net/enic: add simple Rx handler
[dpdk.git] / drivers / net / enic / enic_ethdev.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2008-2017 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  */
5
6 #include <stdio.h>
7 #include <stdint.h>
8
9 #include <rte_dev.h>
10 #include <rte_pci.h>
11 #include <rte_bus_pci.h>
12 #include <rte_ethdev_driver.h>
13 #include <rte_ethdev_pci.h>
14 #include <rte_kvargs.h>
15 #include <rte_string_fns.h>
16
17 #include "vnic_intr.h"
18 #include "vnic_cq.h"
19 #include "vnic_wq.h"
20 #include "vnic_rq.h"
21 #include "vnic_enet.h"
22 #include "enic.h"
23
24 int enicpmd_logtype_init;
25 int enicpmd_logtype_flow;
26
27 #define ENICPMD_FUNC_TRACE() PMD_INIT_LOG(DEBUG, " >>")
28
29 /*
30  * The set of PCI devices this driver supports
31  */
32 #define CISCO_PCI_VENDOR_ID 0x1137
33 static const struct rte_pci_id pci_id_enic_map[] = {
34         { RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET) },
35         { RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
36         {.vendor_id = 0, /* sentinel */},
37 };
38
39 #define ENIC_DEVARG_DISABLE_OVERLAY "disable-overlay"
40 #define ENIC_DEVARG_IG_VLAN_REWRITE "ig-vlan-rewrite"
41
42 RTE_INIT(enicpmd_init_log);
43 static void
44 enicpmd_init_log(void)
45 {
46         enicpmd_logtype_init = rte_log_register("pmd.net.enic.init");
47         if (enicpmd_logtype_init >= 0)
48                 rte_log_set_level(enicpmd_logtype_init, RTE_LOG_NOTICE);
49         enicpmd_logtype_flow = rte_log_register("pmd.net.enic.flow");
50         if (enicpmd_logtype_flow >= 0)
51                 rte_log_set_level(enicpmd_logtype_flow, RTE_LOG_NOTICE);
52 }
53
54 static int
55 enicpmd_fdir_ctrl_func(struct rte_eth_dev *eth_dev,
56                         enum rte_filter_op filter_op, void *arg)
57 {
58         struct enic *enic = pmd_priv(eth_dev);
59         int ret = 0;
60
61         ENICPMD_FUNC_TRACE();
62         if (filter_op == RTE_ETH_FILTER_NOP)
63                 return 0;
64
65         if (arg == NULL && filter_op != RTE_ETH_FILTER_FLUSH)
66                 return -EINVAL;
67
68         switch (filter_op) {
69         case RTE_ETH_FILTER_ADD:
70         case RTE_ETH_FILTER_UPDATE:
71                 ret = enic_fdir_add_fltr(enic,
72                         (struct rte_eth_fdir_filter *)arg);
73                 break;
74
75         case RTE_ETH_FILTER_DELETE:
76                 ret = enic_fdir_del_fltr(enic,
77                         (struct rte_eth_fdir_filter *)arg);
78                 break;
79
80         case RTE_ETH_FILTER_STATS:
81                 enic_fdir_stats_get(enic, (struct rte_eth_fdir_stats *)arg);
82                 break;
83
84         case RTE_ETH_FILTER_FLUSH:
85                 dev_warning(enic, "unsupported operation %u", filter_op);
86                 ret = -ENOTSUP;
87                 break;
88         case RTE_ETH_FILTER_INFO:
89                 enic_fdir_info_get(enic, (struct rte_eth_fdir_info *)arg);
90                 break;
91         default:
92                 dev_err(enic, "unknown operation %u", filter_op);
93                 ret = -EINVAL;
94                 break;
95         }
96         return ret;
97 }
98
99 static int
100 enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
101                      enum rte_filter_type filter_type,
102                      enum rte_filter_op filter_op,
103                      void *arg)
104 {
105         int ret = 0;
106
107         ENICPMD_FUNC_TRACE();
108
109         switch (filter_type) {
110         case RTE_ETH_FILTER_GENERIC:
111                 if (filter_op != RTE_ETH_FILTER_GET)
112                         return -EINVAL;
113                 *(const void **)arg = &enic_flow_ops;
114                 break;
115         case RTE_ETH_FILTER_FDIR:
116                 ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg);
117                 break;
118         default:
119                 dev_warning(enic, "Filter type (%d) not supported",
120                         filter_type);
121                 ret = -EINVAL;
122                 break;
123         }
124
125         return ret;
126 }
127
128 static void enicpmd_dev_tx_queue_release(void *txq)
129 {
130         ENICPMD_FUNC_TRACE();
131
132         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
133                 return;
134
135         enic_free_wq(txq);
136 }
137
138 static int enicpmd_dev_setup_intr(struct enic *enic)
139 {
140         int ret;
141         unsigned int index;
142
143         ENICPMD_FUNC_TRACE();
144
145         /* Are we done with the init of all the queues? */
146         for (index = 0; index < enic->cq_count; index++) {
147                 if (!enic->cq[index].ctrl)
148                         break;
149         }
150         if (enic->cq_count != index)
151                 return 0;
152         for (index = 0; index < enic->wq_count; index++) {
153                 if (!enic->wq[index].ctrl)
154                         break;
155         }
156         if (enic->wq_count != index)
157                 return 0;
158         /* check start of packet (SOP) RQs only in case scatter is disabled. */
159         for (index = 0; index < enic->rq_count; index++) {
160                 if (!enic->rq[enic_rte_rq_idx_to_sop_idx(index)].ctrl)
161                         break;
162         }
163         if (enic->rq_count != index)
164                 return 0;
165
166         ret = enic_alloc_intr_resources(enic);
167         if (ret) {
168                 dev_err(enic, "alloc intr failed\n");
169                 return ret;
170         }
171         enic_init_vnic_resources(enic);
172
173         ret = enic_setup_finish(enic);
174         if (ret)
175                 dev_err(enic, "setup could not be finished\n");
176
177         return ret;
178 }
179
180 static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
181         uint16_t queue_idx,
182         uint16_t nb_desc,
183         unsigned int socket_id,
184         const struct rte_eth_txconf *tx_conf)
185 {
186         int ret;
187         struct enic *enic = pmd_priv(eth_dev);
188         struct vnic_wq *wq;
189
190         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
191                 return -E_RTE_SECONDARY;
192
193         ENICPMD_FUNC_TRACE();
194         RTE_ASSERT(queue_idx < enic->conf_wq_count);
195         wq = &enic->wq[queue_idx];
196         wq->offloads = tx_conf->offloads |
197                 eth_dev->data->dev_conf.txmode.offloads;
198         eth_dev->data->tx_queues[queue_idx] = (void *)wq;
199
200         ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
201         if (ret) {
202                 dev_err(enic, "error in allocating wq\n");
203                 return ret;
204         }
205
206         return enicpmd_dev_setup_intr(enic);
207 }
208
209 static int enicpmd_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
210         uint16_t queue_idx)
211 {
212         struct enic *enic = pmd_priv(eth_dev);
213
214         ENICPMD_FUNC_TRACE();
215
216         enic_start_wq(enic, queue_idx);
217
218         return 0;
219 }
220
221 static int enicpmd_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
222         uint16_t queue_idx)
223 {
224         int ret;
225         struct enic *enic = pmd_priv(eth_dev);
226
227         ENICPMD_FUNC_TRACE();
228
229         ret = enic_stop_wq(enic, queue_idx);
230         if (ret)
231                 dev_err(enic, "error in stopping wq %d\n", queue_idx);
232
233         return ret;
234 }
235
236 static int enicpmd_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
237         uint16_t queue_idx)
238 {
239         struct enic *enic = pmd_priv(eth_dev);
240
241         ENICPMD_FUNC_TRACE();
242
243         enic_start_rq(enic, queue_idx);
244
245         return 0;
246 }
247
248 static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
249         uint16_t queue_idx)
250 {
251         int ret;
252         struct enic *enic = pmd_priv(eth_dev);
253
254         ENICPMD_FUNC_TRACE();
255
256         ret = enic_stop_rq(enic, queue_idx);
257         if (ret)
258                 dev_err(enic, "error in stopping rq %d\n", queue_idx);
259
260         return ret;
261 }
262
263 static void enicpmd_dev_rx_queue_release(void *rxq)
264 {
265         ENICPMD_FUNC_TRACE();
266
267         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
268                 return;
269
270         enic_free_rq(rxq);
271 }
272
273 static uint32_t enicpmd_dev_rx_queue_count(struct rte_eth_dev *dev,
274                                            uint16_t rx_queue_id)
275 {
276         struct enic *enic = pmd_priv(dev);
277         uint32_t queue_count = 0;
278         struct vnic_cq *cq;
279         uint32_t cq_tail;
280         uint16_t cq_idx;
281         int rq_num;
282
283         rq_num = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
284         cq = &enic->cq[enic_cq_rq(enic, rq_num)];
285         cq_idx = cq->to_clean;
286
287         cq_tail = ioread32(&cq->ctrl->cq_tail);
288
289         if (cq_tail < cq_idx)
290                 cq_tail += cq->ring.desc_count;
291
292         queue_count = cq_tail - cq_idx;
293
294         return queue_count;
295 }
296
297 static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
298         uint16_t queue_idx,
299         uint16_t nb_desc,
300         unsigned int socket_id,
301         const struct rte_eth_rxconf *rx_conf,
302         struct rte_mempool *mp)
303 {
304         int ret;
305         struct enic *enic = pmd_priv(eth_dev);
306
307         ENICPMD_FUNC_TRACE();
308
309         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
310                 return -E_RTE_SECONDARY;
311         RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
312         eth_dev->data->rx_queues[queue_idx] =
313                 (void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
314
315         ret = enic_alloc_rq(enic, queue_idx, socket_id, mp, nb_desc,
316                             rx_conf->rx_free_thresh);
317         if (ret) {
318                 dev_err(enic, "error in allocating rq\n");
319                 return ret;
320         }
321
322         return enicpmd_dev_setup_intr(enic);
323 }
324
325 static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
326 {
327         struct enic *enic = pmd_priv(eth_dev);
328         uint64_t offloads;
329
330         ENICPMD_FUNC_TRACE();
331
332         offloads = eth_dev->data->dev_conf.rxmode.offloads;
333         if (mask & ETH_VLAN_STRIP_MASK) {
334                 if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
335                         enic->ig_vlan_strip_en = 1;
336                 else
337                         enic->ig_vlan_strip_en = 0;
338         }
339
340         if ((mask & ETH_VLAN_FILTER_MASK) &&
341             (offloads & DEV_RX_OFFLOAD_VLAN_FILTER)) {
342                 dev_warning(enic,
343                         "Configuration of VLAN filter is not supported\n");
344         }
345
346         if ((mask & ETH_VLAN_EXTEND_MASK) &&
347             (offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)) {
348                 dev_warning(enic,
349                         "Configuration of extended VLAN is not supported\n");
350         }
351
352         return enic_set_vlan_strip(enic);
353 }
354
355 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
356 {
357         int ret;
358         int mask;
359         struct enic *enic = pmd_priv(eth_dev);
360
361         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
362                 return -E_RTE_SECONDARY;
363
364         ENICPMD_FUNC_TRACE();
365         ret = enic_set_vnic_res(enic);
366         if (ret) {
367                 dev_err(enic, "Set vNIC resource num  failed, aborting\n");
368                 return ret;
369         }
370
371         enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
372                                   DEV_RX_OFFLOAD_CHECKSUM);
373         /* All vlan offload masks to apply the current settings */
374         mask = ETH_VLAN_STRIP_MASK |
375                 ETH_VLAN_FILTER_MASK |
376                 ETH_VLAN_EXTEND_MASK;
377         ret = enicpmd_vlan_offload_set(eth_dev, mask);
378         if (ret) {
379                 dev_err(enic, "Failed to configure VLAN offloads\n");
380                 return ret;
381         }
382         /*
383          * Initialize RSS with the default reta and key. If the user key is
384          * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
385          * default key.
386          */
387         return enic_init_rss_nic_cfg(enic);
388 }
389
390 /* Start the device.
391  * It returns 0 on success.
392  */
393 static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
394 {
395         struct enic *enic = pmd_priv(eth_dev);
396
397         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
398                 return -E_RTE_SECONDARY;
399
400         ENICPMD_FUNC_TRACE();
401         return enic_enable(enic);
402 }
403
404 /*
405  * Stop device: disable rx and tx functions to allow for reconfiguring.
406  */
407 static void enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
408 {
409         struct rte_eth_link link;
410         struct enic *enic = pmd_priv(eth_dev);
411
412         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
413                 return;
414
415         ENICPMD_FUNC_TRACE();
416         enic_disable(enic);
417
418         memset(&link, 0, sizeof(link));
419         rte_eth_linkstatus_set(eth_dev, &link);
420 }
421
422 /*
423  * Stop device.
424  */
425 static void enicpmd_dev_close(struct rte_eth_dev *eth_dev)
426 {
427         struct enic *enic = pmd_priv(eth_dev);
428
429         ENICPMD_FUNC_TRACE();
430         enic_remove(enic);
431 }
432
433 static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
434         __rte_unused int wait_to_complete)
435 {
436         struct enic *enic = pmd_priv(eth_dev);
437
438         ENICPMD_FUNC_TRACE();
439         return enic_link_update(enic);
440 }
441
442 static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
443         struct rte_eth_stats *stats)
444 {
445         struct enic *enic = pmd_priv(eth_dev);
446
447         ENICPMD_FUNC_TRACE();
448         return enic_dev_stats_get(enic, stats);
449 }
450
451 static void enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
452 {
453         struct enic *enic = pmd_priv(eth_dev);
454
455         ENICPMD_FUNC_TRACE();
456         enic_dev_stats_clear(enic);
457 }
458
459 static void enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
460         struct rte_eth_dev_info *device_info)
461 {
462         struct enic *enic = pmd_priv(eth_dev);
463
464         ENICPMD_FUNC_TRACE();
465         /* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
466         device_info->max_rx_queues = enic->conf_rq_count / 2;
467         device_info->max_tx_queues = enic->conf_wq_count;
468         device_info->min_rx_bufsize = ENIC_MIN_MTU;
469         /* "Max" mtu is not a typo. HW receives packet sizes up to the
470          * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
471          * a hint to the driver to size receive buffers accordingly so that
472          * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
473          * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
474          * ignoring vNIC mtu.
475          */
476         device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
477         device_info->max_mac_addrs = ENIC_MAX_MAC_ADDR;
478         device_info->rx_offload_capa = enic->rx_offload_capa;
479         device_info->tx_offload_capa = enic->tx_offload_capa;
480         device_info->tx_queue_offload_capa = enic->tx_queue_offload_capa;
481         device_info->default_rxconf = (struct rte_eth_rxconf) {
482                 .rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
483         };
484         device_info->reta_size = enic->reta_size;
485         device_info->hash_key_size = enic->hash_key_size;
486         device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
487         device_info->rx_desc_lim = (struct rte_eth_desc_lim) {
488                 .nb_max = enic->config.rq_desc_count,
489                 .nb_min = ENIC_MIN_RQ_DESCS,
490                 .nb_align = ENIC_ALIGN_DESCS,
491         };
492         device_info->tx_desc_lim = (struct rte_eth_desc_lim) {
493                 .nb_max = enic->config.wq_desc_count,
494                 .nb_min = ENIC_MIN_WQ_DESCS,
495                 .nb_align = ENIC_ALIGN_DESCS,
496                 .nb_seg_max = ENIC_TX_XMIT_MAX,
497                 .nb_mtu_seg_max = ENIC_NON_TSO_MAX_DESC,
498         };
499         device_info->default_rxportconf = (struct rte_eth_dev_portconf) {
500                 .burst_size = ENIC_DEFAULT_RX_BURST,
501                 .ring_size = RTE_MIN(device_info->rx_desc_lim.nb_max,
502                         ENIC_DEFAULT_RX_RING_SIZE),
503                 .nb_queues = ENIC_DEFAULT_RX_RINGS,
504         };
505         device_info->default_txportconf = (struct rte_eth_dev_portconf) {
506                 .burst_size = ENIC_DEFAULT_TX_BURST,
507                 .ring_size = RTE_MIN(device_info->tx_desc_lim.nb_max,
508                         ENIC_DEFAULT_TX_RING_SIZE),
509                 .nb_queues = ENIC_DEFAULT_TX_RINGS,
510         };
511 }
512
513 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
514 {
515         static const uint32_t ptypes[] = {
516                 RTE_PTYPE_L2_ETHER,
517                 RTE_PTYPE_L2_ETHER_VLAN,
518                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
519                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
520                 RTE_PTYPE_L4_TCP,
521                 RTE_PTYPE_L4_UDP,
522                 RTE_PTYPE_L4_FRAG,
523                 RTE_PTYPE_L4_NONFRAG,
524                 RTE_PTYPE_UNKNOWN
525         };
526
527         if (dev->rx_pkt_burst == enic_recv_pkts ||
528             dev->rx_pkt_burst == enic_noscatter_recv_pkts)
529                 return ptypes;
530         return NULL;
531 }
532
533 static void enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
534 {
535         struct enic *enic = pmd_priv(eth_dev);
536
537         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
538                 return;
539
540         ENICPMD_FUNC_TRACE();
541
542         enic->promisc = 1;
543         enic_add_packet_filter(enic);
544 }
545
546 static void enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
547 {
548         struct enic *enic = pmd_priv(eth_dev);
549
550         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
551                 return;
552
553         ENICPMD_FUNC_TRACE();
554         enic->promisc = 0;
555         enic_add_packet_filter(enic);
556 }
557
558 static void enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
559 {
560         struct enic *enic = pmd_priv(eth_dev);
561
562         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
563                 return;
564
565         ENICPMD_FUNC_TRACE();
566         enic->allmulti = 1;
567         enic_add_packet_filter(enic);
568 }
569
570 static void enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
571 {
572         struct enic *enic = pmd_priv(eth_dev);
573
574         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
575                 return;
576
577         ENICPMD_FUNC_TRACE();
578         enic->allmulti = 0;
579         enic_add_packet_filter(enic);
580 }
581
582 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
583         struct ether_addr *mac_addr,
584         __rte_unused uint32_t index, __rte_unused uint32_t pool)
585 {
586         struct enic *enic = pmd_priv(eth_dev);
587
588         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
589                 return -E_RTE_SECONDARY;
590
591         ENICPMD_FUNC_TRACE();
592         return enic_set_mac_address(enic, mac_addr->addr_bytes);
593 }
594
595 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
596 {
597         struct enic *enic = pmd_priv(eth_dev);
598
599         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
600                 return;
601
602         ENICPMD_FUNC_TRACE();
603         if (enic_del_mac_address(enic, index))
604                 dev_err(enic, "del mac addr failed\n");
605 }
606
607 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
608                                 struct ether_addr *addr)
609 {
610         struct enic *enic = pmd_priv(eth_dev);
611         int ret;
612
613         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
614                 return -E_RTE_SECONDARY;
615
616         ENICPMD_FUNC_TRACE();
617         ret = enic_del_mac_address(enic, 0);
618         if (ret)
619                 return ret;
620         return enic_set_mac_address(enic, addr->addr_bytes);
621 }
622
623 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
624 {
625         struct enic *enic = pmd_priv(eth_dev);
626
627         ENICPMD_FUNC_TRACE();
628         return enic_set_mtu(enic, mtu);
629 }
630
631 static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
632                                       struct rte_eth_rss_reta_entry64
633                                       *reta_conf,
634                                       uint16_t reta_size)
635 {
636         struct enic *enic = pmd_priv(dev);
637         uint16_t i, idx, shift;
638
639         ENICPMD_FUNC_TRACE();
640         if (reta_size != ENIC_RSS_RETA_SIZE) {
641                 dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
642                         reta_size, ENIC_RSS_RETA_SIZE);
643                 return -EINVAL;
644         }
645
646         for (i = 0; i < reta_size; i++) {
647                 idx = i / RTE_RETA_GROUP_SIZE;
648                 shift = i % RTE_RETA_GROUP_SIZE;
649                 if (reta_conf[idx].mask & (1ULL << shift))
650                         reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
651                                 enic->rss_cpu.cpu[i / 4].b[i % 4]);
652         }
653
654         return 0;
655 }
656
657 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
658                                        struct rte_eth_rss_reta_entry64
659                                        *reta_conf,
660                                        uint16_t reta_size)
661 {
662         struct enic *enic = pmd_priv(dev);
663         union vnic_rss_cpu rss_cpu;
664         uint16_t i, idx, shift;
665
666         ENICPMD_FUNC_TRACE();
667         if (reta_size != ENIC_RSS_RETA_SIZE) {
668                 dev_err(enic, "reta_update: wrong reta_size. given=%u"
669                         " expected=%u\n",
670                         reta_size, ENIC_RSS_RETA_SIZE);
671                 return -EINVAL;
672         }
673         /*
674          * Start with the current reta and modify it per reta_conf, as we
675          * need to push the entire reta even if we only modify one entry.
676          */
677         rss_cpu = enic->rss_cpu;
678         for (i = 0; i < reta_size; i++) {
679                 idx = i / RTE_RETA_GROUP_SIZE;
680                 shift = i % RTE_RETA_GROUP_SIZE;
681                 if (reta_conf[idx].mask & (1ULL << shift))
682                         rss_cpu.cpu[i / 4].b[i % 4] =
683                                 enic_rte_rq_idx_to_sop_idx(
684                                         reta_conf[idx].reta[shift]);
685         }
686         return enic_set_rss_reta(enic, &rss_cpu);
687 }
688
689 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
690                                        struct rte_eth_rss_conf *rss_conf)
691 {
692         struct enic *enic = pmd_priv(dev);
693
694         ENICPMD_FUNC_TRACE();
695         return enic_set_rss_conf(enic, rss_conf);
696 }
697
698 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
699                                          struct rte_eth_rss_conf *rss_conf)
700 {
701         struct enic *enic = pmd_priv(dev);
702
703         ENICPMD_FUNC_TRACE();
704         if (rss_conf == NULL)
705                 return -EINVAL;
706         if (rss_conf->rss_key != NULL &&
707             rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
708                 dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
709                         " expected=%u+\n",
710                         rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
711                 return -EINVAL;
712         }
713         rss_conf->rss_hf = enic->rss_hf;
714         if (rss_conf->rss_key != NULL) {
715                 int i;
716                 for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
717                         rss_conf->rss_key[i] =
718                                 enic->rss_key.key[i / 10].b[i % 10];
719                 }
720                 rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
721         }
722         return 0;
723 }
724
725 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
726                                      uint16_t rx_queue_id,
727                                      struct rte_eth_rxq_info *qinfo)
728 {
729         struct enic *enic = pmd_priv(dev);
730         struct vnic_rq *rq_sop;
731         struct vnic_rq *rq_data;
732         struct rte_eth_rxconf *conf;
733         uint16_t sop_queue_idx;
734         uint16_t data_queue_idx;
735
736         ENICPMD_FUNC_TRACE();
737         sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
738         data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id);
739         rq_sop = &enic->rq[sop_queue_idx];
740         rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
741         qinfo->mp = rq_sop->mp;
742         qinfo->scattered_rx = rq_sop->data_queue_enable;
743         qinfo->nb_desc = rq_sop->ring.desc_count;
744         if (qinfo->scattered_rx)
745                 qinfo->nb_desc += rq_data->ring.desc_count;
746         conf = &qinfo->conf;
747         memset(conf, 0, sizeof(*conf));
748         conf->rx_free_thresh = rq_sop->rx_free_thresh;
749         conf->rx_drop_en = 1;
750         /*
751          * Except VLAN stripping (port setting), all the checksum offloads
752          * are always enabled.
753          */
754         conf->offloads = enic->rx_offload_capa;
755         if (!enic->ig_vlan_strip_en)
756                 conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
757         /* rx_thresh and other fields are not applicable for enic */
758 }
759
760 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
761                                      uint16_t tx_queue_id,
762                                      struct rte_eth_txq_info *qinfo)
763 {
764         struct enic *enic = pmd_priv(dev);
765         struct vnic_wq *wq = &enic->wq[tx_queue_id];
766
767         ENICPMD_FUNC_TRACE();
768         qinfo->nb_desc = wq->ring.desc_count;
769         memset(&qinfo->conf, 0, sizeof(qinfo->conf));
770         qinfo->conf.offloads = wq->offloads;
771         /* tx_thresh, and all the other fields are not applicable for enic */
772 }
773
774 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
775                                             uint16_t rx_queue_id)
776 {
777         struct enic *enic = pmd_priv(eth_dev);
778
779         ENICPMD_FUNC_TRACE();
780         vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
781         return 0;
782 }
783
784 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
785                                              uint16_t rx_queue_id)
786 {
787         struct enic *enic = pmd_priv(eth_dev);
788
789         ENICPMD_FUNC_TRACE();
790         vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
791         return 0;
792 }
793
794 static int udp_tunnel_common_check(struct enic *enic,
795                                    struct rte_eth_udp_tunnel *tnl)
796 {
797         if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN)
798                 return -ENOTSUP;
799         if (!enic->overlay_offload) {
800                 PMD_INIT_LOG(DEBUG, " vxlan (overlay offload) is not "
801                              "supported\n");
802                 return -ENOTSUP;
803         }
804         return 0;
805 }
806
807 static int update_vxlan_port(struct enic *enic, uint16_t port)
808 {
809         if (vnic_dev_overlay_offload_cfg(enic->vdev,
810                                          OVERLAY_CFG_VXLAN_PORT_UPDATE,
811                                          port)) {
812                 PMD_INIT_LOG(DEBUG, " failed to update vxlan port\n");
813                 return -EINVAL;
814         }
815         PMD_INIT_LOG(DEBUG, " updated vxlan port to %u\n", port);
816         enic->vxlan_port = port;
817         return 0;
818 }
819
820 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
821                                            struct rte_eth_udp_tunnel *tnl)
822 {
823         struct enic *enic = pmd_priv(eth_dev);
824         int ret;
825
826         ENICPMD_FUNC_TRACE();
827         ret = udp_tunnel_common_check(enic, tnl);
828         if (ret)
829                 return ret;
830         /*
831          * The NIC has 1 configurable VXLAN port number. "Adding" a new port
832          * number replaces it.
833          */
834         if (tnl->udp_port == enic->vxlan_port || tnl->udp_port == 0) {
835                 PMD_INIT_LOG(DEBUG, " %u is already configured or invalid\n",
836                              tnl->udp_port);
837                 return -EINVAL;
838         }
839         return update_vxlan_port(enic, tnl->udp_port);
840 }
841
842 static int enicpmd_dev_udp_tunnel_port_del(struct rte_eth_dev *eth_dev,
843                                            struct rte_eth_udp_tunnel *tnl)
844 {
845         struct enic *enic = pmd_priv(eth_dev);
846         int ret;
847
848         ENICPMD_FUNC_TRACE();
849         ret = udp_tunnel_common_check(enic, tnl);
850         if (ret)
851                 return ret;
852         /*
853          * Clear the previously set port number and restore the
854          * hardware default port number. Some drivers disable VXLAN
855          * offloads when there are no configured port numbers. But
856          * enic does not do that as VXLAN is part of overlay offload,
857          * which is tied to inner RSS and TSO.
858          */
859         if (tnl->udp_port != enic->vxlan_port) {
860                 PMD_INIT_LOG(DEBUG, " %u is not a configured vxlan port\n",
861                              tnl->udp_port);
862                 return -EINVAL;
863         }
864         return update_vxlan_port(enic, ENIC_DEFAULT_VXLAN_PORT);
865 }
866
867 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
868         .dev_configure        = enicpmd_dev_configure,
869         .dev_start            = enicpmd_dev_start,
870         .dev_stop             = enicpmd_dev_stop,
871         .dev_set_link_up      = NULL,
872         .dev_set_link_down    = NULL,
873         .dev_close            = enicpmd_dev_close,
874         .promiscuous_enable   = enicpmd_dev_promiscuous_enable,
875         .promiscuous_disable  = enicpmd_dev_promiscuous_disable,
876         .allmulticast_enable  = enicpmd_dev_allmulticast_enable,
877         .allmulticast_disable = enicpmd_dev_allmulticast_disable,
878         .link_update          = enicpmd_dev_link_update,
879         .stats_get            = enicpmd_dev_stats_get,
880         .stats_reset          = enicpmd_dev_stats_reset,
881         .queue_stats_mapping_set = NULL,
882         .dev_infos_get        = enicpmd_dev_info_get,
883         .dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
884         .mtu_set              = enicpmd_mtu_set,
885         .vlan_filter_set      = NULL,
886         .vlan_tpid_set        = NULL,
887         .vlan_offload_set     = enicpmd_vlan_offload_set,
888         .vlan_strip_queue_set = NULL,
889         .rx_queue_start       = enicpmd_dev_rx_queue_start,
890         .rx_queue_stop        = enicpmd_dev_rx_queue_stop,
891         .tx_queue_start       = enicpmd_dev_tx_queue_start,
892         .tx_queue_stop        = enicpmd_dev_tx_queue_stop,
893         .rx_queue_setup       = enicpmd_dev_rx_queue_setup,
894         .rx_queue_release     = enicpmd_dev_rx_queue_release,
895         .rx_queue_count       = enicpmd_dev_rx_queue_count,
896         .rx_descriptor_done   = NULL,
897         .tx_queue_setup       = enicpmd_dev_tx_queue_setup,
898         .tx_queue_release     = enicpmd_dev_tx_queue_release,
899         .rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
900         .rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
901         .rxq_info_get         = enicpmd_dev_rxq_info_get,
902         .txq_info_get         = enicpmd_dev_txq_info_get,
903         .dev_led_on           = NULL,
904         .dev_led_off          = NULL,
905         .flow_ctrl_get        = NULL,
906         .flow_ctrl_set        = NULL,
907         .priority_flow_ctrl_set = NULL,
908         .mac_addr_add         = enicpmd_add_mac_addr,
909         .mac_addr_remove      = enicpmd_remove_mac_addr,
910         .mac_addr_set         = enicpmd_set_mac_addr,
911         .filter_ctrl          = enicpmd_dev_filter_ctrl,
912         .reta_query           = enicpmd_dev_rss_reta_query,
913         .reta_update          = enicpmd_dev_rss_reta_update,
914         .rss_hash_conf_get    = enicpmd_dev_rss_hash_conf_get,
915         .rss_hash_update      = enicpmd_dev_rss_hash_update,
916         .udp_tunnel_port_add  = enicpmd_dev_udp_tunnel_port_add,
917         .udp_tunnel_port_del  = enicpmd_dev_udp_tunnel_port_del,
918 };
919
920 static int enic_parse_disable_overlay(__rte_unused const char *key,
921                                       const char *value,
922                                       void *opaque)
923 {
924         struct enic *enic;
925
926         enic = (struct enic *)opaque;
927         if (strcmp(value, "0") == 0) {
928                 enic->disable_overlay = false;
929         } else if (strcmp(value, "1") == 0) {
930                 enic->disable_overlay = true;
931         } else {
932                 dev_err(enic, "Invalid value for " ENIC_DEVARG_DISABLE_OVERLAY
933                         ": expected=0|1 given=%s\n", value);
934                 return -EINVAL;
935         }
936         return 0;
937 }
938
939 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
940                                       const char *value,
941                                       void *opaque)
942 {
943         struct enic *enic;
944
945         enic = (struct enic *)opaque;
946         if (strcmp(value, "trunk") == 0) {
947                 /* Trunk mode: always tag */
948                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
949         } else if (strcmp(value, "untag") == 0) {
950                 /* Untag default VLAN mode: untag if VLAN = default VLAN */
951                 enic->ig_vlan_rewrite_mode =
952                         IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
953         } else if (strcmp(value, "priority") == 0) {
954                 /*
955                  * Priority-tag default VLAN mode: priority tag (VLAN header
956                  * with ID=0) if VLAN = default
957                  */
958                 enic->ig_vlan_rewrite_mode =
959                         IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
960         } else if (strcmp(value, "pass") == 0) {
961                 /* Pass through mode: do not touch tags */
962                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
963         } else {
964                 dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
965                         ": expected=trunk|untag|priority|pass given=%s\n",
966                         value);
967                 return -EINVAL;
968         }
969         return 0;
970 }
971
972 static int enic_check_devargs(struct rte_eth_dev *dev)
973 {
974         static const char *const valid_keys[] = {
975                 ENIC_DEVARG_DISABLE_OVERLAY,
976                 ENIC_DEVARG_IG_VLAN_REWRITE,
977                 NULL};
978         struct enic *enic = pmd_priv(dev);
979         struct rte_kvargs *kvlist;
980
981         ENICPMD_FUNC_TRACE();
982
983         enic->disable_overlay = false;
984         enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
985         if (!dev->device->devargs)
986                 return 0;
987         kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
988         if (!kvlist)
989                 return -EINVAL;
990         if (rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
991                                enic_parse_disable_overlay, enic) < 0 ||
992             rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
993                                enic_parse_ig_vlan_rewrite, enic) < 0) {
994                 rte_kvargs_free(kvlist);
995                 return -EINVAL;
996         }
997         rte_kvargs_free(kvlist);
998         return 0;
999 }
1000
1001 struct enic *enicpmd_list_head = NULL;
1002 /* Initialize the driver
1003  * It returns 0 on success.
1004  */
1005 static int eth_enicpmd_dev_init(struct rte_eth_dev *eth_dev)
1006 {
1007         struct rte_pci_device *pdev;
1008         struct rte_pci_addr *addr;
1009         struct enic *enic = pmd_priv(eth_dev);
1010         int err;
1011
1012         ENICPMD_FUNC_TRACE();
1013
1014         enic->port_id = eth_dev->data->port_id;
1015         enic->rte_dev = eth_dev;
1016         eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1017         eth_dev->rx_pkt_burst = &enic_recv_pkts;
1018         eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1019         eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1020
1021         pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1022         rte_eth_copy_pci_info(eth_dev, pdev);
1023         enic->pdev = pdev;
1024         addr = &pdev->addr;
1025
1026         snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1027                 addr->domain, addr->bus, addr->devid, addr->function);
1028
1029         err = enic_check_devargs(eth_dev);
1030         if (err)
1031                 return err;
1032         return enic_probe(enic);
1033 }
1034
1035 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1036         struct rte_pci_device *pci_dev)
1037 {
1038         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct enic),
1039                 eth_enicpmd_dev_init);
1040 }
1041
1042 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1043 {
1044         return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1045 }
1046
1047 static struct rte_pci_driver rte_enic_pmd = {
1048         .id_table = pci_id_enic_map,
1049         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1050         .probe = eth_enic_pci_probe,
1051         .remove = eth_enic_pci_remove,
1052 };
1053
1054 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1055 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1056 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1057 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1058         ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1059         ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");