xref: /linux/drivers/net/ethernet/google/gve/gve_main.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: (GPL-2.0 OR MIT)
2 /* Google virtual Ethernet (gve) driver
3  *
4  * Copyright (C) 2015-2024 Google LLC
5  */
6 
7 #include <linux/bpf.h>
8 #include <linux/cpumask.h>
9 #include <linux/etherdevice.h>
10 #include <linux/filter.h>
11 #include <linux/interrupt.h>
12 #include <linux/irq.h>
13 #include <linux/module.h>
14 #include <linux/pci.h>
15 #include <linux/sched.h>
16 #include <linux/timer.h>
17 #include <linux/workqueue.h>
18 #include <linux/utsname.h>
19 #include <linux/version.h>
20 #include <net/netdev_queues.h>
21 #include <net/sch_generic.h>
22 #include <net/xdp_sock_drv.h>
23 #include "gve.h"
24 #include "gve_dqo.h"
25 #include "gve_adminq.h"
26 #include "gve_register.h"
27 #include "gve_utils.h"
28 
29 #define GVE_DEFAULT_RX_COPYBREAK	(256)
30 
31 #define DEFAULT_MSG_LEVEL	(NETIF_MSG_DRV | NETIF_MSG_LINK)
32 #define GVE_VERSION		"1.0.0"
33 #define GVE_VERSION_PREFIX	"GVE-"
34 
35 // Minimum amount of time between queue kicks in msec (10 seconds)
36 #define MIN_TX_TIMEOUT_GAP (1000 * 10)
37 
38 char gve_driver_name[] = "gve";
39 const char gve_version_str[] = GVE_VERSION;
40 static const char gve_version_prefix[] = GVE_VERSION_PREFIX;
41 
42 static int gve_verify_driver_compatibility(struct gve_priv *priv)
43 {
44 	int err;
45 	struct gve_driver_info *driver_info;
46 	dma_addr_t driver_info_bus;
47 
48 	driver_info = dma_alloc_coherent(&priv->pdev->dev,
49 					 sizeof(struct gve_driver_info),
50 					 &driver_info_bus, GFP_KERNEL);
51 	if (!driver_info)
52 		return -ENOMEM;
53 
54 	*driver_info = (struct gve_driver_info) {
55 		.os_type = 1, /* Linux */
56 		.os_version_major = cpu_to_be32(LINUX_VERSION_MAJOR),
57 		.os_version_minor = cpu_to_be32(LINUX_VERSION_SUBLEVEL),
58 		.os_version_sub = cpu_to_be32(LINUX_VERSION_PATCHLEVEL),
59 		.driver_capability_flags = {
60 			cpu_to_be64(GVE_DRIVER_CAPABILITY_FLAGS1),
61 			cpu_to_be64(GVE_DRIVER_CAPABILITY_FLAGS2),
62 			cpu_to_be64(GVE_DRIVER_CAPABILITY_FLAGS3),
63 			cpu_to_be64(GVE_DRIVER_CAPABILITY_FLAGS4),
64 		},
65 	};
66 	strscpy(driver_info->os_version_str1, utsname()->release,
67 		sizeof(driver_info->os_version_str1));
68 	strscpy(driver_info->os_version_str2, utsname()->version,
69 		sizeof(driver_info->os_version_str2));
70 
71 	err = gve_adminq_verify_driver_compatibility(priv,
72 						     sizeof(struct gve_driver_info),
73 						     driver_info_bus);
74 
75 	/* It's ok if the device doesn't support this */
76 	if (err == -EOPNOTSUPP)
77 		err = 0;
78 
79 	dma_free_coherent(&priv->pdev->dev,
80 			  sizeof(struct gve_driver_info),
81 			  driver_info, driver_info_bus);
82 	return err;
83 }
84 
85 static netdev_features_t gve_features_check(struct sk_buff *skb,
86 					    struct net_device *dev,
87 					    netdev_features_t features)
88 {
89 	struct gve_priv *priv = netdev_priv(dev);
90 
91 	if (!gve_is_gqi(priv))
92 		return gve_features_check_dqo(skb, dev, features);
93 
94 	return features;
95 }
96 
97 static netdev_tx_t gve_start_xmit(struct sk_buff *skb, struct net_device *dev)
98 {
99 	struct gve_priv *priv = netdev_priv(dev);
100 
101 	if (gve_is_gqi(priv))
102 		return gve_tx(skb, dev);
103 	else
104 		return gve_tx_dqo(skb, dev);
105 }
106 
107 static void gve_get_stats(struct net_device *dev, struct rtnl_link_stats64 *s)
108 {
109 	struct gve_priv *priv = netdev_priv(dev);
110 	unsigned int start;
111 	u64 packets, bytes;
112 	int num_tx_queues;
113 	int ring;
114 
115 	num_tx_queues = gve_num_tx_queues(priv);
116 	if (priv->rx) {
117 		for (ring = 0; ring < priv->rx_cfg.num_queues; ring++) {
118 			do {
119 				start =
120 				  u64_stats_fetch_begin(&priv->rx[ring].statss);
121 				packets = priv->rx[ring].rpackets;
122 				bytes = priv->rx[ring].rbytes;
123 			} while (u64_stats_fetch_retry(&priv->rx[ring].statss,
124 						       start));
125 			s->rx_packets += packets;
126 			s->rx_bytes += bytes;
127 		}
128 	}
129 	if (priv->tx) {
130 		for (ring = 0; ring < num_tx_queues; ring++) {
131 			do {
132 				start =
133 				  u64_stats_fetch_begin(&priv->tx[ring].statss);
134 				packets = priv->tx[ring].pkt_done;
135 				bytes = priv->tx[ring].bytes_done;
136 			} while (u64_stats_fetch_retry(&priv->tx[ring].statss,
137 						       start));
138 			s->tx_packets += packets;
139 			s->tx_bytes += bytes;
140 		}
141 	}
142 }
143 
144 static int gve_alloc_flow_rule_caches(struct gve_priv *priv)
145 {
146 	struct gve_flow_rules_cache *flow_rules_cache = &priv->flow_rules_cache;
147 	int err = 0;
148 
149 	if (!priv->max_flow_rules)
150 		return 0;
151 
152 	flow_rules_cache->rules_cache =
153 		kvcalloc(GVE_FLOW_RULES_CACHE_SIZE, sizeof(*flow_rules_cache->rules_cache),
154 			 GFP_KERNEL);
155 	if (!flow_rules_cache->rules_cache) {
156 		dev_err(&priv->pdev->dev, "Cannot alloc flow rules cache\n");
157 		return -ENOMEM;
158 	}
159 
160 	flow_rules_cache->rule_ids_cache =
161 		kvcalloc(GVE_FLOW_RULE_IDS_CACHE_SIZE, sizeof(*flow_rules_cache->rule_ids_cache),
162 			 GFP_KERNEL);
163 	if (!flow_rules_cache->rule_ids_cache) {
164 		dev_err(&priv->pdev->dev, "Cannot alloc flow rule ids cache\n");
165 		err = -ENOMEM;
166 		goto free_rules_cache;
167 	}
168 
169 	return 0;
170 
171 free_rules_cache:
172 	kvfree(flow_rules_cache->rules_cache);
173 	flow_rules_cache->rules_cache = NULL;
174 	return err;
175 }
176 
177 static void gve_free_flow_rule_caches(struct gve_priv *priv)
178 {
179 	struct gve_flow_rules_cache *flow_rules_cache = &priv->flow_rules_cache;
180 
181 	kvfree(flow_rules_cache->rule_ids_cache);
182 	flow_rules_cache->rule_ids_cache = NULL;
183 	kvfree(flow_rules_cache->rules_cache);
184 	flow_rules_cache->rules_cache = NULL;
185 }
186 
187 static int gve_alloc_counter_array(struct gve_priv *priv)
188 {
189 	priv->counter_array =
190 		dma_alloc_coherent(&priv->pdev->dev,
191 				   priv->num_event_counters *
192 				   sizeof(*priv->counter_array),
193 				   &priv->counter_array_bus, GFP_KERNEL);
194 	if (!priv->counter_array)
195 		return -ENOMEM;
196 
197 	return 0;
198 }
199 
200 static void gve_free_counter_array(struct gve_priv *priv)
201 {
202 	if (!priv->counter_array)
203 		return;
204 
205 	dma_free_coherent(&priv->pdev->dev,
206 			  priv->num_event_counters *
207 			  sizeof(*priv->counter_array),
208 			  priv->counter_array, priv->counter_array_bus);
209 	priv->counter_array = NULL;
210 }
211 
212 /* NIC requests to report stats */
213 static void gve_stats_report_task(struct work_struct *work)
214 {
215 	struct gve_priv *priv = container_of(work, struct gve_priv,
216 					     stats_report_task);
217 	if (gve_get_do_report_stats(priv)) {
218 		gve_handle_report_stats(priv);
219 		gve_clear_do_report_stats(priv);
220 	}
221 }
222 
223 static void gve_stats_report_schedule(struct gve_priv *priv)
224 {
225 	if (!gve_get_probe_in_progress(priv) &&
226 	    !gve_get_reset_in_progress(priv)) {
227 		gve_set_do_report_stats(priv);
228 		queue_work(priv->gve_wq, &priv->stats_report_task);
229 	}
230 }
231 
232 static void gve_stats_report_timer(struct timer_list *t)
233 {
234 	struct gve_priv *priv = from_timer(priv, t, stats_report_timer);
235 
236 	mod_timer(&priv->stats_report_timer,
237 		  round_jiffies(jiffies +
238 		  msecs_to_jiffies(priv->stats_report_timer_period)));
239 	gve_stats_report_schedule(priv);
240 }
241 
242 static int gve_alloc_stats_report(struct gve_priv *priv)
243 {
244 	int tx_stats_num, rx_stats_num;
245 
246 	tx_stats_num = (GVE_TX_STATS_REPORT_NUM + NIC_TX_STATS_REPORT_NUM) *
247 		       gve_num_tx_queues(priv);
248 	rx_stats_num = (GVE_RX_STATS_REPORT_NUM + NIC_RX_STATS_REPORT_NUM) *
249 		       priv->rx_cfg.num_queues;
250 	priv->stats_report_len = struct_size(priv->stats_report, stats,
251 					     size_add(tx_stats_num, rx_stats_num));
252 	priv->stats_report =
253 		dma_alloc_coherent(&priv->pdev->dev, priv->stats_report_len,
254 				   &priv->stats_report_bus, GFP_KERNEL);
255 	if (!priv->stats_report)
256 		return -ENOMEM;
257 	/* Set up timer for the report-stats task */
258 	timer_setup(&priv->stats_report_timer, gve_stats_report_timer, 0);
259 	priv->stats_report_timer_period = GVE_STATS_REPORT_TIMER_PERIOD;
260 	return 0;
261 }
262 
263 static void gve_free_stats_report(struct gve_priv *priv)
264 {
265 	if (!priv->stats_report)
266 		return;
267 
268 	del_timer_sync(&priv->stats_report_timer);
269 	dma_free_coherent(&priv->pdev->dev, priv->stats_report_len,
270 			  priv->stats_report, priv->stats_report_bus);
271 	priv->stats_report = NULL;
272 }
273 
274 static irqreturn_t gve_mgmnt_intr(int irq, void *arg)
275 {
276 	struct gve_priv *priv = arg;
277 
278 	queue_work(priv->gve_wq, &priv->service_task);
279 	return IRQ_HANDLED;
280 }
281 
282 static irqreturn_t gve_intr(int irq, void *arg)
283 {
284 	struct gve_notify_block *block = arg;
285 	struct gve_priv *priv = block->priv;
286 
287 	iowrite32be(GVE_IRQ_MASK, gve_irq_doorbell(priv, block));
288 	napi_schedule_irqoff(&block->napi);
289 	return IRQ_HANDLED;
290 }
291 
292 static irqreturn_t gve_intr_dqo(int irq, void *arg)
293 {
294 	struct gve_notify_block *block = arg;
295 
296 	/* Interrupts are automatically masked */
297 	napi_schedule_irqoff(&block->napi);
298 	return IRQ_HANDLED;
299 }
300 
301 static int gve_is_napi_on_home_cpu(struct gve_priv *priv, u32 irq)
302 {
303 	int cpu_curr = smp_processor_id();
304 	const struct cpumask *aff_mask;
305 
306 	aff_mask = irq_get_effective_affinity_mask(irq);
307 	if (unlikely(!aff_mask))
308 		return 1;
309 
310 	return cpumask_test_cpu(cpu_curr, aff_mask);
311 }
312 
313 int gve_napi_poll(struct napi_struct *napi, int budget)
314 {
315 	struct gve_notify_block *block;
316 	__be32 __iomem *irq_doorbell;
317 	bool reschedule = false;
318 	struct gve_priv *priv;
319 	int work_done = 0;
320 
321 	block = container_of(napi, struct gve_notify_block, napi);
322 	priv = block->priv;
323 
324 	if (block->tx) {
325 		if (block->tx->q_num < priv->tx_cfg.num_queues)
326 			reschedule |= gve_tx_poll(block, budget);
327 		else if (budget)
328 			reschedule |= gve_xdp_poll(block, budget);
329 	}
330 
331 	if (!budget)
332 		return 0;
333 
334 	if (block->rx) {
335 		work_done = gve_rx_poll(block, budget);
336 		reschedule |= work_done == budget;
337 	}
338 
339 	if (reschedule)
340 		return budget;
341 
342        /* Complete processing - don't unmask irq if busy polling is enabled */
343 	if (likely(napi_complete_done(napi, work_done))) {
344 		irq_doorbell = gve_irq_doorbell(priv, block);
345 		iowrite32be(GVE_IRQ_ACK | GVE_IRQ_EVENT, irq_doorbell);
346 
347 		/* Ensure IRQ ACK is visible before we check pending work.
348 		 * If queue had issued updates, it would be truly visible.
349 		 */
350 		mb();
351 
352 		if (block->tx)
353 			reschedule |= gve_tx_clean_pending(priv, block->tx);
354 		if (block->rx)
355 			reschedule |= gve_rx_work_pending(block->rx);
356 
357 		if (reschedule && napi_schedule(napi))
358 			iowrite32be(GVE_IRQ_MASK, irq_doorbell);
359 	}
360 	return work_done;
361 }
362 
363 int gve_napi_poll_dqo(struct napi_struct *napi, int budget)
364 {
365 	struct gve_notify_block *block =
366 		container_of(napi, struct gve_notify_block, napi);
367 	struct gve_priv *priv = block->priv;
368 	bool reschedule = false;
369 	int work_done = 0;
370 
371 	if (block->tx)
372 		reschedule |= gve_tx_poll_dqo(block, /*do_clean=*/true);
373 
374 	if (!budget)
375 		return 0;
376 
377 	if (block->rx) {
378 		work_done = gve_rx_poll_dqo(block, budget);
379 		reschedule |= work_done == budget;
380 	}
381 
382 	if (reschedule) {
383 		/* Reschedule by returning budget only if already on the correct
384 		 * cpu.
385 		 */
386 		if (likely(gve_is_napi_on_home_cpu(priv, block->irq)))
387 			return budget;
388 
389 		/* If not on the cpu with which this queue's irq has affinity
390 		 * with, we avoid rescheduling napi and arm the irq instead so
391 		 * that napi gets rescheduled back eventually onto the right
392 		 * cpu.
393 		 */
394 		if (work_done == budget)
395 			work_done--;
396 	}
397 
398 	if (likely(napi_complete_done(napi, work_done))) {
399 		/* Enable interrupts again.
400 		 *
401 		 * We don't need to repoll afterwards because HW supports the
402 		 * PCI MSI-X PBA feature.
403 		 *
404 		 * Another interrupt would be triggered if a new event came in
405 		 * since the last one.
406 		 */
407 		gve_write_irq_doorbell_dqo(priv, block,
408 					   GVE_ITR_NO_UPDATE_DQO | GVE_ITR_ENABLE_BIT_DQO);
409 	}
410 
411 	return work_done;
412 }
413 
414 static int gve_alloc_notify_blocks(struct gve_priv *priv)
415 {
416 	int num_vecs_requested = priv->num_ntfy_blks + 1;
417 	unsigned int active_cpus;
418 	int vecs_enabled;
419 	int i, j;
420 	int err;
421 
422 	priv->msix_vectors = kvcalloc(num_vecs_requested,
423 				      sizeof(*priv->msix_vectors), GFP_KERNEL);
424 	if (!priv->msix_vectors)
425 		return -ENOMEM;
426 	for (i = 0; i < num_vecs_requested; i++)
427 		priv->msix_vectors[i].entry = i;
428 	vecs_enabled = pci_enable_msix_range(priv->pdev, priv->msix_vectors,
429 					     GVE_MIN_MSIX, num_vecs_requested);
430 	if (vecs_enabled < 0) {
431 		dev_err(&priv->pdev->dev, "Could not enable min msix %d/%d\n",
432 			GVE_MIN_MSIX, vecs_enabled);
433 		err = vecs_enabled;
434 		goto abort_with_msix_vectors;
435 	}
436 	if (vecs_enabled != num_vecs_requested) {
437 		int new_num_ntfy_blks = (vecs_enabled - 1) & ~0x1;
438 		int vecs_per_type = new_num_ntfy_blks / 2;
439 		int vecs_left = new_num_ntfy_blks % 2;
440 
441 		priv->num_ntfy_blks = new_num_ntfy_blks;
442 		priv->mgmt_msix_idx = priv->num_ntfy_blks;
443 		priv->tx_cfg.max_queues = min_t(int, priv->tx_cfg.max_queues,
444 						vecs_per_type);
445 		priv->rx_cfg.max_queues = min_t(int, priv->rx_cfg.max_queues,
446 						vecs_per_type + vecs_left);
447 		dev_err(&priv->pdev->dev,
448 			"Could not enable desired msix, only enabled %d, adjusting tx max queues to %d, and rx max queues to %d\n",
449 			vecs_enabled, priv->tx_cfg.max_queues,
450 			priv->rx_cfg.max_queues);
451 		if (priv->tx_cfg.num_queues > priv->tx_cfg.max_queues)
452 			priv->tx_cfg.num_queues = priv->tx_cfg.max_queues;
453 		if (priv->rx_cfg.num_queues > priv->rx_cfg.max_queues)
454 			priv->rx_cfg.num_queues = priv->rx_cfg.max_queues;
455 	}
456 	/* Half the notification blocks go to TX and half to RX */
457 	active_cpus = min_t(int, priv->num_ntfy_blks / 2, num_online_cpus());
458 
459 	/* Setup Management Vector  - the last vector */
460 	snprintf(priv->mgmt_msix_name, sizeof(priv->mgmt_msix_name), "gve-mgmnt@pci:%s",
461 		 pci_name(priv->pdev));
462 	err = request_irq(priv->msix_vectors[priv->mgmt_msix_idx].vector,
463 			  gve_mgmnt_intr, 0, priv->mgmt_msix_name, priv);
464 	if (err) {
465 		dev_err(&priv->pdev->dev, "Did not receive management vector.\n");
466 		goto abort_with_msix_enabled;
467 	}
468 	priv->irq_db_indices =
469 		dma_alloc_coherent(&priv->pdev->dev,
470 				   priv->num_ntfy_blks *
471 				   sizeof(*priv->irq_db_indices),
472 				   &priv->irq_db_indices_bus, GFP_KERNEL);
473 	if (!priv->irq_db_indices) {
474 		err = -ENOMEM;
475 		goto abort_with_mgmt_vector;
476 	}
477 
478 	priv->ntfy_blocks = kvzalloc(priv->num_ntfy_blks *
479 				     sizeof(*priv->ntfy_blocks), GFP_KERNEL);
480 	if (!priv->ntfy_blocks) {
481 		err = -ENOMEM;
482 		goto abort_with_irq_db_indices;
483 	}
484 
485 	/* Setup the other blocks - the first n-1 vectors */
486 	for (i = 0; i < priv->num_ntfy_blks; i++) {
487 		struct gve_notify_block *block = &priv->ntfy_blocks[i];
488 		int msix_idx = i;
489 
490 		snprintf(block->name, sizeof(block->name), "gve-ntfy-blk%d@pci:%s",
491 			 i, pci_name(priv->pdev));
492 		block->priv = priv;
493 		err = request_irq(priv->msix_vectors[msix_idx].vector,
494 				  gve_is_gqi(priv) ? gve_intr : gve_intr_dqo,
495 				  0, block->name, block);
496 		if (err) {
497 			dev_err(&priv->pdev->dev,
498 				"Failed to receive msix vector %d\n", i);
499 			goto abort_with_some_ntfy_blocks;
500 		}
501 		block->irq = priv->msix_vectors[msix_idx].vector;
502 		irq_set_affinity_hint(priv->msix_vectors[msix_idx].vector,
503 				      get_cpu_mask(i % active_cpus));
504 		block->irq_db_index = &priv->irq_db_indices[i].index;
505 	}
506 	return 0;
507 abort_with_some_ntfy_blocks:
508 	for (j = 0; j < i; j++) {
509 		struct gve_notify_block *block = &priv->ntfy_blocks[j];
510 		int msix_idx = j;
511 
512 		irq_set_affinity_hint(priv->msix_vectors[msix_idx].vector,
513 				      NULL);
514 		free_irq(priv->msix_vectors[msix_idx].vector, block);
515 		block->irq = 0;
516 	}
517 	kvfree(priv->ntfy_blocks);
518 	priv->ntfy_blocks = NULL;
519 abort_with_irq_db_indices:
520 	dma_free_coherent(&priv->pdev->dev, priv->num_ntfy_blks *
521 			  sizeof(*priv->irq_db_indices),
522 			  priv->irq_db_indices, priv->irq_db_indices_bus);
523 	priv->irq_db_indices = NULL;
524 abort_with_mgmt_vector:
525 	free_irq(priv->msix_vectors[priv->mgmt_msix_idx].vector, priv);
526 abort_with_msix_enabled:
527 	pci_disable_msix(priv->pdev);
528 abort_with_msix_vectors:
529 	kvfree(priv->msix_vectors);
530 	priv->msix_vectors = NULL;
531 	return err;
532 }
533 
534 static void gve_free_notify_blocks(struct gve_priv *priv)
535 {
536 	int i;
537 
538 	if (!priv->msix_vectors)
539 		return;
540 
541 	/* Free the irqs */
542 	for (i = 0; i < priv->num_ntfy_blks; i++) {
543 		struct gve_notify_block *block = &priv->ntfy_blocks[i];
544 		int msix_idx = i;
545 
546 		irq_set_affinity_hint(priv->msix_vectors[msix_idx].vector,
547 				      NULL);
548 		free_irq(priv->msix_vectors[msix_idx].vector, block);
549 		block->irq = 0;
550 	}
551 	free_irq(priv->msix_vectors[priv->mgmt_msix_idx].vector, priv);
552 	kvfree(priv->ntfy_blocks);
553 	priv->ntfy_blocks = NULL;
554 	dma_free_coherent(&priv->pdev->dev, priv->num_ntfy_blks *
555 			  sizeof(*priv->irq_db_indices),
556 			  priv->irq_db_indices, priv->irq_db_indices_bus);
557 	priv->irq_db_indices = NULL;
558 	pci_disable_msix(priv->pdev);
559 	kvfree(priv->msix_vectors);
560 	priv->msix_vectors = NULL;
561 }
562 
563 static int gve_setup_device_resources(struct gve_priv *priv)
564 {
565 	int err;
566 
567 	err = gve_alloc_flow_rule_caches(priv);
568 	if (err)
569 		return err;
570 	err = gve_alloc_counter_array(priv);
571 	if (err)
572 		goto abort_with_flow_rule_caches;
573 	err = gve_alloc_notify_blocks(priv);
574 	if (err)
575 		goto abort_with_counter;
576 	err = gve_alloc_stats_report(priv);
577 	if (err)
578 		goto abort_with_ntfy_blocks;
579 	err = gve_adminq_configure_device_resources(priv,
580 						    priv->counter_array_bus,
581 						    priv->num_event_counters,
582 						    priv->irq_db_indices_bus,
583 						    priv->num_ntfy_blks);
584 	if (unlikely(err)) {
585 		dev_err(&priv->pdev->dev,
586 			"could not setup device_resources: err=%d\n", err);
587 		err = -ENXIO;
588 		goto abort_with_stats_report;
589 	}
590 
591 	if (!gve_is_gqi(priv)) {
592 		priv->ptype_lut_dqo = kvzalloc(sizeof(*priv->ptype_lut_dqo),
593 					       GFP_KERNEL);
594 		if (!priv->ptype_lut_dqo) {
595 			err = -ENOMEM;
596 			goto abort_with_stats_report;
597 		}
598 		err = gve_adminq_get_ptype_map_dqo(priv, priv->ptype_lut_dqo);
599 		if (err) {
600 			dev_err(&priv->pdev->dev,
601 				"Failed to get ptype map: err=%d\n", err);
602 			goto abort_with_ptype_lut;
603 		}
604 	}
605 
606 	err = gve_adminq_report_stats(priv, priv->stats_report_len,
607 				      priv->stats_report_bus,
608 				      GVE_STATS_REPORT_TIMER_PERIOD);
609 	if (err)
610 		dev_err(&priv->pdev->dev,
611 			"Failed to report stats: err=%d\n", err);
612 	gve_set_device_resources_ok(priv);
613 	return 0;
614 
615 abort_with_ptype_lut:
616 	kvfree(priv->ptype_lut_dqo);
617 	priv->ptype_lut_dqo = NULL;
618 abort_with_stats_report:
619 	gve_free_stats_report(priv);
620 abort_with_ntfy_blocks:
621 	gve_free_notify_blocks(priv);
622 abort_with_counter:
623 	gve_free_counter_array(priv);
624 abort_with_flow_rule_caches:
625 	gve_free_flow_rule_caches(priv);
626 
627 	return err;
628 }
629 
630 static void gve_trigger_reset(struct gve_priv *priv);
631 
632 static void gve_teardown_device_resources(struct gve_priv *priv)
633 {
634 	int err;
635 
636 	/* Tell device its resources are being freed */
637 	if (gve_get_device_resources_ok(priv)) {
638 		err = gve_flow_rules_reset(priv);
639 		if (err) {
640 			dev_err(&priv->pdev->dev,
641 				"Failed to reset flow rules: err=%d\n", err);
642 			gve_trigger_reset(priv);
643 		}
644 		/* detach the stats report */
645 		err = gve_adminq_report_stats(priv, 0, 0x0, GVE_STATS_REPORT_TIMER_PERIOD);
646 		if (err) {
647 			dev_err(&priv->pdev->dev,
648 				"Failed to detach stats report: err=%d\n", err);
649 			gve_trigger_reset(priv);
650 		}
651 		err = gve_adminq_deconfigure_device_resources(priv);
652 		if (err) {
653 			dev_err(&priv->pdev->dev,
654 				"Could not deconfigure device resources: err=%d\n",
655 				err);
656 			gve_trigger_reset(priv);
657 		}
658 	}
659 
660 	kvfree(priv->ptype_lut_dqo);
661 	priv->ptype_lut_dqo = NULL;
662 
663 	gve_free_flow_rule_caches(priv);
664 	gve_free_counter_array(priv);
665 	gve_free_notify_blocks(priv);
666 	gve_free_stats_report(priv);
667 	gve_clear_device_resources_ok(priv);
668 }
669 
670 static int gve_unregister_qpl(struct gve_priv *priv,
671 			      struct gve_queue_page_list *qpl)
672 {
673 	int err;
674 
675 	if (!qpl)
676 		return 0;
677 
678 	err = gve_adminq_unregister_page_list(priv, qpl->id);
679 	if (err) {
680 		netif_err(priv, drv, priv->dev,
681 			  "Failed to unregister queue page list %d\n",
682 			  qpl->id);
683 		return err;
684 	}
685 
686 	priv->num_registered_pages -= qpl->num_entries;
687 	return 0;
688 }
689 
690 static int gve_register_qpl(struct gve_priv *priv,
691 			    struct gve_queue_page_list *qpl)
692 {
693 	int pages;
694 	int err;
695 
696 	if (!qpl)
697 		return 0;
698 
699 	pages = qpl->num_entries;
700 
701 	if (pages + priv->num_registered_pages > priv->max_registered_pages) {
702 		netif_err(priv, drv, priv->dev,
703 			  "Reached max number of registered pages %llu > %llu\n",
704 			  pages + priv->num_registered_pages,
705 			  priv->max_registered_pages);
706 		return -EINVAL;
707 	}
708 
709 	err = gve_adminq_register_page_list(priv, qpl);
710 	if (err) {
711 		netif_err(priv, drv, priv->dev,
712 			  "failed to register queue page list %d\n",
713 			  qpl->id);
714 		return err;
715 	}
716 
717 	priv->num_registered_pages += pages;
718 	return 0;
719 }
720 
721 static struct gve_queue_page_list *gve_tx_get_qpl(struct gve_priv *priv, int idx)
722 {
723 	struct gve_tx_ring *tx = &priv->tx[idx];
724 
725 	if (gve_is_gqi(priv))
726 		return tx->tx_fifo.qpl;
727 	else
728 		return tx->dqo.qpl;
729 }
730 
731 static struct gve_queue_page_list *gve_rx_get_qpl(struct gve_priv *priv, int idx)
732 {
733 	struct gve_rx_ring *rx = &priv->rx[idx];
734 
735 	if (gve_is_gqi(priv))
736 		return rx->data.qpl;
737 	else
738 		return rx->dqo.qpl;
739 }
740 
741 static int gve_register_xdp_qpls(struct gve_priv *priv)
742 {
743 	int start_id;
744 	int err;
745 	int i;
746 
747 	start_id = gve_xdp_tx_start_queue_id(priv);
748 	for (i = start_id; i < start_id + gve_num_xdp_qpls(priv); i++) {
749 		err = gve_register_qpl(priv, gve_tx_get_qpl(priv, i));
750 		/* This failure will trigger a reset - no need to clean up */
751 		if (err)
752 			return err;
753 	}
754 	return 0;
755 }
756 
757 static int gve_register_qpls(struct gve_priv *priv)
758 {
759 	int num_tx_qpls, num_rx_qpls;
760 	int err;
761 	int i;
762 
763 	num_tx_qpls = gve_num_tx_qpls(&priv->tx_cfg, gve_num_xdp_qpls(priv),
764 				      gve_is_qpl(priv));
765 	num_rx_qpls = gve_num_rx_qpls(&priv->rx_cfg, gve_is_qpl(priv));
766 
767 	for (i = 0; i < num_tx_qpls; i++) {
768 		err = gve_register_qpl(priv, gve_tx_get_qpl(priv, i));
769 		if (err)
770 			return err;
771 	}
772 
773 	for (i = 0; i < num_rx_qpls; i++) {
774 		err = gve_register_qpl(priv, gve_rx_get_qpl(priv, i));
775 		if (err)
776 			return err;
777 	}
778 
779 	return 0;
780 }
781 
782 static int gve_unregister_xdp_qpls(struct gve_priv *priv)
783 {
784 	int start_id;
785 	int err;
786 	int i;
787 
788 	start_id = gve_xdp_tx_start_queue_id(priv);
789 	for (i = start_id; i < start_id + gve_num_xdp_qpls(priv); i++) {
790 		err = gve_unregister_qpl(priv, gve_tx_get_qpl(priv, i));
791 		/* This failure will trigger a reset - no need to clean */
792 		if (err)
793 			return err;
794 	}
795 	return 0;
796 }
797 
798 static int gve_unregister_qpls(struct gve_priv *priv)
799 {
800 	int num_tx_qpls, num_rx_qpls;
801 	int err;
802 	int i;
803 
804 	num_tx_qpls = gve_num_tx_qpls(&priv->tx_cfg, gve_num_xdp_qpls(priv),
805 				      gve_is_qpl(priv));
806 	num_rx_qpls = gve_num_rx_qpls(&priv->rx_cfg, gve_is_qpl(priv));
807 
808 	for (i = 0; i < num_tx_qpls; i++) {
809 		err = gve_unregister_qpl(priv, gve_tx_get_qpl(priv, i));
810 		/* This failure will trigger a reset - no need to clean */
811 		if (err)
812 			return err;
813 	}
814 
815 	for (i = 0; i < num_rx_qpls; i++) {
816 		err = gve_unregister_qpl(priv, gve_rx_get_qpl(priv, i));
817 		/* This failure will trigger a reset - no need to clean */
818 		if (err)
819 			return err;
820 	}
821 	return 0;
822 }
823 
824 static int gve_create_xdp_rings(struct gve_priv *priv)
825 {
826 	int err;
827 
828 	err = gve_adminq_create_tx_queues(priv,
829 					  gve_xdp_tx_start_queue_id(priv),
830 					  priv->num_xdp_queues);
831 	if (err) {
832 		netif_err(priv, drv, priv->dev, "failed to create %d XDP tx queues\n",
833 			  priv->num_xdp_queues);
834 		/* This failure will trigger a reset - no need to clean
835 		 * up
836 		 */
837 		return err;
838 	}
839 	netif_dbg(priv, drv, priv->dev, "created %d XDP tx queues\n",
840 		  priv->num_xdp_queues);
841 
842 	return 0;
843 }
844 
845 static int gve_create_rings(struct gve_priv *priv)
846 {
847 	int num_tx_queues = gve_num_tx_queues(priv);
848 	int err;
849 	int i;
850 
851 	err = gve_adminq_create_tx_queues(priv, 0, num_tx_queues);
852 	if (err) {
853 		netif_err(priv, drv, priv->dev, "failed to create %d tx queues\n",
854 			  num_tx_queues);
855 		/* This failure will trigger a reset - no need to clean
856 		 * up
857 		 */
858 		return err;
859 	}
860 	netif_dbg(priv, drv, priv->dev, "created %d tx queues\n",
861 		  num_tx_queues);
862 
863 	err = gve_adminq_create_rx_queues(priv, priv->rx_cfg.num_queues);
864 	if (err) {
865 		netif_err(priv, drv, priv->dev, "failed to create %d rx queues\n",
866 			  priv->rx_cfg.num_queues);
867 		/* This failure will trigger a reset - no need to clean
868 		 * up
869 		 */
870 		return err;
871 	}
872 	netif_dbg(priv, drv, priv->dev, "created %d rx queues\n",
873 		  priv->rx_cfg.num_queues);
874 
875 	if (gve_is_gqi(priv)) {
876 		/* Rx data ring has been prefilled with packet buffers at queue
877 		 * allocation time.
878 		 *
879 		 * Write the doorbell to provide descriptor slots and packet
880 		 * buffers to the NIC.
881 		 */
882 		for (i = 0; i < priv->rx_cfg.num_queues; i++)
883 			gve_rx_write_doorbell(priv, &priv->rx[i]);
884 	} else {
885 		for (i = 0; i < priv->rx_cfg.num_queues; i++) {
886 			/* Post buffers and ring doorbell. */
887 			gve_rx_post_buffers_dqo(&priv->rx[i]);
888 		}
889 	}
890 
891 	return 0;
892 }
893 
894 static void init_xdp_sync_stats(struct gve_priv *priv)
895 {
896 	int start_id = gve_xdp_tx_start_queue_id(priv);
897 	int i;
898 
899 	/* Init stats */
900 	for (i = start_id; i < start_id + priv->num_xdp_queues; i++) {
901 		int ntfy_idx = gve_tx_idx_to_ntfy(priv, i);
902 
903 		u64_stats_init(&priv->tx[i].statss);
904 		priv->tx[i].ntfy_id = ntfy_idx;
905 	}
906 }
907 
908 static void gve_init_sync_stats(struct gve_priv *priv)
909 {
910 	int i;
911 
912 	for (i = 0; i < priv->tx_cfg.num_queues; i++)
913 		u64_stats_init(&priv->tx[i].statss);
914 
915 	/* Init stats for XDP TX queues */
916 	init_xdp_sync_stats(priv);
917 
918 	for (i = 0; i < priv->rx_cfg.num_queues; i++)
919 		u64_stats_init(&priv->rx[i].statss);
920 }
921 
922 static void gve_tx_get_curr_alloc_cfg(struct gve_priv *priv,
923 				      struct gve_tx_alloc_rings_cfg *cfg)
924 {
925 	cfg->qcfg = &priv->tx_cfg;
926 	cfg->raw_addressing = !gve_is_qpl(priv);
927 	cfg->ring_size = priv->tx_desc_cnt;
928 	cfg->start_idx = 0;
929 	cfg->num_rings = gve_num_tx_queues(priv);
930 	cfg->tx = priv->tx;
931 }
932 
933 static void gve_tx_stop_rings(struct gve_priv *priv, int start_id, int num_rings)
934 {
935 	int i;
936 
937 	if (!priv->tx)
938 		return;
939 
940 	for (i = start_id; i < start_id + num_rings; i++) {
941 		if (gve_is_gqi(priv))
942 			gve_tx_stop_ring_gqi(priv, i);
943 		else
944 			gve_tx_stop_ring_dqo(priv, i);
945 	}
946 }
947 
948 static void gve_tx_start_rings(struct gve_priv *priv, int start_id,
949 			       int num_rings)
950 {
951 	int i;
952 
953 	for (i = start_id; i < start_id + num_rings; i++) {
954 		if (gve_is_gqi(priv))
955 			gve_tx_start_ring_gqi(priv, i);
956 		else
957 			gve_tx_start_ring_dqo(priv, i);
958 	}
959 }
960 
961 static int gve_alloc_xdp_rings(struct gve_priv *priv)
962 {
963 	struct gve_tx_alloc_rings_cfg cfg = {0};
964 	int err = 0;
965 
966 	if (!priv->num_xdp_queues)
967 		return 0;
968 
969 	gve_tx_get_curr_alloc_cfg(priv, &cfg);
970 	cfg.start_idx = gve_xdp_tx_start_queue_id(priv);
971 	cfg.num_rings = priv->num_xdp_queues;
972 
973 	err = gve_tx_alloc_rings_gqi(priv, &cfg);
974 	if (err)
975 		return err;
976 
977 	gve_tx_start_rings(priv, cfg.start_idx, cfg.num_rings);
978 	init_xdp_sync_stats(priv);
979 
980 	return 0;
981 }
982 
983 static int gve_queues_mem_alloc(struct gve_priv *priv,
984 				struct gve_tx_alloc_rings_cfg *tx_alloc_cfg,
985 				struct gve_rx_alloc_rings_cfg *rx_alloc_cfg)
986 {
987 	int err;
988 
989 	if (gve_is_gqi(priv))
990 		err = gve_tx_alloc_rings_gqi(priv, tx_alloc_cfg);
991 	else
992 		err = gve_tx_alloc_rings_dqo(priv, tx_alloc_cfg);
993 	if (err)
994 		return err;
995 
996 	if (gve_is_gqi(priv))
997 		err = gve_rx_alloc_rings_gqi(priv, rx_alloc_cfg);
998 	else
999 		err = gve_rx_alloc_rings_dqo(priv, rx_alloc_cfg);
1000 	if (err)
1001 		goto free_tx;
1002 
1003 	return 0;
1004 
1005 free_tx:
1006 	if (gve_is_gqi(priv))
1007 		gve_tx_free_rings_gqi(priv, tx_alloc_cfg);
1008 	else
1009 		gve_tx_free_rings_dqo(priv, tx_alloc_cfg);
1010 	return err;
1011 }
1012 
1013 static int gve_destroy_xdp_rings(struct gve_priv *priv)
1014 {
1015 	int start_id;
1016 	int err;
1017 
1018 	start_id = gve_xdp_tx_start_queue_id(priv);
1019 	err = gve_adminq_destroy_tx_queues(priv,
1020 					   start_id,
1021 					   priv->num_xdp_queues);
1022 	if (err) {
1023 		netif_err(priv, drv, priv->dev,
1024 			  "failed to destroy XDP queues\n");
1025 		/* This failure will trigger a reset - no need to clean up */
1026 		return err;
1027 	}
1028 	netif_dbg(priv, drv, priv->dev, "destroyed XDP queues\n");
1029 
1030 	return 0;
1031 }
1032 
1033 static int gve_destroy_rings(struct gve_priv *priv)
1034 {
1035 	int num_tx_queues = gve_num_tx_queues(priv);
1036 	int err;
1037 
1038 	err = gve_adminq_destroy_tx_queues(priv, 0, num_tx_queues);
1039 	if (err) {
1040 		netif_err(priv, drv, priv->dev,
1041 			  "failed to destroy tx queues\n");
1042 		/* This failure will trigger a reset - no need to clean up */
1043 		return err;
1044 	}
1045 	netif_dbg(priv, drv, priv->dev, "destroyed tx queues\n");
1046 	err = gve_adminq_destroy_rx_queues(priv, priv->rx_cfg.num_queues);
1047 	if (err) {
1048 		netif_err(priv, drv, priv->dev,
1049 			  "failed to destroy rx queues\n");
1050 		/* This failure will trigger a reset - no need to clean up */
1051 		return err;
1052 	}
1053 	netif_dbg(priv, drv, priv->dev, "destroyed rx queues\n");
1054 	return 0;
1055 }
1056 
1057 static void gve_free_xdp_rings(struct gve_priv *priv)
1058 {
1059 	struct gve_tx_alloc_rings_cfg cfg = {0};
1060 
1061 	gve_tx_get_curr_alloc_cfg(priv, &cfg);
1062 	cfg.start_idx = gve_xdp_tx_start_queue_id(priv);
1063 	cfg.num_rings = priv->num_xdp_queues;
1064 
1065 	if (priv->tx) {
1066 		gve_tx_stop_rings(priv, cfg.start_idx, cfg.num_rings);
1067 		gve_tx_free_rings_gqi(priv, &cfg);
1068 	}
1069 }
1070 
1071 static void gve_queues_mem_free(struct gve_priv *priv,
1072 				struct gve_tx_alloc_rings_cfg *tx_cfg,
1073 				struct gve_rx_alloc_rings_cfg *rx_cfg)
1074 {
1075 	if (gve_is_gqi(priv)) {
1076 		gve_tx_free_rings_gqi(priv, tx_cfg);
1077 		gve_rx_free_rings_gqi(priv, rx_cfg);
1078 	} else {
1079 		gve_tx_free_rings_dqo(priv, tx_cfg);
1080 		gve_rx_free_rings_dqo(priv, rx_cfg);
1081 	}
1082 }
1083 
1084 int gve_alloc_page(struct gve_priv *priv, struct device *dev,
1085 		   struct page **page, dma_addr_t *dma,
1086 		   enum dma_data_direction dir, gfp_t gfp_flags)
1087 {
1088 	*page = alloc_page(gfp_flags);
1089 	if (!*page) {
1090 		priv->page_alloc_fail++;
1091 		return -ENOMEM;
1092 	}
1093 	*dma = dma_map_page(dev, *page, 0, PAGE_SIZE, dir);
1094 	if (dma_mapping_error(dev, *dma)) {
1095 		priv->dma_mapping_error++;
1096 		put_page(*page);
1097 		return -ENOMEM;
1098 	}
1099 	return 0;
1100 }
1101 
1102 struct gve_queue_page_list *gve_alloc_queue_page_list(struct gve_priv *priv,
1103 						      u32 id, int pages)
1104 {
1105 	struct gve_queue_page_list *qpl;
1106 	int err;
1107 	int i;
1108 
1109 	qpl = kvzalloc(sizeof(*qpl), GFP_KERNEL);
1110 	if (!qpl)
1111 		return NULL;
1112 
1113 	qpl->id = id;
1114 	qpl->num_entries = 0;
1115 	qpl->pages = kvcalloc(pages, sizeof(*qpl->pages), GFP_KERNEL);
1116 	if (!qpl->pages)
1117 		goto abort;
1118 
1119 	qpl->page_buses = kvcalloc(pages, sizeof(*qpl->page_buses), GFP_KERNEL);
1120 	if (!qpl->page_buses)
1121 		goto abort;
1122 
1123 	for (i = 0; i < pages; i++) {
1124 		err = gve_alloc_page(priv, &priv->pdev->dev, &qpl->pages[i],
1125 				     &qpl->page_buses[i],
1126 				     gve_qpl_dma_dir(priv, id), GFP_KERNEL);
1127 		if (err)
1128 			goto abort;
1129 		qpl->num_entries++;
1130 	}
1131 
1132 	return qpl;
1133 
1134 abort:
1135 	gve_free_queue_page_list(priv, qpl, id);
1136 	return NULL;
1137 }
1138 
1139 void gve_free_page(struct device *dev, struct page *page, dma_addr_t dma,
1140 		   enum dma_data_direction dir)
1141 {
1142 	if (!dma_mapping_error(dev, dma))
1143 		dma_unmap_page(dev, dma, PAGE_SIZE, dir);
1144 	if (page)
1145 		put_page(page);
1146 }
1147 
1148 void gve_free_queue_page_list(struct gve_priv *priv,
1149 			      struct gve_queue_page_list *qpl,
1150 			      u32 id)
1151 {
1152 	int i;
1153 
1154 	if (!qpl)
1155 		return;
1156 	if (!qpl->pages)
1157 		goto free_qpl;
1158 	if (!qpl->page_buses)
1159 		goto free_pages;
1160 
1161 	for (i = 0; i < qpl->num_entries; i++)
1162 		gve_free_page(&priv->pdev->dev, qpl->pages[i],
1163 			      qpl->page_buses[i], gve_qpl_dma_dir(priv, id));
1164 
1165 	kvfree(qpl->page_buses);
1166 	qpl->page_buses = NULL;
1167 free_pages:
1168 	kvfree(qpl->pages);
1169 	qpl->pages = NULL;
1170 free_qpl:
1171 	kvfree(qpl);
1172 }
1173 
1174 /* Use this to schedule a reset when the device is capable of continuing
1175  * to handle other requests in its current state. If it is not, do a reset
1176  * in thread instead.
1177  */
1178 void gve_schedule_reset(struct gve_priv *priv)
1179 {
1180 	gve_set_do_reset(priv);
1181 	queue_work(priv->gve_wq, &priv->service_task);
1182 }
1183 
1184 static void gve_reset_and_teardown(struct gve_priv *priv, bool was_up);
1185 static int gve_reset_recovery(struct gve_priv *priv, bool was_up);
1186 static void gve_turndown(struct gve_priv *priv);
1187 static void gve_turnup(struct gve_priv *priv);
1188 
1189 static int gve_reg_xdp_info(struct gve_priv *priv, struct net_device *dev)
1190 {
1191 	struct napi_struct *napi;
1192 	struct gve_rx_ring *rx;
1193 	int err = 0;
1194 	int i, j;
1195 	u32 tx_qid;
1196 
1197 	if (!priv->num_xdp_queues)
1198 		return 0;
1199 
1200 	for (i = 0; i < priv->rx_cfg.num_queues; i++) {
1201 		rx = &priv->rx[i];
1202 		napi = &priv->ntfy_blocks[rx->ntfy_id].napi;
1203 
1204 		err = xdp_rxq_info_reg(&rx->xdp_rxq, dev, i,
1205 				       napi->napi_id);
1206 		if (err)
1207 			goto err;
1208 		err = xdp_rxq_info_reg_mem_model(&rx->xdp_rxq,
1209 						 MEM_TYPE_PAGE_SHARED, NULL);
1210 		if (err)
1211 			goto err;
1212 		rx->xsk_pool = xsk_get_pool_from_qid(dev, i);
1213 		if (rx->xsk_pool) {
1214 			err = xdp_rxq_info_reg(&rx->xsk_rxq, dev, i,
1215 					       napi->napi_id);
1216 			if (err)
1217 				goto err;
1218 			err = xdp_rxq_info_reg_mem_model(&rx->xsk_rxq,
1219 							 MEM_TYPE_XSK_BUFF_POOL, NULL);
1220 			if (err)
1221 				goto err;
1222 			xsk_pool_set_rxq_info(rx->xsk_pool,
1223 					      &rx->xsk_rxq);
1224 		}
1225 	}
1226 
1227 	for (i = 0; i < priv->num_xdp_queues; i++) {
1228 		tx_qid = gve_xdp_tx_queue_id(priv, i);
1229 		priv->tx[tx_qid].xsk_pool = xsk_get_pool_from_qid(dev, i);
1230 	}
1231 	return 0;
1232 
1233 err:
1234 	for (j = i; j >= 0; j--) {
1235 		rx = &priv->rx[j];
1236 		if (xdp_rxq_info_is_reg(&rx->xdp_rxq))
1237 			xdp_rxq_info_unreg(&rx->xdp_rxq);
1238 		if (xdp_rxq_info_is_reg(&rx->xsk_rxq))
1239 			xdp_rxq_info_unreg(&rx->xsk_rxq);
1240 	}
1241 	return err;
1242 }
1243 
1244 static void gve_unreg_xdp_info(struct gve_priv *priv)
1245 {
1246 	int i, tx_qid;
1247 
1248 	if (!priv->num_xdp_queues)
1249 		return;
1250 
1251 	for (i = 0; i < priv->rx_cfg.num_queues; i++) {
1252 		struct gve_rx_ring *rx = &priv->rx[i];
1253 
1254 		xdp_rxq_info_unreg(&rx->xdp_rxq);
1255 		if (rx->xsk_pool) {
1256 			xdp_rxq_info_unreg(&rx->xsk_rxq);
1257 			rx->xsk_pool = NULL;
1258 		}
1259 	}
1260 
1261 	for (i = 0; i < priv->num_xdp_queues; i++) {
1262 		tx_qid = gve_xdp_tx_queue_id(priv, i);
1263 		priv->tx[tx_qid].xsk_pool = NULL;
1264 	}
1265 }
1266 
1267 static void gve_drain_page_cache(struct gve_priv *priv)
1268 {
1269 	int i;
1270 
1271 	for (i = 0; i < priv->rx_cfg.num_queues; i++)
1272 		page_frag_cache_drain(&priv->rx[i].page_cache);
1273 }
1274 
1275 static void gve_rx_get_curr_alloc_cfg(struct gve_priv *priv,
1276 				      struct gve_rx_alloc_rings_cfg *cfg)
1277 {
1278 	cfg->qcfg = &priv->rx_cfg;
1279 	cfg->qcfg_tx = &priv->tx_cfg;
1280 	cfg->raw_addressing = !gve_is_qpl(priv);
1281 	cfg->enable_header_split = priv->header_split_enabled;
1282 	cfg->ring_size = priv->rx_desc_cnt;
1283 	cfg->packet_buffer_size = gve_is_gqi(priv) ?
1284 				  GVE_DEFAULT_RX_BUFFER_SIZE :
1285 				  priv->data_buffer_size_dqo;
1286 	cfg->rx = priv->rx;
1287 }
1288 
1289 void gve_get_curr_alloc_cfgs(struct gve_priv *priv,
1290 			     struct gve_tx_alloc_rings_cfg *tx_alloc_cfg,
1291 			     struct gve_rx_alloc_rings_cfg *rx_alloc_cfg)
1292 {
1293 	gve_tx_get_curr_alloc_cfg(priv, tx_alloc_cfg);
1294 	gve_rx_get_curr_alloc_cfg(priv, rx_alloc_cfg);
1295 }
1296 
1297 static void gve_rx_start_ring(struct gve_priv *priv, int i)
1298 {
1299 	if (gve_is_gqi(priv))
1300 		gve_rx_start_ring_gqi(priv, i);
1301 	else
1302 		gve_rx_start_ring_dqo(priv, i);
1303 }
1304 
1305 static void gve_rx_start_rings(struct gve_priv *priv, int num_rings)
1306 {
1307 	int i;
1308 
1309 	for (i = 0; i < num_rings; i++)
1310 		gve_rx_start_ring(priv, i);
1311 }
1312 
1313 static void gve_rx_stop_ring(struct gve_priv *priv, int i)
1314 {
1315 	if (gve_is_gqi(priv))
1316 		gve_rx_stop_ring_gqi(priv, i);
1317 	else
1318 		gve_rx_stop_ring_dqo(priv, i);
1319 }
1320 
1321 static void gve_rx_stop_rings(struct gve_priv *priv, int num_rings)
1322 {
1323 	int i;
1324 
1325 	if (!priv->rx)
1326 		return;
1327 
1328 	for (i = 0; i < num_rings; i++)
1329 		gve_rx_stop_ring(priv, i);
1330 }
1331 
1332 static void gve_queues_mem_remove(struct gve_priv *priv)
1333 {
1334 	struct gve_tx_alloc_rings_cfg tx_alloc_cfg = {0};
1335 	struct gve_rx_alloc_rings_cfg rx_alloc_cfg = {0};
1336 
1337 	gve_get_curr_alloc_cfgs(priv, &tx_alloc_cfg, &rx_alloc_cfg);
1338 	gve_queues_mem_free(priv, &tx_alloc_cfg, &rx_alloc_cfg);
1339 	priv->tx = NULL;
1340 	priv->rx = NULL;
1341 }
1342 
1343 /* The passed-in queue memory is stored into priv and the queues are made live.
1344  * No memory is allocated. Passed-in memory is freed on errors.
1345  */
1346 static int gve_queues_start(struct gve_priv *priv,
1347 			    struct gve_tx_alloc_rings_cfg *tx_alloc_cfg,
1348 			    struct gve_rx_alloc_rings_cfg *rx_alloc_cfg)
1349 {
1350 	struct net_device *dev = priv->dev;
1351 	int err;
1352 
1353 	/* Record new resources into priv */
1354 	priv->tx = tx_alloc_cfg->tx;
1355 	priv->rx = rx_alloc_cfg->rx;
1356 
1357 	/* Record new configs into priv */
1358 	priv->tx_cfg = *tx_alloc_cfg->qcfg;
1359 	priv->rx_cfg = *rx_alloc_cfg->qcfg;
1360 	priv->tx_desc_cnt = tx_alloc_cfg->ring_size;
1361 	priv->rx_desc_cnt = rx_alloc_cfg->ring_size;
1362 
1363 	if (priv->xdp_prog)
1364 		priv->num_xdp_queues = priv->rx_cfg.num_queues;
1365 	else
1366 		priv->num_xdp_queues = 0;
1367 
1368 	gve_tx_start_rings(priv, 0, tx_alloc_cfg->num_rings);
1369 	gve_rx_start_rings(priv, rx_alloc_cfg->qcfg->num_queues);
1370 	gve_init_sync_stats(priv);
1371 
1372 	err = netif_set_real_num_tx_queues(dev, priv->tx_cfg.num_queues);
1373 	if (err)
1374 		goto stop_and_free_rings;
1375 	err = netif_set_real_num_rx_queues(dev, priv->rx_cfg.num_queues);
1376 	if (err)
1377 		goto stop_and_free_rings;
1378 
1379 	err = gve_reg_xdp_info(priv, dev);
1380 	if (err)
1381 		goto stop_and_free_rings;
1382 
1383 	err = gve_register_qpls(priv);
1384 	if (err)
1385 		goto reset;
1386 
1387 	priv->header_split_enabled = rx_alloc_cfg->enable_header_split;
1388 	priv->data_buffer_size_dqo = rx_alloc_cfg->packet_buffer_size;
1389 
1390 	err = gve_create_rings(priv);
1391 	if (err)
1392 		goto reset;
1393 
1394 	gve_set_device_rings_ok(priv);
1395 
1396 	if (gve_get_report_stats(priv))
1397 		mod_timer(&priv->stats_report_timer,
1398 			  round_jiffies(jiffies +
1399 				msecs_to_jiffies(priv->stats_report_timer_period)));
1400 
1401 	gve_turnup(priv);
1402 	queue_work(priv->gve_wq, &priv->service_task);
1403 	priv->interface_up_cnt++;
1404 	return 0;
1405 
1406 reset:
1407 	if (gve_get_reset_in_progress(priv))
1408 		goto stop_and_free_rings;
1409 	gve_reset_and_teardown(priv, true);
1410 	/* if this fails there is nothing we can do so just ignore the return */
1411 	gve_reset_recovery(priv, false);
1412 	/* return the original error */
1413 	return err;
1414 stop_and_free_rings:
1415 	gve_tx_stop_rings(priv, 0, gve_num_tx_queues(priv));
1416 	gve_rx_stop_rings(priv, priv->rx_cfg.num_queues);
1417 	gve_queues_mem_remove(priv);
1418 	return err;
1419 }
1420 
1421 static int gve_open(struct net_device *dev)
1422 {
1423 	struct gve_tx_alloc_rings_cfg tx_alloc_cfg = {0};
1424 	struct gve_rx_alloc_rings_cfg rx_alloc_cfg = {0};
1425 	struct gve_priv *priv = netdev_priv(dev);
1426 	int err;
1427 
1428 	gve_get_curr_alloc_cfgs(priv, &tx_alloc_cfg, &rx_alloc_cfg);
1429 
1430 	err = gve_queues_mem_alloc(priv, &tx_alloc_cfg, &rx_alloc_cfg);
1431 	if (err)
1432 		return err;
1433 
1434 	/* No need to free on error: ownership of resources is lost after
1435 	 * calling gve_queues_start.
1436 	 */
1437 	err = gve_queues_start(priv, &tx_alloc_cfg, &rx_alloc_cfg);
1438 	if (err)
1439 		return err;
1440 
1441 	return 0;
1442 }
1443 
1444 static int gve_queues_stop(struct gve_priv *priv)
1445 {
1446 	int err;
1447 
1448 	netif_carrier_off(priv->dev);
1449 	if (gve_get_device_rings_ok(priv)) {
1450 		gve_turndown(priv);
1451 		gve_drain_page_cache(priv);
1452 		err = gve_destroy_rings(priv);
1453 		if (err)
1454 			goto err;
1455 		err = gve_unregister_qpls(priv);
1456 		if (err)
1457 			goto err;
1458 		gve_clear_device_rings_ok(priv);
1459 	}
1460 	del_timer_sync(&priv->stats_report_timer);
1461 
1462 	gve_unreg_xdp_info(priv);
1463 
1464 	gve_tx_stop_rings(priv, 0, gve_num_tx_queues(priv));
1465 	gve_rx_stop_rings(priv, priv->rx_cfg.num_queues);
1466 
1467 	priv->interface_down_cnt++;
1468 	return 0;
1469 
1470 err:
1471 	/* This must have been called from a reset due to the rtnl lock
1472 	 * so just return at this point.
1473 	 */
1474 	if (gve_get_reset_in_progress(priv))
1475 		return err;
1476 	/* Otherwise reset before returning */
1477 	gve_reset_and_teardown(priv, true);
1478 	return gve_reset_recovery(priv, false);
1479 }
1480 
1481 static int gve_close(struct net_device *dev)
1482 {
1483 	struct gve_priv *priv = netdev_priv(dev);
1484 	int err;
1485 
1486 	err = gve_queues_stop(priv);
1487 	if (err)
1488 		return err;
1489 
1490 	gve_queues_mem_remove(priv);
1491 	return 0;
1492 }
1493 
1494 static int gve_remove_xdp_queues(struct gve_priv *priv)
1495 {
1496 	int err;
1497 
1498 	err = gve_destroy_xdp_rings(priv);
1499 	if (err)
1500 		return err;
1501 
1502 	err = gve_unregister_xdp_qpls(priv);
1503 	if (err)
1504 		return err;
1505 
1506 	gve_unreg_xdp_info(priv);
1507 	gve_free_xdp_rings(priv);
1508 
1509 	priv->num_xdp_queues = 0;
1510 	return 0;
1511 }
1512 
1513 static int gve_add_xdp_queues(struct gve_priv *priv)
1514 {
1515 	int err;
1516 
1517 	priv->num_xdp_queues = priv->rx_cfg.num_queues;
1518 
1519 	err = gve_alloc_xdp_rings(priv);
1520 	if (err)
1521 		goto err;
1522 
1523 	err = gve_reg_xdp_info(priv, priv->dev);
1524 	if (err)
1525 		goto free_xdp_rings;
1526 
1527 	err = gve_register_xdp_qpls(priv);
1528 	if (err)
1529 		goto free_xdp_rings;
1530 
1531 	err = gve_create_xdp_rings(priv);
1532 	if (err)
1533 		goto free_xdp_rings;
1534 
1535 	return 0;
1536 
1537 free_xdp_rings:
1538 	gve_free_xdp_rings(priv);
1539 err:
1540 	priv->num_xdp_queues = 0;
1541 	return err;
1542 }
1543 
1544 static void gve_handle_link_status(struct gve_priv *priv, bool link_status)
1545 {
1546 	if (!gve_get_napi_enabled(priv))
1547 		return;
1548 
1549 	if (link_status == netif_carrier_ok(priv->dev))
1550 		return;
1551 
1552 	if (link_status) {
1553 		netdev_info(priv->dev, "Device link is up.\n");
1554 		netif_carrier_on(priv->dev);
1555 	} else {
1556 		netdev_info(priv->dev, "Device link is down.\n");
1557 		netif_carrier_off(priv->dev);
1558 	}
1559 }
1560 
1561 static int gve_set_xdp(struct gve_priv *priv, struct bpf_prog *prog,
1562 		       struct netlink_ext_ack *extack)
1563 {
1564 	struct bpf_prog *old_prog;
1565 	int err = 0;
1566 	u32 status;
1567 
1568 	old_prog = READ_ONCE(priv->xdp_prog);
1569 	if (!netif_running(priv->dev)) {
1570 		WRITE_ONCE(priv->xdp_prog, prog);
1571 		if (old_prog)
1572 			bpf_prog_put(old_prog);
1573 		return 0;
1574 	}
1575 
1576 	gve_turndown(priv);
1577 	if (!old_prog && prog) {
1578 		// Allocate XDP TX queues if an XDP program is
1579 		// being installed
1580 		err = gve_add_xdp_queues(priv);
1581 		if (err)
1582 			goto out;
1583 	} else if (old_prog && !prog) {
1584 		// Remove XDP TX queues if an XDP program is
1585 		// being uninstalled
1586 		err = gve_remove_xdp_queues(priv);
1587 		if (err)
1588 			goto out;
1589 	}
1590 	WRITE_ONCE(priv->xdp_prog, prog);
1591 	if (old_prog)
1592 		bpf_prog_put(old_prog);
1593 
1594 out:
1595 	gve_turnup(priv);
1596 	status = ioread32be(&priv->reg_bar0->device_status);
1597 	gve_handle_link_status(priv, GVE_DEVICE_STATUS_LINK_STATUS_MASK & status);
1598 	return err;
1599 }
1600 
1601 static int gve_xsk_pool_enable(struct net_device *dev,
1602 			       struct xsk_buff_pool *pool,
1603 			       u16 qid)
1604 {
1605 	struct gve_priv *priv = netdev_priv(dev);
1606 	struct napi_struct *napi;
1607 	struct gve_rx_ring *rx;
1608 	int tx_qid;
1609 	int err;
1610 
1611 	if (qid >= priv->rx_cfg.num_queues) {
1612 		dev_err(&priv->pdev->dev, "xsk pool invalid qid %d", qid);
1613 		return -EINVAL;
1614 	}
1615 	if (xsk_pool_get_rx_frame_size(pool) <
1616 	     priv->dev->max_mtu + sizeof(struct ethhdr)) {
1617 		dev_err(&priv->pdev->dev, "xsk pool frame_len too small");
1618 		return -EINVAL;
1619 	}
1620 
1621 	err = xsk_pool_dma_map(pool, &priv->pdev->dev,
1622 			       DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING);
1623 	if (err)
1624 		return err;
1625 
1626 	/* If XDP prog is not installed, return */
1627 	if (!priv->xdp_prog)
1628 		return 0;
1629 
1630 	rx = &priv->rx[qid];
1631 	napi = &priv->ntfy_blocks[rx->ntfy_id].napi;
1632 	err = xdp_rxq_info_reg(&rx->xsk_rxq, dev, qid, napi->napi_id);
1633 	if (err)
1634 		goto err;
1635 
1636 	err = xdp_rxq_info_reg_mem_model(&rx->xsk_rxq,
1637 					 MEM_TYPE_XSK_BUFF_POOL, NULL);
1638 	if (err)
1639 		goto err;
1640 
1641 	xsk_pool_set_rxq_info(pool, &rx->xsk_rxq);
1642 	rx->xsk_pool = pool;
1643 
1644 	tx_qid = gve_xdp_tx_queue_id(priv, qid);
1645 	priv->tx[tx_qid].xsk_pool = pool;
1646 
1647 	return 0;
1648 err:
1649 	if (xdp_rxq_info_is_reg(&rx->xsk_rxq))
1650 		xdp_rxq_info_unreg(&rx->xsk_rxq);
1651 
1652 	xsk_pool_dma_unmap(pool,
1653 			   DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING);
1654 	return err;
1655 }
1656 
1657 static int gve_xsk_pool_disable(struct net_device *dev,
1658 				u16 qid)
1659 {
1660 	struct gve_priv *priv = netdev_priv(dev);
1661 	struct napi_struct *napi_rx;
1662 	struct napi_struct *napi_tx;
1663 	struct xsk_buff_pool *pool;
1664 	int tx_qid;
1665 
1666 	pool = xsk_get_pool_from_qid(dev, qid);
1667 	if (!pool)
1668 		return -EINVAL;
1669 	if (qid >= priv->rx_cfg.num_queues)
1670 		return -EINVAL;
1671 
1672 	/* If XDP prog is not installed, unmap DMA and return */
1673 	if (!priv->xdp_prog)
1674 		goto done;
1675 
1676 	tx_qid = gve_xdp_tx_queue_id(priv, qid);
1677 	if (!netif_running(dev)) {
1678 		priv->rx[qid].xsk_pool = NULL;
1679 		xdp_rxq_info_unreg(&priv->rx[qid].xsk_rxq);
1680 		priv->tx[tx_qid].xsk_pool = NULL;
1681 		goto done;
1682 	}
1683 
1684 	napi_rx = &priv->ntfy_blocks[priv->rx[qid].ntfy_id].napi;
1685 	napi_disable(napi_rx); /* make sure current rx poll is done */
1686 
1687 	napi_tx = &priv->ntfy_blocks[priv->tx[tx_qid].ntfy_id].napi;
1688 	napi_disable(napi_tx); /* make sure current tx poll is done */
1689 
1690 	priv->rx[qid].xsk_pool = NULL;
1691 	xdp_rxq_info_unreg(&priv->rx[qid].xsk_rxq);
1692 	priv->tx[tx_qid].xsk_pool = NULL;
1693 	smp_mb(); /* Make sure it is visible to the workers on datapath */
1694 
1695 	napi_enable(napi_rx);
1696 	if (gve_rx_work_pending(&priv->rx[qid]))
1697 		napi_schedule(napi_rx);
1698 
1699 	napi_enable(napi_tx);
1700 	if (gve_tx_clean_pending(priv, &priv->tx[tx_qid]))
1701 		napi_schedule(napi_tx);
1702 
1703 done:
1704 	xsk_pool_dma_unmap(pool,
1705 			   DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING);
1706 	return 0;
1707 }
1708 
1709 static int gve_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
1710 {
1711 	struct gve_priv *priv = netdev_priv(dev);
1712 	int tx_queue_id = gve_xdp_tx_queue_id(priv, queue_id);
1713 
1714 	if (queue_id >= priv->rx_cfg.num_queues || !priv->xdp_prog)
1715 		return -EINVAL;
1716 
1717 	if (flags & XDP_WAKEUP_TX) {
1718 		struct gve_tx_ring *tx = &priv->tx[tx_queue_id];
1719 		struct napi_struct *napi =
1720 			&priv->ntfy_blocks[tx->ntfy_id].napi;
1721 
1722 		if (!napi_if_scheduled_mark_missed(napi)) {
1723 			/* Call local_bh_enable to trigger SoftIRQ processing */
1724 			local_bh_disable();
1725 			napi_schedule(napi);
1726 			local_bh_enable();
1727 		}
1728 
1729 		tx->xdp_xsk_wakeup++;
1730 	}
1731 
1732 	return 0;
1733 }
1734 
1735 static int verify_xdp_configuration(struct net_device *dev)
1736 {
1737 	struct gve_priv *priv = netdev_priv(dev);
1738 
1739 	if (dev->features & NETIF_F_LRO) {
1740 		netdev_warn(dev, "XDP is not supported when LRO is on.\n");
1741 		return -EOPNOTSUPP;
1742 	}
1743 
1744 	if (priv->queue_format != GVE_GQI_QPL_FORMAT) {
1745 		netdev_warn(dev, "XDP is not supported in mode %d.\n",
1746 			    priv->queue_format);
1747 		return -EOPNOTSUPP;
1748 	}
1749 
1750 	if (dev->mtu > GVE_DEFAULT_RX_BUFFER_SIZE - sizeof(struct ethhdr) - GVE_RX_PAD) {
1751 		netdev_warn(dev, "XDP is not supported for mtu %d.\n",
1752 			    dev->mtu);
1753 		return -EOPNOTSUPP;
1754 	}
1755 
1756 	if (priv->rx_cfg.num_queues != priv->tx_cfg.num_queues ||
1757 	    (2 * priv->tx_cfg.num_queues > priv->tx_cfg.max_queues)) {
1758 		netdev_warn(dev, "XDP load failed: The number of configured RX queues %d should be equal to the number of configured TX queues %d and the number of configured RX/TX queues should be less than or equal to half the maximum number of RX/TX queues %d",
1759 			    priv->rx_cfg.num_queues,
1760 			    priv->tx_cfg.num_queues,
1761 			    priv->tx_cfg.max_queues);
1762 		return -EINVAL;
1763 	}
1764 	return 0;
1765 }
1766 
1767 static int gve_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1768 {
1769 	struct gve_priv *priv = netdev_priv(dev);
1770 	int err;
1771 
1772 	err = verify_xdp_configuration(dev);
1773 	if (err)
1774 		return err;
1775 	switch (xdp->command) {
1776 	case XDP_SETUP_PROG:
1777 		return gve_set_xdp(priv, xdp->prog, xdp->extack);
1778 	case XDP_SETUP_XSK_POOL:
1779 		if (xdp->xsk.pool)
1780 			return gve_xsk_pool_enable(dev, xdp->xsk.pool, xdp->xsk.queue_id);
1781 		else
1782 			return gve_xsk_pool_disable(dev, xdp->xsk.queue_id);
1783 	default:
1784 		return -EINVAL;
1785 	}
1786 }
1787 
1788 int gve_flow_rules_reset(struct gve_priv *priv)
1789 {
1790 	if (!priv->max_flow_rules)
1791 		return 0;
1792 
1793 	return gve_adminq_reset_flow_rules(priv);
1794 }
1795 
1796 int gve_adjust_config(struct gve_priv *priv,
1797 		      struct gve_tx_alloc_rings_cfg *tx_alloc_cfg,
1798 		      struct gve_rx_alloc_rings_cfg *rx_alloc_cfg)
1799 {
1800 	int err;
1801 
1802 	/* Allocate resources for the new confiugration */
1803 	err = gve_queues_mem_alloc(priv, tx_alloc_cfg, rx_alloc_cfg);
1804 	if (err) {
1805 		netif_err(priv, drv, priv->dev,
1806 			  "Adjust config failed to alloc new queues");
1807 		return err;
1808 	}
1809 
1810 	/* Teardown the device and free existing resources */
1811 	err = gve_close(priv->dev);
1812 	if (err) {
1813 		netif_err(priv, drv, priv->dev,
1814 			  "Adjust config failed to close old queues");
1815 		gve_queues_mem_free(priv, tx_alloc_cfg, rx_alloc_cfg);
1816 		return err;
1817 	}
1818 
1819 	/* Bring the device back up again with the new resources. */
1820 	err = gve_queues_start(priv, tx_alloc_cfg, rx_alloc_cfg);
1821 	if (err) {
1822 		netif_err(priv, drv, priv->dev,
1823 			  "Adjust config failed to start new queues, !!! DISABLING ALL QUEUES !!!\n");
1824 		/* No need to free on error: ownership of resources is lost after
1825 		 * calling gve_queues_start.
1826 		 */
1827 		gve_turndown(priv);
1828 		return err;
1829 	}
1830 
1831 	return 0;
1832 }
1833 
1834 int gve_adjust_queues(struct gve_priv *priv,
1835 		      struct gve_queue_config new_rx_config,
1836 		      struct gve_queue_config new_tx_config)
1837 {
1838 	struct gve_tx_alloc_rings_cfg tx_alloc_cfg = {0};
1839 	struct gve_rx_alloc_rings_cfg rx_alloc_cfg = {0};
1840 	int err;
1841 
1842 	gve_get_curr_alloc_cfgs(priv, &tx_alloc_cfg, &rx_alloc_cfg);
1843 
1844 	/* Relay the new config from ethtool */
1845 	tx_alloc_cfg.qcfg = &new_tx_config;
1846 	rx_alloc_cfg.qcfg_tx = &new_tx_config;
1847 	rx_alloc_cfg.qcfg = &new_rx_config;
1848 	tx_alloc_cfg.num_rings = new_tx_config.num_queues;
1849 
1850 	if (netif_running(priv->dev)) {
1851 		err = gve_adjust_config(priv, &tx_alloc_cfg, &rx_alloc_cfg);
1852 		return err;
1853 	}
1854 	/* Set the config for the next up. */
1855 	priv->tx_cfg = new_tx_config;
1856 	priv->rx_cfg = new_rx_config;
1857 
1858 	return 0;
1859 }
1860 
1861 static void gve_turndown(struct gve_priv *priv)
1862 {
1863 	int idx;
1864 
1865 	if (netif_carrier_ok(priv->dev))
1866 		netif_carrier_off(priv->dev);
1867 
1868 	if (!gve_get_napi_enabled(priv))
1869 		return;
1870 
1871 	/* Disable napi to prevent more work from coming in */
1872 	for (idx = 0; idx < gve_num_tx_queues(priv); idx++) {
1873 		int ntfy_idx = gve_tx_idx_to_ntfy(priv, idx);
1874 		struct gve_notify_block *block = &priv->ntfy_blocks[ntfy_idx];
1875 
1876 		if (!gve_tx_was_added_to_block(priv, idx))
1877 			continue;
1878 
1879 		if (idx < priv->tx_cfg.num_queues)
1880 			netif_queue_set_napi(priv->dev, idx,
1881 					     NETDEV_QUEUE_TYPE_TX, NULL);
1882 
1883 		napi_disable(&block->napi);
1884 	}
1885 	for (idx = 0; idx < priv->rx_cfg.num_queues; idx++) {
1886 		int ntfy_idx = gve_rx_idx_to_ntfy(priv, idx);
1887 		struct gve_notify_block *block = &priv->ntfy_blocks[ntfy_idx];
1888 
1889 		if (!gve_rx_was_added_to_block(priv, idx))
1890 			continue;
1891 
1892 		netif_queue_set_napi(priv->dev, idx, NETDEV_QUEUE_TYPE_RX,
1893 				     NULL);
1894 		napi_disable(&block->napi);
1895 	}
1896 
1897 	/* Stop tx queues */
1898 	netif_tx_disable(priv->dev);
1899 
1900 	gve_clear_napi_enabled(priv);
1901 	gve_clear_report_stats(priv);
1902 }
1903 
1904 static void gve_turnup(struct gve_priv *priv)
1905 {
1906 	int idx;
1907 
1908 	/* Start the tx queues */
1909 	netif_tx_start_all_queues(priv->dev);
1910 
1911 	/* Enable napi and unmask interrupts for all queues */
1912 	for (idx = 0; idx < gve_num_tx_queues(priv); idx++) {
1913 		int ntfy_idx = gve_tx_idx_to_ntfy(priv, idx);
1914 		struct gve_notify_block *block = &priv->ntfy_blocks[ntfy_idx];
1915 
1916 		if (!gve_tx_was_added_to_block(priv, idx))
1917 			continue;
1918 
1919 		napi_enable(&block->napi);
1920 
1921 		if (idx < priv->tx_cfg.num_queues)
1922 			netif_queue_set_napi(priv->dev, idx,
1923 					     NETDEV_QUEUE_TYPE_TX,
1924 					     &block->napi);
1925 
1926 		if (gve_is_gqi(priv)) {
1927 			iowrite32be(0, gve_irq_doorbell(priv, block));
1928 		} else {
1929 			gve_set_itr_coalesce_usecs_dqo(priv, block,
1930 						       priv->tx_coalesce_usecs);
1931 		}
1932 
1933 		/* Any descs written by the NIC before this barrier will be
1934 		 * handled by the one-off napi schedule below. Whereas any
1935 		 * descs after the barrier will generate interrupts.
1936 		 */
1937 		mb();
1938 		napi_schedule(&block->napi);
1939 	}
1940 	for (idx = 0; idx < priv->rx_cfg.num_queues; idx++) {
1941 		int ntfy_idx = gve_rx_idx_to_ntfy(priv, idx);
1942 		struct gve_notify_block *block = &priv->ntfy_blocks[ntfy_idx];
1943 
1944 		if (!gve_rx_was_added_to_block(priv, idx))
1945 			continue;
1946 
1947 		napi_enable(&block->napi);
1948 		netif_queue_set_napi(priv->dev, idx, NETDEV_QUEUE_TYPE_RX,
1949 				     &block->napi);
1950 
1951 		if (gve_is_gqi(priv)) {
1952 			iowrite32be(0, gve_irq_doorbell(priv, block));
1953 		} else {
1954 			gve_set_itr_coalesce_usecs_dqo(priv, block,
1955 						       priv->rx_coalesce_usecs);
1956 		}
1957 
1958 		/* Any descs written by the NIC before this barrier will be
1959 		 * handled by the one-off napi schedule below. Whereas any
1960 		 * descs after the barrier will generate interrupts.
1961 		 */
1962 		mb();
1963 		napi_schedule(&block->napi);
1964 	}
1965 
1966 	gve_set_napi_enabled(priv);
1967 }
1968 
1969 static void gve_turnup_and_check_status(struct gve_priv *priv)
1970 {
1971 	u32 status;
1972 
1973 	gve_turnup(priv);
1974 	status = ioread32be(&priv->reg_bar0->device_status);
1975 	gve_handle_link_status(priv, GVE_DEVICE_STATUS_LINK_STATUS_MASK & status);
1976 }
1977 
1978 static void gve_tx_timeout(struct net_device *dev, unsigned int txqueue)
1979 {
1980 	struct gve_notify_block *block;
1981 	struct gve_tx_ring *tx = NULL;
1982 	struct gve_priv *priv;
1983 	u32 last_nic_done;
1984 	u32 current_time;
1985 	u32 ntfy_idx;
1986 
1987 	netdev_info(dev, "Timeout on tx queue, %d", txqueue);
1988 	priv = netdev_priv(dev);
1989 	if (txqueue > priv->tx_cfg.num_queues)
1990 		goto reset;
1991 
1992 	ntfy_idx = gve_tx_idx_to_ntfy(priv, txqueue);
1993 	if (ntfy_idx >= priv->num_ntfy_blks)
1994 		goto reset;
1995 
1996 	block = &priv->ntfy_blocks[ntfy_idx];
1997 	tx = block->tx;
1998 
1999 	current_time = jiffies_to_msecs(jiffies);
2000 	if (tx->last_kick_msec + MIN_TX_TIMEOUT_GAP > current_time)
2001 		goto reset;
2002 
2003 	/* Check to see if there are missed completions, which will allow us to
2004 	 * kick the queue.
2005 	 */
2006 	last_nic_done = gve_tx_load_event_counter(priv, tx);
2007 	if (last_nic_done - tx->done) {
2008 		netdev_info(dev, "Kicking queue %d", txqueue);
2009 		iowrite32be(GVE_IRQ_MASK, gve_irq_doorbell(priv, block));
2010 		napi_schedule(&block->napi);
2011 		tx->last_kick_msec = current_time;
2012 		goto out;
2013 	} // Else reset.
2014 
2015 reset:
2016 	gve_schedule_reset(priv);
2017 
2018 out:
2019 	if (tx)
2020 		tx->queue_timeout++;
2021 	priv->tx_timeo_cnt++;
2022 }
2023 
2024 u16 gve_get_pkt_buf_size(const struct gve_priv *priv, bool enable_hsplit)
2025 {
2026 	if (enable_hsplit && priv->max_rx_buffer_size >= GVE_MAX_RX_BUFFER_SIZE)
2027 		return GVE_MAX_RX_BUFFER_SIZE;
2028 	else
2029 		return GVE_DEFAULT_RX_BUFFER_SIZE;
2030 }
2031 
2032 /* header-split is not supported on non-DQO_RDA yet even if device advertises it */
2033 bool gve_header_split_supported(const struct gve_priv *priv)
2034 {
2035 	return priv->header_buf_size && priv->queue_format == GVE_DQO_RDA_FORMAT;
2036 }
2037 
2038 int gve_set_hsplit_config(struct gve_priv *priv, u8 tcp_data_split)
2039 {
2040 	struct gve_tx_alloc_rings_cfg tx_alloc_cfg = {0};
2041 	struct gve_rx_alloc_rings_cfg rx_alloc_cfg = {0};
2042 	bool enable_hdr_split;
2043 	int err = 0;
2044 
2045 	if (tcp_data_split == ETHTOOL_TCP_DATA_SPLIT_UNKNOWN)
2046 		return 0;
2047 
2048 	if (!gve_header_split_supported(priv)) {
2049 		dev_err(&priv->pdev->dev, "Header-split not supported\n");
2050 		return -EOPNOTSUPP;
2051 	}
2052 
2053 	if (tcp_data_split == ETHTOOL_TCP_DATA_SPLIT_ENABLED)
2054 		enable_hdr_split = true;
2055 	else
2056 		enable_hdr_split = false;
2057 
2058 	if (enable_hdr_split == priv->header_split_enabled)
2059 		return 0;
2060 
2061 	gve_get_curr_alloc_cfgs(priv, &tx_alloc_cfg, &rx_alloc_cfg);
2062 
2063 	rx_alloc_cfg.enable_header_split = enable_hdr_split;
2064 	rx_alloc_cfg.packet_buffer_size = gve_get_pkt_buf_size(priv, enable_hdr_split);
2065 
2066 	if (netif_running(priv->dev))
2067 		err = gve_adjust_config(priv, &tx_alloc_cfg, &rx_alloc_cfg);
2068 	return err;
2069 }
2070 
2071 static int gve_set_features(struct net_device *netdev,
2072 			    netdev_features_t features)
2073 {
2074 	const netdev_features_t orig_features = netdev->features;
2075 	struct gve_tx_alloc_rings_cfg tx_alloc_cfg = {0};
2076 	struct gve_rx_alloc_rings_cfg rx_alloc_cfg = {0};
2077 	struct gve_priv *priv = netdev_priv(netdev);
2078 	int err;
2079 
2080 	gve_get_curr_alloc_cfgs(priv, &tx_alloc_cfg, &rx_alloc_cfg);
2081 
2082 	if ((netdev->features & NETIF_F_LRO) != (features & NETIF_F_LRO)) {
2083 		netdev->features ^= NETIF_F_LRO;
2084 		if (netif_running(netdev)) {
2085 			err = gve_adjust_config(priv, &tx_alloc_cfg, &rx_alloc_cfg);
2086 			if (err)
2087 				goto revert_features;
2088 		}
2089 	}
2090 	if ((netdev->features & NETIF_F_NTUPLE) && !(features & NETIF_F_NTUPLE)) {
2091 		err = gve_flow_rules_reset(priv);
2092 		if (err)
2093 			goto revert_features;
2094 	}
2095 
2096 	return 0;
2097 
2098 revert_features:
2099 	netdev->features = orig_features;
2100 	return err;
2101 }
2102 
2103 static const struct net_device_ops gve_netdev_ops = {
2104 	.ndo_start_xmit		=	gve_start_xmit,
2105 	.ndo_features_check	=	gve_features_check,
2106 	.ndo_open		=	gve_open,
2107 	.ndo_stop		=	gve_close,
2108 	.ndo_get_stats64	=	gve_get_stats,
2109 	.ndo_tx_timeout         =       gve_tx_timeout,
2110 	.ndo_set_features	=	gve_set_features,
2111 	.ndo_bpf		=	gve_xdp,
2112 	.ndo_xdp_xmit		=	gve_xdp_xmit,
2113 	.ndo_xsk_wakeup		=	gve_xsk_wakeup,
2114 };
2115 
2116 static void gve_handle_status(struct gve_priv *priv, u32 status)
2117 {
2118 	if (GVE_DEVICE_STATUS_RESET_MASK & status) {
2119 		dev_info(&priv->pdev->dev, "Device requested reset.\n");
2120 		gve_set_do_reset(priv);
2121 	}
2122 	if (GVE_DEVICE_STATUS_REPORT_STATS_MASK & status) {
2123 		priv->stats_report_trigger_cnt++;
2124 		gve_set_do_report_stats(priv);
2125 	}
2126 }
2127 
2128 static void gve_handle_reset(struct gve_priv *priv)
2129 {
2130 	/* A service task will be scheduled at the end of probe to catch any
2131 	 * resets that need to happen, and we don't want to reset until
2132 	 * probe is done.
2133 	 */
2134 	if (gve_get_probe_in_progress(priv))
2135 		return;
2136 
2137 	if (gve_get_do_reset(priv)) {
2138 		rtnl_lock();
2139 		gve_reset(priv, false);
2140 		rtnl_unlock();
2141 	}
2142 }
2143 
2144 void gve_handle_report_stats(struct gve_priv *priv)
2145 {
2146 	struct stats *stats = priv->stats_report->stats;
2147 	int idx, stats_idx = 0;
2148 	unsigned int start = 0;
2149 	u64 tx_bytes;
2150 
2151 	if (!gve_get_report_stats(priv))
2152 		return;
2153 
2154 	be64_add_cpu(&priv->stats_report->written_count, 1);
2155 	/* tx stats */
2156 	if (priv->tx) {
2157 		for (idx = 0; idx < gve_num_tx_queues(priv); idx++) {
2158 			u32 last_completion = 0;
2159 			u32 tx_frames = 0;
2160 
2161 			/* DQO doesn't currently support these metrics. */
2162 			if (gve_is_gqi(priv)) {
2163 				last_completion = priv->tx[idx].done;
2164 				tx_frames = priv->tx[idx].req;
2165 			}
2166 
2167 			do {
2168 				start = u64_stats_fetch_begin(&priv->tx[idx].statss);
2169 				tx_bytes = priv->tx[idx].bytes_done;
2170 			} while (u64_stats_fetch_retry(&priv->tx[idx].statss, start));
2171 			stats[stats_idx++] = (struct stats) {
2172 				.stat_name = cpu_to_be32(TX_WAKE_CNT),
2173 				.value = cpu_to_be64(priv->tx[idx].wake_queue),
2174 				.queue_id = cpu_to_be32(idx),
2175 			};
2176 			stats[stats_idx++] = (struct stats) {
2177 				.stat_name = cpu_to_be32(TX_STOP_CNT),
2178 				.value = cpu_to_be64(priv->tx[idx].stop_queue),
2179 				.queue_id = cpu_to_be32(idx),
2180 			};
2181 			stats[stats_idx++] = (struct stats) {
2182 				.stat_name = cpu_to_be32(TX_FRAMES_SENT),
2183 				.value = cpu_to_be64(tx_frames),
2184 				.queue_id = cpu_to_be32(idx),
2185 			};
2186 			stats[stats_idx++] = (struct stats) {
2187 				.stat_name = cpu_to_be32(TX_BYTES_SENT),
2188 				.value = cpu_to_be64(tx_bytes),
2189 				.queue_id = cpu_to_be32(idx),
2190 			};
2191 			stats[stats_idx++] = (struct stats) {
2192 				.stat_name = cpu_to_be32(TX_LAST_COMPLETION_PROCESSED),
2193 				.value = cpu_to_be64(last_completion),
2194 				.queue_id = cpu_to_be32(idx),
2195 			};
2196 			stats[stats_idx++] = (struct stats) {
2197 				.stat_name = cpu_to_be32(TX_TIMEOUT_CNT),
2198 				.value = cpu_to_be64(priv->tx[idx].queue_timeout),
2199 				.queue_id = cpu_to_be32(idx),
2200 			};
2201 		}
2202 	}
2203 	/* rx stats */
2204 	if (priv->rx) {
2205 		for (idx = 0; idx < priv->rx_cfg.num_queues; idx++) {
2206 			stats[stats_idx++] = (struct stats) {
2207 				.stat_name = cpu_to_be32(RX_NEXT_EXPECTED_SEQUENCE),
2208 				.value = cpu_to_be64(priv->rx[idx].desc.seqno),
2209 				.queue_id = cpu_to_be32(idx),
2210 			};
2211 			stats[stats_idx++] = (struct stats) {
2212 				.stat_name = cpu_to_be32(RX_BUFFERS_POSTED),
2213 				.value = cpu_to_be64(priv->rx[0].fill_cnt),
2214 				.queue_id = cpu_to_be32(idx),
2215 			};
2216 		}
2217 	}
2218 }
2219 
2220 /* Handle NIC status register changes, reset requests and report stats */
2221 static void gve_service_task(struct work_struct *work)
2222 {
2223 	struct gve_priv *priv = container_of(work, struct gve_priv,
2224 					     service_task);
2225 	u32 status = ioread32be(&priv->reg_bar0->device_status);
2226 
2227 	gve_handle_status(priv, status);
2228 
2229 	gve_handle_reset(priv);
2230 	gve_handle_link_status(priv, GVE_DEVICE_STATUS_LINK_STATUS_MASK & status);
2231 }
2232 
2233 static void gve_set_netdev_xdp_features(struct gve_priv *priv)
2234 {
2235 	if (priv->queue_format == GVE_GQI_QPL_FORMAT) {
2236 		priv->dev->xdp_features = NETDEV_XDP_ACT_BASIC;
2237 		priv->dev->xdp_features |= NETDEV_XDP_ACT_REDIRECT;
2238 		priv->dev->xdp_features |= NETDEV_XDP_ACT_NDO_XMIT;
2239 		priv->dev->xdp_features |= NETDEV_XDP_ACT_XSK_ZEROCOPY;
2240 	} else {
2241 		priv->dev->xdp_features = 0;
2242 	}
2243 }
2244 
2245 static int gve_init_priv(struct gve_priv *priv, bool skip_describe_device)
2246 {
2247 	int num_ntfy;
2248 	int err;
2249 
2250 	/* Set up the adminq */
2251 	err = gve_adminq_alloc(&priv->pdev->dev, priv);
2252 	if (err) {
2253 		dev_err(&priv->pdev->dev,
2254 			"Failed to alloc admin queue: err=%d\n", err);
2255 		return err;
2256 	}
2257 
2258 	err = gve_verify_driver_compatibility(priv);
2259 	if (err) {
2260 		dev_err(&priv->pdev->dev,
2261 			"Could not verify driver compatibility: err=%d\n", err);
2262 		goto err;
2263 	}
2264 
2265 	priv->num_registered_pages = 0;
2266 
2267 	if (skip_describe_device)
2268 		goto setup_device;
2269 
2270 	priv->queue_format = GVE_QUEUE_FORMAT_UNSPECIFIED;
2271 	/* Get the initial information we need from the device */
2272 	err = gve_adminq_describe_device(priv);
2273 	if (err) {
2274 		dev_err(&priv->pdev->dev,
2275 			"Could not get device information: err=%d\n", err);
2276 		goto err;
2277 	}
2278 	priv->dev->mtu = priv->dev->max_mtu;
2279 	num_ntfy = pci_msix_vec_count(priv->pdev);
2280 	if (num_ntfy <= 0) {
2281 		dev_err(&priv->pdev->dev,
2282 			"could not count MSI-x vectors: err=%d\n", num_ntfy);
2283 		err = num_ntfy;
2284 		goto err;
2285 	} else if (num_ntfy < GVE_MIN_MSIX) {
2286 		dev_err(&priv->pdev->dev, "gve needs at least %d MSI-x vectors, but only has %d\n",
2287 			GVE_MIN_MSIX, num_ntfy);
2288 		err = -EINVAL;
2289 		goto err;
2290 	}
2291 
2292 	/* Big TCP is only supported on DQ*/
2293 	if (!gve_is_gqi(priv))
2294 		netif_set_tso_max_size(priv->dev, GVE_DQO_TX_MAX);
2295 
2296 	priv->rx_copybreak = GVE_DEFAULT_RX_COPYBREAK;
2297 	/* gvnic has one Notification Block per MSI-x vector, except for the
2298 	 * management vector
2299 	 */
2300 	priv->num_ntfy_blks = (num_ntfy - 1) & ~0x1;
2301 	priv->mgmt_msix_idx = priv->num_ntfy_blks;
2302 
2303 	priv->tx_cfg.max_queues =
2304 		min_t(int, priv->tx_cfg.max_queues, priv->num_ntfy_blks / 2);
2305 	priv->rx_cfg.max_queues =
2306 		min_t(int, priv->rx_cfg.max_queues, priv->num_ntfy_blks / 2);
2307 
2308 	priv->tx_cfg.num_queues = priv->tx_cfg.max_queues;
2309 	priv->rx_cfg.num_queues = priv->rx_cfg.max_queues;
2310 	if (priv->default_num_queues > 0) {
2311 		priv->tx_cfg.num_queues = min_t(int, priv->default_num_queues,
2312 						priv->tx_cfg.num_queues);
2313 		priv->rx_cfg.num_queues = min_t(int, priv->default_num_queues,
2314 						priv->rx_cfg.num_queues);
2315 	}
2316 
2317 	dev_info(&priv->pdev->dev, "TX queues %d, RX queues %d\n",
2318 		 priv->tx_cfg.num_queues, priv->rx_cfg.num_queues);
2319 	dev_info(&priv->pdev->dev, "Max TX queues %d, Max RX queues %d\n",
2320 		 priv->tx_cfg.max_queues, priv->rx_cfg.max_queues);
2321 
2322 	if (!gve_is_gqi(priv)) {
2323 		priv->tx_coalesce_usecs = GVE_TX_IRQ_RATELIMIT_US_DQO;
2324 		priv->rx_coalesce_usecs = GVE_RX_IRQ_RATELIMIT_US_DQO;
2325 	}
2326 
2327 setup_device:
2328 	gve_set_netdev_xdp_features(priv);
2329 	err = gve_setup_device_resources(priv);
2330 	if (!err)
2331 		return 0;
2332 err:
2333 	gve_adminq_free(&priv->pdev->dev, priv);
2334 	return err;
2335 }
2336 
2337 static void gve_teardown_priv_resources(struct gve_priv *priv)
2338 {
2339 	gve_teardown_device_resources(priv);
2340 	gve_adminq_free(&priv->pdev->dev, priv);
2341 }
2342 
2343 static void gve_trigger_reset(struct gve_priv *priv)
2344 {
2345 	/* Reset the device by releasing the AQ */
2346 	gve_adminq_release(priv);
2347 }
2348 
2349 static void gve_reset_and_teardown(struct gve_priv *priv, bool was_up)
2350 {
2351 	gve_trigger_reset(priv);
2352 	/* With the reset having already happened, close cannot fail */
2353 	if (was_up)
2354 		gve_close(priv->dev);
2355 	gve_teardown_priv_resources(priv);
2356 }
2357 
2358 static int gve_reset_recovery(struct gve_priv *priv, bool was_up)
2359 {
2360 	int err;
2361 
2362 	err = gve_init_priv(priv, true);
2363 	if (err)
2364 		goto err;
2365 	if (was_up) {
2366 		err = gve_open(priv->dev);
2367 		if (err)
2368 			goto err;
2369 	}
2370 	return 0;
2371 err:
2372 	dev_err(&priv->pdev->dev, "Reset failed! !!! DISABLING ALL QUEUES !!!\n");
2373 	gve_turndown(priv);
2374 	return err;
2375 }
2376 
2377 int gve_reset(struct gve_priv *priv, bool attempt_teardown)
2378 {
2379 	bool was_up = netif_running(priv->dev);
2380 	int err;
2381 
2382 	dev_info(&priv->pdev->dev, "Performing reset\n");
2383 	gve_clear_do_reset(priv);
2384 	gve_set_reset_in_progress(priv);
2385 	/* If we aren't attempting to teardown normally, just go turndown and
2386 	 * reset right away.
2387 	 */
2388 	if (!attempt_teardown) {
2389 		gve_turndown(priv);
2390 		gve_reset_and_teardown(priv, was_up);
2391 	} else {
2392 		/* Otherwise attempt to close normally */
2393 		if (was_up) {
2394 			err = gve_close(priv->dev);
2395 			/* If that fails reset as we did above */
2396 			if (err)
2397 				gve_reset_and_teardown(priv, was_up);
2398 		}
2399 		/* Clean up any remaining resources */
2400 		gve_teardown_priv_resources(priv);
2401 	}
2402 
2403 	/* Set it all back up */
2404 	err = gve_reset_recovery(priv, was_up);
2405 	gve_clear_reset_in_progress(priv);
2406 	priv->reset_cnt++;
2407 	priv->interface_up_cnt = 0;
2408 	priv->interface_down_cnt = 0;
2409 	priv->stats_report_trigger_cnt = 0;
2410 	return err;
2411 }
2412 
2413 static void gve_write_version(u8 __iomem *driver_version_register)
2414 {
2415 	const char *c = gve_version_prefix;
2416 
2417 	while (*c) {
2418 		writeb(*c, driver_version_register);
2419 		c++;
2420 	}
2421 
2422 	c = gve_version_str;
2423 	while (*c) {
2424 		writeb(*c, driver_version_register);
2425 		c++;
2426 	}
2427 	writeb('\n', driver_version_register);
2428 }
2429 
2430 static int gve_rx_queue_stop(struct net_device *dev, void *per_q_mem, int idx)
2431 {
2432 	struct gve_priv *priv = netdev_priv(dev);
2433 	struct gve_rx_ring *gve_per_q_mem;
2434 	int err;
2435 
2436 	if (!priv->rx)
2437 		return -EAGAIN;
2438 
2439 	/* Destroying queue 0 while other queues exist is not supported in DQO */
2440 	if (!gve_is_gqi(priv) && idx == 0)
2441 		return -ERANGE;
2442 
2443 	/* Single-queue destruction requires quiescence on all queues */
2444 	gve_turndown(priv);
2445 
2446 	/* This failure will trigger a reset - no need to clean up */
2447 	err = gve_adminq_destroy_single_rx_queue(priv, idx);
2448 	if (err)
2449 		return err;
2450 
2451 	if (gve_is_qpl(priv)) {
2452 		/* This failure will trigger a reset - no need to clean up */
2453 		err = gve_unregister_qpl(priv, gve_rx_get_qpl(priv, idx));
2454 		if (err)
2455 			return err;
2456 	}
2457 
2458 	gve_rx_stop_ring(priv, idx);
2459 
2460 	/* Turn the unstopped queues back up */
2461 	gve_turnup_and_check_status(priv);
2462 
2463 	gve_per_q_mem = (struct gve_rx_ring *)per_q_mem;
2464 	*gve_per_q_mem = priv->rx[idx];
2465 	memset(&priv->rx[idx], 0, sizeof(priv->rx[idx]));
2466 	return 0;
2467 }
2468 
2469 static void gve_rx_queue_mem_free(struct net_device *dev, void *per_q_mem)
2470 {
2471 	struct gve_priv *priv = netdev_priv(dev);
2472 	struct gve_rx_alloc_rings_cfg cfg = {0};
2473 	struct gve_rx_ring *gve_per_q_mem;
2474 
2475 	gve_per_q_mem = (struct gve_rx_ring *)per_q_mem;
2476 	gve_rx_get_curr_alloc_cfg(priv, &cfg);
2477 
2478 	if (gve_is_gqi(priv))
2479 		gve_rx_free_ring_gqi(priv, gve_per_q_mem, &cfg);
2480 	else
2481 		gve_rx_free_ring_dqo(priv, gve_per_q_mem, &cfg);
2482 }
2483 
2484 static int gve_rx_queue_mem_alloc(struct net_device *dev, void *per_q_mem,
2485 				  int idx)
2486 {
2487 	struct gve_priv *priv = netdev_priv(dev);
2488 	struct gve_rx_alloc_rings_cfg cfg = {0};
2489 	struct gve_rx_ring *gve_per_q_mem;
2490 	int err;
2491 
2492 	if (!priv->rx)
2493 		return -EAGAIN;
2494 
2495 	gve_per_q_mem = (struct gve_rx_ring *)per_q_mem;
2496 	gve_rx_get_curr_alloc_cfg(priv, &cfg);
2497 
2498 	if (gve_is_gqi(priv))
2499 		err = gve_rx_alloc_ring_gqi(priv, &cfg, gve_per_q_mem, idx);
2500 	else
2501 		err = gve_rx_alloc_ring_dqo(priv, &cfg, gve_per_q_mem, idx);
2502 
2503 	return err;
2504 }
2505 
2506 static int gve_rx_queue_start(struct net_device *dev, void *per_q_mem, int idx)
2507 {
2508 	struct gve_priv *priv = netdev_priv(dev);
2509 	struct gve_rx_ring *gve_per_q_mem;
2510 	int err;
2511 
2512 	if (!priv->rx)
2513 		return -EAGAIN;
2514 
2515 	gve_per_q_mem = (struct gve_rx_ring *)per_q_mem;
2516 	priv->rx[idx] = *gve_per_q_mem;
2517 
2518 	/* Single-queue creation requires quiescence on all queues */
2519 	gve_turndown(priv);
2520 
2521 	gve_rx_start_ring(priv, idx);
2522 
2523 	if (gve_is_qpl(priv)) {
2524 		/* This failure will trigger a reset - no need to clean up */
2525 		err = gve_register_qpl(priv, gve_rx_get_qpl(priv, idx));
2526 		if (err)
2527 			goto abort;
2528 	}
2529 
2530 	/* This failure will trigger a reset - no need to clean up */
2531 	err = gve_adminq_create_single_rx_queue(priv, idx);
2532 	if (err)
2533 		goto abort;
2534 
2535 	if (gve_is_gqi(priv))
2536 		gve_rx_write_doorbell(priv, &priv->rx[idx]);
2537 	else
2538 		gve_rx_post_buffers_dqo(&priv->rx[idx]);
2539 
2540 	/* Turn the unstopped queues back up */
2541 	gve_turnup_and_check_status(priv);
2542 	return 0;
2543 
2544 abort:
2545 	gve_rx_stop_ring(priv, idx);
2546 
2547 	/* All failures in this func result in a reset, by clearing the struct
2548 	 * at idx, we prevent a double free when that reset runs. The reset,
2549 	 * which needs the rtnl lock, will not run till this func returns and
2550 	 * its caller gives up the lock.
2551 	 */
2552 	memset(&priv->rx[idx], 0, sizeof(priv->rx[idx]));
2553 	return err;
2554 }
2555 
2556 static const struct netdev_queue_mgmt_ops gve_queue_mgmt_ops = {
2557 	.ndo_queue_mem_size	=	sizeof(struct gve_rx_ring),
2558 	.ndo_queue_mem_alloc	=	gve_rx_queue_mem_alloc,
2559 	.ndo_queue_mem_free	=	gve_rx_queue_mem_free,
2560 	.ndo_queue_start	=	gve_rx_queue_start,
2561 	.ndo_queue_stop		=	gve_rx_queue_stop,
2562 };
2563 
2564 static void gve_get_rx_queue_stats(struct net_device *dev, int idx,
2565 				   struct netdev_queue_stats_rx *rx_stats)
2566 {
2567 	struct gve_priv *priv = netdev_priv(dev);
2568 	struct gve_rx_ring *rx = &priv->rx[idx];
2569 	unsigned int start;
2570 
2571 	do {
2572 		start = u64_stats_fetch_begin(&rx->statss);
2573 		rx_stats->packets = rx->rpackets;
2574 		rx_stats->bytes = rx->rbytes;
2575 		rx_stats->alloc_fail = rx->rx_skb_alloc_fail +
2576 				       rx->rx_buf_alloc_fail;
2577 	} while (u64_stats_fetch_retry(&rx->statss, start));
2578 }
2579 
2580 static void gve_get_tx_queue_stats(struct net_device *dev, int idx,
2581 				   struct netdev_queue_stats_tx *tx_stats)
2582 {
2583 	struct gve_priv *priv = netdev_priv(dev);
2584 	struct gve_tx_ring *tx = &priv->tx[idx];
2585 	unsigned int start;
2586 
2587 	do {
2588 		start = u64_stats_fetch_begin(&tx->statss);
2589 		tx_stats->packets = tx->pkt_done;
2590 		tx_stats->bytes = tx->bytes_done;
2591 	} while (u64_stats_fetch_retry(&tx->statss, start));
2592 }
2593 
2594 static void gve_get_base_stats(struct net_device *dev,
2595 			       struct netdev_queue_stats_rx *rx,
2596 			       struct netdev_queue_stats_tx *tx)
2597 {
2598 	rx->packets = 0;
2599 	rx->bytes = 0;
2600 	rx->alloc_fail = 0;
2601 
2602 	tx->packets = 0;
2603 	tx->bytes = 0;
2604 }
2605 
2606 static const struct netdev_stat_ops gve_stat_ops = {
2607 	.get_queue_stats_rx	= gve_get_rx_queue_stats,
2608 	.get_queue_stats_tx	= gve_get_tx_queue_stats,
2609 	.get_base_stats		= gve_get_base_stats,
2610 };
2611 
2612 static int gve_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2613 {
2614 	int max_tx_queues, max_rx_queues;
2615 	struct net_device *dev;
2616 	__be32 __iomem *db_bar;
2617 	struct gve_registers __iomem *reg_bar;
2618 	struct gve_priv *priv;
2619 	int err;
2620 
2621 	err = pci_enable_device(pdev);
2622 	if (err)
2623 		return err;
2624 
2625 	err = pci_request_regions(pdev, gve_driver_name);
2626 	if (err)
2627 		goto abort_with_enabled;
2628 
2629 	pci_set_master(pdev);
2630 
2631 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2632 	if (err) {
2633 		dev_err(&pdev->dev, "Failed to set dma mask: err=%d\n", err);
2634 		goto abort_with_pci_region;
2635 	}
2636 
2637 	reg_bar = pci_iomap(pdev, GVE_REGISTER_BAR, 0);
2638 	if (!reg_bar) {
2639 		dev_err(&pdev->dev, "Failed to map pci bar!\n");
2640 		err = -ENOMEM;
2641 		goto abort_with_pci_region;
2642 	}
2643 
2644 	db_bar = pci_iomap(pdev, GVE_DOORBELL_BAR, 0);
2645 	if (!db_bar) {
2646 		dev_err(&pdev->dev, "Failed to map doorbell bar!\n");
2647 		err = -ENOMEM;
2648 		goto abort_with_reg_bar;
2649 	}
2650 
2651 	gve_write_version(&reg_bar->driver_version);
2652 	/* Get max queues to alloc etherdev */
2653 	max_tx_queues = ioread32be(&reg_bar->max_tx_queues);
2654 	max_rx_queues = ioread32be(&reg_bar->max_rx_queues);
2655 	/* Alloc and setup the netdev and priv */
2656 	dev = alloc_etherdev_mqs(sizeof(*priv), max_tx_queues, max_rx_queues);
2657 	if (!dev) {
2658 		dev_err(&pdev->dev, "could not allocate netdev\n");
2659 		err = -ENOMEM;
2660 		goto abort_with_db_bar;
2661 	}
2662 	SET_NETDEV_DEV(dev, &pdev->dev);
2663 	pci_set_drvdata(pdev, dev);
2664 	dev->ethtool_ops = &gve_ethtool_ops;
2665 	dev->netdev_ops = &gve_netdev_ops;
2666 	dev->queue_mgmt_ops = &gve_queue_mgmt_ops;
2667 	dev->stat_ops = &gve_stat_ops;
2668 
2669 	/* Set default and supported features.
2670 	 *
2671 	 * Features might be set in other locations as well (such as
2672 	 * `gve_adminq_describe_device`).
2673 	 */
2674 	dev->hw_features = NETIF_F_HIGHDMA;
2675 	dev->hw_features |= NETIF_F_SG;
2676 	dev->hw_features |= NETIF_F_HW_CSUM;
2677 	dev->hw_features |= NETIF_F_TSO;
2678 	dev->hw_features |= NETIF_F_TSO6;
2679 	dev->hw_features |= NETIF_F_TSO_ECN;
2680 	dev->hw_features |= NETIF_F_RXCSUM;
2681 	dev->hw_features |= NETIF_F_RXHASH;
2682 	dev->features = dev->hw_features;
2683 	dev->watchdog_timeo = 5 * HZ;
2684 	dev->min_mtu = ETH_MIN_MTU;
2685 	netif_carrier_off(dev);
2686 
2687 	priv = netdev_priv(dev);
2688 	priv->dev = dev;
2689 	priv->pdev = pdev;
2690 	priv->msg_enable = DEFAULT_MSG_LEVEL;
2691 	priv->reg_bar0 = reg_bar;
2692 	priv->db_bar2 = db_bar;
2693 	priv->service_task_flags = 0x0;
2694 	priv->state_flags = 0x0;
2695 	priv->ethtool_flags = 0x0;
2696 	priv->data_buffer_size_dqo = GVE_DEFAULT_RX_BUFFER_SIZE;
2697 	priv->max_rx_buffer_size = GVE_DEFAULT_RX_BUFFER_SIZE;
2698 
2699 	gve_set_probe_in_progress(priv);
2700 	priv->gve_wq = alloc_ordered_workqueue("gve", 0);
2701 	if (!priv->gve_wq) {
2702 		dev_err(&pdev->dev, "Could not allocate workqueue");
2703 		err = -ENOMEM;
2704 		goto abort_with_netdev;
2705 	}
2706 	INIT_WORK(&priv->service_task, gve_service_task);
2707 	INIT_WORK(&priv->stats_report_task, gve_stats_report_task);
2708 	priv->tx_cfg.max_queues = max_tx_queues;
2709 	priv->rx_cfg.max_queues = max_rx_queues;
2710 
2711 	err = gve_init_priv(priv, false);
2712 	if (err)
2713 		goto abort_with_wq;
2714 
2715 	err = register_netdev(dev);
2716 	if (err)
2717 		goto abort_with_gve_init;
2718 
2719 	dev_info(&pdev->dev, "GVE version %s\n", gve_version_str);
2720 	dev_info(&pdev->dev, "GVE queue format %d\n", (int)priv->queue_format);
2721 	gve_clear_probe_in_progress(priv);
2722 	queue_work(priv->gve_wq, &priv->service_task);
2723 	return 0;
2724 
2725 abort_with_gve_init:
2726 	gve_teardown_priv_resources(priv);
2727 
2728 abort_with_wq:
2729 	destroy_workqueue(priv->gve_wq);
2730 
2731 abort_with_netdev:
2732 	free_netdev(dev);
2733 
2734 abort_with_db_bar:
2735 	pci_iounmap(pdev, db_bar);
2736 
2737 abort_with_reg_bar:
2738 	pci_iounmap(pdev, reg_bar);
2739 
2740 abort_with_pci_region:
2741 	pci_release_regions(pdev);
2742 
2743 abort_with_enabled:
2744 	pci_disable_device(pdev);
2745 	return err;
2746 }
2747 
2748 static void gve_remove(struct pci_dev *pdev)
2749 {
2750 	struct net_device *netdev = pci_get_drvdata(pdev);
2751 	struct gve_priv *priv = netdev_priv(netdev);
2752 	__be32 __iomem *db_bar = priv->db_bar2;
2753 	void __iomem *reg_bar = priv->reg_bar0;
2754 
2755 	unregister_netdev(netdev);
2756 	gve_teardown_priv_resources(priv);
2757 	destroy_workqueue(priv->gve_wq);
2758 	free_netdev(netdev);
2759 	pci_iounmap(pdev, db_bar);
2760 	pci_iounmap(pdev, reg_bar);
2761 	pci_release_regions(pdev);
2762 	pci_disable_device(pdev);
2763 }
2764 
2765 static void gve_shutdown(struct pci_dev *pdev)
2766 {
2767 	struct net_device *netdev = pci_get_drvdata(pdev);
2768 	struct gve_priv *priv = netdev_priv(netdev);
2769 	bool was_up = netif_running(priv->dev);
2770 
2771 	rtnl_lock();
2772 	if (was_up && gve_close(priv->dev)) {
2773 		/* If the dev was up, attempt to close, if close fails, reset */
2774 		gve_reset_and_teardown(priv, was_up);
2775 	} else {
2776 		/* If the dev wasn't up or close worked, finish tearing down */
2777 		gve_teardown_priv_resources(priv);
2778 	}
2779 	rtnl_unlock();
2780 }
2781 
2782 #ifdef CONFIG_PM
2783 static int gve_suspend(struct pci_dev *pdev, pm_message_t state)
2784 {
2785 	struct net_device *netdev = pci_get_drvdata(pdev);
2786 	struct gve_priv *priv = netdev_priv(netdev);
2787 	bool was_up = netif_running(priv->dev);
2788 
2789 	priv->suspend_cnt++;
2790 	rtnl_lock();
2791 	if (was_up && gve_close(priv->dev)) {
2792 		/* If the dev was up, attempt to close, if close fails, reset */
2793 		gve_reset_and_teardown(priv, was_up);
2794 	} else {
2795 		/* If the dev wasn't up or close worked, finish tearing down */
2796 		gve_teardown_priv_resources(priv);
2797 	}
2798 	priv->up_before_suspend = was_up;
2799 	rtnl_unlock();
2800 	return 0;
2801 }
2802 
2803 static int gve_resume(struct pci_dev *pdev)
2804 {
2805 	struct net_device *netdev = pci_get_drvdata(pdev);
2806 	struct gve_priv *priv = netdev_priv(netdev);
2807 	int err;
2808 
2809 	priv->resume_cnt++;
2810 	rtnl_lock();
2811 	err = gve_reset_recovery(priv, priv->up_before_suspend);
2812 	rtnl_unlock();
2813 	return err;
2814 }
2815 #endif /* CONFIG_PM */
2816 
2817 static const struct pci_device_id gve_id_table[] = {
2818 	{ PCI_DEVICE(PCI_VENDOR_ID_GOOGLE, PCI_DEV_ID_GVNIC) },
2819 	{ }
2820 };
2821 
2822 static struct pci_driver gve_driver = {
2823 	.name		= gve_driver_name,
2824 	.id_table	= gve_id_table,
2825 	.probe		= gve_probe,
2826 	.remove		= gve_remove,
2827 	.shutdown	= gve_shutdown,
2828 #ifdef CONFIG_PM
2829 	.suspend        = gve_suspend,
2830 	.resume         = gve_resume,
2831 #endif
2832 };
2833 
2834 module_pci_driver(gve_driver);
2835 
2836 MODULE_DEVICE_TABLE(pci, gve_id_table);
2837 MODULE_AUTHOR("Google, Inc.");
2838 MODULE_DESCRIPTION("Google Virtual NIC Driver");
2839 MODULE_LICENSE("Dual MIT/GPL");
2840 MODULE_VERSION(GVE_VERSION);
2841