unsigned int *rss_tbl_new;
1249         uint16_t entry;
1250         int rc;
1251
1252
1253         if (port->isolated)
1254                 return -ENOTSUP;
1255
1256         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1257                 sfc_err(sa, "RSS is not available");
1258                 return -ENOTSUP;
1259         }
1260
1261         if (sa->rss_channels == 0) {
1262                 sfc_err(sa, "RSS is not configured");
1263                 return -EINVAL;
1264         }
1265
1266         if (reta_size != EFX_RSS_TBL_SIZE) {
1267                 sfc_err(sa, "RETA size is wrong (should be %u)",
1268                         EFX_RSS_TBL_SIZE);
1269                 return -EINVAL;
1270         }
1271
1272         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1273         if (rss_tbl_new == NULL)
1274                 return -ENOMEM;
1275
1276         sfc_adapter_lock(sa);
1277
1278         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1279
1280         for (entry = 0; entry < reta_size; entry++) {
1281                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1282                 struct rte_eth_rss_reta_entry64 *grp;
1283
1284                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1285
1286                 if (grp->mask & (1ull << grp_idx)) {
1287                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1288                                 rc = EINVAL;
1289                                 goto bad_reta_entry;
1290                         }
1291                         rss_tbl_new[entry] = grp->reta[grp_idx];
1292                 }
1293         }
1294
1295         rc = efx_rx_scale_tbl_set(sa->nic, rss_tbl_new, EFX_RSS_TBL_SIZE);
1296         if (rc == 0)
1297                 rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1298
1299 bad_reta_entry:
1300         sfc_adapter_unlock(sa);
1301
1302         rte_free(rss_tbl_new);
1303
1304         SFC_ASSERT(rc >= 0);
1305         return -rc;
1306 }
1307 #endif
1308
1309 static int
1310 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1311                     enum rte_filter_op filter_op,
1312                     void *arg)
1313 {
1314         struct sfc_adapter *sa = dev->data->dev_private;
1315         int rc = ENOTSUP;
1316
1317         sfc_log_init(sa, "entry");
1318
1319         switch (filter_type) {
1320         case RTE_ETH_FILTER_NONE:
1321                 sfc_err(sa, "Global filters configuration not supported");
1322                 break;
1323         case RTE_ETH_FILTER_MACVLAN:
1324                 sfc_err(sa, "MACVLAN filters not supported");
1325                 break;
1326         case RTE_ETH_FILTER_ETHERTYPE:
1327                 sfc_err(sa, "EtherType filters not supported");
1328                 break;
1329         case RTE_ETH_FILTER_FLEXIBLE:
1330                 sfc_err(sa, "Flexible filters not supported");
1331                 break;
1332         case RTE_ETH_FILTER_SYN:
1333                 sfc_err(sa, "SYN filters not supported");
1334                 break;
1335         case RTE_ETH_FILTER_NTUPLE:
1336                 sfc_err(sa, "NTUPLE filters not supported");
1337                 break;
1338         case RTE_ETH_FILTER_TUNNEL:
1339                 sfc_err(sa, "Tunnel filters not supported");
1340                 break;
1341         case RTE_ETH_FILTER_FDIR:
1342                 sfc_err(sa, "Flow Director filters not supported");
1343                 break;
1344         case RTE_ETH_FILTER_HASH:
1345                 sfc_err(sa, "Hash filters not supported");
1346                 break;
1347         case RTE_ETH_FILTER_GENERIC:
1348                 if (filter_op != RTE_ETH_FILTER_GET) {
1349                         rc = EINVAL;
1350                 } else {
1351                         *(const void **)arg = &sfc_flow_ops;
1352                         rc = 0;
1353                 }
1354                 break;
1355         default:
1356                 sfc_err(sa, "Unknown filter type %u", filter_type);
1357                 break;
1358         }
1359
1360         sfc_log_init(sa, "exit: %d", -rc);
1361         SFC_ASSERT(rc >= 0);
1362         return -rc;
1363 }
1364
1365 static const struct eth_dev_ops sfc_eth_dev_ops = {
1366         .dev_configure                  = sfc_dev_configure,
1367         .dev_start                      = sfc_dev_start,
1368         .dev_stop                       = sfc_dev_stop,
1369         .dev_set_link_up                = sfc_dev_set_link_up,
1370         .dev_set_link_down              = sfc_dev_set_link_down,
1371         .dev_close                      = sfc_dev_close,
1372         .promiscuous_enable             = sfc_dev_promisc_enable,
1373         .promiscuous_disable            = sfc_dev_promisc_disable,
1374         .allmulticast_enable            = sfc_dev_allmulti_enable,
1375         .allmulticast_disable           = sfc_dev_allmulti_disable,
1376         .link_update                    = sfc_dev_link_update,
1377         .stats_get                      = sfc_stats_get,
1378         .stats_reset                    = sfc_stats_reset,
1379         .xstats_get                     = sfc_xstats_get,
1380         .xstats_reset                   = sfc_stats_reset,
1381         .xstats_get_names               = sfc_xstats_get_names,
1382         .dev_infos_get                  = sfc_dev_infos_get,
1383         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1384         .mtu_set                        = sfc_dev_set_mtu,
1385         .rx_queue_start                 = sfc_rx_queue_start,
1386         .rx_queue_stop                  = sfc_rx_queue_stop,
1387         .tx_queue_start                 = sfc_tx_queue_start,
1388         .tx_queue_stop                  = sfc_tx_queue_stop,
1389         .rx_queue_setup                 = sfc_rx_queue_setup,
1390         .rx_queue_release               = sfc_rx_queue_release,
1391         .rx_queue_count                 = sfc_rx_queue_count,
1392         .rx_descriptor_done             = sfc_rx_descriptor_done,
1393         .tx_queue_setup                 = sfc_tx_queue_setup,
1394         .tx_queue_release               = sfc_tx_queue_release,
1395         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1396         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1397         .mac_addr_set                   = sfc_mac_addr_set,
1398 #if EFSYS_OPT_RX_SCALE
1399         .reta_update                    = sfc_dev_rss_reta_update,
1400         .reta_query                     = sfc_dev_rss_reta_query,
1401         .rss_hash_update                = sfc_dev_rss_hash_update,
1402         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1403 #endif
1404         .filter_ctrl                    = sfc_dev_filter_ctrl,
1405         .set_mc_addr_list               = sfc_set_mc_addr_list,
1406         .rxq_info_get                   = sfc_rx_queue_info_get,
1407         .txq_info_get                   = sfc_tx_queue_info_get,
1408         .fw_version_get                 = sfc_fw_version_get,
1409 };
1410
1411 /**
1412  * Duplicate a string in potentially shared memory required for
1413  * multi-process support.
1414  *
1415  * strdup() allocates from process-local heap/memory.
1416  */
1417 static char *
1418 sfc_strdup(const char *str)
1419 {
1420         size_t size;
1421         char *copy;
1422
1423         if (str == NULL)
1424                 return NULL;
1425
1426         size = strlen(str) + 1;
1427         copy = rte_malloc(__func__, size, 0);
1428         if (copy != NULL)
1429                 rte_memcpy(copy, str, size);
1430
1431         return copy;
1432 }
1433
1434 static int
1435 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1436 {
1437         struct sfc_adapter *sa = dev->data->dev_private;
1438         unsigned int avail_caps = 0;
1439         const char *rx_name = NULL;
1440         const char *tx_name = NULL;
1441         int rc;
1442
1443         switch (sa->family) {
1444         case EFX_FAMILY_HUNTINGTON:
1445         case EFX_FAMILY_MEDFORD:
1446                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1447                 break;
1448         default:
1449                 break;
1450         }
1451
1452         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1453                                 sfc_kvarg_string_handler, &rx_name);
1454         if (rc != 0)
1455                 goto fail_kvarg_rx_datapath;
1456
1457         if (rx_name != NULL) {
1458                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1459                 if (sa->dp_rx == NULL) {
1460                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1461                         rc = ENOENT;
1462                         goto fail_dp_rx;
1463                 }
1464                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1465                         sfc_err(sa,
1466                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1467                                 rx_name);
1468                         rc = EINVAL;
1469                         goto fail_dp_rx_caps;
1470                 }
1471         } else {
1472                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1473                 if (sa->dp_rx == NULL) {
1474                         sfc_err(sa, "Rx datapath by caps %#x not found",
1475                                 avail_caps);
1476                         rc = ENOENT;
1477                         goto fail_dp_rx;
1478                 }
1479         }
1480
1481         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1482         if (sa->dp_rx_name == NULL) {
1483                 rc = ENOMEM;
1484                 goto fail_dp_rx_name;
1485         }
1486
1487         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1488
1489         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1490
1491         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1492                                 sfc_kvarg_string_handler, &tx_name);
1493         if (rc != 0)
1494                 goto fail_kvarg_tx_datapath;
1495
1496         if (tx_name != NULL) {
1497                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1498                 if (sa->dp_tx == NULL) {
1499                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1500                         rc = ENOENT;
1501                         goto fail_dp_tx;
1502                 }
1503                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1504                         sfc_err(sa,
1505                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1506                                 tx_name);
1507                         rc = EINVAL;
1508                         goto fail_dp_tx_caps;
1509                 }
1510         } else {
1511                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1512                 if (sa->dp_tx == NULL) {
1513                         sfc_err(sa, "Tx datapath by caps %#x not found",
1514                                 avail_caps);
1515                         rc = ENOENT;
1516                         goto fail_dp_tx;
1517                 }
1518         }
1519
1520         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1521         if (sa->dp_tx_name == NULL) {
1522                 rc = ENOMEM;
1523                 goto fail_dp_tx_name;
1524         }
1525
1526         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1527
1528         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1529
1530         dev->dev_ops = &sfc_eth_dev_ops;
1531
1532         return 0;
1533
1534 fail_dp_tx_name:
1535 fail_dp_tx_caps:
1536         sa->dp_tx = NULL;
1537
1538 fail_dp_tx:
1539 fail_kvarg_tx_datapath:
1540         rte_free(sa->dp_rx_name);
1541         sa->dp_rx_name = NULL;
1542
1543 fail_dp_rx_name:
1544 fail_dp_rx_caps:
1545         sa->dp_rx = NULL;
1546
1547 fail_dp_rx:
1548 fail_kvarg_rx_datapath:
1549         return rc;
1550 }
1551
1552 static void
1553 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1554 {
1555         struct sfc_adapter *sa = dev->data->dev_private;
1556
1557         dev->dev_ops = NULL;
1558         dev->rx_pkt_burst = NULL;
1559         dev->tx_pkt_burst = NULL;
1560
1561         rte_free(sa->dp_tx_name);
1562         sa->dp_tx_name = NULL;
1563         sa->dp_tx = NULL;
1564
1565         rte_free(sa->dp_rx_name);
1566         sa->dp_rx_name = NULL;
1567         sa->dp_rx = NULL;
1568 }
1569
1570 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1571         .rxq_info_get                   = sfc_rx_queue_info_get,
1572         .txq_info_get                   = sfc_tx_queue_info_get,
1573 };
1574
1575 static int
1576 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1577 {
1578         /*
1579          * Device private data has really many process-local pointers.
1580          * Below code should be extremely careful to use data located
1581          * in shared memory only.
1582          */
1583         struct sfc_adapter *sa = dev->data->dev_private;
1584         const struct sfc_dp_rx *dp_rx;
1585         const struct sfc_dp_tx *dp_tx;
1586         int rc;
1587
1588         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1589         if (dp_rx == NULL) {
1590                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1591                 rc = ENOENT;
1592                 goto fail_dp_rx;
1593         }
1594         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1595                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1596                         sa->dp_tx_name);
1597                 rc = EINVAL;
1598                 goto fail_dp_rx_multi_process;
1599         }
1600
1601         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1602         if (dp_tx == NULL) {
1603                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1604                 rc = ENOENT;
1605                 goto fail_dp_tx;
1606         }
1607         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1608                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1609                         sa->dp_tx_name);
1610                 rc = EINVAL;
1611                 goto fail_dp_tx_multi_process;
1612         }
1613
1614         dev->rx_pkt_burst = dp_rx->pkt_burst;
1615         dev->tx_pkt_burst = dp_tx->pkt_burst;
1616         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1617
1618         return 0;
1619
1620 fail_dp_tx_multi_process:
1621 fail_dp_tx:
1622 fail_dp_rx_multi_process:
1623 fail_dp_rx:
1624         return rc;
1625 }
1626
1627 static void
1628 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1629 {
1630         dev->dev_ops = NULL;
1631         dev->tx_pkt_burst = NULL;
1632         dev->rx_pkt_burst = NULL;
1633 }
1634
1635 static void
1636 sfc_register_dp(void)
1637 {
1638         /* Register once */
1639         if (TAILQ_EMPTY(&sfc_dp_head)) {
1640                 /* Prefer EF10 datapath */
1641                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1642                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1643
1644                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1645                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1646                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1647         }
1648 }
1649
1650 static int
1651 sfc_eth_dev_init(struct rte_eth_dev *dev)
1652 {
1653         struct sfc_adapter *sa = dev->data->dev_private;
1654         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1655         int rc;
1656         const efx_nic_cfg_t *encp;
1657         const struct ether_addr *from;
1658
1659         sfc_register_dp();
1660
1661         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1662                 return -sfc_eth_dev_secondary_set_ops(dev);
1663
1664         /* Required for logging */
1665         sa->pci_addr = pci_dev->addr;
1666         sa->port_id = dev->data->port_id;
1667
1668         sa->eth_dev = dev;
1669
1670         /* Copy PCI device info to the dev->data */
1671         rte_eth_copy_pci_info(dev, pci_dev);
1672
1673         dev->data->dev_flags |= RTE_ETH_DEV_DETACHABLE;
1674
1675         rc = sfc_kvargs_parse(sa);
1676         if (rc != 0)
1677                 goto fail_kvargs_parse;
1678
1679         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1680                                 sfc_kvarg_bool_handler, &sa->debug_init);
1681         if (rc != 0)
1682                 goto fail_kvarg_debug_init;
1683
1684         sfc_log_init(sa, "entry");
1685
1686         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1687         if (dev->data->mac_addrs == NULL) {
1688                 rc = ENOMEM;
1689                 goto fail_mac_addrs;
1690         }
1691
1692         sfc_adapter_lock_init(sa);
1693         sfc_adapter_lock(sa);
1694
1695         sfc_log_init(sa, "probing");
1696         rc = sfc_probe(sa);
1697         if (rc != 0)
1698                 goto fail_probe;
1699
1700         sfc_log_init(sa, "set device ops");
1701         rc = sfc_eth_dev_set_ops(dev);
1702         if (rc != 0)
1703                 goto fail_set_ops;
1704
1705         sfc_log_init(sa, "attaching");
1706         rc = sfc_attach(sa);
1707         if (rc != 0)
1708                 goto fail_attach;
1709
1710         encp = efx_nic_cfg_get(sa->nic);
1711
1712         /*
1713          * The arguments are really reverse order in comparison to
1714          * Linux kernel. Copy from NIC config to Ethernet device data.
1715          */
1716         from = (const struct ether_addr *)(encp->enc_mac_addr);
1717         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1718
1719         sfc_adapter_unlock(sa);
1720
1721         sfc_log_init(sa, "done");
1722         return 0;
1723
1724 fail_attach:
1725         sfc_eth_dev_clear_ops(dev);
1726
1727 fail_set_ops:
1728         sfc_unprobe(sa);
1729
1730 fail_probe:
1731         sfc_adapter_unlock(sa);
1732         sfc_adapter_lock_fini(sa);
1733         rte_free(dev->data->mac_addrs);
1734         dev->data->mac_addrs = NULL;
1735
1736 fail_mac_addrs:
1737 fail_kvarg_debug_init:
1738         sfc_kvargs_cleanup(sa);
1739
1740 fail_kvargs_parse:
1741         sfc_log_init(sa, "failed %d", rc);
1742         SFC_ASSERT(rc > 0);
1743         return -rc;
1744 }
1745
1746 static int
1747 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1748 {
1749         struct sfc_adapter *sa;
1750
1751         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1752                 sfc_eth_dev_secondary_clear_ops(dev);
1753                 return 0;
1754         }
1755
1756         sa = dev->data->dev_private;
1757         sfc_log_init(sa, "entry");
1758
1759         sfc_adapter_lock(sa);
1760
1761         sfc_eth_dev_clear_ops(dev);
1762
1763         sfc_detach(sa);
1764         sfc_unprobe(sa);
1765
1766         rte_free(dev->data->mac_addrs);
1767         dev->data->mac_addrs = NULL;
1768
1769         sfc_kvargs_cleanup(sa);
1770
1771         sfc_adapter_unlock(sa);
1772         sfc_adapter_lock_fini(sa);
1773
1774         sfc_log_init(sa, "done");
1775
1776         /* Required for logging, so cleanup last */
1777         sa->eth_dev = NULL;
1778         return 0;
1779 }
1780
1781 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
1782         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
1783         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
1784         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
1785         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
1786         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
1787         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
1788         { .vendor_id = 0 /* sentinel */ }
1789 };
1790
1791 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1792         struct rte_pci_device *pci_dev)
1793 {
1794         return rte_eth_dev_pci_generic_probe(pci_dev,
1795                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
1796 }
1797
1798 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
1799 {
1800         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
1801 }
1802
1803 static struct rte_pci_driver sfc_efx_pmd = {
1804         .id_table = pci_id_sfc_efx_map,
1805         .drv_flags =
1806                 RTE_PCI_DRV_INTR_LSC |
1807                 RTE_PCI_DRV_NEED_MAPPING,
1808         .probe = sfc_eth_dev_pci_probe,
1809         .remove = sfc_eth_dev_pci_remove,
1810 };
1811
1812 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
1813 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
1814 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
1815 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
1816         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
1817         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
1818         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
1819         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
1820         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
1821         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);