xref: /linux/drivers/net/ethernet/meta/fbnic/fbnic_netdev.c (revision 07cf71bf25cd4e5735ff13468e7b86f02c3665cb)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) Meta Platforms, Inc. and affiliates. */
3 
4 #include <linux/etherdevice.h>
5 #include <linux/ipv6.h>
6 #include <linux/types.h>
7 #include <net/netdev_queues.h>
8 
9 #include "fbnic.h"
10 #include "fbnic_netdev.h"
11 #include "fbnic_txrx.h"
12 
13 int __fbnic_open(struct fbnic_net *fbn)
14 {
15 	struct fbnic_dev *fbd = fbn->fbd;
16 	int err;
17 
18 	err = fbnic_alloc_napi_vectors(fbn);
19 	if (err)
20 		return err;
21 
22 	err = fbnic_alloc_resources(fbn);
23 	if (err)
24 		goto free_napi_vectors;
25 
26 	err = fbnic_set_netif_queues(fbn);
27 	if (err)
28 		goto free_resources;
29 
30 	/* Send ownership message and flush to verify FW has seen it */
31 	err = fbnic_fw_xmit_ownership_msg(fbd, true);
32 	if (err) {
33 		dev_warn(fbd->dev,
34 			 "Error %d sending host ownership message to the firmware\n",
35 			 err);
36 		goto err_reset_queues;
37 	}
38 
39 	err = fbnic_time_start(fbn);
40 	if (err)
41 		goto release_ownership;
42 
43 	err = fbnic_fw_init_heartbeat(fbd, false);
44 	if (err)
45 		goto time_stop;
46 
47 	err = fbnic_pcs_request_irq(fbd);
48 	if (err)
49 		goto time_stop;
50 
51 	/* Pull the BMC config and initialize the RPC */
52 	fbnic_bmc_rpc_init(fbd);
53 	fbnic_rss_reinit(fbd, fbn);
54 
55 	return 0;
56 time_stop:
57 	fbnic_time_stop(fbn);
58 release_ownership:
59 	fbnic_fw_xmit_ownership_msg(fbn->fbd, false);
60 err_reset_queues:
61 	fbnic_reset_netif_queues(fbn);
62 free_resources:
63 	fbnic_free_resources(fbn);
64 free_napi_vectors:
65 	fbnic_free_napi_vectors(fbn);
66 	return err;
67 }
68 
69 static int fbnic_open(struct net_device *netdev)
70 {
71 	struct fbnic_net *fbn = netdev_priv(netdev);
72 	int err;
73 
74 	fbnic_napi_name_irqs(fbn->fbd);
75 
76 	err = __fbnic_open(fbn);
77 	if (!err)
78 		fbnic_up(fbn);
79 
80 	return err;
81 }
82 
83 static int fbnic_stop(struct net_device *netdev)
84 {
85 	struct fbnic_net *fbn = netdev_priv(netdev);
86 
87 	fbnic_down(fbn);
88 	fbnic_pcs_free_irq(fbn->fbd);
89 
90 	fbnic_time_stop(fbn);
91 	fbnic_fw_xmit_ownership_msg(fbn->fbd, false);
92 
93 	fbnic_reset_netif_queues(fbn);
94 	fbnic_free_resources(fbn);
95 	fbnic_free_napi_vectors(fbn);
96 
97 	return 0;
98 }
99 
100 static int fbnic_uc_sync(struct net_device *netdev, const unsigned char *addr)
101 {
102 	struct fbnic_net *fbn = netdev_priv(netdev);
103 	struct fbnic_mac_addr *avail_addr;
104 
105 	if (WARN_ON(!is_valid_ether_addr(addr)))
106 		return -EADDRNOTAVAIL;
107 
108 	avail_addr = __fbnic_uc_sync(fbn->fbd, addr);
109 	if (!avail_addr)
110 		return -ENOSPC;
111 
112 	/* Add type flag indicating this address is in use by the host */
113 	set_bit(FBNIC_MAC_ADDR_T_UNICAST, avail_addr->act_tcam);
114 
115 	return 0;
116 }
117 
118 static int fbnic_uc_unsync(struct net_device *netdev, const unsigned char *addr)
119 {
120 	struct fbnic_net *fbn = netdev_priv(netdev);
121 	struct fbnic_dev *fbd = fbn->fbd;
122 	int i, ret;
123 
124 	/* Scan from middle of list to bottom, filling bottom up.
125 	 * Skip the first entry which is reserved for dev_addr and
126 	 * leave the last entry to use for promiscuous filtering.
127 	 */
128 	for (i = fbd->mac_addr_boundary, ret = -ENOENT;
129 	     i < FBNIC_RPC_TCAM_MACDA_HOST_ADDR_IDX && ret; i++) {
130 		struct fbnic_mac_addr *mac_addr = &fbd->mac_addr[i];
131 
132 		if (!ether_addr_equal(mac_addr->value.addr8, addr))
133 			continue;
134 
135 		ret = __fbnic_uc_unsync(mac_addr);
136 	}
137 
138 	return ret;
139 }
140 
141 static int fbnic_mc_sync(struct net_device *netdev, const unsigned char *addr)
142 {
143 	struct fbnic_net *fbn = netdev_priv(netdev);
144 	struct fbnic_mac_addr *avail_addr;
145 
146 	if (WARN_ON(!is_multicast_ether_addr(addr)))
147 		return -EADDRNOTAVAIL;
148 
149 	avail_addr = __fbnic_mc_sync(fbn->fbd, addr);
150 	if (!avail_addr)
151 		return -ENOSPC;
152 
153 	/* Add type flag indicating this address is in use by the host */
154 	set_bit(FBNIC_MAC_ADDR_T_MULTICAST, avail_addr->act_tcam);
155 
156 	return 0;
157 }
158 
159 static int fbnic_mc_unsync(struct net_device *netdev, const unsigned char *addr)
160 {
161 	struct fbnic_net *fbn = netdev_priv(netdev);
162 	struct fbnic_dev *fbd = fbn->fbd;
163 	int i, ret;
164 
165 	/* Scan from middle of list to top, filling top down.
166 	 * Skip over the address reserved for the BMC MAC and
167 	 * exclude index 0 as that belongs to the broadcast address
168 	 */
169 	for (i = fbd->mac_addr_boundary, ret = -ENOENT;
170 	     --i > FBNIC_RPC_TCAM_MACDA_BROADCAST_IDX && ret;) {
171 		struct fbnic_mac_addr *mac_addr = &fbd->mac_addr[i];
172 
173 		if (!ether_addr_equal(mac_addr->value.addr8, addr))
174 			continue;
175 
176 		ret = __fbnic_mc_unsync(mac_addr);
177 	}
178 
179 	return ret;
180 }
181 
182 void __fbnic_set_rx_mode(struct net_device *netdev)
183 {
184 	struct fbnic_net *fbn = netdev_priv(netdev);
185 	bool uc_promisc = false, mc_promisc = false;
186 	struct fbnic_dev *fbd = fbn->fbd;
187 	struct fbnic_mac_addr *mac_addr;
188 	int err;
189 
190 	/* Populate host address from dev_addr */
191 	mac_addr = &fbd->mac_addr[FBNIC_RPC_TCAM_MACDA_HOST_ADDR_IDX];
192 	if (!ether_addr_equal(mac_addr->value.addr8, netdev->dev_addr) ||
193 	    mac_addr->state != FBNIC_TCAM_S_VALID) {
194 		ether_addr_copy(mac_addr->value.addr8, netdev->dev_addr);
195 		mac_addr->state = FBNIC_TCAM_S_UPDATE;
196 		set_bit(FBNIC_MAC_ADDR_T_UNICAST, mac_addr->act_tcam);
197 	}
198 
199 	/* Populate broadcast address if broadcast is enabled */
200 	mac_addr = &fbd->mac_addr[FBNIC_RPC_TCAM_MACDA_BROADCAST_IDX];
201 	if (netdev->flags & IFF_BROADCAST) {
202 		if (!is_broadcast_ether_addr(mac_addr->value.addr8) ||
203 		    mac_addr->state != FBNIC_TCAM_S_VALID) {
204 			eth_broadcast_addr(mac_addr->value.addr8);
205 			mac_addr->state = FBNIC_TCAM_S_ADD;
206 		}
207 		set_bit(FBNIC_MAC_ADDR_T_BROADCAST, mac_addr->act_tcam);
208 	} else if (mac_addr->state == FBNIC_TCAM_S_VALID) {
209 		__fbnic_xc_unsync(mac_addr, FBNIC_MAC_ADDR_T_BROADCAST);
210 	}
211 
212 	/* Synchronize unicast and multicast address lists */
213 	err = __dev_uc_sync(netdev, fbnic_uc_sync, fbnic_uc_unsync);
214 	if (err == -ENOSPC)
215 		uc_promisc = true;
216 	err = __dev_mc_sync(netdev, fbnic_mc_sync, fbnic_mc_unsync);
217 	if (err == -ENOSPC)
218 		mc_promisc = true;
219 
220 	uc_promisc |= !!(netdev->flags & IFF_PROMISC);
221 	mc_promisc |= !!(netdev->flags & IFF_ALLMULTI) || uc_promisc;
222 
223 	/* Populate last TCAM entry with promiscuous entry and 0/1 bit mask */
224 	mac_addr = &fbd->mac_addr[FBNIC_RPC_TCAM_MACDA_PROMISC_IDX];
225 	if (uc_promisc) {
226 		if (!is_zero_ether_addr(mac_addr->value.addr8) ||
227 		    mac_addr->state != FBNIC_TCAM_S_VALID) {
228 			eth_zero_addr(mac_addr->value.addr8);
229 			eth_broadcast_addr(mac_addr->mask.addr8);
230 			clear_bit(FBNIC_MAC_ADDR_T_ALLMULTI,
231 				  mac_addr->act_tcam);
232 			set_bit(FBNIC_MAC_ADDR_T_PROMISC,
233 				mac_addr->act_tcam);
234 			mac_addr->state = FBNIC_TCAM_S_ADD;
235 		}
236 	} else if (mc_promisc &&
237 		   (!fbnic_bmc_present(fbd) || !fbd->fw_cap.all_multi)) {
238 		/* We have to add a special handler for multicast as the
239 		 * BMC may have an all-multi rule already in place. As such
240 		 * adding a rule ourselves won't do any good so we will have
241 		 * to modify the rules for the ALL MULTI below if the BMC
242 		 * already has the rule in place.
243 		 */
244 		if (!is_multicast_ether_addr(mac_addr->value.addr8) ||
245 		    mac_addr->state != FBNIC_TCAM_S_VALID) {
246 			eth_zero_addr(mac_addr->value.addr8);
247 			eth_broadcast_addr(mac_addr->mask.addr8);
248 			mac_addr->value.addr8[0] ^= 1;
249 			mac_addr->mask.addr8[0] ^= 1;
250 			set_bit(FBNIC_MAC_ADDR_T_ALLMULTI,
251 				mac_addr->act_tcam);
252 			clear_bit(FBNIC_MAC_ADDR_T_PROMISC,
253 				  mac_addr->act_tcam);
254 			mac_addr->state = FBNIC_TCAM_S_ADD;
255 		}
256 	} else if (mac_addr->state == FBNIC_TCAM_S_VALID) {
257 		if (test_bit(FBNIC_MAC_ADDR_T_BMC, mac_addr->act_tcam)) {
258 			clear_bit(FBNIC_MAC_ADDR_T_ALLMULTI,
259 				  mac_addr->act_tcam);
260 			clear_bit(FBNIC_MAC_ADDR_T_PROMISC,
261 				  mac_addr->act_tcam);
262 		} else {
263 			mac_addr->state = FBNIC_TCAM_S_DELETE;
264 		}
265 	}
266 
267 	/* Add rules for BMC all multicast if it is enabled */
268 	fbnic_bmc_rpc_all_multi_config(fbd, mc_promisc);
269 
270 	/* Sift out any unshared BMC rules and place them in BMC only section */
271 	fbnic_sift_macda(fbd);
272 
273 	/* Write updates to hardware */
274 	fbnic_write_rules(fbd);
275 	fbnic_write_macda(fbd);
276 	fbnic_write_tce_tcam(fbd);
277 }
278 
279 static void fbnic_set_rx_mode(struct net_device *netdev)
280 {
281 	/* No need to update the hardware if we are not running */
282 	if (netif_running(netdev))
283 		__fbnic_set_rx_mode(netdev);
284 }
285 
286 static int fbnic_set_mac(struct net_device *netdev, void *p)
287 {
288 	struct sockaddr *addr = p;
289 
290 	if (!is_valid_ether_addr(addr->sa_data))
291 		return -EADDRNOTAVAIL;
292 
293 	eth_hw_addr_set(netdev, addr->sa_data);
294 
295 	fbnic_set_rx_mode(netdev);
296 
297 	return 0;
298 }
299 
300 void fbnic_clear_rx_mode(struct net_device *netdev)
301 {
302 	struct fbnic_net *fbn = netdev_priv(netdev);
303 	struct fbnic_dev *fbd = fbn->fbd;
304 	int idx;
305 
306 	for (idx = ARRAY_SIZE(fbd->mac_addr); idx--;) {
307 		struct fbnic_mac_addr *mac_addr = &fbd->mac_addr[idx];
308 
309 		if (mac_addr->state != FBNIC_TCAM_S_VALID)
310 			continue;
311 
312 		bitmap_clear(mac_addr->act_tcam,
313 			     FBNIC_MAC_ADDR_T_HOST_START,
314 			     FBNIC_MAC_ADDR_T_HOST_LEN);
315 
316 		if (bitmap_empty(mac_addr->act_tcam,
317 				 FBNIC_RPC_TCAM_ACT_NUM_ENTRIES))
318 			mac_addr->state = FBNIC_TCAM_S_DELETE;
319 	}
320 
321 	/* Write updates to hardware */
322 	fbnic_write_macda(fbd);
323 
324 	__dev_uc_unsync(netdev, NULL);
325 	__dev_mc_unsync(netdev, NULL);
326 }
327 
328 static int fbnic_hwtstamp_get(struct net_device *netdev,
329 			      struct kernel_hwtstamp_config *config)
330 {
331 	struct fbnic_net *fbn = netdev_priv(netdev);
332 
333 	*config = fbn->hwtstamp_config;
334 
335 	return 0;
336 }
337 
338 static int fbnic_hwtstamp_set(struct net_device *netdev,
339 			      struct kernel_hwtstamp_config *config,
340 			      struct netlink_ext_ack *extack)
341 {
342 	struct fbnic_net *fbn = netdev_priv(netdev);
343 	int old_rx_filter;
344 
345 	if (config->source != HWTSTAMP_SOURCE_NETDEV)
346 		return -EOPNOTSUPP;
347 
348 	if (!kernel_hwtstamp_config_changed(config, &fbn->hwtstamp_config))
349 		return 0;
350 
351 	/* Upscale the filters */
352 	switch (config->rx_filter) {
353 	case HWTSTAMP_FILTER_NONE:
354 	case HWTSTAMP_FILTER_ALL:
355 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
356 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
357 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
358 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
359 		break;
360 	case HWTSTAMP_FILTER_NTP_ALL:
361 		config->rx_filter = HWTSTAMP_FILTER_ALL;
362 		break;
363 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
364 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
365 		config->rx_filter = HWTSTAMP_FILTER_PTP_V1_L4_EVENT;
366 		break;
367 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
368 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
369 		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_L4_EVENT;
370 		break;
371 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
372 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
373 		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_L2_EVENT;
374 		break;
375 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
376 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
377 		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
378 		break;
379 	default:
380 		return -ERANGE;
381 	}
382 
383 	/* Configure */
384 	old_rx_filter = fbn->hwtstamp_config.rx_filter;
385 	memcpy(&fbn->hwtstamp_config, config, sizeof(*config));
386 
387 	if (old_rx_filter != config->rx_filter && netif_running(fbn->netdev)) {
388 		fbnic_rss_reinit(fbn->fbd, fbn);
389 		fbnic_write_rules(fbn->fbd);
390 	}
391 
392 	/* Save / report back filter configuration
393 	 * Note that our filter configuration is inexact. Instead of
394 	 * filtering for a specific UDP port or L2 Ethertype we are
395 	 * filtering in all UDP or all non-IP packets for timestamping. So
396 	 * if anything other than FILTER_ALL is requested we report
397 	 * FILTER_SOME indicating that we will be timestamping a few
398 	 * additional packets.
399 	 */
400 	if (config->rx_filter > HWTSTAMP_FILTER_ALL)
401 		config->rx_filter = HWTSTAMP_FILTER_SOME;
402 
403 	return 0;
404 }
405 
406 static void fbnic_get_stats64(struct net_device *dev,
407 			      struct rtnl_link_stats64 *stats64)
408 {
409 	u64 rx_bytes, rx_packets, rx_dropped = 0, rx_errors = 0;
410 	u64 rx_over = 0, rx_missed = 0, rx_length = 0;
411 	u64 tx_bytes, tx_packets, tx_dropped = 0;
412 	struct fbnic_net *fbn = netdev_priv(dev);
413 	struct fbnic_dev *fbd = fbn->fbd;
414 	struct fbnic_queue_stats *stats;
415 
416 	unsigned int start, i;
417 
418 	fbnic_get_hw_stats(fbd);
419 
420 	stats = &fbn->tx_stats;
421 
422 	tx_bytes = stats->bytes;
423 	tx_packets = stats->packets;
424 	tx_dropped = stats->dropped;
425 
426 	/* Record drops from Tx HW Datapath */
427 	spin_lock(&fbd->hw_stats_lock);
428 	tx_dropped += fbd->hw_stats.tmi.drop.frames.value +
429 		      fbd->hw_stats.tti.cm_drop.frames.value +
430 		      fbd->hw_stats.tti.frame_drop.frames.value +
431 		      fbd->hw_stats.tti.tbi_drop.frames.value;
432 	spin_unlock(&fbd->hw_stats_lock);
433 
434 	stats64->tx_bytes = tx_bytes;
435 	stats64->tx_packets = tx_packets;
436 	stats64->tx_dropped = tx_dropped;
437 
438 	for (i = 0; i < fbn->num_tx_queues; i++) {
439 		struct fbnic_ring *txr = fbn->tx[i];
440 
441 		if (!txr)
442 			continue;
443 
444 		stats = &txr->stats;
445 		do {
446 			start = u64_stats_fetch_begin(&stats->syncp);
447 			tx_bytes = stats->bytes;
448 			tx_packets = stats->packets;
449 			tx_dropped = stats->dropped;
450 		} while (u64_stats_fetch_retry(&stats->syncp, start));
451 
452 		stats64->tx_bytes += tx_bytes;
453 		stats64->tx_packets += tx_packets;
454 		stats64->tx_dropped += tx_dropped;
455 	}
456 
457 	stats = &fbn->rx_stats;
458 
459 	rx_bytes = stats->bytes;
460 	rx_packets = stats->packets;
461 	rx_dropped = stats->dropped;
462 
463 	spin_lock(&fbd->hw_stats_lock);
464 	/* Record drops for the host FIFOs.
465 	 * 4: network to Host,	6: BMC to Host
466 	 * Exclude the BMC and MC FIFOs as those stats may contain drops
467 	 * due to unrelated items such as TCAM misses. They are still
468 	 * accessible through the ethtool stats.
469 	 */
470 	i = FBNIC_RXB_FIFO_HOST;
471 	rx_missed += fbd->hw_stats.rxb.fifo[i].drop.frames.value;
472 	i = FBNIC_RXB_FIFO_BMC_TO_HOST;
473 	rx_missed += fbd->hw_stats.rxb.fifo[i].drop.frames.value;
474 
475 	for (i = 0; i < fbd->max_num_queues; i++) {
476 		/* Report packets dropped due to CQ/BDQ being full/empty */
477 		rx_over += fbd->hw_stats.hw_q[i].rde_pkt_cq_drop.value;
478 		rx_over += fbd->hw_stats.hw_q[i].rde_pkt_bdq_drop.value;
479 
480 		/* Report packets with errors */
481 		rx_errors += fbd->hw_stats.hw_q[i].rde_pkt_err.value;
482 	}
483 	spin_unlock(&fbd->hw_stats_lock);
484 
485 	stats64->rx_bytes = rx_bytes;
486 	stats64->rx_packets = rx_packets;
487 	stats64->rx_dropped = rx_dropped;
488 	stats64->rx_over_errors = rx_over;
489 	stats64->rx_errors = rx_errors;
490 	stats64->rx_missed_errors = rx_missed;
491 
492 	for (i = 0; i < fbn->num_rx_queues; i++) {
493 		struct fbnic_ring *xdpr = fbn->tx[FBNIC_MAX_TXQS + i];
494 		struct fbnic_ring *rxr = fbn->rx[i];
495 
496 		if (!rxr)
497 			continue;
498 
499 		stats = &rxr->stats;
500 		do {
501 			start = u64_stats_fetch_begin(&stats->syncp);
502 			rx_bytes = stats->bytes;
503 			rx_packets = stats->packets;
504 			rx_dropped = stats->dropped;
505 			rx_length = stats->rx.length_errors;
506 		} while (u64_stats_fetch_retry(&stats->syncp, start));
507 
508 		stats64->rx_bytes += rx_bytes;
509 		stats64->rx_packets += rx_packets;
510 		stats64->rx_dropped += rx_dropped;
511 		stats64->rx_errors += rx_length;
512 		stats64->rx_length_errors += rx_length;
513 
514 		if (!xdpr)
515 			continue;
516 
517 		stats = &xdpr->stats;
518 		do {
519 			start = u64_stats_fetch_begin(&stats->syncp);
520 			tx_bytes = stats->bytes;
521 			tx_packets = stats->packets;
522 			tx_dropped = stats->dropped;
523 		} while (u64_stats_fetch_retry(&stats->syncp, start));
524 
525 		stats64->tx_bytes += tx_bytes;
526 		stats64->tx_packets += tx_packets;
527 		stats64->tx_dropped += tx_dropped;
528 	}
529 }
530 
531 bool fbnic_check_split_frames(struct bpf_prog *prog, unsigned int mtu,
532 			      u32 hds_thresh)
533 {
534 	if (!prog)
535 		return false;
536 
537 	if (prog->aux->xdp_has_frags)
538 		return false;
539 
540 	return mtu + ETH_HLEN > hds_thresh;
541 }
542 
543 static int fbnic_bpf(struct net_device *netdev, struct netdev_bpf *bpf)
544 {
545 	struct bpf_prog *prog = bpf->prog, *prev_prog;
546 	struct fbnic_net *fbn = netdev_priv(netdev);
547 
548 	if (bpf->command != XDP_SETUP_PROG)
549 		return -EINVAL;
550 
551 	if (fbnic_check_split_frames(prog, netdev->mtu,
552 				     fbn->hds_thresh)) {
553 		NL_SET_ERR_MSG_MOD(bpf->extack,
554 				   "MTU too high, or HDS threshold is too low for single buffer XDP");
555 		return -EOPNOTSUPP;
556 	}
557 
558 	prev_prog = xchg(&fbn->xdp_prog, prog);
559 	if (prev_prog)
560 		bpf_prog_put(prev_prog);
561 
562 	return 0;
563 }
564 
565 static const struct net_device_ops fbnic_netdev_ops = {
566 	.ndo_open		= fbnic_open,
567 	.ndo_stop		= fbnic_stop,
568 	.ndo_validate_addr	= eth_validate_addr,
569 	.ndo_start_xmit		= fbnic_xmit_frame,
570 	.ndo_features_check	= fbnic_features_check,
571 	.ndo_set_mac_address	= fbnic_set_mac,
572 	.ndo_set_rx_mode	= fbnic_set_rx_mode,
573 	.ndo_get_stats64	= fbnic_get_stats64,
574 	.ndo_bpf		= fbnic_bpf,
575 	.ndo_hwtstamp_get	= fbnic_hwtstamp_get,
576 	.ndo_hwtstamp_set	= fbnic_hwtstamp_set,
577 };
578 
579 static void fbnic_get_queue_stats_rx(struct net_device *dev, int idx,
580 				     struct netdev_queue_stats_rx *rx)
581 {
582 	struct fbnic_net *fbn = netdev_priv(dev);
583 	struct fbnic_ring *rxr = fbn->rx[idx];
584 	struct fbnic_dev *fbd = fbn->fbd;
585 	struct fbnic_queue_stats *stats;
586 	u64 bytes, packets, alloc_fail;
587 	u64 csum_complete, csum_none;
588 	unsigned int start;
589 
590 	if (!rxr)
591 		return;
592 
593 	stats = &rxr->stats;
594 	do {
595 		start = u64_stats_fetch_begin(&stats->syncp);
596 		bytes = stats->bytes;
597 		packets = stats->packets;
598 		alloc_fail = stats->rx.alloc_failed;
599 		csum_complete = stats->rx.csum_complete;
600 		csum_none = stats->rx.csum_none;
601 	} while (u64_stats_fetch_retry(&stats->syncp, start));
602 
603 	rx->bytes = bytes;
604 	rx->packets = packets;
605 	rx->alloc_fail = alloc_fail;
606 	rx->csum_complete = csum_complete;
607 	rx->csum_none = csum_none;
608 
609 	fbnic_get_hw_q_stats(fbd, fbd->hw_stats.hw_q);
610 
611 	spin_lock(&fbd->hw_stats_lock);
612 	rx->hw_drop_overruns = fbd->hw_stats.hw_q[idx].rde_pkt_cq_drop.value +
613 			       fbd->hw_stats.hw_q[idx].rde_pkt_bdq_drop.value;
614 	rx->hw_drops = fbd->hw_stats.hw_q[idx].rde_pkt_err.value +
615 		       rx->hw_drop_overruns;
616 	spin_unlock(&fbd->hw_stats_lock);
617 }
618 
619 static void fbnic_get_queue_stats_tx(struct net_device *dev, int idx,
620 				     struct netdev_queue_stats_tx *tx)
621 {
622 	struct fbnic_net *fbn = netdev_priv(dev);
623 	struct fbnic_ring *txr = fbn->tx[idx];
624 	struct fbnic_queue_stats *stats;
625 	u64 stop, wake, csum, lso;
626 	struct fbnic_ring *xdpr;
627 	unsigned int start;
628 	u64 bytes, packets;
629 
630 	if (!txr)
631 		return;
632 
633 	stats = &txr->stats;
634 	do {
635 		start = u64_stats_fetch_begin(&stats->syncp);
636 		bytes = stats->bytes;
637 		packets = stats->packets;
638 		csum = stats->twq.csum_partial;
639 		lso = stats->twq.lso;
640 		stop = stats->twq.stop;
641 		wake = stats->twq.wake;
642 	} while (u64_stats_fetch_retry(&stats->syncp, start));
643 
644 	tx->bytes = bytes;
645 	tx->packets = packets;
646 	tx->needs_csum = csum + lso;
647 	tx->hw_gso_wire_packets = lso;
648 	tx->stop = stop;
649 	tx->wake = wake;
650 
651 	xdpr = fbn->tx[FBNIC_MAX_TXQS + idx];
652 	if (xdpr) {
653 		stats = &xdpr->stats;
654 		do {
655 			start = u64_stats_fetch_begin(&stats->syncp);
656 			bytes = stats->bytes;
657 			packets = stats->packets;
658 		} while (u64_stats_fetch_retry(&stats->syncp, start));
659 
660 		tx->bytes += bytes;
661 		tx->packets += packets;
662 	}
663 }
664 
665 static void fbnic_get_base_stats(struct net_device *dev,
666 				 struct netdev_queue_stats_rx *rx,
667 				 struct netdev_queue_stats_tx *tx)
668 {
669 	struct fbnic_net *fbn = netdev_priv(dev);
670 
671 	tx->bytes = fbn->tx_stats.bytes;
672 	tx->packets = fbn->tx_stats.packets;
673 	tx->needs_csum = fbn->tx_stats.twq.csum_partial + fbn->tx_stats.twq.lso;
674 	tx->hw_gso_wire_packets = fbn->tx_stats.twq.lso;
675 	tx->stop = fbn->tx_stats.twq.stop;
676 	tx->wake = fbn->tx_stats.twq.wake;
677 
678 	rx->bytes = fbn->rx_stats.bytes;
679 	rx->packets = fbn->rx_stats.packets;
680 	rx->alloc_fail = fbn->rx_stats.rx.alloc_failed;
681 	rx->csum_complete = fbn->rx_stats.rx.csum_complete;
682 	rx->csum_none = fbn->rx_stats.rx.csum_none;
683 }
684 
685 static const struct netdev_stat_ops fbnic_stat_ops = {
686 	.get_queue_stats_rx	= fbnic_get_queue_stats_rx,
687 	.get_queue_stats_tx	= fbnic_get_queue_stats_tx,
688 	.get_base_stats		= fbnic_get_base_stats,
689 };
690 
691 void fbnic_reset_queues(struct fbnic_net *fbn,
692 			unsigned int tx, unsigned int rx)
693 {
694 	struct fbnic_dev *fbd = fbn->fbd;
695 	unsigned int max_napis;
696 
697 	max_napis = fbd->num_irqs - FBNIC_NON_NAPI_VECTORS;
698 
699 	tx = min(tx, max_napis);
700 	fbn->num_tx_queues = tx;
701 
702 	rx = min(rx, max_napis);
703 	fbn->num_rx_queues = rx;
704 
705 	fbn->num_napi = max(tx, rx);
706 }
707 
708 /**
709  * fbnic_netdev_free - Free the netdev associate with fbnic
710  * @fbd: Driver specific structure to free netdev from
711  *
712  * Allocate and initialize the netdev and netdev private structure. Bind
713  * together the hardware, netdev, and pci data structures.
714  **/
715 void fbnic_netdev_free(struct fbnic_dev *fbd)
716 {
717 	struct fbnic_net *fbn = netdev_priv(fbd->netdev);
718 
719 	if (fbn->phylink)
720 		phylink_destroy(fbn->phylink);
721 
722 	free_netdev(fbd->netdev);
723 	fbd->netdev = NULL;
724 }
725 
726 /**
727  * fbnic_netdev_alloc - Allocate a netdev and associate with fbnic
728  * @fbd: Driver specific structure to associate netdev with
729  *
730  * Allocate and initialize the netdev and netdev private structure. Bind
731  * together the hardware, netdev, and pci data structures.
732  *
733  *  Return: Pointer to net_device on success, NULL on failure
734  **/
735 struct net_device *fbnic_netdev_alloc(struct fbnic_dev *fbd)
736 {
737 	struct net_device *netdev;
738 	struct fbnic_net *fbn;
739 	int default_queues;
740 
741 	netdev = alloc_etherdev_mq(sizeof(*fbn), FBNIC_MAX_RXQS);
742 	if (!netdev)
743 		return NULL;
744 
745 	SET_NETDEV_DEV(netdev, fbd->dev);
746 	fbd->netdev = netdev;
747 
748 	netdev->netdev_ops = &fbnic_netdev_ops;
749 	netdev->stat_ops = &fbnic_stat_ops;
750 
751 	fbnic_set_ethtool_ops(netdev);
752 
753 	fbn = netdev_priv(netdev);
754 
755 	fbn->netdev = netdev;
756 	fbn->fbd = fbd;
757 
758 	fbn->txq_size = FBNIC_TXQ_SIZE_DEFAULT;
759 	fbn->hpq_size = FBNIC_HPQ_SIZE_DEFAULT;
760 	fbn->ppq_size = FBNIC_PPQ_SIZE_DEFAULT;
761 	fbn->rcq_size = FBNIC_RCQ_SIZE_DEFAULT;
762 
763 	fbn->tx_usecs = FBNIC_TX_USECS_DEFAULT;
764 	fbn->rx_usecs = FBNIC_RX_USECS_DEFAULT;
765 	fbn->rx_max_frames = FBNIC_RX_FRAMES_DEFAULT;
766 
767 	/* Initialize the hds_thresh */
768 	netdev->cfg->hds_thresh = FBNIC_HDS_THRESH_DEFAULT;
769 	fbn->hds_thresh = FBNIC_HDS_THRESH_DEFAULT;
770 
771 	default_queues = netif_get_num_default_rss_queues();
772 	if (default_queues > fbd->max_num_queues)
773 		default_queues = fbd->max_num_queues;
774 
775 	fbnic_reset_queues(fbn, default_queues, default_queues);
776 
777 	fbnic_reset_indir_tbl(fbn);
778 	fbnic_rss_key_fill(fbn->rss_key);
779 	fbnic_rss_init_en_mask(fbn);
780 
781 	netdev->priv_flags |= IFF_UNICAST_FLT;
782 
783 	netdev->gso_partial_features =
784 		NETIF_F_GSO_GRE |
785 		NETIF_F_GSO_GRE_CSUM |
786 		NETIF_F_GSO_IPXIP4 |
787 		NETIF_F_GSO_UDP_TUNNEL |
788 		NETIF_F_GSO_UDP_TUNNEL_CSUM;
789 
790 	netdev->features |=
791 		netdev->gso_partial_features |
792 		FBNIC_TUN_GSO_FEATURES |
793 		NETIF_F_RXHASH |
794 		NETIF_F_SG |
795 		NETIF_F_HW_CSUM |
796 		NETIF_F_RXCSUM |
797 		NETIF_F_TSO |
798 		NETIF_F_TSO_ECN |
799 		NETIF_F_TSO6 |
800 		NETIF_F_GSO_PARTIAL |
801 		NETIF_F_GSO_UDP_L4;
802 
803 	netdev->hw_features |= netdev->features;
804 	netdev->vlan_features |= netdev->features;
805 	netdev->hw_enc_features |= netdev->features;
806 	netdev->features |= NETIF_F_NTUPLE;
807 
808 	netdev->min_mtu = IPV6_MIN_MTU;
809 	netdev->max_mtu = FBNIC_MAX_JUMBO_FRAME_SIZE - ETH_HLEN;
810 
811 	/* TBD: This is workaround for BMC as phylink doesn't have support
812 	 * for leavling the link enabled if a BMC is present.
813 	 */
814 	netdev->ethtool->wol_enabled = true;
815 
816 	netif_carrier_off(netdev);
817 
818 	netif_tx_stop_all_queues(netdev);
819 
820 	if (fbnic_phylink_init(netdev)) {
821 		fbnic_netdev_free(fbd);
822 		return NULL;
823 	}
824 
825 	return netdev;
826 }
827 
828 static int fbnic_dsn_to_mac_addr(u64 dsn, char *addr)
829 {
830 	addr[0] = (dsn >> 56) & 0xFF;
831 	addr[1] = (dsn >> 48) & 0xFF;
832 	addr[2] = (dsn >> 40) & 0xFF;
833 	addr[3] = (dsn >> 16) & 0xFF;
834 	addr[4] = (dsn >> 8) & 0xFF;
835 	addr[5] = dsn & 0xFF;
836 
837 	return is_valid_ether_addr(addr) ? 0 : -EINVAL;
838 }
839 
840 /**
841  * fbnic_netdev_register - Initialize general software structures
842  * @netdev: Netdev containing structure to initialize and register
843  *
844  * Initialize the MAC address for the netdev and register it.
845  *
846  *  Return: 0 on success, negative on failure
847  **/
848 int fbnic_netdev_register(struct net_device *netdev)
849 {
850 	struct fbnic_net *fbn = netdev_priv(netdev);
851 	struct fbnic_dev *fbd = fbn->fbd;
852 	u64 dsn = fbd->dsn;
853 	u8 addr[ETH_ALEN];
854 	int err;
855 
856 	err = fbnic_dsn_to_mac_addr(dsn, addr);
857 	if (!err) {
858 		ether_addr_copy(netdev->perm_addr, addr);
859 		eth_hw_addr_set(netdev, addr);
860 	} else {
861 		/* A randomly assigned MAC address will cause provisioning
862 		 * issues so instead just fail to spawn the netdev and
863 		 * avoid any confusion.
864 		 */
865 		dev_err(fbd->dev, "MAC addr %pM invalid\n", addr);
866 		return err;
867 	}
868 
869 	return register_netdev(netdev);
870 }
871 
872 void fbnic_netdev_unregister(struct net_device *netdev)
873 {
874 	unregister_netdev(netdev);
875 }
876