xref: /linux/drivers/net/ethernet/meta/fbnic/fbnic_netdev.c (revision 90e63d5354951d37fa2b3b91e6f17b95d2bf9bee)
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_mac_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 	phylink_resume(fbn->phylink);
56 
57 	return 0;
58 time_stop:
59 	fbnic_time_stop(fbn);
60 release_ownership:
61 	fbnic_fw_xmit_ownership_msg(fbn->fbd, false);
62 err_reset_queues:
63 	fbnic_reset_netif_queues(fbn);
64 free_resources:
65 	fbnic_free_resources(fbn);
66 free_napi_vectors:
67 	fbnic_free_napi_vectors(fbn);
68 	return err;
69 }
70 
71 static int fbnic_open(struct net_device *netdev)
72 {
73 	struct fbnic_net *fbn = netdev_priv(netdev);
74 	int err;
75 
76 	fbnic_napi_name_irqs(fbn->fbd);
77 
78 	err = __fbnic_open(fbn);
79 	if (!err)
80 		fbnic_up(fbn);
81 
82 	return err;
83 }
84 
85 static int fbnic_stop(struct net_device *netdev)
86 {
87 	struct fbnic_net *fbn = netdev_priv(netdev);
88 
89 	fbnic_mac_free_irq(fbn->fbd);
90 	phylink_suspend(fbn->phylink, fbnic_bmc_present(fbn->fbd));
91 
92 	fbnic_down(fbn);
93 
94 	fbnic_time_stop(fbn);
95 	fbnic_fw_xmit_ownership_msg(fbn->fbd, false);
96 
97 	fbnic_reset_netif_queues(fbn);
98 	fbnic_free_resources(fbn);
99 	fbnic_free_napi_vectors(fbn);
100 
101 	return 0;
102 }
103 
104 static int fbnic_uc_sync(struct net_device *netdev, const unsigned char *addr)
105 {
106 	struct fbnic_net *fbn = netdev_priv(netdev);
107 	struct fbnic_mac_addr *avail_addr;
108 
109 	if (WARN_ON(!is_valid_ether_addr(addr)))
110 		return -EADDRNOTAVAIL;
111 
112 	avail_addr = __fbnic_uc_sync(fbn->fbd, addr);
113 	if (!avail_addr)
114 		return -ENOSPC;
115 
116 	/* Add type flag indicating this address is in use by the host */
117 	set_bit(FBNIC_MAC_ADDR_T_UNICAST, avail_addr->act_tcam);
118 
119 	return 0;
120 }
121 
122 static int fbnic_uc_unsync(struct net_device *netdev, const unsigned char *addr)
123 {
124 	struct fbnic_net *fbn = netdev_priv(netdev);
125 	struct fbnic_dev *fbd = fbn->fbd;
126 	int i, ret;
127 
128 	/* Scan from middle of list to bottom, filling bottom up.
129 	 * Skip the first entry which is reserved for dev_addr and
130 	 * leave the last entry to use for promiscuous filtering.
131 	 */
132 	for (i = fbd->mac_addr_boundary, ret = -ENOENT;
133 	     i < FBNIC_RPC_TCAM_MACDA_HOST_ADDR_IDX && ret; i++) {
134 		struct fbnic_mac_addr *mac_addr = &fbd->mac_addr[i];
135 
136 		if (!ether_addr_equal(mac_addr->value.addr8, addr))
137 			continue;
138 
139 		ret = __fbnic_uc_unsync(mac_addr);
140 	}
141 
142 	return ret;
143 }
144 
145 static int fbnic_mc_sync(struct net_device *netdev, const unsigned char *addr)
146 {
147 	struct fbnic_net *fbn = netdev_priv(netdev);
148 	struct fbnic_mac_addr *avail_addr;
149 
150 	if (WARN_ON(!is_multicast_ether_addr(addr)))
151 		return -EADDRNOTAVAIL;
152 
153 	avail_addr = __fbnic_mc_sync(fbn->fbd, addr);
154 	if (!avail_addr)
155 		return -ENOSPC;
156 
157 	/* Add type flag indicating this address is in use by the host */
158 	set_bit(FBNIC_MAC_ADDR_T_MULTICAST, avail_addr->act_tcam);
159 
160 	return 0;
161 }
162 
163 static int fbnic_mc_unsync(struct net_device *netdev, const unsigned char *addr)
164 {
165 	struct fbnic_net *fbn = netdev_priv(netdev);
166 	struct fbnic_dev *fbd = fbn->fbd;
167 	int i, ret;
168 
169 	/* Scan from middle of list to top, filling top down.
170 	 * Skip over the address reserved for the BMC MAC and
171 	 * exclude index 0 as that belongs to the broadcast address
172 	 */
173 	for (i = fbd->mac_addr_boundary, ret = -ENOENT;
174 	     --i > FBNIC_RPC_TCAM_MACDA_BROADCAST_IDX && ret;) {
175 		struct fbnic_mac_addr *mac_addr = &fbd->mac_addr[i];
176 
177 		if (!ether_addr_equal(mac_addr->value.addr8, addr))
178 			continue;
179 
180 		ret = __fbnic_mc_unsync(mac_addr);
181 	}
182 
183 	return ret;
184 }
185 
186 void __fbnic_set_rx_mode(struct fbnic_dev *fbd,
187 			 struct netdev_hw_addr_list *uc,
188 			 struct netdev_hw_addr_list *mc)
189 {
190 	bool uc_promisc = false, mc_promisc = false;
191 	struct net_device *netdev = fbd->netdev;
192 	struct fbnic_mac_addr *mac_addr;
193 	int err;
194 
195 	/* Populate host address from dev_addr */
196 	mac_addr = &fbd->mac_addr[FBNIC_RPC_TCAM_MACDA_HOST_ADDR_IDX];
197 	if (!ether_addr_equal(mac_addr->value.addr8, netdev->dev_addr) ||
198 	    mac_addr->state != FBNIC_TCAM_S_VALID) {
199 		ether_addr_copy(mac_addr->value.addr8, netdev->dev_addr);
200 		mac_addr->state = FBNIC_TCAM_S_UPDATE;
201 		set_bit(FBNIC_MAC_ADDR_T_UNICAST, mac_addr->act_tcam);
202 	}
203 
204 	/* Populate broadcast address if broadcast is enabled */
205 	mac_addr = &fbd->mac_addr[FBNIC_RPC_TCAM_MACDA_BROADCAST_IDX];
206 	if (netdev->flags & IFF_BROADCAST) {
207 		if (!is_broadcast_ether_addr(mac_addr->value.addr8) ||
208 		    mac_addr->state != FBNIC_TCAM_S_VALID) {
209 			eth_broadcast_addr(mac_addr->value.addr8);
210 			mac_addr->state = FBNIC_TCAM_S_ADD;
211 		}
212 		set_bit(FBNIC_MAC_ADDR_T_BROADCAST, mac_addr->act_tcam);
213 	} else if (mac_addr->state == FBNIC_TCAM_S_VALID) {
214 		__fbnic_xc_unsync(mac_addr, FBNIC_MAC_ADDR_T_BROADCAST);
215 	}
216 
217 	/* Synchronize unicast and multicast address lists */
218 	err = __hw_addr_sync_dev(uc, netdev, fbnic_uc_sync, fbnic_uc_unsync);
219 	if (err == -ENOSPC)
220 		uc_promisc = true;
221 	err = __hw_addr_sync_dev(mc, netdev, fbnic_mc_sync, fbnic_mc_unsync);
222 	if (err == -ENOSPC)
223 		mc_promisc = true;
224 
225 	uc_promisc |= !!(netdev->flags & IFF_PROMISC);
226 	mc_promisc |= !!(netdev->flags & IFF_ALLMULTI) || uc_promisc;
227 
228 	/* Update the promiscuous rules */
229 	fbnic_promisc_sync(fbd, uc_promisc, mc_promisc);
230 
231 	/* Add rules for BMC all multicast if it is enabled */
232 	fbnic_bmc_rpc_all_multi_config(fbd, mc_promisc);
233 
234 	/* Sift out any unshared BMC rules and place them in BMC only section */
235 	fbnic_sift_macda(fbd);
236 
237 	/* Write updates to hardware */
238 	fbnic_write_rules(fbd);
239 	fbnic_write_macda(fbd);
240 	fbnic_write_tce_tcam(fbd);
241 }
242 
243 static int fbnic_set_rx_mode(struct net_device *netdev,
244 			     struct netdev_hw_addr_list *uc,
245 			     struct netdev_hw_addr_list *mc)
246 {
247 	struct fbnic_net *fbn = netdev_priv(netdev);
248 	struct fbnic_dev *fbd = fbn->fbd;
249 
250 	/* No need to update the hardware if we are not running */
251 	if (netif_running(netdev))
252 		__fbnic_set_rx_mode(fbd, uc, mc);
253 
254 	return 0;
255 }
256 
257 static int fbnic_set_mac(struct net_device *netdev, void *p)
258 {
259 	struct fbnic_net *fbn = netdev_priv(netdev);
260 	struct sockaddr *addr = p;
261 
262 	if (!is_valid_ether_addr(addr->sa_data))
263 		return -EADDRNOTAVAIL;
264 
265 	eth_hw_addr_set(netdev, addr->sa_data);
266 
267 	if (netif_running(netdev))
268 		__fbnic_set_rx_mode(fbn->fbd, &netdev->uc, &netdev->mc);
269 
270 	return 0;
271 }
272 
273 static int fbnic_change_mtu(struct net_device *dev, int new_mtu)
274 {
275 	struct fbnic_net *fbn = netdev_priv(dev);
276 
277 	if (fbnic_check_split_frames(fbn->xdp_prog, new_mtu, fbn->hds_thresh)) {
278 		dev_err(&dev->dev,
279 			"MTU %d is larger than HDS threshold %d in XDP mode\n",
280 			new_mtu, fbn->hds_thresh);
281 
282 		return -EINVAL;
283 	}
284 
285 	WRITE_ONCE(dev->mtu, new_mtu);
286 
287 	return 0;
288 }
289 
290 void fbnic_clear_rx_mode(struct fbnic_dev *fbd)
291 {
292 	struct net_device *netdev = fbd->netdev;
293 	int idx;
294 
295 	for (idx = ARRAY_SIZE(fbd->mac_addr); idx--;) {
296 		struct fbnic_mac_addr *mac_addr = &fbd->mac_addr[idx];
297 
298 		if (mac_addr->state != FBNIC_TCAM_S_VALID)
299 			continue;
300 
301 		bitmap_clear(mac_addr->act_tcam,
302 			     FBNIC_MAC_ADDR_T_HOST_START,
303 			     FBNIC_MAC_ADDR_T_HOST_LEN);
304 
305 		if (bitmap_empty(mac_addr->act_tcam,
306 				 FBNIC_RPC_TCAM_ACT_NUM_ENTRIES))
307 			mac_addr->state = FBNIC_TCAM_S_DELETE;
308 	}
309 
310 	/* Write updates to hardware */
311 	fbnic_write_macda(fbd);
312 
313 	__dev_uc_unsync(netdev, NULL);
314 	__dev_mc_unsync(netdev, NULL);
315 }
316 
317 static int fbnic_hwtstamp_get(struct net_device *netdev,
318 			      struct kernel_hwtstamp_config *config)
319 {
320 	struct fbnic_net *fbn = netdev_priv(netdev);
321 
322 	*config = fbn->hwtstamp_config;
323 
324 	return 0;
325 }
326 
327 static int fbnic_hwtstamp_set(struct net_device *netdev,
328 			      struct kernel_hwtstamp_config *config,
329 			      struct netlink_ext_ack *extack)
330 {
331 	struct fbnic_net *fbn = netdev_priv(netdev);
332 	int old_rx_filter;
333 
334 	if (config->source != HWTSTAMP_SOURCE_NETDEV)
335 		return -EOPNOTSUPP;
336 
337 	if (!kernel_hwtstamp_config_changed(config, &fbn->hwtstamp_config))
338 		return 0;
339 
340 	/* Upscale the filters */
341 	switch (config->rx_filter) {
342 	case HWTSTAMP_FILTER_NONE:
343 	case HWTSTAMP_FILTER_ALL:
344 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
345 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
346 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
347 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
348 		break;
349 	case HWTSTAMP_FILTER_NTP_ALL:
350 		config->rx_filter = HWTSTAMP_FILTER_ALL;
351 		break;
352 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
353 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
354 		config->rx_filter = HWTSTAMP_FILTER_PTP_V1_L4_EVENT;
355 		break;
356 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
357 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
358 		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_L4_EVENT;
359 		break;
360 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
361 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
362 		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_L2_EVENT;
363 		break;
364 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
365 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
366 		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
367 		break;
368 	default:
369 		return -ERANGE;
370 	}
371 
372 	/* Configure */
373 	old_rx_filter = fbn->hwtstamp_config.rx_filter;
374 	memcpy(&fbn->hwtstamp_config, config, sizeof(*config));
375 
376 	if (old_rx_filter != config->rx_filter && netif_running(fbn->netdev)) {
377 		fbnic_rss_reinit(fbn->fbd, fbn);
378 		fbnic_write_rules(fbn->fbd);
379 	}
380 
381 	/* Save / report back filter configuration
382 	 * Note that our filter configuration is inexact. Instead of
383 	 * filtering for a specific UDP port or L2 Ethertype we are
384 	 * filtering in all UDP or all non-IP packets for timestamping. So
385 	 * if anything other than FILTER_ALL is requested we report
386 	 * FILTER_SOME indicating that we will be timestamping a few
387 	 * additional packets.
388 	 */
389 	if (config->rx_filter > HWTSTAMP_FILTER_ALL)
390 		config->rx_filter = HWTSTAMP_FILTER_SOME;
391 
392 	return 0;
393 }
394 
395 static void fbnic_get_stats64(struct net_device *dev,
396 			      struct rtnl_link_stats64 *stats64)
397 {
398 	u64 rx_bytes, rx_packets, rx_dropped = 0, rx_errors = 0;
399 	u64 rx_over = 0, rx_missed = 0, rx_length = 0;
400 	u64 tx_bytes, tx_packets, tx_dropped = 0;
401 	struct fbnic_net *fbn = netdev_priv(dev);
402 	struct fbnic_dev *fbd = fbn->fbd;
403 	struct fbnic_queue_stats *stats;
404 
405 	unsigned int start, i;
406 
407 	fbnic_get_hw_stats(fbd);
408 
409 	stats = &fbn->tx_stats;
410 
411 	tx_bytes = stats->bytes;
412 	tx_packets = stats->packets;
413 	tx_dropped = stats->dropped;
414 
415 	/* Record drops from Tx HW Datapath */
416 	spin_lock(&fbd->hw_stats.lock);
417 	tx_dropped += fbd->hw_stats.tmi.drop.frames.value +
418 		      fbd->hw_stats.tti.cm_drop.frames.value +
419 		      fbd->hw_stats.tti.frame_drop.frames.value +
420 		      fbd->hw_stats.tti.tbi_drop.frames.value;
421 	spin_unlock(&fbd->hw_stats.lock);
422 
423 	stats64->tx_bytes = tx_bytes;
424 	stats64->tx_packets = tx_packets;
425 	stats64->tx_dropped = tx_dropped;
426 
427 	for (i = 0; i < fbn->num_tx_queues; i++) {
428 		struct fbnic_ring *txr = fbn->tx[i];
429 
430 		if (!txr)
431 			continue;
432 
433 		stats = &txr->stats;
434 		do {
435 			start = u64_stats_fetch_begin(&stats->syncp);
436 			tx_bytes = stats->bytes;
437 			tx_packets = stats->packets;
438 			tx_dropped = stats->dropped;
439 		} while (u64_stats_fetch_retry(&stats->syncp, start));
440 
441 		stats64->tx_bytes += tx_bytes;
442 		stats64->tx_packets += tx_packets;
443 		stats64->tx_dropped += tx_dropped;
444 	}
445 
446 	stats = &fbn->rx_stats;
447 
448 	rx_bytes = stats->bytes;
449 	rx_packets = stats->packets;
450 	rx_dropped = stats->dropped;
451 
452 	spin_lock(&fbd->hw_stats.lock);
453 	/* Record drops for the host FIFOs.
454 	 * 4: network to Host,	6: BMC to Host
455 	 * Exclude the BMC and MC FIFOs as those stats may contain drops
456 	 * due to unrelated items such as TCAM misses. They are still
457 	 * accessible through the ethtool stats.
458 	 */
459 	i = FBNIC_RXB_FIFO_HOST;
460 	rx_missed += fbd->hw_stats.rxb.fifo[i].drop.frames.value;
461 	i = FBNIC_RXB_FIFO_BMC_TO_HOST;
462 	rx_missed += fbd->hw_stats.rxb.fifo[i].drop.frames.value;
463 
464 	for (i = 0; i < fbd->max_num_queues; i++) {
465 		/* Report packets dropped due to CQ/BDQ being full/empty */
466 		rx_over += fbd->hw_stats.hw_q[i].rde_pkt_cq_drop.value;
467 		rx_over += fbd->hw_stats.hw_q[i].rde_pkt_bdq_drop.value;
468 
469 		/* Report packets with errors */
470 		rx_errors += fbd->hw_stats.hw_q[i].rde_pkt_err.value;
471 	}
472 	spin_unlock(&fbd->hw_stats.lock);
473 
474 	stats64->rx_bytes = rx_bytes;
475 	stats64->rx_packets = rx_packets;
476 	stats64->rx_dropped = rx_dropped;
477 	stats64->rx_over_errors = rx_over;
478 	stats64->rx_errors = rx_errors;
479 	stats64->rx_missed_errors = rx_missed;
480 
481 	for (i = 0; i < fbn->num_rx_queues; i++) {
482 		struct fbnic_ring *xdpr = fbn->tx[FBNIC_MAX_TXQS + i];
483 		struct fbnic_ring *rxr = fbn->rx[i];
484 
485 		if (!rxr)
486 			continue;
487 
488 		stats = &rxr->stats;
489 		do {
490 			start = u64_stats_fetch_begin(&stats->syncp);
491 			rx_bytes = stats->bytes;
492 			rx_packets = stats->packets;
493 			rx_dropped = stats->dropped;
494 			rx_length = stats->rx.length_errors;
495 		} while (u64_stats_fetch_retry(&stats->syncp, start));
496 
497 		stats64->rx_bytes += rx_bytes;
498 		stats64->rx_packets += rx_packets;
499 		stats64->rx_dropped += rx_dropped;
500 		stats64->rx_errors += rx_length;
501 		stats64->rx_length_errors += rx_length;
502 
503 		if (!xdpr)
504 			continue;
505 
506 		stats = &xdpr->stats;
507 		do {
508 			start = u64_stats_fetch_begin(&stats->syncp);
509 			tx_bytes = stats->bytes;
510 			tx_packets = stats->packets;
511 			tx_dropped = stats->dropped;
512 		} while (u64_stats_fetch_retry(&stats->syncp, start));
513 
514 		stats64->tx_bytes += tx_bytes;
515 		stats64->tx_packets += tx_packets;
516 		stats64->tx_dropped += tx_dropped;
517 	}
518 }
519 
520 bool fbnic_check_split_frames(struct bpf_prog *prog, unsigned int mtu,
521 			      u32 hds_thresh)
522 {
523 	if (!prog)
524 		return false;
525 
526 	if (prog->aux->xdp_has_frags)
527 		return false;
528 
529 	return mtu + ETH_HLEN > hds_thresh;
530 }
531 
532 static int fbnic_bpf(struct net_device *netdev, struct netdev_bpf *bpf)
533 {
534 	struct bpf_prog *prog = bpf->prog, *prev_prog;
535 	struct fbnic_net *fbn = netdev_priv(netdev);
536 
537 	if (bpf->command != XDP_SETUP_PROG)
538 		return -EINVAL;
539 
540 	if (fbnic_check_split_frames(prog, netdev->mtu,
541 				     fbn->hds_thresh)) {
542 		NL_SET_ERR_MSG_MOD(bpf->extack,
543 				   "MTU too high, or HDS threshold is too low for single buffer XDP");
544 		return -EOPNOTSUPP;
545 	}
546 
547 	prev_prog = xchg(&fbn->xdp_prog, prog);
548 	if (prev_prog)
549 		bpf_prog_put(prev_prog);
550 
551 	return 0;
552 }
553 
554 static const struct net_device_ops fbnic_netdev_ops = {
555 	.ndo_open		= fbnic_open,
556 	.ndo_stop		= fbnic_stop,
557 	.ndo_validate_addr	= eth_validate_addr,
558 	.ndo_start_xmit		= fbnic_xmit_frame,
559 	.ndo_features_check	= fbnic_features_check,
560 	.ndo_set_mac_address	= fbnic_set_mac,
561 	.ndo_change_mtu		= fbnic_change_mtu,
562 	.ndo_set_rx_mode_async	= fbnic_set_rx_mode,
563 	.ndo_get_stats64	= fbnic_get_stats64,
564 	.ndo_bpf		= fbnic_bpf,
565 	.ndo_hwtstamp_get	= fbnic_hwtstamp_get,
566 	.ndo_hwtstamp_set	= fbnic_hwtstamp_set,
567 };
568 
569 static void fbnic_get_queue_stats_rx(struct net_device *dev, int idx,
570 				     struct netdev_queue_stats_rx *rx)
571 {
572 	u64 bytes, packets, alloc_fail, alloc_fail_bdq;
573 	struct fbnic_net *fbn = netdev_priv(dev);
574 	struct fbnic_ring *rxr = fbn->rx[idx];
575 	struct fbnic_dev *fbd = fbn->fbd;
576 	struct fbnic_queue_stats *stats;
577 	u64 csum_complete, csum_none;
578 	struct fbnic_q_triad *qt;
579 	unsigned int start;
580 
581 	if (!rxr)
582 		return;
583 
584 	/* fbn->rx points to completion queues */
585 	qt = container_of(rxr, struct fbnic_q_triad, cmpl);
586 
587 	stats = &rxr->stats;
588 	do {
589 		start = u64_stats_fetch_begin(&stats->syncp);
590 		bytes = stats->bytes;
591 		packets = stats->packets;
592 		alloc_fail = stats->rx.alloc_failed;
593 		csum_complete = stats->rx.csum_complete;
594 		csum_none = stats->rx.csum_none;
595 	} while (u64_stats_fetch_retry(&stats->syncp, start));
596 
597 	stats = &qt->sub0.stats;
598 	do {
599 		start = u64_stats_fetch_begin(&stats->syncp);
600 		alloc_fail_bdq = stats->bdq.alloc_failed;
601 	} while (u64_stats_fetch_retry(&stats->syncp, start));
602 	alloc_fail += alloc_fail_bdq;
603 
604 	stats = &qt->sub1.stats;
605 	do {
606 		start = u64_stats_fetch_begin(&stats->syncp);
607 		alloc_fail_bdq = stats->bdq.alloc_failed;
608 	} while (u64_stats_fetch_retry(&stats->syncp, start));
609 	alloc_fail += alloc_fail_bdq;
610 
611 	rx->bytes = bytes;
612 	rx->packets = packets;
613 	rx->alloc_fail = alloc_fail;
614 	rx->csum_complete = csum_complete;
615 	rx->csum_none = csum_none;
616 
617 	fbnic_get_hw_q_stats(fbd, fbd->hw_stats.hw_q);
618 
619 	spin_lock(&fbd->hw_stats.lock);
620 	rx->hw_drop_overruns = fbd->hw_stats.hw_q[idx].rde_pkt_cq_drop.value +
621 			       fbd->hw_stats.hw_q[idx].rde_pkt_bdq_drop.value;
622 	rx->hw_drops = fbd->hw_stats.hw_q[idx].rde_pkt_err.value +
623 		       rx->hw_drop_overruns;
624 	spin_unlock(&fbd->hw_stats.lock);
625 }
626 
627 static void fbnic_get_queue_stats_tx(struct net_device *dev, int idx,
628 				     struct netdev_queue_stats_tx *tx)
629 {
630 	struct fbnic_net *fbn = netdev_priv(dev);
631 	struct fbnic_ring *txr = fbn->tx[idx];
632 	struct fbnic_queue_stats *stats;
633 	u64 stop, wake, csum, lso;
634 	struct fbnic_ring *xdpr;
635 	unsigned int start;
636 	u64 bytes, packets;
637 
638 	if (!txr)
639 		return;
640 
641 	stats = &txr->stats;
642 	do {
643 		start = u64_stats_fetch_begin(&stats->syncp);
644 		bytes = stats->bytes;
645 		packets = stats->packets;
646 		csum = stats->twq.csum_partial;
647 		lso = stats->twq.lso;
648 		stop = stats->twq.stop;
649 		wake = stats->twq.wake;
650 	} while (u64_stats_fetch_retry(&stats->syncp, start));
651 
652 	tx->bytes = bytes;
653 	tx->packets = packets;
654 	tx->needs_csum = csum + lso;
655 	tx->hw_gso_wire_packets = lso;
656 	tx->stop = stop;
657 	tx->wake = wake;
658 
659 	xdpr = fbn->tx[FBNIC_MAX_TXQS + idx];
660 	if (xdpr) {
661 		stats = &xdpr->stats;
662 		do {
663 			start = u64_stats_fetch_begin(&stats->syncp);
664 			bytes = stats->bytes;
665 			packets = stats->packets;
666 		} while (u64_stats_fetch_retry(&stats->syncp, start));
667 
668 		tx->bytes += bytes;
669 		tx->packets += packets;
670 	}
671 }
672 
673 static void fbnic_get_base_stats(struct net_device *dev,
674 				 struct netdev_queue_stats_rx *rx,
675 				 struct netdev_queue_stats_tx *tx)
676 {
677 	struct fbnic_net *fbn = netdev_priv(dev);
678 
679 	tx->bytes = fbn->tx_stats.bytes;
680 	tx->packets = fbn->tx_stats.packets;
681 	tx->needs_csum = fbn->tx_stats.twq.csum_partial + fbn->tx_stats.twq.lso;
682 	tx->hw_gso_wire_packets = fbn->tx_stats.twq.lso;
683 	tx->stop = fbn->tx_stats.twq.stop;
684 	tx->wake = fbn->tx_stats.twq.wake;
685 
686 	rx->bytes = fbn->rx_stats.bytes;
687 	rx->packets = fbn->rx_stats.packets;
688 	rx->alloc_fail = fbn->rx_stats.rx.alloc_failed +
689 		fbn->bdq_stats.bdq.alloc_failed;
690 	rx->csum_complete = fbn->rx_stats.rx.csum_complete;
691 	rx->csum_none = fbn->rx_stats.rx.csum_none;
692 }
693 
694 static const struct netdev_stat_ops fbnic_stat_ops = {
695 	.get_queue_stats_rx	= fbnic_get_queue_stats_rx,
696 	.get_queue_stats_tx	= fbnic_get_queue_stats_tx,
697 	.get_base_stats		= fbnic_get_base_stats,
698 };
699 
700 void fbnic_reset_queues(struct fbnic_net *fbn,
701 			unsigned int tx, unsigned int rx)
702 {
703 	struct fbnic_dev *fbd = fbn->fbd;
704 	unsigned int max_napis;
705 
706 	max_napis = fbd->num_irqs - FBNIC_NON_NAPI_VECTORS;
707 
708 	tx = min(tx, max_napis);
709 	fbn->num_tx_queues = tx;
710 
711 	rx = min(rx, max_napis);
712 	fbn->num_rx_queues = rx;
713 
714 	fbn->num_napi = max(tx, rx);
715 }
716 
717 /**
718  * fbnic_netdev_free - Free the netdev associate with fbnic
719  * @fbd: Driver specific structure to free netdev from
720  *
721  * Allocate and initialize the netdev and netdev private structure. Bind
722  * together the hardware, netdev, and pci data structures.
723  **/
724 void fbnic_netdev_free(struct fbnic_dev *fbd)
725 {
726 	fbnic_phylink_destroy(fbd->netdev);
727 
728 	free_netdev(fbd->netdev);
729 	fbd->netdev = NULL;
730 }
731 
732 /**
733  * fbnic_netdev_alloc - Allocate a netdev and associate with fbnic
734  * @fbd: Driver specific structure to associate netdev with
735  *
736  * Allocate and initialize the netdev and netdev private structure. Bind
737  * together the hardware, netdev, and pci data structures.
738  *
739  *  Return: Pointer to net_device on success, NULL on failure
740  **/
741 struct net_device *fbnic_netdev_alloc(struct fbnic_dev *fbd)
742 {
743 	struct net_device *netdev;
744 	struct fbnic_net *fbn;
745 	int default_queues;
746 
747 	netdev = alloc_etherdev_mq(sizeof(*fbn), FBNIC_MAX_RXQS);
748 	if (!netdev)
749 		return NULL;
750 
751 	SET_NETDEV_DEV(netdev, fbd->dev);
752 	fbd->netdev = netdev;
753 
754 	netdev->netdev_ops = &fbnic_netdev_ops;
755 	netdev->stat_ops = &fbnic_stat_ops;
756 	netdev->queue_mgmt_ops = &fbnic_queue_mgmt_ops;
757 	netdev->netmem_tx = NETMEM_TX_DMA;
758 
759 	fbnic_set_ethtool_ops(netdev);
760 
761 	fbn = netdev_priv(netdev);
762 
763 	fbn->netdev = netdev;
764 	fbn->fbd = fbd;
765 
766 	fbn->txq_size = FBNIC_TXQ_SIZE_DEFAULT;
767 	fbn->hpq_size = FBNIC_HPQ_SIZE_DEFAULT;
768 	fbn->ppq_size = FBNIC_PPQ_SIZE_DEFAULT;
769 	fbn->rcq_size = FBNIC_RCQ_SIZE_DEFAULT;
770 
771 	fbn->tx_usecs = FBNIC_TX_USECS_DEFAULT;
772 	fbn->rx_usecs = FBNIC_RX_USECS_DEFAULT;
773 	fbn->rx_max_frames = FBNIC_RX_FRAMES_DEFAULT;
774 
775 	/* Initialize the hds_thresh */
776 	netdev->cfg->hds_thresh = FBNIC_HDS_THRESH_DEFAULT;
777 	fbn->hds_thresh = FBNIC_HDS_THRESH_DEFAULT;
778 
779 	default_queues = netif_get_num_default_rss_queues();
780 	if (default_queues > fbd->max_num_queues)
781 		default_queues = fbd->max_num_queues;
782 
783 	fbnic_reset_queues(fbn, default_queues, default_queues);
784 
785 	fbnic_reset_indir_tbl(fbn);
786 	fbnic_rss_key_fill(fbn->rss_key);
787 	fbnic_rss_init_en_mask(fbn);
788 
789 	netdev->priv_flags |= IFF_UNICAST_FLT;
790 
791 	netdev->gso_partial_features =
792 		NETIF_F_GSO_GRE |
793 		NETIF_F_GSO_GRE_CSUM |
794 		NETIF_F_GSO_IPXIP4 |
795 		NETIF_F_GSO_UDP_TUNNEL |
796 		NETIF_F_GSO_UDP_TUNNEL_CSUM;
797 
798 	netdev->features |=
799 		netdev->gso_partial_features |
800 		FBNIC_TUN_GSO_FEATURES |
801 		NETIF_F_RXHASH |
802 		NETIF_F_SG |
803 		NETIF_F_HW_CSUM |
804 		NETIF_F_RXCSUM |
805 		NETIF_F_TSO |
806 		NETIF_F_TSO_ECN |
807 		NETIF_F_TSO6 |
808 		NETIF_F_GSO_PARTIAL |
809 		NETIF_F_GSO_UDP_L4;
810 
811 	netdev->hw_features |= netdev->features;
812 	netdev->vlan_features |= netdev->features;
813 	netdev->hw_enc_features |= netdev->features;
814 	netdev->features |= NETIF_F_NTUPLE;
815 
816 	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_RX_SG;
817 
818 	netdev->min_mtu = IPV6_MIN_MTU;
819 	netdev->max_mtu = FBNIC_MAX_JUMBO_FRAME_SIZE - ETH_HLEN;
820 
821 	/* TBD: This is workaround for BMC as phylink doesn't have support
822 	 * for leavling the link enabled if a BMC is present.
823 	 */
824 	netdev->ethtool->wol_enabled = true;
825 
826 	netif_carrier_off(netdev);
827 
828 	netif_tx_stop_all_queues(netdev);
829 
830 	if (fbnic_phylink_create(netdev)) {
831 		free_netdev(netdev);
832 		fbd->netdev = NULL;
833 		return NULL;
834 	}
835 
836 	return netdev;
837 }
838 
839 static int fbnic_dsn_to_mac_addr(u64 dsn, char *addr)
840 {
841 	addr[0] = (dsn >> 56) & 0xFF;
842 	addr[1] = (dsn >> 48) & 0xFF;
843 	addr[2] = (dsn >> 40) & 0xFF;
844 	addr[3] = (dsn >> 16) & 0xFF;
845 	addr[4] = (dsn >> 8) & 0xFF;
846 	addr[5] = dsn & 0xFF;
847 
848 	return is_valid_ether_addr(addr) ? 0 : -EINVAL;
849 }
850 
851 /**
852  * fbnic_netdev_register - Initialize general software structures
853  * @netdev: Netdev containing structure to initialize and register
854  *
855  * Initialize the MAC address for the netdev and register it.
856  *
857  *  Return: 0 on success, negative on failure
858  **/
859 int fbnic_netdev_register(struct net_device *netdev)
860 {
861 	struct fbnic_net *fbn = netdev_priv(netdev);
862 	struct fbnic_dev *fbd = fbn->fbd;
863 	u64 dsn = fbd->dsn;
864 	u8 addr[ETH_ALEN];
865 	int err;
866 
867 	err = fbnic_dsn_to_mac_addr(dsn, addr);
868 	if (!err) {
869 		ether_addr_copy(netdev->perm_addr, addr);
870 		eth_hw_addr_set(netdev, addr);
871 	} else {
872 		/* A randomly assigned MAC address will cause provisioning
873 		 * issues so instead just fail to spawn the netdev and
874 		 * avoid any confusion.
875 		 */
876 		dev_err(fbd->dev, "MAC addr %pM invalid\n", addr);
877 		return err;
878 	}
879 
880 	return register_netdev(netdev);
881 }
882 
883 void fbnic_netdev_unregister(struct net_device *netdev)
884 {
885 	unregister_netdev(netdev);
886 }
887