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