xref: /openbmc/linux/drivers/net/xen-netfront.c (revision b8265621)
1 /*
2  * Virtual network driver for conversing with remote driver backends.
3  *
4  * Copyright (c) 2002-2005, K A Fraser
5  * Copyright (c) 2005, XenSource Ltd
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License version 2
9  * as published by the Free Software Foundation; or, when distributed
10  * separately from the Linux kernel or incorporated into other
11  * software packages, subject to the following license:
12  *
13  * Permission is hereby granted, free of charge, to any person obtaining a copy
14  * of this source file (the "Software"), to deal in the Software without
15  * restriction, including without limitation the rights to use, copy, modify,
16  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17  * and to permit persons to whom the Software is furnished to do so, subject to
18  * the following conditions:
19  *
20  * The above copyright notice and this permission notice shall be included in
21  * all copies or substantial portions of the Software.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29  * IN THE SOFTWARE.
30  */
31 
32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33 
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_ether.h>
41 #include <net/tcp.h>
42 #include <linux/udp.h>
43 #include <linux/moduleparam.h>
44 #include <linux/mm.h>
45 #include <linux/slab.h>
46 #include <net/ip.h>
47 #include <linux/bpf.h>
48 #include <net/page_pool.h>
49 #include <linux/bpf_trace.h>
50 
51 #include <xen/xen.h>
52 #include <xen/xenbus.h>
53 #include <xen/events.h>
54 #include <xen/page.h>
55 #include <xen/platform_pci.h>
56 #include <xen/grant_table.h>
57 
58 #include <xen/interface/io/netif.h>
59 #include <xen/interface/memory.h>
60 #include <xen/interface/grant_table.h>
61 
62 /* Module parameters */
63 #define MAX_QUEUES_DEFAULT 8
64 static unsigned int xennet_max_queues;
65 module_param_named(max_queues, xennet_max_queues, uint, 0644);
66 MODULE_PARM_DESC(max_queues,
67 		 "Maximum number of queues per virtual interface");
68 
69 #define XENNET_TIMEOUT  (5 * HZ)
70 
71 static const struct ethtool_ops xennet_ethtool_ops;
72 
73 struct netfront_cb {
74 	int pull_to;
75 };
76 
77 #define NETFRONT_SKB_CB(skb)	((struct netfront_cb *)((skb)->cb))
78 
79 #define RX_COPY_THRESHOLD 256
80 
81 #define GRANT_INVALID_REF	0
82 
83 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
84 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
85 
86 /* Minimum number of Rx slots (includes slot for GSO metadata). */
87 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
88 
89 /* Queue name is interface name with "-qNNN" appended */
90 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
91 
92 /* IRQ name is queue name with "-tx" or "-rx" appended */
93 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
94 
95 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
96 
97 struct netfront_stats {
98 	u64			packets;
99 	u64			bytes;
100 	struct u64_stats_sync	syncp;
101 };
102 
103 struct netfront_info;
104 
105 struct netfront_queue {
106 	unsigned int id; /* Queue ID, 0-based */
107 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
108 	struct netfront_info *info;
109 
110 	struct bpf_prog __rcu *xdp_prog;
111 
112 	struct napi_struct napi;
113 
114 	/* Split event channels support, tx_* == rx_* when using
115 	 * single event channel.
116 	 */
117 	unsigned int tx_evtchn, rx_evtchn;
118 	unsigned int tx_irq, rx_irq;
119 	/* Only used when split event channels support is enabled */
120 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
121 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
122 
123 	spinlock_t   tx_lock;
124 	struct xen_netif_tx_front_ring tx;
125 	int tx_ring_ref;
126 
127 	/*
128 	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
129 	 * are linked from tx_skb_freelist through skb_entry.link.
130 	 *
131 	 *  NB. Freelist index entries are always going to be less than
132 	 *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
133 	 *  greater than PAGE_OFFSET: we use this property to distinguish
134 	 *  them.
135 	 */
136 	union skb_entry {
137 		struct sk_buff *skb;
138 		unsigned long link;
139 	} tx_skbs[NET_TX_RING_SIZE];
140 	grant_ref_t gref_tx_head;
141 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
142 	struct page *grant_tx_page[NET_TX_RING_SIZE];
143 	unsigned tx_skb_freelist;
144 
145 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
146 	struct xen_netif_rx_front_ring rx;
147 	int rx_ring_ref;
148 
149 	struct timer_list rx_refill_timer;
150 
151 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
152 	grant_ref_t gref_rx_head;
153 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
154 
155 	struct page_pool *page_pool;
156 	struct xdp_rxq_info xdp_rxq;
157 };
158 
159 struct netfront_info {
160 	struct list_head list;
161 	struct net_device *netdev;
162 
163 	struct xenbus_device *xbdev;
164 
165 	/* Multi-queue support */
166 	struct netfront_queue *queues;
167 
168 	/* Statistics */
169 	struct netfront_stats __percpu *rx_stats;
170 	struct netfront_stats __percpu *tx_stats;
171 
172 	/* XDP state */
173 	bool netback_has_xdp_headroom;
174 	bool netfront_xdp_enabled;
175 
176 	atomic_t rx_gso_checksum_fixup;
177 };
178 
179 struct netfront_rx_info {
180 	struct xen_netif_rx_response rx;
181 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
182 };
183 
184 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
185 {
186 	list->link = id;
187 }
188 
189 static int skb_entry_is_link(const union skb_entry *list)
190 {
191 	BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
192 	return (unsigned long)list->skb < PAGE_OFFSET;
193 }
194 
195 /*
196  * Access macros for acquiring freeing slots in tx_skbs[].
197  */
198 
199 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
200 			       unsigned short id)
201 {
202 	skb_entry_set_link(&list[id], *head);
203 	*head = id;
204 }
205 
206 static unsigned short get_id_from_freelist(unsigned *head,
207 					   union skb_entry *list)
208 {
209 	unsigned int id = *head;
210 	*head = list[id].link;
211 	return id;
212 }
213 
214 static int xennet_rxidx(RING_IDX idx)
215 {
216 	return idx & (NET_RX_RING_SIZE - 1);
217 }
218 
219 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
220 					 RING_IDX ri)
221 {
222 	int i = xennet_rxidx(ri);
223 	struct sk_buff *skb = queue->rx_skbs[i];
224 	queue->rx_skbs[i] = NULL;
225 	return skb;
226 }
227 
228 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
229 					    RING_IDX ri)
230 {
231 	int i = xennet_rxidx(ri);
232 	grant_ref_t ref = queue->grant_rx_ref[i];
233 	queue->grant_rx_ref[i] = GRANT_INVALID_REF;
234 	return ref;
235 }
236 
237 #ifdef CONFIG_SYSFS
238 static const struct attribute_group xennet_dev_group;
239 #endif
240 
241 static bool xennet_can_sg(struct net_device *dev)
242 {
243 	return dev->features & NETIF_F_SG;
244 }
245 
246 
247 static void rx_refill_timeout(struct timer_list *t)
248 {
249 	struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
250 	napi_schedule(&queue->napi);
251 }
252 
253 static int netfront_tx_slot_available(struct netfront_queue *queue)
254 {
255 	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
256 		(NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
257 }
258 
259 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
260 {
261 	struct net_device *dev = queue->info->netdev;
262 	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
263 
264 	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
265 	    netfront_tx_slot_available(queue) &&
266 	    likely(netif_running(dev)))
267 		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
268 }
269 
270 
271 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
272 {
273 	struct sk_buff *skb;
274 	struct page *page;
275 
276 	skb = __netdev_alloc_skb(queue->info->netdev,
277 				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
278 				 GFP_ATOMIC | __GFP_NOWARN);
279 	if (unlikely(!skb))
280 		return NULL;
281 
282 	page = page_pool_dev_alloc_pages(queue->page_pool);
283 	if (unlikely(!page)) {
284 		kfree_skb(skb);
285 		return NULL;
286 	}
287 	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
288 
289 	/* Align ip header to a 16 bytes boundary */
290 	skb_reserve(skb, NET_IP_ALIGN);
291 	skb->dev = queue->info->netdev;
292 
293 	return skb;
294 }
295 
296 
297 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
298 {
299 	RING_IDX req_prod = queue->rx.req_prod_pvt;
300 	int notify;
301 	int err = 0;
302 
303 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
304 		return;
305 
306 	for (req_prod = queue->rx.req_prod_pvt;
307 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
308 	     req_prod++) {
309 		struct sk_buff *skb;
310 		unsigned short id;
311 		grant_ref_t ref;
312 		struct page *page;
313 		struct xen_netif_rx_request *req;
314 
315 		skb = xennet_alloc_one_rx_buffer(queue);
316 		if (!skb) {
317 			err = -ENOMEM;
318 			break;
319 		}
320 
321 		id = xennet_rxidx(req_prod);
322 
323 		BUG_ON(queue->rx_skbs[id]);
324 		queue->rx_skbs[id] = skb;
325 
326 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
327 		WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
328 		queue->grant_rx_ref[id] = ref;
329 
330 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
331 
332 		req = RING_GET_REQUEST(&queue->rx, req_prod);
333 		gnttab_page_grant_foreign_access_ref_one(ref,
334 							 queue->info->xbdev->otherend_id,
335 							 page,
336 							 0);
337 		req->id = id;
338 		req->gref = ref;
339 	}
340 
341 	queue->rx.req_prod_pvt = req_prod;
342 
343 	/* Try again later if there are not enough requests or skb allocation
344 	 * failed.
345 	 * Enough requests is quantified as the sum of newly created slots and
346 	 * the unconsumed slots at the backend.
347 	 */
348 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
349 	    unlikely(err)) {
350 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
351 		return;
352 	}
353 
354 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
355 	if (notify)
356 		notify_remote_via_irq(queue->rx_irq);
357 }
358 
359 static int xennet_open(struct net_device *dev)
360 {
361 	struct netfront_info *np = netdev_priv(dev);
362 	unsigned int num_queues = dev->real_num_tx_queues;
363 	unsigned int i = 0;
364 	struct netfront_queue *queue = NULL;
365 
366 	if (!np->queues)
367 		return -ENODEV;
368 
369 	for (i = 0; i < num_queues; ++i) {
370 		queue = &np->queues[i];
371 		napi_enable(&queue->napi);
372 
373 		spin_lock_bh(&queue->rx_lock);
374 		if (netif_carrier_ok(dev)) {
375 			xennet_alloc_rx_buffers(queue);
376 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
377 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
378 				napi_schedule(&queue->napi);
379 		}
380 		spin_unlock_bh(&queue->rx_lock);
381 	}
382 
383 	netif_tx_start_all_queues(dev);
384 
385 	return 0;
386 }
387 
388 static void xennet_tx_buf_gc(struct netfront_queue *queue)
389 {
390 	RING_IDX cons, prod;
391 	unsigned short id;
392 	struct sk_buff *skb;
393 	bool more_to_do;
394 
395 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
396 
397 	do {
398 		prod = queue->tx.sring->rsp_prod;
399 		rmb(); /* Ensure we see responses up to 'rp'. */
400 
401 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
402 			struct xen_netif_tx_response *txrsp;
403 
404 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
405 			if (txrsp->status == XEN_NETIF_RSP_NULL)
406 				continue;
407 
408 			id  = txrsp->id;
409 			skb = queue->tx_skbs[id].skb;
410 			if (unlikely(gnttab_query_foreign_access(
411 				queue->grant_tx_ref[id]) != 0)) {
412 				pr_alert("%s: warning -- grant still in use by backend domain\n",
413 					 __func__);
414 				BUG();
415 			}
416 			gnttab_end_foreign_access_ref(
417 				queue->grant_tx_ref[id], GNTMAP_readonly);
418 			gnttab_release_grant_reference(
419 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
420 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
421 			queue->grant_tx_page[id] = NULL;
422 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
423 			dev_kfree_skb_irq(skb);
424 		}
425 
426 		queue->tx.rsp_cons = prod;
427 
428 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
429 	} while (more_to_do);
430 
431 	xennet_maybe_wake_tx(queue);
432 }
433 
434 struct xennet_gnttab_make_txreq {
435 	struct netfront_queue *queue;
436 	struct sk_buff *skb;
437 	struct page *page;
438 	struct xen_netif_tx_request *tx; /* Last request */
439 	unsigned int size;
440 };
441 
442 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
443 				  unsigned int len, void *data)
444 {
445 	struct xennet_gnttab_make_txreq *info = data;
446 	unsigned int id;
447 	struct xen_netif_tx_request *tx;
448 	grant_ref_t ref;
449 	/* convenient aliases */
450 	struct page *page = info->page;
451 	struct netfront_queue *queue = info->queue;
452 	struct sk_buff *skb = info->skb;
453 
454 	id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
455 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
456 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
457 	WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
458 
459 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
460 					gfn, GNTMAP_readonly);
461 
462 	queue->tx_skbs[id].skb = skb;
463 	queue->grant_tx_page[id] = page;
464 	queue->grant_tx_ref[id] = ref;
465 
466 	tx->id = id;
467 	tx->gref = ref;
468 	tx->offset = offset;
469 	tx->size = len;
470 	tx->flags = 0;
471 
472 	info->tx = tx;
473 	info->size += tx->size;
474 }
475 
476 static struct xen_netif_tx_request *xennet_make_first_txreq(
477 	struct netfront_queue *queue, struct sk_buff *skb,
478 	struct page *page, unsigned int offset, unsigned int len)
479 {
480 	struct xennet_gnttab_make_txreq info = {
481 		.queue = queue,
482 		.skb = skb,
483 		.page = page,
484 		.size = 0,
485 	};
486 
487 	gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
488 
489 	return info.tx;
490 }
491 
492 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
493 				  unsigned int len, void *data)
494 {
495 	struct xennet_gnttab_make_txreq *info = data;
496 
497 	info->tx->flags |= XEN_NETTXF_more_data;
498 	skb_get(info->skb);
499 	xennet_tx_setup_grant(gfn, offset, len, data);
500 }
501 
502 static struct xen_netif_tx_request *xennet_make_txreqs(
503 	struct netfront_queue *queue, struct xen_netif_tx_request *tx,
504 	struct sk_buff *skb, struct page *page,
505 	unsigned int offset, unsigned int len)
506 {
507 	struct xennet_gnttab_make_txreq info = {
508 		.queue = queue,
509 		.skb = skb,
510 		.tx = tx,
511 	};
512 
513 	/* Skip unused frames from start of page */
514 	page += offset >> PAGE_SHIFT;
515 	offset &= ~PAGE_MASK;
516 
517 	while (len) {
518 		info.page = page;
519 		info.size = 0;
520 
521 		gnttab_foreach_grant_in_range(page, offset, len,
522 					      xennet_make_one_txreq,
523 					      &info);
524 
525 		page++;
526 		offset = 0;
527 		len -= info.size;
528 	}
529 
530 	return info.tx;
531 }
532 
533 /*
534  * Count how many ring slots are required to send this skb. Each frag
535  * might be a compound page.
536  */
537 static int xennet_count_skb_slots(struct sk_buff *skb)
538 {
539 	int i, frags = skb_shinfo(skb)->nr_frags;
540 	int slots;
541 
542 	slots = gnttab_count_grant(offset_in_page(skb->data),
543 				   skb_headlen(skb));
544 
545 	for (i = 0; i < frags; i++) {
546 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
547 		unsigned long size = skb_frag_size(frag);
548 		unsigned long offset = skb_frag_off(frag);
549 
550 		/* Skip unused frames from start of page */
551 		offset &= ~PAGE_MASK;
552 
553 		slots += gnttab_count_grant(offset, size);
554 	}
555 
556 	return slots;
557 }
558 
559 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
560 			       struct net_device *sb_dev)
561 {
562 	unsigned int num_queues = dev->real_num_tx_queues;
563 	u32 hash;
564 	u16 queue_idx;
565 
566 	/* First, check if there is only one queue */
567 	if (num_queues == 1) {
568 		queue_idx = 0;
569 	} else {
570 		hash = skb_get_hash(skb);
571 		queue_idx = hash % num_queues;
572 	}
573 
574 	return queue_idx;
575 }
576 
577 static int xennet_xdp_xmit_one(struct net_device *dev,
578 			       struct netfront_queue *queue,
579 			       struct xdp_frame *xdpf)
580 {
581 	struct netfront_info *np = netdev_priv(dev);
582 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
583 	int notify;
584 
585 	xennet_make_first_txreq(queue, NULL,
586 				virt_to_page(xdpf->data),
587 				offset_in_page(xdpf->data),
588 				xdpf->len);
589 
590 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
591 	if (notify)
592 		notify_remote_via_irq(queue->tx_irq);
593 
594 	u64_stats_update_begin(&tx_stats->syncp);
595 	tx_stats->bytes += xdpf->len;
596 	tx_stats->packets++;
597 	u64_stats_update_end(&tx_stats->syncp);
598 
599 	xennet_tx_buf_gc(queue);
600 
601 	return 0;
602 }
603 
604 static int xennet_xdp_xmit(struct net_device *dev, int n,
605 			   struct xdp_frame **frames, u32 flags)
606 {
607 	unsigned int num_queues = dev->real_num_tx_queues;
608 	struct netfront_info *np = netdev_priv(dev);
609 	struct netfront_queue *queue = NULL;
610 	unsigned long irq_flags;
611 	int drops = 0;
612 	int i, err;
613 
614 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
615 		return -EINVAL;
616 
617 	queue = &np->queues[smp_processor_id() % num_queues];
618 
619 	spin_lock_irqsave(&queue->tx_lock, irq_flags);
620 	for (i = 0; i < n; i++) {
621 		struct xdp_frame *xdpf = frames[i];
622 
623 		if (!xdpf)
624 			continue;
625 		err = xennet_xdp_xmit_one(dev, queue, xdpf);
626 		if (err) {
627 			xdp_return_frame_rx_napi(xdpf);
628 			drops++;
629 		}
630 	}
631 	spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
632 
633 	return n - drops;
634 }
635 
636 
637 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
638 
639 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
640 {
641 	struct netfront_info *np = netdev_priv(dev);
642 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
643 	struct xen_netif_tx_request *tx, *first_tx;
644 	unsigned int i;
645 	int notify;
646 	int slots;
647 	struct page *page;
648 	unsigned int offset;
649 	unsigned int len;
650 	unsigned long flags;
651 	struct netfront_queue *queue = NULL;
652 	unsigned int num_queues = dev->real_num_tx_queues;
653 	u16 queue_index;
654 	struct sk_buff *nskb;
655 
656 	/* Drop the packet if no queues are set up */
657 	if (num_queues < 1)
658 		goto drop;
659 	/* Determine which queue to transmit this SKB on */
660 	queue_index = skb_get_queue_mapping(skb);
661 	queue = &np->queues[queue_index];
662 
663 	/* If skb->len is too big for wire format, drop skb and alert
664 	 * user about misconfiguration.
665 	 */
666 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
667 		net_alert_ratelimited(
668 			"xennet: skb->len = %u, too big for wire format\n",
669 			skb->len);
670 		goto drop;
671 	}
672 
673 	slots = xennet_count_skb_slots(skb);
674 	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
675 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
676 				    slots, skb->len);
677 		if (skb_linearize(skb))
678 			goto drop;
679 	}
680 
681 	page = virt_to_page(skb->data);
682 	offset = offset_in_page(skb->data);
683 
684 	/* The first req should be at least ETH_HLEN size or the packet will be
685 	 * dropped by netback.
686 	 */
687 	if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
688 		nskb = skb_copy(skb, GFP_ATOMIC);
689 		if (!nskb)
690 			goto drop;
691 		dev_consume_skb_any(skb);
692 		skb = nskb;
693 		page = virt_to_page(skb->data);
694 		offset = offset_in_page(skb->data);
695 	}
696 
697 	len = skb_headlen(skb);
698 
699 	spin_lock_irqsave(&queue->tx_lock, flags);
700 
701 	if (unlikely(!netif_carrier_ok(dev) ||
702 		     (slots > 1 && !xennet_can_sg(dev)) ||
703 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
704 		spin_unlock_irqrestore(&queue->tx_lock, flags);
705 		goto drop;
706 	}
707 
708 	/* First request for the linear area. */
709 	first_tx = tx = xennet_make_first_txreq(queue, skb,
710 						page, offset, len);
711 	offset += tx->size;
712 	if (offset == PAGE_SIZE) {
713 		page++;
714 		offset = 0;
715 	}
716 	len -= tx->size;
717 
718 	if (skb->ip_summed == CHECKSUM_PARTIAL)
719 		/* local packet? */
720 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
721 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
722 		/* remote but checksummed. */
723 		tx->flags |= XEN_NETTXF_data_validated;
724 
725 	/* Optional extra info after the first request. */
726 	if (skb_shinfo(skb)->gso_size) {
727 		struct xen_netif_extra_info *gso;
728 
729 		gso = (struct xen_netif_extra_info *)
730 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
731 
732 		tx->flags |= XEN_NETTXF_extra_info;
733 
734 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
735 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
736 			XEN_NETIF_GSO_TYPE_TCPV6 :
737 			XEN_NETIF_GSO_TYPE_TCPV4;
738 		gso->u.gso.pad = 0;
739 		gso->u.gso.features = 0;
740 
741 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
742 		gso->flags = 0;
743 	}
744 
745 	/* Requests for the rest of the linear area. */
746 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
747 
748 	/* Requests for all the frags. */
749 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
750 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
751 		tx = xennet_make_txreqs(queue, tx, skb, skb_frag_page(frag),
752 					skb_frag_off(frag),
753 					skb_frag_size(frag));
754 	}
755 
756 	/* First request has the packet length. */
757 	first_tx->size = skb->len;
758 
759 	/* timestamp packet in software */
760 	skb_tx_timestamp(skb);
761 
762 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
763 	if (notify)
764 		notify_remote_via_irq(queue->tx_irq);
765 
766 	u64_stats_update_begin(&tx_stats->syncp);
767 	tx_stats->bytes += skb->len;
768 	tx_stats->packets++;
769 	u64_stats_update_end(&tx_stats->syncp);
770 
771 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
772 	xennet_tx_buf_gc(queue);
773 
774 	if (!netfront_tx_slot_available(queue))
775 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
776 
777 	spin_unlock_irqrestore(&queue->tx_lock, flags);
778 
779 	return NETDEV_TX_OK;
780 
781  drop:
782 	dev->stats.tx_dropped++;
783 	dev_kfree_skb_any(skb);
784 	return NETDEV_TX_OK;
785 }
786 
787 static int xennet_close(struct net_device *dev)
788 {
789 	struct netfront_info *np = netdev_priv(dev);
790 	unsigned int num_queues = dev->real_num_tx_queues;
791 	unsigned int i;
792 	struct netfront_queue *queue;
793 	netif_tx_stop_all_queues(np->netdev);
794 	for (i = 0; i < num_queues; ++i) {
795 		queue = &np->queues[i];
796 		napi_disable(&queue->napi);
797 	}
798 	return 0;
799 }
800 
801 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
802 				grant_ref_t ref)
803 {
804 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
805 
806 	BUG_ON(queue->rx_skbs[new]);
807 	queue->rx_skbs[new] = skb;
808 	queue->grant_rx_ref[new] = ref;
809 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
810 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
811 	queue->rx.req_prod_pvt++;
812 }
813 
814 static int xennet_get_extras(struct netfront_queue *queue,
815 			     struct xen_netif_extra_info *extras,
816 			     RING_IDX rp)
817 
818 {
819 	struct xen_netif_extra_info *extra;
820 	struct device *dev = &queue->info->netdev->dev;
821 	RING_IDX cons = queue->rx.rsp_cons;
822 	int err = 0;
823 
824 	do {
825 		struct sk_buff *skb;
826 		grant_ref_t ref;
827 
828 		if (unlikely(cons + 1 == rp)) {
829 			if (net_ratelimit())
830 				dev_warn(dev, "Missing extra info\n");
831 			err = -EBADR;
832 			break;
833 		}
834 
835 		extra = (struct xen_netif_extra_info *)
836 			RING_GET_RESPONSE(&queue->rx, ++cons);
837 
838 		if (unlikely(!extra->type ||
839 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
840 			if (net_ratelimit())
841 				dev_warn(dev, "Invalid extra type: %d\n",
842 					extra->type);
843 			err = -EINVAL;
844 		} else {
845 			memcpy(&extras[extra->type - 1], extra,
846 			       sizeof(*extra));
847 		}
848 
849 		skb = xennet_get_rx_skb(queue, cons);
850 		ref = xennet_get_rx_ref(queue, cons);
851 		xennet_move_rx_slot(queue, skb, ref);
852 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
853 
854 	queue->rx.rsp_cons = cons;
855 	return err;
856 }
857 
858 static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
859 		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
860 		   struct xdp_buff *xdp, bool *need_xdp_flush)
861 {
862 	struct xdp_frame *xdpf;
863 	u32 len = rx->status;
864 	u32 act;
865 	int err;
866 
867 	xdp->data_hard_start = page_address(pdata);
868 	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
869 	xdp_set_data_meta_invalid(xdp);
870 	xdp->data_end = xdp->data + len;
871 	xdp->rxq = &queue->xdp_rxq;
872 	xdp->frame_sz = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
873 
874 	act = bpf_prog_run_xdp(prog, xdp);
875 	switch (act) {
876 	case XDP_TX:
877 		get_page(pdata);
878 		xdpf = xdp_convert_buff_to_frame(xdp);
879 		err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
880 		if (unlikely(err < 0))
881 			trace_xdp_exception(queue->info->netdev, prog, act);
882 		break;
883 	case XDP_REDIRECT:
884 		get_page(pdata);
885 		err = xdp_do_redirect(queue->info->netdev, xdp, prog);
886 		*need_xdp_flush = true;
887 		if (unlikely(err))
888 			trace_xdp_exception(queue->info->netdev, prog, act);
889 		break;
890 	case XDP_PASS:
891 	case XDP_DROP:
892 		break;
893 
894 	case XDP_ABORTED:
895 		trace_xdp_exception(queue->info->netdev, prog, act);
896 		break;
897 
898 	default:
899 		bpf_warn_invalid_xdp_action(act);
900 	}
901 
902 	return act;
903 }
904 
905 static int xennet_get_responses(struct netfront_queue *queue,
906 				struct netfront_rx_info *rinfo, RING_IDX rp,
907 				struct sk_buff_head *list,
908 				bool *need_xdp_flush)
909 {
910 	struct xen_netif_rx_response *rx = &rinfo->rx;
911 	int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
912 	RING_IDX cons = queue->rx.rsp_cons;
913 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
914 	struct xen_netif_extra_info *extras = rinfo->extras;
915 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
916 	struct device *dev = &queue->info->netdev->dev;
917 	struct bpf_prog *xdp_prog;
918 	struct xdp_buff xdp;
919 	unsigned long ret;
920 	int slots = 1;
921 	int err = 0;
922 	u32 verdict;
923 
924 	if (rx->flags & XEN_NETRXF_extra_info) {
925 		err = xennet_get_extras(queue, extras, rp);
926 		if (!err) {
927 			if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
928 				struct xen_netif_extra_info *xdp;
929 
930 				xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
931 				rx->offset = xdp->u.xdp.headroom;
932 			}
933 		}
934 		cons = queue->rx.rsp_cons;
935 	}
936 
937 	for (;;) {
938 		if (unlikely(rx->status < 0 ||
939 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
940 			if (net_ratelimit())
941 				dev_warn(dev, "rx->offset: %u, size: %d\n",
942 					 rx->offset, rx->status);
943 			xennet_move_rx_slot(queue, skb, ref);
944 			err = -EINVAL;
945 			goto next;
946 		}
947 
948 		/*
949 		 * This definitely indicates a bug, either in this driver or in
950 		 * the backend driver. In future this should flag the bad
951 		 * situation to the system controller to reboot the backend.
952 		 */
953 		if (ref == GRANT_INVALID_REF) {
954 			if (net_ratelimit())
955 				dev_warn(dev, "Bad rx response id %d.\n",
956 					 rx->id);
957 			err = -EINVAL;
958 			goto next;
959 		}
960 
961 		ret = gnttab_end_foreign_access_ref(ref, 0);
962 		BUG_ON(!ret);
963 
964 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
965 
966 		rcu_read_lock();
967 		xdp_prog = rcu_dereference(queue->xdp_prog);
968 		if (xdp_prog) {
969 			if (!(rx->flags & XEN_NETRXF_more_data)) {
970 				/* currently only a single page contains data */
971 				verdict = xennet_run_xdp(queue,
972 							 skb_frag_page(&skb_shinfo(skb)->frags[0]),
973 							 rx, xdp_prog, &xdp, need_xdp_flush);
974 				if (verdict != XDP_PASS)
975 					err = -EINVAL;
976 			} else {
977 				/* drop the frame */
978 				err = -EINVAL;
979 			}
980 		}
981 		rcu_read_unlock();
982 next:
983 		__skb_queue_tail(list, skb);
984 		if (!(rx->flags & XEN_NETRXF_more_data))
985 			break;
986 
987 		if (cons + slots == rp) {
988 			if (net_ratelimit())
989 				dev_warn(dev, "Need more slots\n");
990 			err = -ENOENT;
991 			break;
992 		}
993 
994 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
995 		skb = xennet_get_rx_skb(queue, cons + slots);
996 		ref = xennet_get_rx_ref(queue, cons + slots);
997 		slots++;
998 	}
999 
1000 	if (unlikely(slots > max)) {
1001 		if (net_ratelimit())
1002 			dev_warn(dev, "Too many slots\n");
1003 		err = -E2BIG;
1004 	}
1005 
1006 	if (unlikely(err))
1007 		queue->rx.rsp_cons = cons + slots;
1008 
1009 	return err;
1010 }
1011 
1012 static int xennet_set_skb_gso(struct sk_buff *skb,
1013 			      struct xen_netif_extra_info *gso)
1014 {
1015 	if (!gso->u.gso.size) {
1016 		if (net_ratelimit())
1017 			pr_warn("GSO size must not be zero\n");
1018 		return -EINVAL;
1019 	}
1020 
1021 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1022 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
1023 		if (net_ratelimit())
1024 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
1025 		return -EINVAL;
1026 	}
1027 
1028 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
1029 	skb_shinfo(skb)->gso_type =
1030 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1031 		SKB_GSO_TCPV4 :
1032 		SKB_GSO_TCPV6;
1033 
1034 	/* Header must be checked, and gso_segs computed. */
1035 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1036 	skb_shinfo(skb)->gso_segs = 0;
1037 
1038 	return 0;
1039 }
1040 
1041 static int xennet_fill_frags(struct netfront_queue *queue,
1042 			     struct sk_buff *skb,
1043 			     struct sk_buff_head *list)
1044 {
1045 	RING_IDX cons = queue->rx.rsp_cons;
1046 	struct sk_buff *nskb;
1047 
1048 	while ((nskb = __skb_dequeue(list))) {
1049 		struct xen_netif_rx_response *rx =
1050 			RING_GET_RESPONSE(&queue->rx, ++cons);
1051 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
1052 
1053 		if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
1054 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1055 
1056 			BUG_ON(pull_to < skb_headlen(skb));
1057 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1058 		}
1059 		if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
1060 			queue->rx.rsp_cons = ++cons + skb_queue_len(list);
1061 			kfree_skb(nskb);
1062 			return -ENOENT;
1063 		}
1064 
1065 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1066 				skb_frag_page(nfrag),
1067 				rx->offset, rx->status, PAGE_SIZE);
1068 
1069 		skb_shinfo(nskb)->nr_frags = 0;
1070 		kfree_skb(nskb);
1071 	}
1072 
1073 	queue->rx.rsp_cons = cons;
1074 
1075 	return 0;
1076 }
1077 
1078 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
1079 {
1080 	bool recalculate_partial_csum = false;
1081 
1082 	/*
1083 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1084 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
1085 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1086 	 * recalculate the partial checksum.
1087 	 */
1088 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1089 		struct netfront_info *np = netdev_priv(dev);
1090 		atomic_inc(&np->rx_gso_checksum_fixup);
1091 		skb->ip_summed = CHECKSUM_PARTIAL;
1092 		recalculate_partial_csum = true;
1093 	}
1094 
1095 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1096 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1097 		return 0;
1098 
1099 	return skb_checksum_setup(skb, recalculate_partial_csum);
1100 }
1101 
1102 static int handle_incoming_queue(struct netfront_queue *queue,
1103 				 struct sk_buff_head *rxq)
1104 {
1105 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
1106 	int packets_dropped = 0;
1107 	struct sk_buff *skb;
1108 
1109 	while ((skb = __skb_dequeue(rxq)) != NULL) {
1110 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1111 
1112 		if (pull_to > skb_headlen(skb))
1113 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1114 
1115 		/* Ethernet work: Delayed to here as it peeks the header. */
1116 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
1117 		skb_reset_network_header(skb);
1118 
1119 		if (checksum_setup(queue->info->netdev, skb)) {
1120 			kfree_skb(skb);
1121 			packets_dropped++;
1122 			queue->info->netdev->stats.rx_errors++;
1123 			continue;
1124 		}
1125 
1126 		u64_stats_update_begin(&rx_stats->syncp);
1127 		rx_stats->packets++;
1128 		rx_stats->bytes += skb->len;
1129 		u64_stats_update_end(&rx_stats->syncp);
1130 
1131 		/* Pass it up. */
1132 		napi_gro_receive(&queue->napi, skb);
1133 	}
1134 
1135 	return packets_dropped;
1136 }
1137 
1138 static int xennet_poll(struct napi_struct *napi, int budget)
1139 {
1140 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1141 	struct net_device *dev = queue->info->netdev;
1142 	struct sk_buff *skb;
1143 	struct netfront_rx_info rinfo;
1144 	struct xen_netif_rx_response *rx = &rinfo.rx;
1145 	struct xen_netif_extra_info *extras = rinfo.extras;
1146 	RING_IDX i, rp;
1147 	int work_done;
1148 	struct sk_buff_head rxq;
1149 	struct sk_buff_head errq;
1150 	struct sk_buff_head tmpq;
1151 	int err;
1152 	bool need_xdp_flush = false;
1153 
1154 	spin_lock(&queue->rx_lock);
1155 
1156 	skb_queue_head_init(&rxq);
1157 	skb_queue_head_init(&errq);
1158 	skb_queue_head_init(&tmpq);
1159 
1160 	rp = queue->rx.sring->rsp_prod;
1161 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1162 
1163 	i = queue->rx.rsp_cons;
1164 	work_done = 0;
1165 	while ((i != rp) && (work_done < budget)) {
1166 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1167 		memset(extras, 0, sizeof(rinfo.extras));
1168 
1169 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1170 					   &need_xdp_flush);
1171 
1172 		if (unlikely(err)) {
1173 err:
1174 			while ((skb = __skb_dequeue(&tmpq)))
1175 				__skb_queue_tail(&errq, skb);
1176 			dev->stats.rx_errors++;
1177 			i = queue->rx.rsp_cons;
1178 			continue;
1179 		}
1180 
1181 		skb = __skb_dequeue(&tmpq);
1182 
1183 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1184 			struct xen_netif_extra_info *gso;
1185 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1186 
1187 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1188 				__skb_queue_head(&tmpq, skb);
1189 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1190 				goto err;
1191 			}
1192 		}
1193 
1194 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1195 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1196 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1197 
1198 		skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1199 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1200 		skb->data_len = rx->status;
1201 		skb->len += rx->status;
1202 
1203 		if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1204 			goto err;
1205 
1206 		if (rx->flags & XEN_NETRXF_csum_blank)
1207 			skb->ip_summed = CHECKSUM_PARTIAL;
1208 		else if (rx->flags & XEN_NETRXF_data_validated)
1209 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1210 
1211 		__skb_queue_tail(&rxq, skb);
1212 
1213 		i = ++queue->rx.rsp_cons;
1214 		work_done++;
1215 	}
1216 	if (need_xdp_flush)
1217 		xdp_do_flush();
1218 
1219 	__skb_queue_purge(&errq);
1220 
1221 	work_done -= handle_incoming_queue(queue, &rxq);
1222 
1223 	xennet_alloc_rx_buffers(queue);
1224 
1225 	if (work_done < budget) {
1226 		int more_to_do = 0;
1227 
1228 		napi_complete_done(napi, work_done);
1229 
1230 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1231 		if (more_to_do)
1232 			napi_schedule(napi);
1233 	}
1234 
1235 	spin_unlock(&queue->rx_lock);
1236 
1237 	return work_done;
1238 }
1239 
1240 static int xennet_change_mtu(struct net_device *dev, int mtu)
1241 {
1242 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1243 
1244 	if (mtu > max)
1245 		return -EINVAL;
1246 	dev->mtu = mtu;
1247 	return 0;
1248 }
1249 
1250 static void xennet_get_stats64(struct net_device *dev,
1251 			       struct rtnl_link_stats64 *tot)
1252 {
1253 	struct netfront_info *np = netdev_priv(dev);
1254 	int cpu;
1255 
1256 	for_each_possible_cpu(cpu) {
1257 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1258 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1259 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1260 		unsigned int start;
1261 
1262 		do {
1263 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1264 			tx_packets = tx_stats->packets;
1265 			tx_bytes = tx_stats->bytes;
1266 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1267 
1268 		do {
1269 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1270 			rx_packets = rx_stats->packets;
1271 			rx_bytes = rx_stats->bytes;
1272 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1273 
1274 		tot->rx_packets += rx_packets;
1275 		tot->tx_packets += tx_packets;
1276 		tot->rx_bytes   += rx_bytes;
1277 		tot->tx_bytes   += tx_bytes;
1278 	}
1279 
1280 	tot->rx_errors  = dev->stats.rx_errors;
1281 	tot->tx_dropped = dev->stats.tx_dropped;
1282 }
1283 
1284 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1285 {
1286 	struct sk_buff *skb;
1287 	int i;
1288 
1289 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1290 		/* Skip over entries which are actually freelist references */
1291 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1292 			continue;
1293 
1294 		skb = queue->tx_skbs[i].skb;
1295 		get_page(queue->grant_tx_page[i]);
1296 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1297 					  GNTMAP_readonly,
1298 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1299 		queue->grant_tx_page[i] = NULL;
1300 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1301 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1302 		dev_kfree_skb_irq(skb);
1303 	}
1304 }
1305 
1306 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1307 {
1308 	int id, ref;
1309 
1310 	spin_lock_bh(&queue->rx_lock);
1311 
1312 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1313 		struct sk_buff *skb;
1314 		struct page *page;
1315 
1316 		skb = queue->rx_skbs[id];
1317 		if (!skb)
1318 			continue;
1319 
1320 		ref = queue->grant_rx_ref[id];
1321 		if (ref == GRANT_INVALID_REF)
1322 			continue;
1323 
1324 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1325 
1326 		/* gnttab_end_foreign_access() needs a page ref until
1327 		 * foreign access is ended (which may be deferred).
1328 		 */
1329 		get_page(page);
1330 		gnttab_end_foreign_access(ref, 0,
1331 					  (unsigned long)page_address(page));
1332 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1333 
1334 		kfree_skb(skb);
1335 	}
1336 
1337 	spin_unlock_bh(&queue->rx_lock);
1338 }
1339 
1340 static netdev_features_t xennet_fix_features(struct net_device *dev,
1341 	netdev_features_t features)
1342 {
1343 	struct netfront_info *np = netdev_priv(dev);
1344 
1345 	if (features & NETIF_F_SG &&
1346 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1347 		features &= ~NETIF_F_SG;
1348 
1349 	if (features & NETIF_F_IPV6_CSUM &&
1350 	    !xenbus_read_unsigned(np->xbdev->otherend,
1351 				  "feature-ipv6-csum-offload", 0))
1352 		features &= ~NETIF_F_IPV6_CSUM;
1353 
1354 	if (features & NETIF_F_TSO &&
1355 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1356 		features &= ~NETIF_F_TSO;
1357 
1358 	if (features & NETIF_F_TSO6 &&
1359 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1360 		features &= ~NETIF_F_TSO6;
1361 
1362 	return features;
1363 }
1364 
1365 static int xennet_set_features(struct net_device *dev,
1366 	netdev_features_t features)
1367 {
1368 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1369 		netdev_info(dev, "Reducing MTU because no SG offload");
1370 		dev->mtu = ETH_DATA_LEN;
1371 	}
1372 
1373 	return 0;
1374 }
1375 
1376 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1377 {
1378 	struct netfront_queue *queue = dev_id;
1379 	unsigned long flags;
1380 
1381 	spin_lock_irqsave(&queue->tx_lock, flags);
1382 	xennet_tx_buf_gc(queue);
1383 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1384 
1385 	return IRQ_HANDLED;
1386 }
1387 
1388 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1389 {
1390 	struct netfront_queue *queue = dev_id;
1391 	struct net_device *dev = queue->info->netdev;
1392 
1393 	if (likely(netif_carrier_ok(dev) &&
1394 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1395 		napi_schedule(&queue->napi);
1396 
1397 	return IRQ_HANDLED;
1398 }
1399 
1400 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1401 {
1402 	xennet_tx_interrupt(irq, dev_id);
1403 	xennet_rx_interrupt(irq, dev_id);
1404 	return IRQ_HANDLED;
1405 }
1406 
1407 #ifdef CONFIG_NET_POLL_CONTROLLER
1408 static void xennet_poll_controller(struct net_device *dev)
1409 {
1410 	/* Poll each queue */
1411 	struct netfront_info *info = netdev_priv(dev);
1412 	unsigned int num_queues = dev->real_num_tx_queues;
1413 	unsigned int i;
1414 	for (i = 0; i < num_queues; ++i)
1415 		xennet_interrupt(0, &info->queues[i]);
1416 }
1417 #endif
1418 
1419 #define NETBACK_XDP_HEADROOM_DISABLE	0
1420 #define NETBACK_XDP_HEADROOM_ENABLE	1
1421 
1422 static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1423 {
1424 	int err;
1425 	unsigned short headroom;
1426 
1427 	headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1428 	err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1429 			    "xdp-headroom", "%hu",
1430 			    headroom);
1431 	if (err)
1432 		pr_warn("Error writing xdp-headroom\n");
1433 
1434 	return err;
1435 }
1436 
1437 static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1438 			  struct netlink_ext_ack *extack)
1439 {
1440 	unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1441 	struct netfront_info *np = netdev_priv(dev);
1442 	struct bpf_prog *old_prog;
1443 	unsigned int i, err;
1444 
1445 	if (dev->mtu > max_mtu) {
1446 		netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1447 		return -EINVAL;
1448 	}
1449 
1450 	if (!np->netback_has_xdp_headroom)
1451 		return 0;
1452 
1453 	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1454 
1455 	err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1456 				  NETBACK_XDP_HEADROOM_DISABLE);
1457 	if (err)
1458 		return err;
1459 
1460 	/* avoid the race with XDP headroom adjustment */
1461 	wait_event(module_wq,
1462 		   xenbus_read_driver_state(np->xbdev->otherend) ==
1463 		   XenbusStateReconfigured);
1464 	np->netfront_xdp_enabled = true;
1465 
1466 	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1467 
1468 	if (prog)
1469 		bpf_prog_add(prog, dev->real_num_tx_queues);
1470 
1471 	for (i = 0; i < dev->real_num_tx_queues; ++i)
1472 		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1473 
1474 	if (old_prog)
1475 		for (i = 0; i < dev->real_num_tx_queues; ++i)
1476 			bpf_prog_put(old_prog);
1477 
1478 	xenbus_switch_state(np->xbdev, XenbusStateConnected);
1479 
1480 	return 0;
1481 }
1482 
1483 static u32 xennet_xdp_query(struct net_device *dev)
1484 {
1485 	unsigned int num_queues = dev->real_num_tx_queues;
1486 	struct netfront_info *np = netdev_priv(dev);
1487 	const struct bpf_prog *xdp_prog;
1488 	struct netfront_queue *queue;
1489 	unsigned int i;
1490 
1491 	for (i = 0; i < num_queues; ++i) {
1492 		queue = &np->queues[i];
1493 		xdp_prog = rtnl_dereference(queue->xdp_prog);
1494 		if (xdp_prog)
1495 			return xdp_prog->aux->id;
1496 	}
1497 
1498 	return 0;
1499 }
1500 
1501 static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1502 {
1503 	switch (xdp->command) {
1504 	case XDP_SETUP_PROG:
1505 		return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1506 	case XDP_QUERY_PROG:
1507 		xdp->prog_id = xennet_xdp_query(dev);
1508 		return 0;
1509 	default:
1510 		return -EINVAL;
1511 	}
1512 }
1513 
1514 static const struct net_device_ops xennet_netdev_ops = {
1515 	.ndo_open            = xennet_open,
1516 	.ndo_stop            = xennet_close,
1517 	.ndo_start_xmit      = xennet_start_xmit,
1518 	.ndo_change_mtu	     = xennet_change_mtu,
1519 	.ndo_get_stats64     = xennet_get_stats64,
1520 	.ndo_set_mac_address = eth_mac_addr,
1521 	.ndo_validate_addr   = eth_validate_addr,
1522 	.ndo_fix_features    = xennet_fix_features,
1523 	.ndo_set_features    = xennet_set_features,
1524 	.ndo_select_queue    = xennet_select_queue,
1525 	.ndo_bpf            = xennet_xdp,
1526 	.ndo_xdp_xmit	    = xennet_xdp_xmit,
1527 #ifdef CONFIG_NET_POLL_CONTROLLER
1528 	.ndo_poll_controller = xennet_poll_controller,
1529 #endif
1530 };
1531 
1532 static void xennet_free_netdev(struct net_device *netdev)
1533 {
1534 	struct netfront_info *np = netdev_priv(netdev);
1535 
1536 	free_percpu(np->rx_stats);
1537 	free_percpu(np->tx_stats);
1538 	free_netdev(netdev);
1539 }
1540 
1541 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1542 {
1543 	int err;
1544 	struct net_device *netdev;
1545 	struct netfront_info *np;
1546 
1547 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1548 	if (!netdev)
1549 		return ERR_PTR(-ENOMEM);
1550 
1551 	np                   = netdev_priv(netdev);
1552 	np->xbdev            = dev;
1553 
1554 	np->queues = NULL;
1555 
1556 	err = -ENOMEM;
1557 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1558 	if (np->rx_stats == NULL)
1559 		goto exit;
1560 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1561 	if (np->tx_stats == NULL)
1562 		goto exit;
1563 
1564 	netdev->netdev_ops	= &xennet_netdev_ops;
1565 
1566 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1567 				  NETIF_F_GSO_ROBUST;
1568 	netdev->hw_features	= NETIF_F_SG |
1569 				  NETIF_F_IPV6_CSUM |
1570 				  NETIF_F_TSO | NETIF_F_TSO6;
1571 
1572 	/*
1573          * Assume that all hw features are available for now. This set
1574          * will be adjusted by the call to netdev_update_features() in
1575          * xennet_connect() which is the earliest point where we can
1576          * negotiate with the backend regarding supported features.
1577          */
1578 	netdev->features |= netdev->hw_features;
1579 
1580 	netdev->ethtool_ops = &xennet_ethtool_ops;
1581 	netdev->min_mtu = ETH_MIN_MTU;
1582 	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1583 	SET_NETDEV_DEV(netdev, &dev->dev);
1584 
1585 	np->netdev = netdev;
1586 	np->netfront_xdp_enabled = false;
1587 
1588 	netif_carrier_off(netdev);
1589 
1590 	do {
1591 		xenbus_switch_state(dev, XenbusStateInitialising);
1592 		err = wait_event_timeout(module_wq,
1593 				 xenbus_read_driver_state(dev->otherend) !=
1594 				 XenbusStateClosed &&
1595 				 xenbus_read_driver_state(dev->otherend) !=
1596 				 XenbusStateUnknown, XENNET_TIMEOUT);
1597 	} while (!err);
1598 
1599 	return netdev;
1600 
1601  exit:
1602 	xennet_free_netdev(netdev);
1603 	return ERR_PTR(err);
1604 }
1605 
1606 /**
1607  * Entry point to this code when a new device is created.  Allocate the basic
1608  * structures and the ring buffers for communication with the backend, and
1609  * inform the backend of the appropriate details for those.
1610  */
1611 static int netfront_probe(struct xenbus_device *dev,
1612 			  const struct xenbus_device_id *id)
1613 {
1614 	int err;
1615 	struct net_device *netdev;
1616 	struct netfront_info *info;
1617 
1618 	netdev = xennet_create_dev(dev);
1619 	if (IS_ERR(netdev)) {
1620 		err = PTR_ERR(netdev);
1621 		xenbus_dev_fatal(dev, err, "creating netdev");
1622 		return err;
1623 	}
1624 
1625 	info = netdev_priv(netdev);
1626 	dev_set_drvdata(&dev->dev, info);
1627 #ifdef CONFIG_SYSFS
1628 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1629 #endif
1630 
1631 	return 0;
1632 }
1633 
1634 static void xennet_end_access(int ref, void *page)
1635 {
1636 	/* This frees the page as a side-effect */
1637 	if (ref != GRANT_INVALID_REF)
1638 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1639 }
1640 
1641 static void xennet_disconnect_backend(struct netfront_info *info)
1642 {
1643 	unsigned int i = 0;
1644 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1645 
1646 	netif_carrier_off(info->netdev);
1647 
1648 	for (i = 0; i < num_queues && info->queues; ++i) {
1649 		struct netfront_queue *queue = &info->queues[i];
1650 
1651 		del_timer_sync(&queue->rx_refill_timer);
1652 
1653 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1654 			unbind_from_irqhandler(queue->tx_irq, queue);
1655 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1656 			unbind_from_irqhandler(queue->tx_irq, queue);
1657 			unbind_from_irqhandler(queue->rx_irq, queue);
1658 		}
1659 		queue->tx_evtchn = queue->rx_evtchn = 0;
1660 		queue->tx_irq = queue->rx_irq = 0;
1661 
1662 		if (netif_running(info->netdev))
1663 			napi_synchronize(&queue->napi);
1664 
1665 		xennet_release_tx_bufs(queue);
1666 		xennet_release_rx_bufs(queue);
1667 		gnttab_free_grant_references(queue->gref_tx_head);
1668 		gnttab_free_grant_references(queue->gref_rx_head);
1669 
1670 		/* End access and free the pages */
1671 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1672 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1673 
1674 		queue->tx_ring_ref = GRANT_INVALID_REF;
1675 		queue->rx_ring_ref = GRANT_INVALID_REF;
1676 		queue->tx.sring = NULL;
1677 		queue->rx.sring = NULL;
1678 
1679 		page_pool_destroy(queue->page_pool);
1680 	}
1681 }
1682 
1683 /**
1684  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1685  * driver restart.  We tear down our netif structure and recreate it, but
1686  * leave the device-layer structures intact so that this is transparent to the
1687  * rest of the kernel.
1688  */
1689 static int netfront_resume(struct xenbus_device *dev)
1690 {
1691 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1692 
1693 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1694 
1695 	xennet_disconnect_backend(info);
1696 	return 0;
1697 }
1698 
1699 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1700 {
1701 	char *s, *e, *macstr;
1702 	int i;
1703 
1704 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1705 	if (IS_ERR(macstr))
1706 		return PTR_ERR(macstr);
1707 
1708 	for (i = 0; i < ETH_ALEN; i++) {
1709 		mac[i] = simple_strtoul(s, &e, 16);
1710 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1711 			kfree(macstr);
1712 			return -ENOENT;
1713 		}
1714 		s = e+1;
1715 	}
1716 
1717 	kfree(macstr);
1718 	return 0;
1719 }
1720 
1721 static int setup_netfront_single(struct netfront_queue *queue)
1722 {
1723 	int err;
1724 
1725 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1726 	if (err < 0)
1727 		goto fail;
1728 
1729 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1730 					xennet_interrupt,
1731 					0, queue->info->netdev->name, queue);
1732 	if (err < 0)
1733 		goto bind_fail;
1734 	queue->rx_evtchn = queue->tx_evtchn;
1735 	queue->rx_irq = queue->tx_irq = err;
1736 
1737 	return 0;
1738 
1739 bind_fail:
1740 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1741 	queue->tx_evtchn = 0;
1742 fail:
1743 	return err;
1744 }
1745 
1746 static int setup_netfront_split(struct netfront_queue *queue)
1747 {
1748 	int err;
1749 
1750 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1751 	if (err < 0)
1752 		goto fail;
1753 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1754 	if (err < 0)
1755 		goto alloc_rx_evtchn_fail;
1756 
1757 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1758 		 "%s-tx", queue->name);
1759 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1760 					xennet_tx_interrupt,
1761 					0, queue->tx_irq_name, queue);
1762 	if (err < 0)
1763 		goto bind_tx_fail;
1764 	queue->tx_irq = err;
1765 
1766 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1767 		 "%s-rx", queue->name);
1768 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1769 					xennet_rx_interrupt,
1770 					0, queue->rx_irq_name, queue);
1771 	if (err < 0)
1772 		goto bind_rx_fail;
1773 	queue->rx_irq = err;
1774 
1775 	return 0;
1776 
1777 bind_rx_fail:
1778 	unbind_from_irqhandler(queue->tx_irq, queue);
1779 	queue->tx_irq = 0;
1780 bind_tx_fail:
1781 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1782 	queue->rx_evtchn = 0;
1783 alloc_rx_evtchn_fail:
1784 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1785 	queue->tx_evtchn = 0;
1786 fail:
1787 	return err;
1788 }
1789 
1790 static int setup_netfront(struct xenbus_device *dev,
1791 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1792 {
1793 	struct xen_netif_tx_sring *txs;
1794 	struct xen_netif_rx_sring *rxs;
1795 	grant_ref_t gref;
1796 	int err;
1797 
1798 	queue->tx_ring_ref = GRANT_INVALID_REF;
1799 	queue->rx_ring_ref = GRANT_INVALID_REF;
1800 	queue->rx.sring = NULL;
1801 	queue->tx.sring = NULL;
1802 
1803 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1804 	if (!txs) {
1805 		err = -ENOMEM;
1806 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1807 		goto fail;
1808 	}
1809 	SHARED_RING_INIT(txs);
1810 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1811 
1812 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1813 	if (err < 0)
1814 		goto grant_tx_ring_fail;
1815 	queue->tx_ring_ref = gref;
1816 
1817 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1818 	if (!rxs) {
1819 		err = -ENOMEM;
1820 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1821 		goto alloc_rx_ring_fail;
1822 	}
1823 	SHARED_RING_INIT(rxs);
1824 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1825 
1826 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1827 	if (err < 0)
1828 		goto grant_rx_ring_fail;
1829 	queue->rx_ring_ref = gref;
1830 
1831 	if (feature_split_evtchn)
1832 		err = setup_netfront_split(queue);
1833 	/* setup single event channel if
1834 	 *  a) feature-split-event-channels == 0
1835 	 *  b) feature-split-event-channels == 1 but failed to setup
1836 	 */
1837 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1838 		err = setup_netfront_single(queue);
1839 
1840 	if (err)
1841 		goto alloc_evtchn_fail;
1842 
1843 	return 0;
1844 
1845 	/* If we fail to setup netfront, it is safe to just revoke access to
1846 	 * granted pages because backend is not accessing it at this point.
1847 	 */
1848 alloc_evtchn_fail:
1849 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1850 grant_rx_ring_fail:
1851 	free_page((unsigned long)rxs);
1852 alloc_rx_ring_fail:
1853 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1854 grant_tx_ring_fail:
1855 	free_page((unsigned long)txs);
1856 fail:
1857 	return err;
1858 }
1859 
1860 /* Queue-specific initialisation
1861  * This used to be done in xennet_create_dev() but must now
1862  * be run per-queue.
1863  */
1864 static int xennet_init_queue(struct netfront_queue *queue)
1865 {
1866 	unsigned short i;
1867 	int err = 0;
1868 	char *devid;
1869 
1870 	spin_lock_init(&queue->tx_lock);
1871 	spin_lock_init(&queue->rx_lock);
1872 
1873 	timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1874 
1875 	devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1876 	snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1877 		 devid, queue->id);
1878 
1879 	/* Initialise tx_skbs as a free chain containing every entry. */
1880 	queue->tx_skb_freelist = 0;
1881 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1882 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1883 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1884 		queue->grant_tx_page[i] = NULL;
1885 	}
1886 
1887 	/* Clear out rx_skbs */
1888 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1889 		queue->rx_skbs[i] = NULL;
1890 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1891 	}
1892 
1893 	/* A grant for every tx ring slot */
1894 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1895 					  &queue->gref_tx_head) < 0) {
1896 		pr_alert("can't alloc tx grant refs\n");
1897 		err = -ENOMEM;
1898 		goto exit;
1899 	}
1900 
1901 	/* A grant for every rx ring slot */
1902 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1903 					  &queue->gref_rx_head) < 0) {
1904 		pr_alert("can't alloc rx grant refs\n");
1905 		err = -ENOMEM;
1906 		goto exit_free_tx;
1907 	}
1908 
1909 	return 0;
1910 
1911  exit_free_tx:
1912 	gnttab_free_grant_references(queue->gref_tx_head);
1913  exit:
1914 	return err;
1915 }
1916 
1917 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1918 			   struct xenbus_transaction *xbt, int write_hierarchical)
1919 {
1920 	/* Write the queue-specific keys into XenStore in the traditional
1921 	 * way for a single queue, or in a queue subkeys for multiple
1922 	 * queues.
1923 	 */
1924 	struct xenbus_device *dev = queue->info->xbdev;
1925 	int err;
1926 	const char *message;
1927 	char *path;
1928 	size_t pathsize;
1929 
1930 	/* Choose the correct place to write the keys */
1931 	if (write_hierarchical) {
1932 		pathsize = strlen(dev->nodename) + 10;
1933 		path = kzalloc(pathsize, GFP_KERNEL);
1934 		if (!path) {
1935 			err = -ENOMEM;
1936 			message = "out of memory while writing ring references";
1937 			goto error;
1938 		}
1939 		snprintf(path, pathsize, "%s/queue-%u",
1940 				dev->nodename, queue->id);
1941 	} else {
1942 		path = (char *)dev->nodename;
1943 	}
1944 
1945 	/* Write ring references */
1946 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1947 			queue->tx_ring_ref);
1948 	if (err) {
1949 		message = "writing tx-ring-ref";
1950 		goto error;
1951 	}
1952 
1953 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1954 			queue->rx_ring_ref);
1955 	if (err) {
1956 		message = "writing rx-ring-ref";
1957 		goto error;
1958 	}
1959 
1960 	/* Write event channels; taking into account both shared
1961 	 * and split event channel scenarios.
1962 	 */
1963 	if (queue->tx_evtchn == queue->rx_evtchn) {
1964 		/* Shared event channel */
1965 		err = xenbus_printf(*xbt, path,
1966 				"event-channel", "%u", queue->tx_evtchn);
1967 		if (err) {
1968 			message = "writing event-channel";
1969 			goto error;
1970 		}
1971 	} else {
1972 		/* Split event channels */
1973 		err = xenbus_printf(*xbt, path,
1974 				"event-channel-tx", "%u", queue->tx_evtchn);
1975 		if (err) {
1976 			message = "writing event-channel-tx";
1977 			goto error;
1978 		}
1979 
1980 		err = xenbus_printf(*xbt, path,
1981 				"event-channel-rx", "%u", queue->rx_evtchn);
1982 		if (err) {
1983 			message = "writing event-channel-rx";
1984 			goto error;
1985 		}
1986 	}
1987 
1988 	if (write_hierarchical)
1989 		kfree(path);
1990 	return 0;
1991 
1992 error:
1993 	if (write_hierarchical)
1994 		kfree(path);
1995 	xenbus_dev_fatal(dev, err, "%s", message);
1996 	return err;
1997 }
1998 
1999 static void xennet_destroy_queues(struct netfront_info *info)
2000 {
2001 	unsigned int i;
2002 
2003 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
2004 		struct netfront_queue *queue = &info->queues[i];
2005 
2006 		if (netif_running(info->netdev))
2007 			napi_disable(&queue->napi);
2008 		netif_napi_del(&queue->napi);
2009 	}
2010 
2011 	kfree(info->queues);
2012 	info->queues = NULL;
2013 }
2014 
2015 
2016 
2017 static int xennet_create_page_pool(struct netfront_queue *queue)
2018 {
2019 	int err;
2020 	struct page_pool_params pp_params = {
2021 		.order = 0,
2022 		.flags = 0,
2023 		.pool_size = NET_RX_RING_SIZE,
2024 		.nid = NUMA_NO_NODE,
2025 		.dev = &queue->info->netdev->dev,
2026 		.offset = XDP_PACKET_HEADROOM,
2027 		.max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2028 	};
2029 
2030 	queue->page_pool = page_pool_create(&pp_params);
2031 	if (IS_ERR(queue->page_pool)) {
2032 		err = PTR_ERR(queue->page_pool);
2033 		queue->page_pool = NULL;
2034 		return err;
2035 	}
2036 
2037 	err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2038 			       queue->id);
2039 	if (err) {
2040 		netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2041 		goto err_free_pp;
2042 	}
2043 
2044 	err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2045 					 MEM_TYPE_PAGE_POOL, queue->page_pool);
2046 	if (err) {
2047 		netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2048 		goto err_unregister_rxq;
2049 	}
2050 	return 0;
2051 
2052 err_unregister_rxq:
2053 	xdp_rxq_info_unreg(&queue->xdp_rxq);
2054 err_free_pp:
2055 	page_pool_destroy(queue->page_pool);
2056 	queue->page_pool = NULL;
2057 	return err;
2058 }
2059 
2060 static int xennet_create_queues(struct netfront_info *info,
2061 				unsigned int *num_queues)
2062 {
2063 	unsigned int i;
2064 	int ret;
2065 
2066 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2067 			       GFP_KERNEL);
2068 	if (!info->queues)
2069 		return -ENOMEM;
2070 
2071 	for (i = 0; i < *num_queues; i++) {
2072 		struct netfront_queue *queue = &info->queues[i];
2073 
2074 		queue->id = i;
2075 		queue->info = info;
2076 
2077 		ret = xennet_init_queue(queue);
2078 		if (ret < 0) {
2079 			dev_warn(&info->xbdev->dev,
2080 				 "only created %d queues\n", i);
2081 			*num_queues = i;
2082 			break;
2083 		}
2084 
2085 		/* use page pool recycling instead of buddy allocator */
2086 		ret = xennet_create_page_pool(queue);
2087 		if (ret < 0) {
2088 			dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2089 			*num_queues = i;
2090 			return ret;
2091 		}
2092 
2093 		netif_napi_add(queue->info->netdev, &queue->napi,
2094 			       xennet_poll, 64);
2095 		if (netif_running(info->netdev))
2096 			napi_enable(&queue->napi);
2097 	}
2098 
2099 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
2100 
2101 	if (*num_queues == 0) {
2102 		dev_err(&info->xbdev->dev, "no queues\n");
2103 		return -EINVAL;
2104 	}
2105 	return 0;
2106 }
2107 
2108 /* Common code used when first setting up, and when resuming. */
2109 static int talk_to_netback(struct xenbus_device *dev,
2110 			   struct netfront_info *info)
2111 {
2112 	const char *message;
2113 	struct xenbus_transaction xbt;
2114 	int err;
2115 	unsigned int feature_split_evtchn;
2116 	unsigned int i = 0;
2117 	unsigned int max_queues = 0;
2118 	struct netfront_queue *queue = NULL;
2119 	unsigned int num_queues = 1;
2120 
2121 	info->netdev->irq = 0;
2122 
2123 	/* Check if backend supports multiple queues */
2124 	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2125 					  "multi-queue-max-queues", 1);
2126 	num_queues = min(max_queues, xennet_max_queues);
2127 
2128 	/* Check feature-split-event-channels */
2129 	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2130 					"feature-split-event-channels", 0);
2131 
2132 	/* Read mac addr. */
2133 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
2134 	if (err) {
2135 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2136 		goto out_unlocked;
2137 	}
2138 
2139 	info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2140 							      "feature-xdp-headroom", 0);
2141 	if (info->netback_has_xdp_headroom) {
2142 		/* set the current xen-netfront xdp state */
2143 		err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2144 					  NETBACK_XDP_HEADROOM_ENABLE :
2145 					  NETBACK_XDP_HEADROOM_DISABLE);
2146 		if (err)
2147 			goto out_unlocked;
2148 	}
2149 
2150 	rtnl_lock();
2151 	if (info->queues)
2152 		xennet_destroy_queues(info);
2153 
2154 	err = xennet_create_queues(info, &num_queues);
2155 	if (err < 0) {
2156 		xenbus_dev_fatal(dev, err, "creating queues");
2157 		kfree(info->queues);
2158 		info->queues = NULL;
2159 		goto out;
2160 	}
2161 	rtnl_unlock();
2162 
2163 	/* Create shared ring, alloc event channel -- for each queue */
2164 	for (i = 0; i < num_queues; ++i) {
2165 		queue = &info->queues[i];
2166 		err = setup_netfront(dev, queue, feature_split_evtchn);
2167 		if (err)
2168 			goto destroy_ring;
2169 	}
2170 
2171 again:
2172 	err = xenbus_transaction_start(&xbt);
2173 	if (err) {
2174 		xenbus_dev_fatal(dev, err, "starting transaction");
2175 		goto destroy_ring;
2176 	}
2177 
2178 	if (xenbus_exists(XBT_NIL,
2179 			  info->xbdev->otherend, "multi-queue-max-queues")) {
2180 		/* Write the number of queues */
2181 		err = xenbus_printf(xbt, dev->nodename,
2182 				    "multi-queue-num-queues", "%u", num_queues);
2183 		if (err) {
2184 			message = "writing multi-queue-num-queues";
2185 			goto abort_transaction_no_dev_fatal;
2186 		}
2187 	}
2188 
2189 	if (num_queues == 1) {
2190 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2191 		if (err)
2192 			goto abort_transaction_no_dev_fatal;
2193 	} else {
2194 		/* Write the keys for each queue */
2195 		for (i = 0; i < num_queues; ++i) {
2196 			queue = &info->queues[i];
2197 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2198 			if (err)
2199 				goto abort_transaction_no_dev_fatal;
2200 		}
2201 	}
2202 
2203 	/* The remaining keys are not queue-specific */
2204 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2205 			    1);
2206 	if (err) {
2207 		message = "writing request-rx-copy";
2208 		goto abort_transaction;
2209 	}
2210 
2211 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2212 	if (err) {
2213 		message = "writing feature-rx-notify";
2214 		goto abort_transaction;
2215 	}
2216 
2217 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2218 	if (err) {
2219 		message = "writing feature-sg";
2220 		goto abort_transaction;
2221 	}
2222 
2223 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2224 	if (err) {
2225 		message = "writing feature-gso-tcpv4";
2226 		goto abort_transaction;
2227 	}
2228 
2229 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2230 	if (err) {
2231 		message = "writing feature-gso-tcpv6";
2232 		goto abort_transaction;
2233 	}
2234 
2235 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2236 			   "1");
2237 	if (err) {
2238 		message = "writing feature-ipv6-csum-offload";
2239 		goto abort_transaction;
2240 	}
2241 
2242 	err = xenbus_transaction_end(xbt, 0);
2243 	if (err) {
2244 		if (err == -EAGAIN)
2245 			goto again;
2246 		xenbus_dev_fatal(dev, err, "completing transaction");
2247 		goto destroy_ring;
2248 	}
2249 
2250 	return 0;
2251 
2252  abort_transaction:
2253 	xenbus_dev_fatal(dev, err, "%s", message);
2254 abort_transaction_no_dev_fatal:
2255 	xenbus_transaction_end(xbt, 1);
2256  destroy_ring:
2257 	xennet_disconnect_backend(info);
2258 	rtnl_lock();
2259 	xennet_destroy_queues(info);
2260  out:
2261 	rtnl_unlock();
2262 out_unlocked:
2263 	device_unregister(&dev->dev);
2264 	return err;
2265 }
2266 
2267 static int xennet_connect(struct net_device *dev)
2268 {
2269 	struct netfront_info *np = netdev_priv(dev);
2270 	unsigned int num_queues = 0;
2271 	int err;
2272 	unsigned int j = 0;
2273 	struct netfront_queue *queue = NULL;
2274 
2275 	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2276 		dev_info(&dev->dev,
2277 			 "backend does not support copying receive path\n");
2278 		return -ENODEV;
2279 	}
2280 
2281 	err = talk_to_netback(np->xbdev, np);
2282 	if (err)
2283 		return err;
2284 	if (np->netback_has_xdp_headroom)
2285 		pr_info("backend supports XDP headroom\n");
2286 
2287 	/* talk_to_netback() sets the correct number of queues */
2288 	num_queues = dev->real_num_tx_queues;
2289 
2290 	if (dev->reg_state == NETREG_UNINITIALIZED) {
2291 		err = register_netdev(dev);
2292 		if (err) {
2293 			pr_warn("%s: register_netdev err=%d\n", __func__, err);
2294 			device_unregister(&np->xbdev->dev);
2295 			return err;
2296 		}
2297 	}
2298 
2299 	rtnl_lock();
2300 	netdev_update_features(dev);
2301 	rtnl_unlock();
2302 
2303 	/*
2304 	 * All public and private state should now be sane.  Get
2305 	 * ready to start sending and receiving packets and give the driver
2306 	 * domain a kick because we've probably just requeued some
2307 	 * packets.
2308 	 */
2309 	netif_carrier_on(np->netdev);
2310 	for (j = 0; j < num_queues; ++j) {
2311 		queue = &np->queues[j];
2312 
2313 		notify_remote_via_irq(queue->tx_irq);
2314 		if (queue->tx_irq != queue->rx_irq)
2315 			notify_remote_via_irq(queue->rx_irq);
2316 
2317 		spin_lock_irq(&queue->tx_lock);
2318 		xennet_tx_buf_gc(queue);
2319 		spin_unlock_irq(&queue->tx_lock);
2320 
2321 		spin_lock_bh(&queue->rx_lock);
2322 		xennet_alloc_rx_buffers(queue);
2323 		spin_unlock_bh(&queue->rx_lock);
2324 	}
2325 
2326 	return 0;
2327 }
2328 
2329 /**
2330  * Callback received when the backend's state changes.
2331  */
2332 static void netback_changed(struct xenbus_device *dev,
2333 			    enum xenbus_state backend_state)
2334 {
2335 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2336 	struct net_device *netdev = np->netdev;
2337 
2338 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2339 
2340 	wake_up_all(&module_wq);
2341 
2342 	switch (backend_state) {
2343 	case XenbusStateInitialising:
2344 	case XenbusStateInitialised:
2345 	case XenbusStateReconfiguring:
2346 	case XenbusStateReconfigured:
2347 	case XenbusStateUnknown:
2348 		break;
2349 
2350 	case XenbusStateInitWait:
2351 		if (dev->state != XenbusStateInitialising)
2352 			break;
2353 		if (xennet_connect(netdev) != 0)
2354 			break;
2355 		xenbus_switch_state(dev, XenbusStateConnected);
2356 		break;
2357 
2358 	case XenbusStateConnected:
2359 		netdev_notify_peers(netdev);
2360 		break;
2361 
2362 	case XenbusStateClosed:
2363 		if (dev->state == XenbusStateClosed)
2364 			break;
2365 		/* Fall through - Missed the backend's CLOSING state. */
2366 	case XenbusStateClosing:
2367 		xenbus_frontend_closed(dev);
2368 		break;
2369 	}
2370 }
2371 
2372 static const struct xennet_stat {
2373 	char name[ETH_GSTRING_LEN];
2374 	u16 offset;
2375 } xennet_stats[] = {
2376 	{
2377 		"rx_gso_checksum_fixup",
2378 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2379 	},
2380 };
2381 
2382 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2383 {
2384 	switch (string_set) {
2385 	case ETH_SS_STATS:
2386 		return ARRAY_SIZE(xennet_stats);
2387 	default:
2388 		return -EINVAL;
2389 	}
2390 }
2391 
2392 static void xennet_get_ethtool_stats(struct net_device *dev,
2393 				     struct ethtool_stats *stats, u64 * data)
2394 {
2395 	void *np = netdev_priv(dev);
2396 	int i;
2397 
2398 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2399 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2400 }
2401 
2402 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2403 {
2404 	int i;
2405 
2406 	switch (stringset) {
2407 	case ETH_SS_STATS:
2408 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2409 			memcpy(data + i * ETH_GSTRING_LEN,
2410 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2411 		break;
2412 	}
2413 }
2414 
2415 static const struct ethtool_ops xennet_ethtool_ops =
2416 {
2417 	.get_link = ethtool_op_get_link,
2418 
2419 	.get_sset_count = xennet_get_sset_count,
2420 	.get_ethtool_stats = xennet_get_ethtool_stats,
2421 	.get_strings = xennet_get_strings,
2422 	.get_ts_info = ethtool_op_get_ts_info,
2423 };
2424 
2425 #ifdef CONFIG_SYSFS
2426 static ssize_t show_rxbuf(struct device *dev,
2427 			  struct device_attribute *attr, char *buf)
2428 {
2429 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2430 }
2431 
2432 static ssize_t store_rxbuf(struct device *dev,
2433 			   struct device_attribute *attr,
2434 			   const char *buf, size_t len)
2435 {
2436 	char *endp;
2437 	unsigned long target;
2438 
2439 	if (!capable(CAP_NET_ADMIN))
2440 		return -EPERM;
2441 
2442 	target = simple_strtoul(buf, &endp, 0);
2443 	if (endp == buf)
2444 		return -EBADMSG;
2445 
2446 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2447 
2448 	return len;
2449 }
2450 
2451 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2452 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2453 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2454 
2455 static struct attribute *xennet_dev_attrs[] = {
2456 	&dev_attr_rxbuf_min.attr,
2457 	&dev_attr_rxbuf_max.attr,
2458 	&dev_attr_rxbuf_cur.attr,
2459 	NULL
2460 };
2461 
2462 static const struct attribute_group xennet_dev_group = {
2463 	.attrs = xennet_dev_attrs
2464 };
2465 #endif /* CONFIG_SYSFS */
2466 
2467 static void xennet_bus_close(struct xenbus_device *dev)
2468 {
2469 	int ret;
2470 
2471 	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2472 		return;
2473 	do {
2474 		xenbus_switch_state(dev, XenbusStateClosing);
2475 		ret = wait_event_timeout(module_wq,
2476 				   xenbus_read_driver_state(dev->otherend) ==
2477 				   XenbusStateClosing ||
2478 				   xenbus_read_driver_state(dev->otherend) ==
2479 				   XenbusStateClosed ||
2480 				   xenbus_read_driver_state(dev->otherend) ==
2481 				   XenbusStateUnknown,
2482 				   XENNET_TIMEOUT);
2483 	} while (!ret);
2484 
2485 	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2486 		return;
2487 
2488 	do {
2489 		xenbus_switch_state(dev, XenbusStateClosed);
2490 		ret = wait_event_timeout(module_wq,
2491 				   xenbus_read_driver_state(dev->otherend) ==
2492 				   XenbusStateClosed ||
2493 				   xenbus_read_driver_state(dev->otherend) ==
2494 				   XenbusStateUnknown,
2495 				   XENNET_TIMEOUT);
2496 	} while (!ret);
2497 }
2498 
2499 static int xennet_remove(struct xenbus_device *dev)
2500 {
2501 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2502 
2503 	xennet_bus_close(dev);
2504 	xennet_disconnect_backend(info);
2505 
2506 	if (info->netdev->reg_state == NETREG_REGISTERED)
2507 		unregister_netdev(info->netdev);
2508 
2509 	if (info->queues) {
2510 		rtnl_lock();
2511 		xennet_destroy_queues(info);
2512 		rtnl_unlock();
2513 	}
2514 	xennet_free_netdev(info->netdev);
2515 
2516 	return 0;
2517 }
2518 
2519 static const struct xenbus_device_id netfront_ids[] = {
2520 	{ "vif" },
2521 	{ "" }
2522 };
2523 
2524 static struct xenbus_driver netfront_driver = {
2525 	.ids = netfront_ids,
2526 	.probe = netfront_probe,
2527 	.remove = xennet_remove,
2528 	.resume = netfront_resume,
2529 	.otherend_changed = netback_changed,
2530 };
2531 
2532 static int __init netif_init(void)
2533 {
2534 	if (!xen_domain())
2535 		return -ENODEV;
2536 
2537 	if (!xen_has_pv_nic_devices())
2538 		return -ENODEV;
2539 
2540 	pr_info("Initialising Xen virtual ethernet driver\n");
2541 
2542 	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2543 	 * specified a value.
2544 	 */
2545 	if (xennet_max_queues == 0)
2546 		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2547 					  num_online_cpus());
2548 
2549 	return xenbus_register_frontend(&netfront_driver);
2550 }
2551 module_init(netif_init);
2552 
2553 
2554 static void __exit netif_exit(void)
2555 {
2556 	xenbus_unregister_driver(&netfront_driver);
2557 }
2558 module_exit(netif_exit);
2559 
2560 MODULE_DESCRIPTION("Xen virtual network device frontend");
2561 MODULE_LICENSE("GPL");
2562 MODULE_ALIAS("xen:vif");
2563 MODULE_ALIAS("xennet");
2564