ethdev: allow drivers to return error on close
[dpdk.git] / drivers / net / sfc / sfc_ethdev.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  * Copyright(c) 2019-2020 Xilinx, Inc.
4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
5  *
6  * This software was jointly developed between OKTET Labs (under contract
7  * for Solarflare) and Solarflare Communications, Inc.
8  */
9
10 #include <rte_dev.h>
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_errno.h>
16 #include <rte_string_fns.h>
17 #include <rte_ether.h>
18
19 #include "efx.h"
20
21 #include "sfc.h"
22 #include "sfc_debug.h"
23 #include "sfc_log.h"
24 #include "sfc_kvargs.h"
25 #include "sfc_ev.h"
26 #include "sfc_rx.h"
27 #include "sfc_tx.h"
28 #include "sfc_flow.h"
29 #include "sfc_dp.h"
30 #include "sfc_dp_rx.h"
31
32 uint32_t sfc_logtype_driver;
33
34 static struct sfc_dp_list sfc_dp_head =
35         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
36
37
38 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
39
40
41 static int
42 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
43 {
44         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
45         efx_nic_fw_info_t enfi;
46         int ret;
47         int rc;
48
49         /*
50          * Return value of the callback is likely supposed to be
51          * equal to or greater than 0, nevertheless, if an error
52          * occurs, it will be desirable to pass it to the caller
53          */
54         if ((fw_version == NULL) || (fw_size == 0))
55                 return -EINVAL;
56
57         rc = efx_nic_get_fw_version(sa->nic, &enfi);
58         if (rc != 0)
59                 return -rc;
60
61         ret = snprintf(fw_version, fw_size,
62                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
63                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
64                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
65         if (ret < 0)
66                 return ret;
67
68         if (enfi.enfi_dpcpu_fw_ids_valid) {
69                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
70                 int ret_extra;
71
72                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
73                                      fw_size - dpcpu_fw_ids_offset,
74                                      " rx%" PRIx16 " tx%" PRIx16,
75                                      enfi.enfi_rx_dpcpu_fw_id,
76                                      enfi.enfi_tx_dpcpu_fw_id);
77                 if (ret_extra < 0)
78                         return ret_extra;
79
80                 ret += ret_extra;
81         }
82
83         if (fw_size < (size_t)(++ret))
84                 return ret;
85         else
86                 return 0;
87 }
88
89 static int
90 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
91 {
92         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
93         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
94         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
95         struct sfc_rss *rss = &sas->rss;
96         uint64_t txq_offloads_def = 0;
97
98         sfc_log_init(sa, "entry");
99
100         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
101         dev_info->max_mtu = EFX_MAC_SDU_MAX;
102
103         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
104
105         /* Autonegotiation may be disabled */
106         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
107         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
108                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
109         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
110                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
111         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
112                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
113         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
114                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
115         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
116                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
117         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
118                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
119
120         dev_info->max_rx_queues = sa->rxq_max;
121         dev_info->max_tx_queues = sa->txq_max;
122
123         /* By default packets are dropped if no descriptors are available */
124         dev_info->default_rxconf.rx_drop_en = 1;
125
126         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
127
128         /*
129          * rx_offload_capa includes both device and queue offloads since
130          * the latter may be requested on a per device basis which makes
131          * sense when some offloads are needed to be set on all queues.
132          */
133         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
134                                     dev_info->rx_queue_offload_capa;
135
136         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
137
138         /*
139          * tx_offload_capa includes both device and queue offloads since
140          * the latter may be requested on a per device basis which makes
141          * sense when some offloads are needed to be set on all queues.
142          */
143         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
144                                     dev_info->tx_queue_offload_capa;
145
146         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
147                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
148
149         dev_info->default_txconf.offloads |= txq_offloads_def;
150
151         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
152                 uint64_t rte_hf = 0;
153                 unsigned int i;
154
155                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
156                         rte_hf |= rss->hf_map[i].rte;
157
158                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
159                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
160                 dev_info->flow_type_rss_offloads = rte_hf;
161         }
162
163         /* Initialize to hardware limits */
164         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
165         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
166         /* The RXQ hardware requires that the descriptor count is a power
167          * of 2, but rx_desc_lim cannot properly describe that constraint.
168          */
169         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
170
171         /* Initialize to hardware limits */
172         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
173         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
174         /*
175          * The TXQ hardware requires that the descriptor count is a power
176          * of 2, but tx_desc_lim cannot properly describe that constraint
177          */
178         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
179
180         if (sap->dp_rx->get_dev_info != NULL)
181                 sap->dp_rx->get_dev_info(dev_info);
182         if (sap->dp_tx->get_dev_info != NULL)
183                 sap->dp_tx->get_dev_info(dev_info);
184
185         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
186                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
187
188         return 0;
189 }
190
191 static const uint32_t *
192 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
193 {
194         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
195
196         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
197 }
198
199 static int
200 sfc_dev_configure(struct rte_eth_dev *dev)
201 {
202         struct rte_eth_dev_data *dev_data = dev->data;
203         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
204         int rc;
205
206         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
207                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
208
209         sfc_adapter_lock(sa);
210         switch (sa->state) {
211         case SFC_ADAPTER_CONFIGURED:
212                 /* FALLTHROUGH */
213         case SFC_ADAPTER_INITIALIZED:
214                 rc = sfc_configure(sa);
215                 break;
216         default:
217                 sfc_err(sa, "unexpected adapter state %u to configure",
218                         sa->state);
219                 rc = EINVAL;
220                 break;
221         }
222         sfc_adapter_unlock(sa);
223
224         sfc_log_init(sa, "done %d", rc);
225         SFC_ASSERT(rc >= 0);
226         return -rc;
227 }
228
229 static int
230 sfc_dev_start(struct rte_eth_dev *dev)
231 {
232         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
233         int rc;
234
235         sfc_log_init(sa, "entry");
236
237         sfc_adapter_lock(sa);
238         rc = sfc_start(sa);
239         sfc_adapter_unlock(sa);
240
241         sfc_log_init(sa, "done %d", rc);
242         SFC_ASSERT(rc >= 0);
243         return -rc;
244 }
245
246 static int
247 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
248 {
249         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
250         struct rte_eth_link current_link;
251         int ret;
252
253         sfc_log_init(sa, "entry");
254
255         if (sa->state != SFC_ADAPTER_STARTED) {
256                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
257         } else if (wait_to_complete) {
258                 efx_link_mode_t link_mode;
259
260                 if (efx_port_poll(sa->nic, &link_mode) != 0)
261                         link_mode = EFX_LINK_UNKNOWN;
262                 sfc_port_link_mode_to_info(link_mode, &current_link);
263
264         } else {
265                 sfc_ev_mgmt_qpoll(sa);
266                 rte_eth_linkstatus_get(dev, &current_link);
267         }
268
269         ret = rte_eth_linkstatus_set(dev, &current_link);
270         if (ret == 0)
271                 sfc_notice(sa, "Link status is %s",
272                            current_link.link_status ? "UP" : "DOWN");
273
274         return ret;
275 }
276
277 static void
278 sfc_dev_stop(struct rte_eth_dev *dev)
279 {
280         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
281
282         sfc_log_init(sa, "entry");
283
284         sfc_adapter_lock(sa);
285         sfc_stop(sa);
286         sfc_adapter_unlock(sa);
287
288         sfc_log_init(sa, "done");
289 }
290
291 static int
292 sfc_dev_set_link_up(struct rte_eth_dev *dev)
293 {
294         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
295         int rc;
296
297         sfc_log_init(sa, "entry");
298
299         sfc_adapter_lock(sa);
300         rc = sfc_start(sa);
301         sfc_adapter_unlock(sa);
302
303         SFC_ASSERT(rc >= 0);
304         return -rc;
305 }
306
307 static int
308 sfc_dev_set_link_down(struct rte_eth_dev *dev)
309 {
310         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
311
312         sfc_log_init(sa, "entry");
313
314         sfc_adapter_lock(sa);
315         sfc_stop(sa);
316         sfc_adapter_unlock(sa);
317
318         return 0;
319 }
320
321 static int
322 sfc_dev_close(struct rte_eth_dev *dev)
323 {
324         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
325
326         sfc_log_init(sa, "entry");
327
328         sfc_adapter_lock(sa);
329         switch (sa->state) {
330         case SFC_ADAPTER_STARTED:
331                 sfc_stop(sa);
332                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
333                 /* FALLTHROUGH */
334         case SFC_ADAPTER_CONFIGURED:
335                 sfc_close(sa);
336                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
337                 /* FALLTHROUGH */
338         case SFC_ADAPTER_INITIALIZED:
339                 break;
340         default:
341                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
342                 break;
343         }
344
345         /*
346          * Cleanup all resources in accordance with RTE_ETH_DEV_CLOSE_REMOVE.
347          * Rollback primary process sfc_eth_dev_init() below.
348          */
349
350         sfc_eth_dev_clear_ops(dev);
351
352         sfc_detach(sa);
353         sfc_unprobe(sa);
354
355         sfc_kvargs_cleanup(sa);
356
357         sfc_adapter_unlock(sa);
358         sfc_adapter_lock_fini(sa);
359
360         sfc_log_init(sa, "done");
361
362         /* Required for logging, so cleanup last */
363         sa->eth_dev = NULL;
364
365         dev->process_private = NULL;
366         free(sa);
367
368         return 0;
369 }
370
371 static int
372 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
373                    boolean_t enabled)
374 {
375         struct sfc_port *port;
376         boolean_t *toggle;
377         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
378         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
379         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
380         int rc = 0;
381
382         sfc_adapter_lock(sa);
383
384         port = &sa->port;
385         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
386
387         if (*toggle != enabled) {
388                 *toggle = enabled;
389
390                 if (sfc_sa2shared(sa)->isolated) {
391                         sfc_warn(sa, "isolated mode is active on the port");
392                         sfc_warn(sa, "the change is to be applied on the next "
393                                      "start provided that isolated mode is "
394                                      "disabled prior the next start");
395                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
396                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
397                         *toggle = !(enabled);
398                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
399                                  ((enabled) ? "enable" : "disable"), desc, rc);
400
401                         /*
402                          * For promiscuous and all-multicast filters a
403                          * permission failure should be reported as an
404                          * unsupported filter.
405                          */
406                         if (rc == EPERM)
407                                 rc = ENOTSUP;
408                 }
409         }
410
411         sfc_adapter_unlock(sa);
412         return rc;
413 }
414
415 static int
416 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
417 {
418         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
419
420         SFC_ASSERT(rc >= 0);
421         return -rc;
422 }
423
424 static int
425 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
426 {
427         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
428
429         SFC_ASSERT(rc >= 0);
430         return -rc;
431 }
432
433 static int
434 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
435 {
436         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
437
438         SFC_ASSERT(rc >= 0);
439         return -rc;
440 }
441
442 static int
443 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
444 {
445         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
446
447         SFC_ASSERT(rc >= 0);
448         return -rc;
449 }
450
451 static int
452 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
453                    uint16_t nb_rx_desc, unsigned int socket_id,
454                    const struct rte_eth_rxconf *rx_conf,
455                    struct rte_mempool *mb_pool)
456 {
457         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
458         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
459         int rc;
460
461         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
462                      rx_queue_id, nb_rx_desc, socket_id);
463
464         sfc_adapter_lock(sa);
465
466         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
467                           rx_conf, mb_pool);
468         if (rc != 0)
469                 goto fail_rx_qinit;
470
471         dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
472
473         sfc_adapter_unlock(sa);
474
475         return 0;
476
477 fail_rx_qinit:
478         sfc_adapter_unlock(sa);
479         SFC_ASSERT(rc > 0);
480         return -rc;
481 }
482
483 static void
484 sfc_rx_queue_release(void *queue)
485 {
486         struct sfc_dp_rxq *dp_rxq = queue;
487         struct sfc_rxq *rxq;
488         struct sfc_adapter *sa;
489         unsigned int sw_index;
490
491         if (dp_rxq == NULL)
492                 return;
493
494         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
495         sa = rxq->evq->sa;
496         sfc_adapter_lock(sa);
497
498         sw_index = dp_rxq->dpq.queue_id;
499
500         sfc_log_init(sa, "RxQ=%u", sw_index);
501
502         sfc_rx_qfini(sa, sw_index);
503
504         sfc_adapter_unlock(sa);
505 }
506
507 static int
508 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
509                    uint16_t nb_tx_desc, unsigned int socket_id,
510                    const struct rte_eth_txconf *tx_conf)
511 {
512         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
513         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
514         int rc;
515
516         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
517                      tx_queue_id, nb_tx_desc, socket_id);
518
519         sfc_adapter_lock(sa);
520
521         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
522         if (rc != 0)
523                 goto fail_tx_qinit;
524
525         dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
526
527         sfc_adapter_unlock(sa);
528         return 0;
529
530 fail_tx_qinit:
531         sfc_adapter_unlock(sa);
532         SFC_ASSERT(rc > 0);
533         return -rc;
534 }
535
536 static void
537 sfc_tx_queue_release(void *queue)
538 {
539         struct sfc_dp_txq *dp_txq = queue;
540         struct sfc_txq *txq;
541         unsigned int sw_index;
542         struct sfc_adapter *sa;
543
544         if (dp_txq == NULL)
545                 return;
546
547         txq = sfc_txq_by_dp_txq(dp_txq);
548         sw_index = dp_txq->dpq.queue_id;
549
550         SFC_ASSERT(txq->evq != NULL);
551         sa = txq->evq->sa;
552
553         sfc_log_init(sa, "TxQ = %u", sw_index);
554
555         sfc_adapter_lock(sa);
556
557         sfc_tx_qfini(sa, sw_index);
558
559         sfc_adapter_unlock(sa);
560 }
561
562 /*
563  * Some statistics are computed as A - B where A and B each increase
564  * monotonically with some hardware counter(s) and the counters are read
565  * asynchronously.
566  *
567  * If packet X is counted in A, but not counted in B yet, computed value is
568  * greater than real.
569  *
570  * If packet X is not counted in A at the moment of reading the counter,
571  * but counted in B at the moment of reading the counter, computed value
572  * is less than real.
573  *
574  * However, counter which grows backward is worse evil than slightly wrong
575  * value. So, let's try to guarantee that it never happens except may be
576  * the case when the MAC stats are zeroed as a result of a NIC reset.
577  */
578 static void
579 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
580 {
581         if ((int64_t)(newval - *stat) > 0 || newval == 0)
582                 *stat = newval;
583 }
584
585 static int
586 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
587 {
588         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
589         struct sfc_port *port = &sa->port;
590         uint64_t *mac_stats;
591         int ret;
592
593         rte_spinlock_lock(&port->mac_stats_lock);
594
595         ret = sfc_port_update_mac_stats(sa);
596         if (ret != 0)
597                 goto unlock;
598
599         mac_stats = port->mac_stats_buf;
600
601         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
602                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
603                 stats->ipackets =
604                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
605                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
606                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
607                 stats->opackets =
608                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
609                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
610                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
611                 stats->ibytes =
612                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
613                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
614                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
615                 stats->obytes =
616                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
617                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
618                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
619                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
620                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
621         } else {
622                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
623                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
624                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
625                 /*
626                  * Take into account stats which are whenever supported
627                  * on EF10. If some stat is not supported by current
628                  * firmware variant or HW revision, it is guaranteed
629                  * to be zero in mac_stats.
630                  */
631                 stats->imissed =
632                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
633                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
634                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
635                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
636                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
637                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
638                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
639                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
640                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
641                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
642                 stats->ierrors =
643                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
644                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
645                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
646                 /* no oerrors counters supported on EF10 */
647
648                 /* Exclude missed, errors and pauses from Rx packets */
649                 sfc_update_diff_stat(&port->ipackets,
650                         mac_stats[EFX_MAC_RX_PKTS] -
651                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
652                         stats->imissed - stats->ierrors);
653                 stats->ipackets = port->ipackets;
654         }
655
656 unlock:
657         rte_spinlock_unlock(&port->mac_stats_lock);
658         SFC_ASSERT(ret >= 0);
659         return -ret;
660 }
661
662 static int
663 sfc_stats_reset(struct rte_eth_dev *dev)
664 {
665         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
666         struct sfc_port *port = &sa->port;
667         int rc;
668
669         if (sa->state != SFC_ADAPTER_STARTED) {
670                 /*
671                  * The operation cannot be done if port is not started; it
672                  * will be scheduled to be done during the next port start
673                  */
674                 port->mac_stats_reset_pending = B_TRUE;
675                 return 0;
676         }
677
678         rc = sfc_port_reset_mac_stats(sa);
679         if (rc != 0)
680                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
681
682         SFC_ASSERT(rc >= 0);
683         return -rc;
684 }
685
686 static int
687 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
688                unsigned int xstats_count)
689 {
690         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
691         struct sfc_port *port = &sa->port;
692         uint64_t *mac_stats;
693         int rc;
694         unsigned int i;
695         int nstats = 0;
696
697         rte_spinlock_lock(&port->mac_stats_lock);
698
699         rc = sfc_port_update_mac_stats(sa);
700         if (rc != 0) {
701                 SFC_ASSERT(rc > 0);
702                 nstats = -rc;
703                 goto unlock;
704         }
705
706         mac_stats = port->mac_stats_buf;
707
708         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
709                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
710                         if (xstats != NULL && nstats < (int)xstats_count) {
711                                 xstats[nstats].id = nstats;
712                                 xstats[nstats].value = mac_stats[i];
713                         }
714                         nstats++;
715                 }
716         }
717
718 unlock:
719         rte_spinlock_unlock(&port->mac_stats_lock);
720
721         return nstats;
722 }
723
724 static int
725 sfc_xstats_get_names(struct rte_eth_dev *dev,
726                      struct rte_eth_xstat_name *xstats_names,
727                      unsigned int xstats_count)
728 {
729         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
730         struct sfc_port *port = &sa->port;
731         unsigned int i;
732         unsigned int nstats = 0;
733
734         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
735                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
736                         if (xstats_names != NULL && nstats < xstats_count)
737                                 strlcpy(xstats_names[nstats].name,
738                                         efx_mac_stat_name(sa->nic, i),
739                                         sizeof(xstats_names[0].name));
740                         nstats++;
741                 }
742         }
743
744         return nstats;
745 }
746
747 static int
748 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
749                      uint64_t *values, unsigned int n)
750 {
751         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
752         struct sfc_port *port = &sa->port;
753         uint64_t *mac_stats;
754         unsigned int nb_supported = 0;
755         unsigned int nb_written = 0;
756         unsigned int i;
757         int ret;
758         int rc;
759
760         if (unlikely(values == NULL) ||
761             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
762                 return port->mac_stats_nb_supported;
763
764         rte_spinlock_lock(&port->mac_stats_lock);
765
766         rc = sfc_port_update_mac_stats(sa);
767         if (rc != 0) {
768                 SFC_ASSERT(rc > 0);
769                 ret = -rc;
770                 goto unlock;
771         }
772
773         mac_stats = port->mac_stats_buf;
774
775         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
776                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
777                         continue;
778
779                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
780                         values[nb_written++] = mac_stats[i];
781
782                 ++nb_supported;
783         }
784
785         ret = nb_written;
786
787 unlock:
788         rte_spinlock_unlock(&port->mac_stats_lock);
789
790         return ret;
791 }
792
793 static int
794 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
795                            struct rte_eth_xstat_name *xstats_names,
796                            const uint64_t *ids, unsigned int size)
797 {
798         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
799         struct sfc_port *port = &sa->port;
800         unsigned int nb_supported = 0;
801         unsigned int nb_written = 0;
802         unsigned int i;
803
804         if (unlikely(xstats_names == NULL) ||
805             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
806                 return port->mac_stats_nb_supported;
807
808         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
809                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
810                         continue;
811
812                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
813                         char *name = xstats_names[nb_written++].name;
814
815                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
816                                 sizeof(xstats_names[0].name));
817                 }
818
819                 ++nb_supported;
820         }
821
822         return nb_written;
823 }
824
825 static int
826 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
827 {
828         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
829         unsigned int wanted_fc, link_fc;
830
831         memset(fc_conf, 0, sizeof(*fc_conf));
832
833         sfc_adapter_lock(sa);
834
835         if (sa->state == SFC_ADAPTER_STARTED)
836                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
837         else
838                 link_fc = sa->port.flow_ctrl;
839
840         switch (link_fc) {
841         case 0:
842                 fc_conf->mode = RTE_FC_NONE;
843                 break;
844         case EFX_FCNTL_RESPOND:
845                 fc_conf->mode = RTE_FC_RX_PAUSE;
846                 break;
847         case EFX_FCNTL_GENERATE:
848                 fc_conf->mode = RTE_FC_TX_PAUSE;
849                 break;
850         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
851                 fc_conf->mode = RTE_FC_FULL;
852                 break;
853         default:
854                 sfc_err(sa, "%s: unexpected flow control value %#x",
855                         __func__, link_fc);
856         }
857
858         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
859
860         sfc_adapter_unlock(sa);
861
862         return 0;
863 }
864
865 static int
866 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
867 {
868         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
869         struct sfc_port *port = &sa->port;
870         unsigned int fcntl;
871         int rc;
872
873         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
874             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
875             fc_conf->mac_ctrl_frame_fwd != 0) {
876                 sfc_err(sa, "unsupported flow control settings specified");
877                 rc = EINVAL;
878                 goto fail_inval;
879         }
880
881         switch (fc_conf->mode) {
882         case RTE_FC_NONE:
883                 fcntl = 0;
884                 break;
885         case RTE_FC_RX_PAUSE:
886                 fcntl = EFX_FCNTL_RESPOND;
887                 break;
888         case RTE_FC_TX_PAUSE:
889                 fcntl = EFX_FCNTL_GENERATE;
890                 break;
891         case RTE_FC_FULL:
892                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
893                 break;
894         default:
895                 rc = EINVAL;
896                 goto fail_inval;
897         }
898
899         sfc_adapter_lock(sa);
900
901         if (sa->state == SFC_ADAPTER_STARTED) {
902                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
903                 if (rc != 0)
904                         goto fail_mac_fcntl_set;
905         }
906
907         port->flow_ctrl = fcntl;
908         port->flow_ctrl_autoneg = fc_conf->autoneg;
909
910         sfc_adapter_unlock(sa);
911
912         return 0;
913
914 fail_mac_fcntl_set:
915         sfc_adapter_unlock(sa);
916 fail_inval:
917         SFC_ASSERT(rc > 0);
918         return -rc;
919 }
920
921 static int
922 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
923 {
924         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
925         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
926         boolean_t scatter_enabled;
927         const char *error;
928         unsigned int i;
929
930         for (i = 0; i < sas->rxq_count; i++) {
931                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
932                         continue;
933
934                 scatter_enabled = (sas->rxq_info[i].type_flags &
935                                    EFX_RXQ_FLAG_SCATTER);
936
937                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
938                                           encp->enc_rx_prefix_size,
939                                           scatter_enabled, &error)) {
940                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
941                                 error);
942                         return EINVAL;
943                 }
944         }
945
946         return 0;
947 }
948
949 static int
950 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
951 {
952         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
953         size_t pdu = EFX_MAC_PDU(mtu);
954         size_t old_pdu;
955         int rc;
956
957         sfc_log_init(sa, "mtu=%u", mtu);
958
959         rc = EINVAL;
960         if (pdu < EFX_MAC_PDU_MIN) {
961                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
962                         (unsigned int)mtu, (unsigned int)pdu,
963                         EFX_MAC_PDU_MIN);
964                 goto fail_inval;
965         }
966         if (pdu > EFX_MAC_PDU_MAX) {
967                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
968                         (unsigned int)mtu, (unsigned int)pdu,
969                         (unsigned int)EFX_MAC_PDU_MAX);
970                 goto fail_inval;
971         }
972
973         sfc_adapter_lock(sa);
974
975         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
976         if (rc != 0)
977                 goto fail_check_scatter;
978
979         if (pdu != sa->port.pdu) {
980                 if (sa->state == SFC_ADAPTER_STARTED) {
981                         sfc_stop(sa);
982
983                         old_pdu = sa->port.pdu;
984                         sa->port.pdu = pdu;
985                         rc = sfc_start(sa);
986                         if (rc != 0)
987                                 goto fail_start;
988                 } else {
989                         sa->port.pdu = pdu;
990                 }
991         }
992
993         /*
994          * The driver does not use it, but other PMDs update jumbo frame
995          * flag and max_rx_pkt_len when MTU is set.
996          */
997         if (mtu > RTE_ETHER_MAX_LEN) {
998                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
999                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1000         }
1001
1002         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1003
1004         sfc_adapter_unlock(sa);
1005
1006         sfc_log_init(sa, "done");
1007         return 0;
1008
1009 fail_start:
1010         sa->port.pdu = old_pdu;
1011         if (sfc_start(sa) != 0)
1012                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1013                         "PDU max size - port is stopped",
1014                         (unsigned int)pdu, (unsigned int)old_pdu);
1015
1016 fail_check_scatter:
1017         sfc_adapter_unlock(sa);
1018
1019 fail_inval:
1020         sfc_log_init(sa, "failed %d", rc);
1021         SFC_ASSERT(rc > 0);
1022         return -rc;
1023 }
1024 static int
1025 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1026 {
1027         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1028         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1029         struct sfc_port *port = &sa->port;
1030         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1031         int rc = 0;
1032
1033         sfc_adapter_lock(sa);
1034
1035         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1036                 goto unlock;
1037
1038         /*
1039          * Copy the address to the device private data so that
1040          * it could be recalled in the case of adapter restart.
1041          */
1042         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1043
1044         /*
1045          * Neither of the two following checks can return
1046          * an error. The new MAC address is preserved in
1047          * the device private data and can be activated
1048          * on the next port start if the user prevents
1049          * isolated mode from being enabled.
1050          */
1051         if (sfc_sa2shared(sa)->isolated) {
1052                 sfc_warn(sa, "isolated mode is active on the port");
1053                 sfc_warn(sa, "will not set MAC address");
1054                 goto unlock;
1055         }
1056
1057         if (sa->state != SFC_ADAPTER_STARTED) {
1058                 sfc_notice(sa, "the port is not started");
1059                 sfc_notice(sa, "the new MAC address will be set on port start");
1060
1061                 goto unlock;
1062         }
1063
1064         if (encp->enc_allow_set_mac_with_installed_filters) {
1065                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1066                 if (rc != 0) {
1067                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1068                         goto unlock;
1069                 }
1070
1071                 /*
1072                  * Changing the MAC address by means of MCDI request
1073                  * has no effect on received traffic, therefore
1074                  * we also need to update unicast filters
1075                  */
1076                 rc = sfc_set_rx_mode_unchecked(sa);
1077                 if (rc != 0) {
1078                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1079                         /* Rollback the old address */
1080                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1081                         (void)sfc_set_rx_mode_unchecked(sa);
1082                 }
1083         } else {
1084                 sfc_warn(sa, "cannot set MAC address with filters installed");
1085                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1086                 sfc_warn(sa, "(some traffic may be dropped)");
1087
1088                 /*
1089                  * Since setting MAC address with filters installed is not
1090                  * allowed on the adapter, the new MAC address will be set
1091                  * by means of adapter restart. sfc_start() shall retrieve
1092                  * the new address from the device private data and set it.
1093                  */
1094                 sfc_stop(sa);
1095                 rc = sfc_start(sa);
1096                 if (rc != 0)
1097                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1098         }
1099
1100 unlock:
1101         if (rc != 0)
1102                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1103
1104         sfc_adapter_unlock(sa);
1105
1106         SFC_ASSERT(rc >= 0);
1107         return -rc;
1108 }
1109
1110
1111 static int
1112 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1113                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1114 {
1115         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1116         struct sfc_port *port = &sa->port;
1117         uint8_t *mc_addrs = port->mcast_addrs;
1118         int rc;
1119         unsigned int i;
1120
1121         if (sfc_sa2shared(sa)->isolated) {
1122                 sfc_err(sa, "isolated mode is active on the port");
1123                 sfc_err(sa, "will not set multicast address list");
1124                 return -ENOTSUP;
1125         }
1126
1127         if (mc_addrs == NULL)
1128                 return -ENOBUFS;
1129
1130         if (nb_mc_addr > port->max_mcast_addrs) {
1131                 sfc_err(sa, "too many multicast addresses: %u > %u",
1132                          nb_mc_addr, port->max_mcast_addrs);
1133                 return -EINVAL;
1134         }
1135
1136         for (i = 0; i < nb_mc_addr; ++i) {
1137                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1138                                  EFX_MAC_ADDR_LEN);
1139                 mc_addrs += EFX_MAC_ADDR_LEN;
1140         }
1141
1142         port->nb_mcast_addrs = nb_mc_addr;
1143
1144         if (sa->state != SFC_ADAPTER_STARTED)
1145                 return 0;
1146
1147         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1148                                         port->nb_mcast_addrs);
1149         if (rc != 0)
1150                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1151
1152         SFC_ASSERT(rc >= 0);
1153         return -rc;
1154 }
1155
1156 /*
1157  * The function is used by the secondary process as well. It must not
1158  * use any process-local pointers from the adapter data.
1159  */
1160 static void
1161 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1162                       struct rte_eth_rxq_info *qinfo)
1163 {
1164         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1165         struct sfc_rxq_info *rxq_info;
1166
1167         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1168
1169         rxq_info = &sas->rxq_info[rx_queue_id];
1170
1171         qinfo->mp = rxq_info->refill_mb_pool;
1172         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1173         qinfo->conf.rx_drop_en = 1;
1174         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1175         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1176         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1177                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1178                 qinfo->scattered_rx = 1;
1179         }
1180         qinfo->nb_desc = rxq_info->entries;
1181 }
1182
1183 /*
1184  * The function is used by the secondary process as well. It must not
1185  * use any process-local pointers from the adapter data.
1186  */
1187 static void
1188 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1189                       struct rte_eth_txq_info *qinfo)
1190 {
1191         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1192         struct sfc_txq_info *txq_info;
1193
1194         SFC_ASSERT(tx_queue_id < sas->txq_count);
1195
1196         txq_info = &sas->txq_info[tx_queue_id];
1197
1198         memset(qinfo, 0, sizeof(*qinfo));
1199
1200         qinfo->conf.offloads = txq_info->offloads;
1201         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1202         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1203         qinfo->nb_desc = txq_info->entries;
1204 }
1205
1206 /*
1207  * The function is used by the secondary process as well. It must not
1208  * use any process-local pointers from the adapter data.
1209  */
1210 static uint32_t
1211 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1212 {
1213         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1214         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1215         struct sfc_rxq_info *rxq_info;
1216
1217         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1218         rxq_info = &sas->rxq_info[rx_queue_id];
1219
1220         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1221                 return 0;
1222
1223         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1224 }
1225
1226 /*
1227  * The function is used by the secondary process as well. It must not
1228  * use any process-local pointers from the adapter data.
1229  */
1230 static int
1231 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1232 {
1233         struct sfc_dp_rxq *dp_rxq = queue;
1234         const struct sfc_dp_rx *dp_rx;
1235
1236         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1237
1238         return offset < dp_rx->qdesc_npending(dp_rxq);
1239 }
1240
1241 /*
1242  * The function is used by the secondary process as well. It must not
1243  * use any process-local pointers from the adapter data.
1244  */
1245 static int
1246 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1247 {
1248         struct sfc_dp_rxq *dp_rxq = queue;
1249         const struct sfc_dp_rx *dp_rx;
1250
1251         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1252
1253         return dp_rx->qdesc_status(dp_rxq, offset);
1254 }
1255
1256 /*
1257  * The function is used by the secondary process as well. It must not
1258  * use any process-local pointers from the adapter data.
1259  */
1260 static int
1261 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1262 {
1263         struct sfc_dp_txq *dp_txq = queue;
1264         const struct sfc_dp_tx *dp_tx;
1265
1266         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1267
1268         return dp_tx->qdesc_status(dp_txq, offset);
1269 }
1270
1271 static int
1272 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1273 {
1274         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1275         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1276         int rc;
1277
1278         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1279
1280         sfc_adapter_lock(sa);
1281
1282         rc = EINVAL;
1283         if (sa->state != SFC_ADAPTER_STARTED)
1284                 goto fail_not_started;
1285
1286         if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1287                 goto fail_not_setup;
1288
1289         rc = sfc_rx_qstart(sa, rx_queue_id);
1290         if (rc != 0)
1291                 goto fail_rx_qstart;
1292
1293         sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1294
1295         sfc_adapter_unlock(sa);
1296
1297         return 0;
1298
1299 fail_rx_qstart:
1300 fail_not_setup:
1301 fail_not_started:
1302         sfc_adapter_unlock(sa);
1303         SFC_ASSERT(rc > 0);
1304         return -rc;
1305 }
1306
1307 static int
1308 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1309 {
1310         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1311         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1312
1313         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1314
1315         sfc_adapter_lock(sa);
1316         sfc_rx_qstop(sa, rx_queue_id);
1317
1318         sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1319
1320         sfc_adapter_unlock(sa);
1321
1322         return 0;
1323 }
1324
1325 static int
1326 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1327 {
1328         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1329         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1330         int rc;
1331
1332         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1333
1334         sfc_adapter_lock(sa);
1335
1336         rc = EINVAL;
1337         if (sa->state != SFC_ADAPTER_STARTED)
1338                 goto fail_not_started;
1339
1340         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1341                 goto fail_not_setup;
1342
1343         rc = sfc_tx_qstart(sa, tx_queue_id);
1344         if (rc != 0)
1345                 goto fail_tx_qstart;
1346
1347         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1348
1349         sfc_adapter_unlock(sa);
1350         return 0;
1351
1352 fail_tx_qstart:
1353
1354 fail_not_setup:
1355 fail_not_started:
1356         sfc_adapter_unlock(sa);
1357         SFC_ASSERT(rc > 0);
1358         return -rc;
1359 }
1360
1361 static int
1362 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1363 {
1364         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1365         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1366
1367         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1368
1369         sfc_adapter_lock(sa);
1370
1371         sfc_tx_qstop(sa, tx_queue_id);
1372
1373         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1374
1375         sfc_adapter_unlock(sa);
1376         return 0;
1377 }
1378
1379 static efx_tunnel_protocol_t
1380 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1381 {
1382         switch (rte_type) {
1383         case RTE_TUNNEL_TYPE_VXLAN:
1384                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1385         case RTE_TUNNEL_TYPE_GENEVE:
1386                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1387         default:
1388                 return EFX_TUNNEL_NPROTOS;
1389         }
1390 }
1391
1392 enum sfc_udp_tunnel_op_e {
1393         SFC_UDP_TUNNEL_ADD_PORT,
1394         SFC_UDP_TUNNEL_DEL_PORT,
1395 };
1396
1397 static int
1398 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1399                       struct rte_eth_udp_tunnel *tunnel_udp,
1400                       enum sfc_udp_tunnel_op_e op)
1401 {
1402         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1403         efx_tunnel_protocol_t tunnel_proto;
1404         int rc;
1405
1406         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1407                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1408                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1409                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1410
1411         tunnel_proto =
1412                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1413         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1414                 rc = ENOTSUP;
1415                 goto fail_bad_proto;
1416         }
1417
1418         sfc_adapter_lock(sa);
1419
1420         switch (op) {
1421         case SFC_UDP_TUNNEL_ADD_PORT:
1422                 rc = efx_tunnel_config_udp_add(sa->nic,
1423                                                tunnel_udp->udp_port,
1424                                                tunnel_proto);
1425                 break;
1426         case SFC_UDP_TUNNEL_DEL_PORT:
1427                 rc = efx_tunnel_config_udp_remove(sa->nic,
1428                                                   tunnel_udp->udp_port,
1429                                                   tunnel_proto);
1430                 break;
1431         default:
1432                 rc = EINVAL;
1433                 goto fail_bad_op;
1434         }
1435
1436         if (rc != 0)
1437                 goto fail_op;
1438
1439         if (sa->state == SFC_ADAPTER_STARTED) {
1440                 rc = efx_tunnel_reconfigure(sa->nic);
1441                 if (rc == EAGAIN) {
1442                         /*
1443                          * Configuration is accepted by FW and MC reboot
1444                          * is initiated to apply the changes. MC reboot
1445                          * will be handled in a usual way (MC reboot
1446                          * event on management event queue and adapter
1447                          * restart).
1448                          */
1449                         rc = 0;
1450                 } else if (rc != 0) {
1451                         goto fail_reconfigure;
1452                 }
1453         }
1454
1455         sfc_adapter_unlock(sa);
1456         return 0;
1457
1458 fail_reconfigure:
1459         /* Remove/restore entry since the change makes the trouble */
1460         switch (op) {
1461         case SFC_UDP_TUNNEL_ADD_PORT:
1462                 (void)efx_tunnel_config_udp_remove(sa->nic,
1463                                                    tunnel_udp->udp_port,
1464                                                    tunnel_proto);
1465                 break;
1466         case SFC_UDP_TUNNEL_DEL_PORT:
1467                 (void)efx_tunnel_config_udp_add(sa->nic,
1468                                                 tunnel_udp->udp_port,
1469                                                 tunnel_proto);
1470                 break;
1471         }
1472
1473 fail_op:
1474 fail_bad_op:
1475         sfc_adapter_unlock(sa);
1476
1477 fail_bad_proto:
1478         SFC_ASSERT(rc > 0);
1479         return -rc;
1480 }
1481
1482 static int
1483 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1484                             struct rte_eth_udp_tunnel *tunnel_udp)
1485 {
1486         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1487 }
1488
1489 static int
1490 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1491                             struct rte_eth_udp_tunnel *tunnel_udp)
1492 {
1493         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1494 }
1495
1496 /*
1497  * The function is used by the secondary process as well. It must not
1498  * use any process-local pointers from the adapter data.
1499  */
1500 static int
1501 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1502                           struct rte_eth_rss_conf *rss_conf)
1503 {
1504         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1505         struct sfc_rss *rss = &sas->rss;
1506
1507         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1508                 return -ENOTSUP;
1509
1510         /*
1511          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1512          * hence, conversion is done here to derive a correct set of ETH_RSS
1513          * flags which corresponds to the active EFX configuration stored
1514          * locally in 'sfc_adapter' and kept up-to-date
1515          */
1516         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1517         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1518         if (rss_conf->rss_key != NULL)
1519                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1520
1521         return 0;
1522 }
1523
1524 static int
1525 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1526                         struct rte_eth_rss_conf *rss_conf)
1527 {
1528         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1529         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1530         unsigned int efx_hash_types;
1531         int rc = 0;
1532
1533         if (sfc_sa2shared(sa)->isolated)
1534                 return -ENOTSUP;
1535
1536         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1537                 sfc_err(sa, "RSS is not available");
1538                 return -ENOTSUP;
1539         }
1540
1541         if (rss->channels == 0) {
1542                 sfc_err(sa, "RSS is not configured");
1543                 return -EINVAL;
1544         }
1545
1546         if ((rss_conf->rss_key != NULL) &&
1547             (rss_conf->rss_key_len != sizeof(rss->key))) {
1548                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1549                         sizeof(rss->key));
1550                 return -EINVAL;
1551         }
1552
1553         sfc_adapter_lock(sa);
1554
1555         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1556         if (rc != 0)
1557                 goto fail_rx_hf_rte_to_efx;
1558
1559         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1560                                    rss->hash_alg, efx_hash_types, B_TRUE);
1561         if (rc != 0)
1562                 goto fail_scale_mode_set;
1563
1564         if (rss_conf->rss_key != NULL) {
1565                 if (sa->state == SFC_ADAPTER_STARTED) {
1566                         rc = efx_rx_scale_key_set(sa->nic,
1567                                                   EFX_RSS_CONTEXT_DEFAULT,
1568                                                   rss_conf->rss_key,
1569                                                   sizeof(rss->key));
1570                         if (rc != 0)
1571                                 goto fail_scale_key_set;
1572                 }
1573
1574                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1575         }
1576
1577         rss->hash_types = efx_hash_types;
1578
1579         sfc_adapter_unlock(sa);
1580
1581         return 0;
1582
1583 fail_scale_key_set:
1584         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1585                                   EFX_RX_HASHALG_TOEPLITZ,
1586                                   rss->hash_types, B_TRUE) != 0)
1587                 sfc_err(sa, "failed to restore RSS mode");
1588
1589 fail_scale_mode_set:
1590 fail_rx_hf_rte_to_efx:
1591         sfc_adapter_unlock(sa);
1592         return -rc;
1593 }
1594
1595 /*
1596  * The function is used by the secondary process as well. It must not
1597  * use any process-local pointers from the adapter data.
1598  */
1599 static int
1600 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1601                        struct rte_eth_rss_reta_entry64 *reta_conf,
1602                        uint16_t reta_size)
1603 {
1604         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1605         struct sfc_rss *rss = &sas->rss;
1606         int entry;
1607
1608         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1609                 return -ENOTSUP;
1610
1611         if (rss->channels == 0)
1612                 return -EINVAL;
1613
1614         if (reta_size != EFX_RSS_TBL_SIZE)
1615                 return -EINVAL;
1616
1617         for (entry = 0; entry < reta_size; entry++) {
1618                 int grp = entry / RTE_RETA_GROUP_SIZE;
1619                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1620
1621                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1622                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1623         }
1624
1625         return 0;
1626 }
1627
1628 static int
1629 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1630                         struct rte_eth_rss_reta_entry64 *reta_conf,
1631                         uint16_t reta_size)
1632 {
1633         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1634         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1635         unsigned int *rss_tbl_new;
1636         uint16_t entry;
1637         int rc = 0;
1638
1639
1640         if (sfc_sa2shared(sa)->isolated)
1641                 return -ENOTSUP;
1642
1643         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1644                 sfc_err(sa, "RSS is not available");
1645                 return -ENOTSUP;
1646         }
1647
1648         if (rss->channels == 0) {
1649                 sfc_err(sa, "RSS is not configured");
1650                 return -EINVAL;
1651         }
1652
1653         if (reta_size != EFX_RSS_TBL_SIZE) {
1654                 sfc_err(sa, "RETA size is wrong (should be %u)",
1655                         EFX_RSS_TBL_SIZE);
1656                 return -EINVAL;
1657         }
1658
1659         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1660         if (rss_tbl_new == NULL)
1661                 return -ENOMEM;
1662
1663         sfc_adapter_lock(sa);
1664
1665         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1666
1667         for (entry = 0; entry < reta_size; entry++) {
1668                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1669                 struct rte_eth_rss_reta_entry64 *grp;
1670
1671                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1672
1673                 if (grp->mask & (1ull << grp_idx)) {
1674                         if (grp->reta[grp_idx] >= rss->channels) {
1675                                 rc = EINVAL;
1676                                 goto bad_reta_entry;
1677                         }
1678                         rss_tbl_new[entry] = grp->reta[grp_idx];
1679                 }
1680         }
1681
1682         if (sa->state == SFC_ADAPTER_STARTED) {
1683                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1684                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1685                 if (rc != 0)
1686                         goto fail_scale_tbl_set;
1687         }
1688
1689         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1690
1691 fail_scale_tbl_set:
1692 bad_reta_entry:
1693         sfc_adapter_unlock(sa);
1694
1695         rte_free(rss_tbl_new);
1696
1697         SFC_ASSERT(rc >= 0);
1698         return -rc;
1699 }
1700
1701 static int
1702 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1703                     enum rte_filter_op filter_op,
1704                     void *arg)
1705 {
1706         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1707         int rc = ENOTSUP;
1708
1709         sfc_log_init(sa, "entry");
1710
1711         switch (filter_type) {
1712         case RTE_ETH_FILTER_NONE:
1713                 sfc_err(sa, "Global filters configuration not supported");
1714                 break;
1715         case RTE_ETH_FILTER_MACVLAN:
1716                 sfc_err(sa, "MACVLAN filters not supported");
1717                 break;
1718         case RTE_ETH_FILTER_ETHERTYPE:
1719                 sfc_err(sa, "EtherType filters not supported");
1720                 break;
1721         case RTE_ETH_FILTER_FLEXIBLE:
1722                 sfc_err(sa, "Flexible filters not supported");
1723                 break;
1724         case RTE_ETH_FILTER_SYN:
1725                 sfc_err(sa, "SYN filters not supported");
1726                 break;
1727         case RTE_ETH_FILTER_NTUPLE:
1728                 sfc_err(sa, "NTUPLE filters not supported");
1729                 break;
1730         case RTE_ETH_FILTER_TUNNEL:
1731                 sfc_err(sa, "Tunnel filters not supported");
1732                 break;
1733         case RTE_ETH_FILTER_FDIR:
1734                 sfc_err(sa, "Flow Director filters not supported");
1735                 break;
1736         case RTE_ETH_FILTER_HASH:
1737                 sfc_err(sa, "Hash filters not supported");
1738                 break;
1739         case RTE_ETH_FILTER_GENERIC:
1740                 if (filter_op != RTE_ETH_FILTER_GET) {
1741                         rc = EINVAL;
1742                 } else {
1743                         *(const void **)arg = &sfc_flow_ops;
1744                         rc = 0;
1745                 }
1746                 break;
1747         default:
1748                 sfc_err(sa, "Unknown filter type %u", filter_type);
1749                 break;
1750         }
1751
1752         sfc_log_init(sa, "exit: %d", -rc);
1753         SFC_ASSERT(rc >= 0);
1754         return -rc;
1755 }
1756
1757 static int
1758 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1759 {
1760         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1761
1762         /*
1763          * If Rx datapath does not provide callback to check mempool,
1764          * all pools are supported.
1765          */
1766         if (sap->dp_rx->pool_ops_supported == NULL)
1767                 return 1;
1768
1769         return sap->dp_rx->pool_ops_supported(pool);
1770 }
1771
1772 static int
1773 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1774 {
1775         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1776         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1777         struct sfc_rxq_info *rxq_info;
1778
1779         SFC_ASSERT(queue_id < sas->rxq_count);
1780         rxq_info = &sas->rxq_info[queue_id];
1781
1782         return sap->dp_rx->intr_enable(rxq_info->dp);
1783 }
1784
1785 static int
1786 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1787 {
1788         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1789         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1790         struct sfc_rxq_info *rxq_info;
1791
1792         SFC_ASSERT(queue_id < sas->rxq_count);
1793         rxq_info = &sas->rxq_info[queue_id];
1794
1795         return sap->dp_rx->intr_disable(rxq_info->dp);
1796 }
1797
1798 static const struct eth_dev_ops sfc_eth_dev_ops = {
1799         .dev_configure                  = sfc_dev_configure,
1800         .dev_start                      = sfc_dev_start,
1801         .dev_stop                       = sfc_dev_stop,
1802         .dev_set_link_up                = sfc_dev_set_link_up,
1803         .dev_set_link_down              = sfc_dev_set_link_down,
1804         .dev_close                      = sfc_dev_close,
1805         .promiscuous_enable             = sfc_dev_promisc_enable,
1806         .promiscuous_disable            = sfc_dev_promisc_disable,
1807         .allmulticast_enable            = sfc_dev_allmulti_enable,
1808         .allmulticast_disable           = sfc_dev_allmulti_disable,
1809         .link_update                    = sfc_dev_link_update,
1810         .stats_get                      = sfc_stats_get,
1811         .stats_reset                    = sfc_stats_reset,
1812         .xstats_get                     = sfc_xstats_get,
1813         .xstats_reset                   = sfc_stats_reset,
1814         .xstats_get_names               = sfc_xstats_get_names,
1815         .dev_infos_get                  = sfc_dev_infos_get,
1816         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1817         .mtu_set                        = sfc_dev_set_mtu,
1818         .rx_queue_start                 = sfc_rx_queue_start,
1819         .rx_queue_stop                  = sfc_rx_queue_stop,
1820         .tx_queue_start                 = sfc_tx_queue_start,
1821         .tx_queue_stop                  = sfc_tx_queue_stop,
1822         .rx_queue_setup                 = sfc_rx_queue_setup,
1823         .rx_queue_release               = sfc_rx_queue_release,
1824         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1825         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1826         .tx_queue_setup                 = sfc_tx_queue_setup,
1827         .tx_queue_release               = sfc_tx_queue_release,
1828         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1829         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1830         .mac_addr_set                   = sfc_mac_addr_set,
1831         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1832         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1833         .reta_update                    = sfc_dev_rss_reta_update,
1834         .reta_query                     = sfc_dev_rss_reta_query,
1835         .rss_hash_update                = sfc_dev_rss_hash_update,
1836         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1837         .filter_ctrl                    = sfc_dev_filter_ctrl,
1838         .set_mc_addr_list               = sfc_set_mc_addr_list,
1839         .rxq_info_get                   = sfc_rx_queue_info_get,
1840         .txq_info_get                   = sfc_tx_queue_info_get,
1841         .fw_version_get                 = sfc_fw_version_get,
1842         .xstats_get_by_id               = sfc_xstats_get_by_id,
1843         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1844         .pool_ops_supported             = sfc_pool_ops_supported,
1845 };
1846
1847 /**
1848  * Duplicate a string in potentially shared memory required for
1849  * multi-process support.
1850  *
1851  * strdup() allocates from process-local heap/memory.
1852  */
1853 static char *
1854 sfc_strdup(const char *str)
1855 {
1856         size_t size;
1857         char *copy;
1858
1859         if (str == NULL)
1860                 return NULL;
1861
1862         size = strlen(str) + 1;
1863         copy = rte_malloc(__func__, size, 0);
1864         if (copy != NULL)
1865                 rte_memcpy(copy, str, size);
1866
1867         return copy;
1868 }
1869
1870 static int
1871 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1872 {
1873         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1874         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1875         const struct sfc_dp_rx *dp_rx;
1876         const struct sfc_dp_tx *dp_tx;
1877         const efx_nic_cfg_t *encp;
1878         unsigned int avail_caps = 0;
1879         const char *rx_name = NULL;
1880         const char *tx_name = NULL;
1881         int rc;
1882
1883         switch (sa->family) {
1884         case EFX_FAMILY_HUNTINGTON:
1885         case EFX_FAMILY_MEDFORD:
1886         case EFX_FAMILY_MEDFORD2:
1887                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1888                 break;
1889         default:
1890                 break;
1891         }
1892
1893         encp = efx_nic_cfg_get(sa->nic);
1894         if (encp->enc_rx_es_super_buffer_supported)
1895                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1896
1897         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1898                                 sfc_kvarg_string_handler, &rx_name);
1899         if (rc != 0)
1900                 goto fail_kvarg_rx_datapath;
1901
1902         if (rx_name != NULL) {
1903                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1904                 if (dp_rx == NULL) {
1905                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1906                         rc = ENOENT;
1907                         goto fail_dp_rx;
1908                 }
1909                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1910                         sfc_err(sa,
1911                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1912                                 rx_name);
1913                         rc = EINVAL;
1914                         goto fail_dp_rx_caps;
1915                 }
1916         } else {
1917                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1918                 if (dp_rx == NULL) {
1919                         sfc_err(sa, "Rx datapath by caps %#x not found",
1920                                 avail_caps);
1921                         rc = ENOENT;
1922                         goto fail_dp_rx;
1923                 }
1924         }
1925
1926         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1927         if (sas->dp_rx_name == NULL) {
1928                 rc = ENOMEM;
1929                 goto fail_dp_rx_name;
1930         }
1931
1932         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1933
1934         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1935                                 sfc_kvarg_string_handler, &tx_name);
1936         if (rc != 0)
1937                 goto fail_kvarg_tx_datapath;
1938
1939         if (tx_name != NULL) {
1940                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1941                 if (dp_tx == NULL) {
1942                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1943                         rc = ENOENT;
1944                         goto fail_dp_tx;
1945                 }
1946                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1947                         sfc_err(sa,
1948                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1949                                 tx_name);
1950                         rc = EINVAL;
1951                         goto fail_dp_tx_caps;
1952                 }
1953         } else {
1954                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1955                 if (dp_tx == NULL) {
1956                         sfc_err(sa, "Tx datapath by caps %#x not found",
1957                                 avail_caps);
1958                         rc = ENOENT;
1959                         goto fail_dp_tx;
1960                 }
1961         }
1962
1963         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1964         if (sas->dp_tx_name == NULL) {
1965                 rc = ENOMEM;
1966                 goto fail_dp_tx_name;
1967         }
1968
1969         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1970
1971         sa->priv.dp_rx = dp_rx;
1972         sa->priv.dp_tx = dp_tx;
1973
1974         dev->rx_pkt_burst = dp_rx->pkt_burst;
1975         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1976         dev->tx_pkt_burst = dp_tx->pkt_burst;
1977
1978         dev->rx_queue_count = sfc_rx_queue_count;
1979         dev->rx_descriptor_done = sfc_rx_descriptor_done;
1980         dev->rx_descriptor_status = sfc_rx_descriptor_status;
1981         dev->tx_descriptor_status = sfc_tx_descriptor_status;
1982         dev->dev_ops = &sfc_eth_dev_ops;
1983
1984         return 0;
1985
1986 fail_dp_tx_name:
1987 fail_dp_tx_caps:
1988 fail_dp_tx:
1989 fail_kvarg_tx_datapath:
1990         rte_free(sas->dp_rx_name);
1991         sas->dp_rx_name = NULL;
1992
1993 fail_dp_rx_name:
1994 fail_dp_rx_caps:
1995 fail_dp_rx:
1996 fail_kvarg_rx_datapath:
1997         return rc;
1998 }
1999
2000 static void
2001 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2002 {
2003         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2004         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2005
2006         dev->dev_ops = NULL;
2007         dev->tx_pkt_prepare = NULL;
2008         dev->rx_pkt_burst = NULL;
2009         dev->tx_pkt_burst = NULL;
2010
2011         rte_free(sas->dp_tx_name);
2012         sas->dp_tx_name = NULL;
2013         sa->priv.dp_tx = NULL;
2014
2015         rte_free(sas->dp_rx_name);
2016         sas->dp_rx_name = NULL;
2017         sa->priv.dp_rx = NULL;
2018 }
2019
2020 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2021         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2022         .reta_query                     = sfc_dev_rss_reta_query,
2023         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2024         .rxq_info_get                   = sfc_rx_queue_info_get,
2025         .txq_info_get                   = sfc_tx_queue_info_get,
2026 };
2027
2028 static int
2029 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2030 {
2031         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2032         struct sfc_adapter_priv *sap;
2033         const struct sfc_dp_rx *dp_rx;
2034         const struct sfc_dp_tx *dp_tx;
2035         int rc;
2036
2037         /*
2038          * Allocate process private data from heap, since it should not
2039          * be located in shared memory allocated using rte_malloc() API.
2040          */
2041         sap = calloc(1, sizeof(*sap));
2042         if (sap == NULL) {
2043                 rc = ENOMEM;
2044                 goto fail_alloc_priv;
2045         }
2046
2047         sap->logtype_main = logtype_main;
2048
2049         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2050         if (dp_rx == NULL) {
2051                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2052                         "cannot find %s Rx datapath", sas->dp_rx_name);
2053                 rc = ENOENT;
2054                 goto fail_dp_rx;
2055         }
2056         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2057                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2058                         "%s Rx datapath does not support multi-process",
2059                         sas->dp_rx_name);
2060                 rc = EINVAL;
2061                 goto fail_dp_rx_multi_process;
2062         }
2063
2064         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2065         if (dp_tx == NULL) {
2066                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2067                         "cannot find %s Tx datapath", sas->dp_tx_name);
2068                 rc = ENOENT;
2069                 goto fail_dp_tx;
2070         }
2071         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2072                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2073                         "%s Tx datapath does not support multi-process",
2074                         sas->dp_tx_name);
2075                 rc = EINVAL;
2076                 goto fail_dp_tx_multi_process;
2077         }
2078
2079         sap->dp_rx = dp_rx;
2080         sap->dp_tx = dp_tx;
2081
2082         dev->process_private = sap;
2083         dev->rx_pkt_burst = dp_rx->pkt_burst;
2084         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2085         dev->tx_pkt_burst = dp_tx->pkt_burst;
2086         dev->rx_queue_count = sfc_rx_queue_count;
2087         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2088         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2089         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2090         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2091
2092         return 0;
2093
2094 fail_dp_tx_multi_process:
2095 fail_dp_tx:
2096 fail_dp_rx_multi_process:
2097 fail_dp_rx:
2098         free(sap);
2099
2100 fail_alloc_priv:
2101         return rc;
2102 }
2103
2104 static void
2105 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
2106 {
2107         free(dev->process_private);
2108         dev->process_private = NULL;
2109         dev->dev_ops = NULL;
2110         dev->tx_pkt_prepare = NULL;
2111         dev->tx_pkt_burst = NULL;
2112         dev->rx_pkt_burst = NULL;
2113 }
2114
2115 static void
2116 sfc_register_dp(void)
2117 {
2118         /* Register once */
2119         if (TAILQ_EMPTY(&sfc_dp_head)) {
2120                 /* Prefer EF10 datapath */
2121                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2122                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2123                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2124
2125                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2126                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2127                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2128         }
2129 }
2130
2131 static int
2132 sfc_eth_dev_init(struct rte_eth_dev *dev)
2133 {
2134         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2135         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2136         uint32_t logtype_main;
2137         struct sfc_adapter *sa;
2138         int rc;
2139         const efx_nic_cfg_t *encp;
2140         const struct rte_ether_addr *from;
2141         int ret;
2142
2143         sfc_register_dp();
2144
2145         logtype_main = sfc_register_logtype(&pci_dev->addr,
2146                                             SFC_LOGTYPE_MAIN_STR,
2147                                             RTE_LOG_NOTICE);
2148
2149         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2150                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2151
2152         /* Required for logging */
2153         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2154                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2155                         pci_dev->addr.domain, pci_dev->addr.bus,
2156                         pci_dev->addr.devid, pci_dev->addr.function,
2157                         dev->data->port_id);
2158         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2159                 SFC_GENERIC_LOG(ERR,
2160                         "reserved log prefix is too short for " PCI_PRI_FMT,
2161                         pci_dev->addr.domain, pci_dev->addr.bus,
2162                         pci_dev->addr.devid, pci_dev->addr.function);
2163                 return -EINVAL;
2164         }
2165         sas->pci_addr = pci_dev->addr;
2166         sas->port_id = dev->data->port_id;
2167
2168         /*
2169          * Allocate process private data from heap, since it should not
2170          * be located in shared memory allocated using rte_malloc() API.
2171          */
2172         sa = calloc(1, sizeof(*sa));
2173         if (sa == NULL) {
2174                 rc = ENOMEM;
2175                 goto fail_alloc_sa;
2176         }
2177
2178         dev->process_private = sa;
2179
2180         /* Required for logging */
2181         sa->priv.shared = sas;
2182         sa->priv.logtype_main = logtype_main;
2183
2184         sa->eth_dev = dev;
2185
2186         /* Copy PCI device info to the dev->data */
2187         rte_eth_copy_pci_info(dev, pci_dev);
2188
2189         rc = sfc_kvargs_parse(sa);
2190         if (rc != 0)
2191                 goto fail_kvargs_parse;
2192
2193         sfc_log_init(sa, "entry");
2194
2195         dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
2196
2197         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2198         if (dev->data->mac_addrs == NULL) {
2199                 rc = ENOMEM;
2200                 goto fail_mac_addrs;
2201         }
2202
2203         sfc_adapter_lock_init(sa);
2204         sfc_adapter_lock(sa);
2205
2206         sfc_log_init(sa, "probing");
2207         rc = sfc_probe(sa);
2208         if (rc != 0)
2209                 goto fail_probe;
2210
2211         sfc_log_init(sa, "set device ops");
2212         rc = sfc_eth_dev_set_ops(dev);
2213         if (rc != 0)
2214                 goto fail_set_ops;
2215
2216         sfc_log_init(sa, "attaching");
2217         rc = sfc_attach(sa);
2218         if (rc != 0)
2219                 goto fail_attach;
2220
2221         encp = efx_nic_cfg_get(sa->nic);
2222
2223         /*
2224          * The arguments are really reverse order in comparison to
2225          * Linux kernel. Copy from NIC config to Ethernet device data.
2226          */
2227         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2228         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2229
2230         sfc_adapter_unlock(sa);
2231
2232         sfc_log_init(sa, "done");
2233         return 0;
2234
2235 fail_attach:
2236         sfc_eth_dev_clear_ops(dev);
2237
2238 fail_set_ops:
2239         sfc_unprobe(sa);
2240
2241 fail_probe:
2242         sfc_adapter_unlock(sa);
2243         sfc_adapter_lock_fini(sa);
2244         rte_free(dev->data->mac_addrs);
2245         dev->data->mac_addrs = NULL;
2246
2247 fail_mac_addrs:
2248         sfc_kvargs_cleanup(sa);
2249
2250 fail_kvargs_parse:
2251         sfc_log_init(sa, "failed %d", rc);
2252         dev->process_private = NULL;
2253         free(sa);
2254
2255 fail_alloc_sa:
2256         SFC_ASSERT(rc > 0);
2257         return -rc;
2258 }
2259
2260 static int
2261 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2262 {
2263         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2264                 sfc_eth_dev_secondary_clear_ops(dev);
2265                 return 0;
2266         }
2267
2268         sfc_dev_close(dev);
2269
2270         return 0;
2271 }
2272
2273 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2274         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2275         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2276         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2277         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2278         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2279         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2280         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2281         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2282         { .vendor_id = 0 /* sentinel */ }
2283 };
2284
2285 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2286         struct rte_pci_device *pci_dev)
2287 {
2288         return rte_eth_dev_pci_generic_probe(pci_dev,
2289                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2290 }
2291
2292 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2293 {
2294         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2295 }
2296
2297 static struct rte_pci_driver sfc_efx_pmd = {
2298         .id_table = pci_id_sfc_efx_map,
2299         .drv_flags =
2300                 RTE_PCI_DRV_INTR_LSC |
2301                 RTE_PCI_DRV_NEED_MAPPING,
2302         .probe = sfc_eth_dev_pci_probe,
2303         .remove = sfc_eth_dev_pci_remove,
2304 };
2305
2306 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2307 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2308 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2309 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2310         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2311         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2312         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2313         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2314         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2315         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2316
2317 RTE_INIT(sfc_driver_register_logtype)
2318 {
2319         int ret;
2320
2321         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2322                                                    RTE_LOG_NOTICE);
2323         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2324 }