xref: /linux/drivers/net/xen-netfront.c (revision 93a3545d812ae7cfe4426374e00a7d8f64ac02e0)
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 static const struct ethtool_ops xennet_ethtool_ops;
70 
71 struct netfront_cb {
72 	int pull_to;
73 };
74 
75 #define NETFRONT_SKB_CB(skb)	((struct netfront_cb *)((skb)->cb))
76 
77 #define RX_COPY_THRESHOLD 256
78 
79 #define GRANT_INVALID_REF	0
80 
81 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
82 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
83 
84 /* Minimum number of Rx slots (includes slot for GSO metadata). */
85 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
86 
87 /* Queue name is interface name with "-qNNN" appended */
88 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
89 
90 /* IRQ name is queue name with "-tx" or "-rx" appended */
91 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
92 
93 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
94 
95 struct netfront_stats {
96 	u64			packets;
97 	u64			bytes;
98 	struct u64_stats_sync	syncp;
99 };
100 
101 struct netfront_info;
102 
103 struct netfront_queue {
104 	unsigned int id; /* Queue ID, 0-based */
105 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
106 	struct netfront_info *info;
107 
108 	struct bpf_prog __rcu *xdp_prog;
109 
110 	struct napi_struct napi;
111 
112 	/* Split event channels support, tx_* == rx_* when using
113 	 * single event channel.
114 	 */
115 	unsigned int tx_evtchn, rx_evtchn;
116 	unsigned int tx_irq, rx_irq;
117 	/* Only used when split event channels support is enabled */
118 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
119 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
120 
121 	spinlock_t   tx_lock;
122 	struct xen_netif_tx_front_ring tx;
123 	int tx_ring_ref;
124 
125 	/*
126 	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
127 	 * are linked from tx_skb_freelist through skb_entry.link.
128 	 *
129 	 *  NB. Freelist index entries are always going to be less than
130 	 *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
131 	 *  greater than PAGE_OFFSET: we use this property to distinguish
132 	 *  them.
133 	 */
134 	union skb_entry {
135 		struct sk_buff *skb;
136 		unsigned long link;
137 	} tx_skbs[NET_TX_RING_SIZE];
138 	grant_ref_t gref_tx_head;
139 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
140 	struct page *grant_tx_page[NET_TX_RING_SIZE];
141 	unsigned tx_skb_freelist;
142 
143 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
144 	struct xen_netif_rx_front_ring rx;
145 	int rx_ring_ref;
146 
147 	struct timer_list rx_refill_timer;
148 
149 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
150 	grant_ref_t gref_rx_head;
151 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
152 
153 	struct page_pool *page_pool;
154 	struct xdp_rxq_info xdp_rxq;
155 };
156 
157 struct netfront_info {
158 	struct list_head list;
159 	struct net_device *netdev;
160 
161 	struct xenbus_device *xbdev;
162 
163 	/* Multi-queue support */
164 	struct netfront_queue *queues;
165 
166 	/* Statistics */
167 	struct netfront_stats __percpu *rx_stats;
168 	struct netfront_stats __percpu *tx_stats;
169 
170 	/* XDP state */
171 	bool netback_has_xdp_headroom;
172 	bool netfront_xdp_enabled;
173 
174 	atomic_t rx_gso_checksum_fixup;
175 };
176 
177 struct netfront_rx_info {
178 	struct xen_netif_rx_response rx;
179 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
180 };
181 
182 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
183 {
184 	list->link = id;
185 }
186 
187 static int skb_entry_is_link(const union skb_entry *list)
188 {
189 	BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
190 	return (unsigned long)list->skb < PAGE_OFFSET;
191 }
192 
193 /*
194  * Access macros for acquiring freeing slots in tx_skbs[].
195  */
196 
197 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
198 			       unsigned short id)
199 {
200 	skb_entry_set_link(&list[id], *head);
201 	*head = id;
202 }
203 
204 static unsigned short get_id_from_freelist(unsigned *head,
205 					   union skb_entry *list)
206 {
207 	unsigned int id = *head;
208 	*head = list[id].link;
209 	return id;
210 }
211 
212 static int xennet_rxidx(RING_IDX idx)
213 {
214 	return idx & (NET_RX_RING_SIZE - 1);
215 }
216 
217 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
218 					 RING_IDX ri)
219 {
220 	int i = xennet_rxidx(ri);
221 	struct sk_buff *skb = queue->rx_skbs[i];
222 	queue->rx_skbs[i] = NULL;
223 	return skb;
224 }
225 
226 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
227 					    RING_IDX ri)
228 {
229 	int i = xennet_rxidx(ri);
230 	grant_ref_t ref = queue->grant_rx_ref[i];
231 	queue->grant_rx_ref[i] = GRANT_INVALID_REF;
232 	return ref;
233 }
234 
235 #ifdef CONFIG_SYSFS
236 static const struct attribute_group xennet_dev_group;
237 #endif
238 
239 static bool xennet_can_sg(struct net_device *dev)
240 {
241 	return dev->features & NETIF_F_SG;
242 }
243 
244 
245 static void rx_refill_timeout(struct timer_list *t)
246 {
247 	struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
248 	napi_schedule(&queue->napi);
249 }
250 
251 static int netfront_tx_slot_available(struct netfront_queue *queue)
252 {
253 	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
254 		(NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
255 }
256 
257 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
258 {
259 	struct net_device *dev = queue->info->netdev;
260 	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
261 
262 	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
263 	    netfront_tx_slot_available(queue) &&
264 	    likely(netif_running(dev)))
265 		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
266 }
267 
268 
269 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
270 {
271 	struct sk_buff *skb;
272 	struct page *page;
273 
274 	skb = __netdev_alloc_skb(queue->info->netdev,
275 				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
276 				 GFP_ATOMIC | __GFP_NOWARN);
277 	if (unlikely(!skb))
278 		return NULL;
279 
280 	page = page_pool_dev_alloc_pages(queue->page_pool);
281 	if (unlikely(!page)) {
282 		kfree_skb(skb);
283 		return NULL;
284 	}
285 	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
286 
287 	/* Align ip header to a 16 bytes boundary */
288 	skb_reserve(skb, NET_IP_ALIGN);
289 	skb->dev = queue->info->netdev;
290 
291 	return skb;
292 }
293 
294 
295 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
296 {
297 	RING_IDX req_prod = queue->rx.req_prod_pvt;
298 	int notify;
299 	int err = 0;
300 
301 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
302 		return;
303 
304 	for (req_prod = queue->rx.req_prod_pvt;
305 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
306 	     req_prod++) {
307 		struct sk_buff *skb;
308 		unsigned short id;
309 		grant_ref_t ref;
310 		struct page *page;
311 		struct xen_netif_rx_request *req;
312 
313 		skb = xennet_alloc_one_rx_buffer(queue);
314 		if (!skb) {
315 			err = -ENOMEM;
316 			break;
317 		}
318 
319 		id = xennet_rxidx(req_prod);
320 
321 		BUG_ON(queue->rx_skbs[id]);
322 		queue->rx_skbs[id] = skb;
323 
324 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
325 		WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
326 		queue->grant_rx_ref[id] = ref;
327 
328 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
329 
330 		req = RING_GET_REQUEST(&queue->rx, req_prod);
331 		gnttab_page_grant_foreign_access_ref_one(ref,
332 							 queue->info->xbdev->otherend_id,
333 							 page,
334 							 0);
335 		req->id = id;
336 		req->gref = ref;
337 	}
338 
339 	queue->rx.req_prod_pvt = req_prod;
340 
341 	/* Try again later if there are not enough requests or skb allocation
342 	 * failed.
343 	 * Enough requests is quantified as the sum of newly created slots and
344 	 * the unconsumed slots at the backend.
345 	 */
346 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
347 	    unlikely(err)) {
348 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
349 		return;
350 	}
351 
352 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
353 	if (notify)
354 		notify_remote_via_irq(queue->rx_irq);
355 }
356 
357 static int xennet_open(struct net_device *dev)
358 {
359 	struct netfront_info *np = netdev_priv(dev);
360 	unsigned int num_queues = dev->real_num_tx_queues;
361 	unsigned int i = 0;
362 	struct netfront_queue *queue = NULL;
363 
364 	if (!np->queues)
365 		return -ENODEV;
366 
367 	for (i = 0; i < num_queues; ++i) {
368 		queue = &np->queues[i];
369 		napi_enable(&queue->napi);
370 
371 		spin_lock_bh(&queue->rx_lock);
372 		if (netif_carrier_ok(dev)) {
373 			xennet_alloc_rx_buffers(queue);
374 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
375 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
376 				napi_schedule(&queue->napi);
377 		}
378 		spin_unlock_bh(&queue->rx_lock);
379 	}
380 
381 	netif_tx_start_all_queues(dev);
382 
383 	return 0;
384 }
385 
386 static void xennet_tx_buf_gc(struct netfront_queue *queue)
387 {
388 	RING_IDX cons, prod;
389 	unsigned short id;
390 	struct sk_buff *skb;
391 	bool more_to_do;
392 
393 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
394 
395 	do {
396 		prod = queue->tx.sring->rsp_prod;
397 		rmb(); /* Ensure we see responses up to 'rp'. */
398 
399 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
400 			struct xen_netif_tx_response *txrsp;
401 
402 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
403 			if (txrsp->status == XEN_NETIF_RSP_NULL)
404 				continue;
405 
406 			id  = txrsp->id;
407 			skb = queue->tx_skbs[id].skb;
408 			if (unlikely(gnttab_query_foreign_access(
409 				queue->grant_tx_ref[id]) != 0)) {
410 				pr_alert("%s: warning -- grant still in use by backend domain\n",
411 					 __func__);
412 				BUG();
413 			}
414 			gnttab_end_foreign_access_ref(
415 				queue->grant_tx_ref[id], GNTMAP_readonly);
416 			gnttab_release_grant_reference(
417 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
418 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
419 			queue->grant_tx_page[id] = NULL;
420 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
421 			dev_kfree_skb_irq(skb);
422 		}
423 
424 		queue->tx.rsp_cons = prod;
425 
426 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
427 	} while (more_to_do);
428 
429 	xennet_maybe_wake_tx(queue);
430 }
431 
432 struct xennet_gnttab_make_txreq {
433 	struct netfront_queue *queue;
434 	struct sk_buff *skb;
435 	struct page *page;
436 	struct xen_netif_tx_request *tx; /* Last request */
437 	unsigned int size;
438 };
439 
440 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
441 				  unsigned int len, void *data)
442 {
443 	struct xennet_gnttab_make_txreq *info = data;
444 	unsigned int id;
445 	struct xen_netif_tx_request *tx;
446 	grant_ref_t ref;
447 	/* convenient aliases */
448 	struct page *page = info->page;
449 	struct netfront_queue *queue = info->queue;
450 	struct sk_buff *skb = info->skb;
451 
452 	id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
453 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
454 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
455 	WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
456 
457 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
458 					gfn, GNTMAP_readonly);
459 
460 	queue->tx_skbs[id].skb = skb;
461 	queue->grant_tx_page[id] = page;
462 	queue->grant_tx_ref[id] = ref;
463 
464 	tx->id = id;
465 	tx->gref = ref;
466 	tx->offset = offset;
467 	tx->size = len;
468 	tx->flags = 0;
469 
470 	info->tx = tx;
471 	info->size += tx->size;
472 }
473 
474 static struct xen_netif_tx_request *xennet_make_first_txreq(
475 	struct netfront_queue *queue, struct sk_buff *skb,
476 	struct page *page, unsigned int offset, unsigned int len)
477 {
478 	struct xennet_gnttab_make_txreq info = {
479 		.queue = queue,
480 		.skb = skb,
481 		.page = page,
482 		.size = 0,
483 	};
484 
485 	gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
486 
487 	return info.tx;
488 }
489 
490 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
491 				  unsigned int len, void *data)
492 {
493 	struct xennet_gnttab_make_txreq *info = data;
494 
495 	info->tx->flags |= XEN_NETTXF_more_data;
496 	skb_get(info->skb);
497 	xennet_tx_setup_grant(gfn, offset, len, data);
498 }
499 
500 static struct xen_netif_tx_request *xennet_make_txreqs(
501 	struct netfront_queue *queue, struct xen_netif_tx_request *tx,
502 	struct sk_buff *skb, struct page *page,
503 	unsigned int offset, unsigned int len)
504 {
505 	struct xennet_gnttab_make_txreq info = {
506 		.queue = queue,
507 		.skb = skb,
508 		.tx = tx,
509 	};
510 
511 	/* Skip unused frames from start of page */
512 	page += offset >> PAGE_SHIFT;
513 	offset &= ~PAGE_MASK;
514 
515 	while (len) {
516 		info.page = page;
517 		info.size = 0;
518 
519 		gnttab_foreach_grant_in_range(page, offset, len,
520 					      xennet_make_one_txreq,
521 					      &info);
522 
523 		page++;
524 		offset = 0;
525 		len -= info.size;
526 	}
527 
528 	return info.tx;
529 }
530 
531 /*
532  * Count how many ring slots are required to send this skb. Each frag
533  * might be a compound page.
534  */
535 static int xennet_count_skb_slots(struct sk_buff *skb)
536 {
537 	int i, frags = skb_shinfo(skb)->nr_frags;
538 	int slots;
539 
540 	slots = gnttab_count_grant(offset_in_page(skb->data),
541 				   skb_headlen(skb));
542 
543 	for (i = 0; i < frags; i++) {
544 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
545 		unsigned long size = skb_frag_size(frag);
546 		unsigned long offset = skb_frag_off(frag);
547 
548 		/* Skip unused frames from start of page */
549 		offset &= ~PAGE_MASK;
550 
551 		slots += gnttab_count_grant(offset, size);
552 	}
553 
554 	return slots;
555 }
556 
557 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
558 			       struct net_device *sb_dev)
559 {
560 	unsigned int num_queues = dev->real_num_tx_queues;
561 	u32 hash;
562 	u16 queue_idx;
563 
564 	/* First, check if there is only one queue */
565 	if (num_queues == 1) {
566 		queue_idx = 0;
567 	} else {
568 		hash = skb_get_hash(skb);
569 		queue_idx = hash % num_queues;
570 	}
571 
572 	return queue_idx;
573 }
574 
575 static int xennet_xdp_xmit_one(struct net_device *dev,
576 			       struct netfront_queue *queue,
577 			       struct xdp_frame *xdpf)
578 {
579 	struct netfront_info *np = netdev_priv(dev);
580 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
581 	int notify;
582 
583 	xennet_make_first_txreq(queue, NULL,
584 				virt_to_page(xdpf->data),
585 				offset_in_page(xdpf->data),
586 				xdpf->len);
587 
588 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
589 	if (notify)
590 		notify_remote_via_irq(queue->tx_irq);
591 
592 	u64_stats_update_begin(&tx_stats->syncp);
593 	tx_stats->bytes += xdpf->len;
594 	tx_stats->packets++;
595 	u64_stats_update_end(&tx_stats->syncp);
596 
597 	xennet_tx_buf_gc(queue);
598 
599 	return 0;
600 }
601 
602 static int xennet_xdp_xmit(struct net_device *dev, int n,
603 			   struct xdp_frame **frames, u32 flags)
604 {
605 	unsigned int num_queues = dev->real_num_tx_queues;
606 	struct netfront_info *np = netdev_priv(dev);
607 	struct netfront_queue *queue = NULL;
608 	unsigned long irq_flags;
609 	int drops = 0;
610 	int i, err;
611 
612 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
613 		return -EINVAL;
614 
615 	queue = &np->queues[smp_processor_id() % num_queues];
616 
617 	spin_lock_irqsave(&queue->tx_lock, irq_flags);
618 	for (i = 0; i < n; i++) {
619 		struct xdp_frame *xdpf = frames[i];
620 
621 		if (!xdpf)
622 			continue;
623 		err = xennet_xdp_xmit_one(dev, queue, xdpf);
624 		if (err) {
625 			xdp_return_frame_rx_napi(xdpf);
626 			drops++;
627 		}
628 	}
629 	spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
630 
631 	return n - drops;
632 }
633 
634 
635 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
636 
637 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
638 {
639 	struct netfront_info *np = netdev_priv(dev);
640 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
641 	struct xen_netif_tx_request *tx, *first_tx;
642 	unsigned int i;
643 	int notify;
644 	int slots;
645 	struct page *page;
646 	unsigned int offset;
647 	unsigned int len;
648 	unsigned long flags;
649 	struct netfront_queue *queue = NULL;
650 	unsigned int num_queues = dev->real_num_tx_queues;
651 	u16 queue_index;
652 	struct sk_buff *nskb;
653 
654 	/* Drop the packet if no queues are set up */
655 	if (num_queues < 1)
656 		goto drop;
657 	/* Determine which queue to transmit this SKB on */
658 	queue_index = skb_get_queue_mapping(skb);
659 	queue = &np->queues[queue_index];
660 
661 	/* If skb->len is too big for wire format, drop skb and alert
662 	 * user about misconfiguration.
663 	 */
664 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
665 		net_alert_ratelimited(
666 			"xennet: skb->len = %u, too big for wire format\n",
667 			skb->len);
668 		goto drop;
669 	}
670 
671 	slots = xennet_count_skb_slots(skb);
672 	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
673 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
674 				    slots, skb->len);
675 		if (skb_linearize(skb))
676 			goto drop;
677 	}
678 
679 	page = virt_to_page(skb->data);
680 	offset = offset_in_page(skb->data);
681 
682 	/* The first req should be at least ETH_HLEN size or the packet will be
683 	 * dropped by netback.
684 	 */
685 	if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
686 		nskb = skb_copy(skb, GFP_ATOMIC);
687 		if (!nskb)
688 			goto drop;
689 		dev_consume_skb_any(skb);
690 		skb = nskb;
691 		page = virt_to_page(skb->data);
692 		offset = offset_in_page(skb->data);
693 	}
694 
695 	len = skb_headlen(skb);
696 
697 	spin_lock_irqsave(&queue->tx_lock, flags);
698 
699 	if (unlikely(!netif_carrier_ok(dev) ||
700 		     (slots > 1 && !xennet_can_sg(dev)) ||
701 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
702 		spin_unlock_irqrestore(&queue->tx_lock, flags);
703 		goto drop;
704 	}
705 
706 	/* First request for the linear area. */
707 	first_tx = tx = xennet_make_first_txreq(queue, skb,
708 						page, offset, len);
709 	offset += tx->size;
710 	if (offset == PAGE_SIZE) {
711 		page++;
712 		offset = 0;
713 	}
714 	len -= tx->size;
715 
716 	if (skb->ip_summed == CHECKSUM_PARTIAL)
717 		/* local packet? */
718 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
719 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
720 		/* remote but checksummed. */
721 		tx->flags |= XEN_NETTXF_data_validated;
722 
723 	/* Optional extra info after the first request. */
724 	if (skb_shinfo(skb)->gso_size) {
725 		struct xen_netif_extra_info *gso;
726 
727 		gso = (struct xen_netif_extra_info *)
728 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
729 
730 		tx->flags |= XEN_NETTXF_extra_info;
731 
732 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
733 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
734 			XEN_NETIF_GSO_TYPE_TCPV6 :
735 			XEN_NETIF_GSO_TYPE_TCPV4;
736 		gso->u.gso.pad = 0;
737 		gso->u.gso.features = 0;
738 
739 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
740 		gso->flags = 0;
741 	}
742 
743 	/* Requests for the rest of the linear area. */
744 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
745 
746 	/* Requests for all the frags. */
747 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
748 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
749 		tx = xennet_make_txreqs(queue, tx, skb, skb_frag_page(frag),
750 					skb_frag_off(frag),
751 					skb_frag_size(frag));
752 	}
753 
754 	/* First request has the packet length. */
755 	first_tx->size = skb->len;
756 
757 	/* timestamp packet in software */
758 	skb_tx_timestamp(skb);
759 
760 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
761 	if (notify)
762 		notify_remote_via_irq(queue->tx_irq);
763 
764 	u64_stats_update_begin(&tx_stats->syncp);
765 	tx_stats->bytes += skb->len;
766 	tx_stats->packets++;
767 	u64_stats_update_end(&tx_stats->syncp);
768 
769 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
770 	xennet_tx_buf_gc(queue);
771 
772 	if (!netfront_tx_slot_available(queue))
773 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
774 
775 	spin_unlock_irqrestore(&queue->tx_lock, flags);
776 
777 	return NETDEV_TX_OK;
778 
779  drop:
780 	dev->stats.tx_dropped++;
781 	dev_kfree_skb_any(skb);
782 	return NETDEV_TX_OK;
783 }
784 
785 static int xennet_close(struct net_device *dev)
786 {
787 	struct netfront_info *np = netdev_priv(dev);
788 	unsigned int num_queues = dev->real_num_tx_queues;
789 	unsigned int i;
790 	struct netfront_queue *queue;
791 	netif_tx_stop_all_queues(np->netdev);
792 	for (i = 0; i < num_queues; ++i) {
793 		queue = &np->queues[i];
794 		napi_disable(&queue->napi);
795 	}
796 	return 0;
797 }
798 
799 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
800 				grant_ref_t ref)
801 {
802 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
803 
804 	BUG_ON(queue->rx_skbs[new]);
805 	queue->rx_skbs[new] = skb;
806 	queue->grant_rx_ref[new] = ref;
807 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
808 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
809 	queue->rx.req_prod_pvt++;
810 }
811 
812 static int xennet_get_extras(struct netfront_queue *queue,
813 			     struct xen_netif_extra_info *extras,
814 			     RING_IDX rp)
815 
816 {
817 	struct xen_netif_extra_info *extra;
818 	struct device *dev = &queue->info->netdev->dev;
819 	RING_IDX cons = queue->rx.rsp_cons;
820 	int err = 0;
821 
822 	do {
823 		struct sk_buff *skb;
824 		grant_ref_t ref;
825 
826 		if (unlikely(cons + 1 == rp)) {
827 			if (net_ratelimit())
828 				dev_warn(dev, "Missing extra info\n");
829 			err = -EBADR;
830 			break;
831 		}
832 
833 		extra = (struct xen_netif_extra_info *)
834 			RING_GET_RESPONSE(&queue->rx, ++cons);
835 
836 		if (unlikely(!extra->type ||
837 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
838 			if (net_ratelimit())
839 				dev_warn(dev, "Invalid extra type: %d\n",
840 					extra->type);
841 			err = -EINVAL;
842 		} else {
843 			memcpy(&extras[extra->type - 1], extra,
844 			       sizeof(*extra));
845 		}
846 
847 		skb = xennet_get_rx_skb(queue, cons);
848 		ref = xennet_get_rx_ref(queue, cons);
849 		xennet_move_rx_slot(queue, skb, ref);
850 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
851 
852 	queue->rx.rsp_cons = cons;
853 	return err;
854 }
855 
856 static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
857 		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
858 		   struct xdp_buff *xdp, bool *need_xdp_flush)
859 {
860 	struct xdp_frame *xdpf;
861 	u32 len = rx->status;
862 	u32 act;
863 	int err;
864 
865 	xdp->data_hard_start = page_address(pdata);
866 	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
867 	xdp_set_data_meta_invalid(xdp);
868 	xdp->data_end = xdp->data + len;
869 	xdp->rxq = &queue->xdp_rxq;
870 	xdp->frame_sz = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
871 
872 	act = bpf_prog_run_xdp(prog, xdp);
873 	switch (act) {
874 	case XDP_TX:
875 		get_page(pdata);
876 		xdpf = xdp_convert_buff_to_frame(xdp);
877 		err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
878 		if (unlikely(err < 0))
879 			trace_xdp_exception(queue->info->netdev, prog, act);
880 		break;
881 	case XDP_REDIRECT:
882 		get_page(pdata);
883 		err = xdp_do_redirect(queue->info->netdev, xdp, prog);
884 		*need_xdp_flush = true;
885 		if (unlikely(err))
886 			trace_xdp_exception(queue->info->netdev, prog, act);
887 		break;
888 	case XDP_PASS:
889 	case XDP_DROP:
890 		break;
891 
892 	case XDP_ABORTED:
893 		trace_xdp_exception(queue->info->netdev, prog, act);
894 		break;
895 
896 	default:
897 		bpf_warn_invalid_xdp_action(act);
898 	}
899 
900 	return act;
901 }
902 
903 static int xennet_get_responses(struct netfront_queue *queue,
904 				struct netfront_rx_info *rinfo, RING_IDX rp,
905 				struct sk_buff_head *list,
906 				bool *need_xdp_flush)
907 {
908 	struct xen_netif_rx_response *rx = &rinfo->rx;
909 	int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
910 	RING_IDX cons = queue->rx.rsp_cons;
911 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
912 	struct xen_netif_extra_info *extras = rinfo->extras;
913 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
914 	struct device *dev = &queue->info->netdev->dev;
915 	struct bpf_prog *xdp_prog;
916 	struct xdp_buff xdp;
917 	unsigned long ret;
918 	int slots = 1;
919 	int err = 0;
920 	u32 verdict;
921 
922 	if (rx->flags & XEN_NETRXF_extra_info) {
923 		err = xennet_get_extras(queue, extras, rp);
924 		if (!err) {
925 			if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
926 				struct xen_netif_extra_info *xdp;
927 
928 				xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
929 				rx->offset = xdp->u.xdp.headroom;
930 			}
931 		}
932 		cons = queue->rx.rsp_cons;
933 	}
934 
935 	for (;;) {
936 		if (unlikely(rx->status < 0 ||
937 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
938 			if (net_ratelimit())
939 				dev_warn(dev, "rx->offset: %u, size: %d\n",
940 					 rx->offset, rx->status);
941 			xennet_move_rx_slot(queue, skb, ref);
942 			err = -EINVAL;
943 			goto next;
944 		}
945 
946 		/*
947 		 * This definitely indicates a bug, either in this driver or in
948 		 * the backend driver. In future this should flag the bad
949 		 * situation to the system controller to reboot the backend.
950 		 */
951 		if (ref == GRANT_INVALID_REF) {
952 			if (net_ratelimit())
953 				dev_warn(dev, "Bad rx response id %d.\n",
954 					 rx->id);
955 			err = -EINVAL;
956 			goto next;
957 		}
958 
959 		ret = gnttab_end_foreign_access_ref(ref, 0);
960 		BUG_ON(!ret);
961 
962 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
963 
964 		rcu_read_lock();
965 		xdp_prog = rcu_dereference(queue->xdp_prog);
966 		if (xdp_prog) {
967 			if (!(rx->flags & XEN_NETRXF_more_data)) {
968 				/* currently only a single page contains data */
969 				verdict = xennet_run_xdp(queue,
970 							 skb_frag_page(&skb_shinfo(skb)->frags[0]),
971 							 rx, xdp_prog, &xdp, need_xdp_flush);
972 				if (verdict != XDP_PASS)
973 					err = -EINVAL;
974 			} else {
975 				/* drop the frame */
976 				err = -EINVAL;
977 			}
978 		}
979 		rcu_read_unlock();
980 next:
981 		__skb_queue_tail(list, skb);
982 		if (!(rx->flags & XEN_NETRXF_more_data))
983 			break;
984 
985 		if (cons + slots == rp) {
986 			if (net_ratelimit())
987 				dev_warn(dev, "Need more slots\n");
988 			err = -ENOENT;
989 			break;
990 		}
991 
992 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
993 		skb = xennet_get_rx_skb(queue, cons + slots);
994 		ref = xennet_get_rx_ref(queue, cons + slots);
995 		slots++;
996 	}
997 
998 	if (unlikely(slots > max)) {
999 		if (net_ratelimit())
1000 			dev_warn(dev, "Too many slots\n");
1001 		err = -E2BIG;
1002 	}
1003 
1004 	if (unlikely(err))
1005 		queue->rx.rsp_cons = cons + slots;
1006 
1007 	return err;
1008 }
1009 
1010 static int xennet_set_skb_gso(struct sk_buff *skb,
1011 			      struct xen_netif_extra_info *gso)
1012 {
1013 	if (!gso->u.gso.size) {
1014 		if (net_ratelimit())
1015 			pr_warn("GSO size must not be zero\n");
1016 		return -EINVAL;
1017 	}
1018 
1019 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1020 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
1021 		if (net_ratelimit())
1022 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
1023 		return -EINVAL;
1024 	}
1025 
1026 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
1027 	skb_shinfo(skb)->gso_type =
1028 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1029 		SKB_GSO_TCPV4 :
1030 		SKB_GSO_TCPV6;
1031 
1032 	/* Header must be checked, and gso_segs computed. */
1033 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1034 	skb_shinfo(skb)->gso_segs = 0;
1035 
1036 	return 0;
1037 }
1038 
1039 static int xennet_fill_frags(struct netfront_queue *queue,
1040 			     struct sk_buff *skb,
1041 			     struct sk_buff_head *list)
1042 {
1043 	RING_IDX cons = queue->rx.rsp_cons;
1044 	struct sk_buff *nskb;
1045 
1046 	while ((nskb = __skb_dequeue(list))) {
1047 		struct xen_netif_rx_response *rx =
1048 			RING_GET_RESPONSE(&queue->rx, ++cons);
1049 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
1050 
1051 		if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
1052 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1053 
1054 			BUG_ON(pull_to < skb_headlen(skb));
1055 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1056 		}
1057 		if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
1058 			queue->rx.rsp_cons = ++cons + skb_queue_len(list);
1059 			kfree_skb(nskb);
1060 			return -ENOENT;
1061 		}
1062 
1063 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1064 				skb_frag_page(nfrag),
1065 				rx->offset, rx->status, PAGE_SIZE);
1066 
1067 		skb_shinfo(nskb)->nr_frags = 0;
1068 		kfree_skb(nskb);
1069 	}
1070 
1071 	queue->rx.rsp_cons = cons;
1072 
1073 	return 0;
1074 }
1075 
1076 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
1077 {
1078 	bool recalculate_partial_csum = false;
1079 
1080 	/*
1081 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1082 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
1083 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1084 	 * recalculate the partial checksum.
1085 	 */
1086 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1087 		struct netfront_info *np = netdev_priv(dev);
1088 		atomic_inc(&np->rx_gso_checksum_fixup);
1089 		skb->ip_summed = CHECKSUM_PARTIAL;
1090 		recalculate_partial_csum = true;
1091 	}
1092 
1093 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1094 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1095 		return 0;
1096 
1097 	return skb_checksum_setup(skb, recalculate_partial_csum);
1098 }
1099 
1100 static int handle_incoming_queue(struct netfront_queue *queue,
1101 				 struct sk_buff_head *rxq)
1102 {
1103 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
1104 	int packets_dropped = 0;
1105 	struct sk_buff *skb;
1106 
1107 	while ((skb = __skb_dequeue(rxq)) != NULL) {
1108 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1109 
1110 		if (pull_to > skb_headlen(skb))
1111 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1112 
1113 		/* Ethernet work: Delayed to here as it peeks the header. */
1114 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
1115 		skb_reset_network_header(skb);
1116 
1117 		if (checksum_setup(queue->info->netdev, skb)) {
1118 			kfree_skb(skb);
1119 			packets_dropped++;
1120 			queue->info->netdev->stats.rx_errors++;
1121 			continue;
1122 		}
1123 
1124 		u64_stats_update_begin(&rx_stats->syncp);
1125 		rx_stats->packets++;
1126 		rx_stats->bytes += skb->len;
1127 		u64_stats_update_end(&rx_stats->syncp);
1128 
1129 		/* Pass it up. */
1130 		napi_gro_receive(&queue->napi, skb);
1131 	}
1132 
1133 	return packets_dropped;
1134 }
1135 
1136 static int xennet_poll(struct napi_struct *napi, int budget)
1137 {
1138 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1139 	struct net_device *dev = queue->info->netdev;
1140 	struct sk_buff *skb;
1141 	struct netfront_rx_info rinfo;
1142 	struct xen_netif_rx_response *rx = &rinfo.rx;
1143 	struct xen_netif_extra_info *extras = rinfo.extras;
1144 	RING_IDX i, rp;
1145 	int work_done;
1146 	struct sk_buff_head rxq;
1147 	struct sk_buff_head errq;
1148 	struct sk_buff_head tmpq;
1149 	int err;
1150 	bool need_xdp_flush = false;
1151 
1152 	spin_lock(&queue->rx_lock);
1153 
1154 	skb_queue_head_init(&rxq);
1155 	skb_queue_head_init(&errq);
1156 	skb_queue_head_init(&tmpq);
1157 
1158 	rp = queue->rx.sring->rsp_prod;
1159 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1160 
1161 	i = queue->rx.rsp_cons;
1162 	work_done = 0;
1163 	while ((i != rp) && (work_done < budget)) {
1164 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1165 		memset(extras, 0, sizeof(rinfo.extras));
1166 
1167 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1168 					   &need_xdp_flush);
1169 
1170 		if (unlikely(err)) {
1171 err:
1172 			while ((skb = __skb_dequeue(&tmpq)))
1173 				__skb_queue_tail(&errq, skb);
1174 			dev->stats.rx_errors++;
1175 			i = queue->rx.rsp_cons;
1176 			continue;
1177 		}
1178 
1179 		skb = __skb_dequeue(&tmpq);
1180 
1181 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1182 			struct xen_netif_extra_info *gso;
1183 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1184 
1185 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1186 				__skb_queue_head(&tmpq, skb);
1187 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1188 				goto err;
1189 			}
1190 		}
1191 
1192 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1193 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1194 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1195 
1196 		skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1197 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1198 		skb->data_len = rx->status;
1199 		skb->len += rx->status;
1200 
1201 		if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1202 			goto err;
1203 
1204 		if (rx->flags & XEN_NETRXF_csum_blank)
1205 			skb->ip_summed = CHECKSUM_PARTIAL;
1206 		else if (rx->flags & XEN_NETRXF_data_validated)
1207 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1208 
1209 		__skb_queue_tail(&rxq, skb);
1210 
1211 		i = ++queue->rx.rsp_cons;
1212 		work_done++;
1213 	}
1214 	if (need_xdp_flush)
1215 		xdp_do_flush();
1216 
1217 	__skb_queue_purge(&errq);
1218 
1219 	work_done -= handle_incoming_queue(queue, &rxq);
1220 
1221 	xennet_alloc_rx_buffers(queue);
1222 
1223 	if (work_done < budget) {
1224 		int more_to_do = 0;
1225 
1226 		napi_complete_done(napi, work_done);
1227 
1228 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1229 		if (more_to_do)
1230 			napi_schedule(napi);
1231 	}
1232 
1233 	spin_unlock(&queue->rx_lock);
1234 
1235 	return work_done;
1236 }
1237 
1238 static int xennet_change_mtu(struct net_device *dev, int mtu)
1239 {
1240 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1241 
1242 	if (mtu > max)
1243 		return -EINVAL;
1244 	dev->mtu = mtu;
1245 	return 0;
1246 }
1247 
1248 static void xennet_get_stats64(struct net_device *dev,
1249 			       struct rtnl_link_stats64 *tot)
1250 {
1251 	struct netfront_info *np = netdev_priv(dev);
1252 	int cpu;
1253 
1254 	for_each_possible_cpu(cpu) {
1255 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1256 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1257 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1258 		unsigned int start;
1259 
1260 		do {
1261 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1262 			tx_packets = tx_stats->packets;
1263 			tx_bytes = tx_stats->bytes;
1264 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1265 
1266 		do {
1267 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1268 			rx_packets = rx_stats->packets;
1269 			rx_bytes = rx_stats->bytes;
1270 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1271 
1272 		tot->rx_packets += rx_packets;
1273 		tot->tx_packets += tx_packets;
1274 		tot->rx_bytes   += rx_bytes;
1275 		tot->tx_bytes   += tx_bytes;
1276 	}
1277 
1278 	tot->rx_errors  = dev->stats.rx_errors;
1279 	tot->tx_dropped = dev->stats.tx_dropped;
1280 }
1281 
1282 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1283 {
1284 	struct sk_buff *skb;
1285 	int i;
1286 
1287 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1288 		/* Skip over entries which are actually freelist references */
1289 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1290 			continue;
1291 
1292 		skb = queue->tx_skbs[i].skb;
1293 		get_page(queue->grant_tx_page[i]);
1294 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1295 					  GNTMAP_readonly,
1296 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1297 		queue->grant_tx_page[i] = NULL;
1298 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1299 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1300 		dev_kfree_skb_irq(skb);
1301 	}
1302 }
1303 
1304 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1305 {
1306 	int id, ref;
1307 
1308 	spin_lock_bh(&queue->rx_lock);
1309 
1310 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1311 		struct sk_buff *skb;
1312 		struct page *page;
1313 
1314 		skb = queue->rx_skbs[id];
1315 		if (!skb)
1316 			continue;
1317 
1318 		ref = queue->grant_rx_ref[id];
1319 		if (ref == GRANT_INVALID_REF)
1320 			continue;
1321 
1322 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1323 
1324 		/* gnttab_end_foreign_access() needs a page ref until
1325 		 * foreign access is ended (which may be deferred).
1326 		 */
1327 		get_page(page);
1328 		gnttab_end_foreign_access(ref, 0,
1329 					  (unsigned long)page_address(page));
1330 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1331 
1332 		kfree_skb(skb);
1333 	}
1334 
1335 	spin_unlock_bh(&queue->rx_lock);
1336 }
1337 
1338 static netdev_features_t xennet_fix_features(struct net_device *dev,
1339 	netdev_features_t features)
1340 {
1341 	struct netfront_info *np = netdev_priv(dev);
1342 
1343 	if (features & NETIF_F_SG &&
1344 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1345 		features &= ~NETIF_F_SG;
1346 
1347 	if (features & NETIF_F_IPV6_CSUM &&
1348 	    !xenbus_read_unsigned(np->xbdev->otherend,
1349 				  "feature-ipv6-csum-offload", 0))
1350 		features &= ~NETIF_F_IPV6_CSUM;
1351 
1352 	if (features & NETIF_F_TSO &&
1353 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1354 		features &= ~NETIF_F_TSO;
1355 
1356 	if (features & NETIF_F_TSO6 &&
1357 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1358 		features &= ~NETIF_F_TSO6;
1359 
1360 	return features;
1361 }
1362 
1363 static int xennet_set_features(struct net_device *dev,
1364 	netdev_features_t features)
1365 {
1366 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1367 		netdev_info(dev, "Reducing MTU because no SG offload");
1368 		dev->mtu = ETH_DATA_LEN;
1369 	}
1370 
1371 	return 0;
1372 }
1373 
1374 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1375 {
1376 	struct netfront_queue *queue = dev_id;
1377 	unsigned long flags;
1378 
1379 	spin_lock_irqsave(&queue->tx_lock, flags);
1380 	xennet_tx_buf_gc(queue);
1381 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1382 
1383 	return IRQ_HANDLED;
1384 }
1385 
1386 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1387 {
1388 	struct netfront_queue *queue = dev_id;
1389 	struct net_device *dev = queue->info->netdev;
1390 
1391 	if (likely(netif_carrier_ok(dev) &&
1392 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1393 		napi_schedule(&queue->napi);
1394 
1395 	return IRQ_HANDLED;
1396 }
1397 
1398 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1399 {
1400 	xennet_tx_interrupt(irq, dev_id);
1401 	xennet_rx_interrupt(irq, dev_id);
1402 	return IRQ_HANDLED;
1403 }
1404 
1405 #ifdef CONFIG_NET_POLL_CONTROLLER
1406 static void xennet_poll_controller(struct net_device *dev)
1407 {
1408 	/* Poll each queue */
1409 	struct netfront_info *info = netdev_priv(dev);
1410 	unsigned int num_queues = dev->real_num_tx_queues;
1411 	unsigned int i;
1412 	for (i = 0; i < num_queues; ++i)
1413 		xennet_interrupt(0, &info->queues[i]);
1414 }
1415 #endif
1416 
1417 #define NETBACK_XDP_HEADROOM_DISABLE	0
1418 #define NETBACK_XDP_HEADROOM_ENABLE	1
1419 
1420 static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1421 {
1422 	int err;
1423 	unsigned short headroom;
1424 
1425 	headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1426 	err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1427 			    "xdp-headroom", "%hu",
1428 			    headroom);
1429 	if (err)
1430 		pr_warn("Error writing xdp-headroom\n");
1431 
1432 	return err;
1433 }
1434 
1435 static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1436 			  struct netlink_ext_ack *extack)
1437 {
1438 	unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1439 	struct netfront_info *np = netdev_priv(dev);
1440 	struct bpf_prog *old_prog;
1441 	unsigned int i, err;
1442 
1443 	if (dev->mtu > max_mtu) {
1444 		netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1445 		return -EINVAL;
1446 	}
1447 
1448 	if (!np->netback_has_xdp_headroom)
1449 		return 0;
1450 
1451 	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1452 
1453 	err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1454 				  NETBACK_XDP_HEADROOM_DISABLE);
1455 	if (err)
1456 		return err;
1457 
1458 	/* avoid the race with XDP headroom adjustment */
1459 	wait_event(module_wq,
1460 		   xenbus_read_driver_state(np->xbdev->otherend) ==
1461 		   XenbusStateReconfigured);
1462 	np->netfront_xdp_enabled = true;
1463 
1464 	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1465 
1466 	if (prog)
1467 		bpf_prog_add(prog, dev->real_num_tx_queues);
1468 
1469 	for (i = 0; i < dev->real_num_tx_queues; ++i)
1470 		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1471 
1472 	if (old_prog)
1473 		for (i = 0; i < dev->real_num_tx_queues; ++i)
1474 			bpf_prog_put(old_prog);
1475 
1476 	xenbus_switch_state(np->xbdev, XenbusStateConnected);
1477 
1478 	return 0;
1479 }
1480 
1481 static u32 xennet_xdp_query(struct net_device *dev)
1482 {
1483 	unsigned int num_queues = dev->real_num_tx_queues;
1484 	struct netfront_info *np = netdev_priv(dev);
1485 	const struct bpf_prog *xdp_prog;
1486 	struct netfront_queue *queue;
1487 	unsigned int i;
1488 
1489 	for (i = 0; i < num_queues; ++i) {
1490 		queue = &np->queues[i];
1491 		xdp_prog = rtnl_dereference(queue->xdp_prog);
1492 		if (xdp_prog)
1493 			return xdp_prog->aux->id;
1494 	}
1495 
1496 	return 0;
1497 }
1498 
1499 static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1500 {
1501 	switch (xdp->command) {
1502 	case XDP_SETUP_PROG:
1503 		return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1504 	case XDP_QUERY_PROG:
1505 		xdp->prog_id = xennet_xdp_query(dev);
1506 		return 0;
1507 	default:
1508 		return -EINVAL;
1509 	}
1510 }
1511 
1512 static const struct net_device_ops xennet_netdev_ops = {
1513 	.ndo_open            = xennet_open,
1514 	.ndo_stop            = xennet_close,
1515 	.ndo_start_xmit      = xennet_start_xmit,
1516 	.ndo_change_mtu	     = xennet_change_mtu,
1517 	.ndo_get_stats64     = xennet_get_stats64,
1518 	.ndo_set_mac_address = eth_mac_addr,
1519 	.ndo_validate_addr   = eth_validate_addr,
1520 	.ndo_fix_features    = xennet_fix_features,
1521 	.ndo_set_features    = xennet_set_features,
1522 	.ndo_select_queue    = xennet_select_queue,
1523 	.ndo_bpf            = xennet_xdp,
1524 	.ndo_xdp_xmit	    = xennet_xdp_xmit,
1525 #ifdef CONFIG_NET_POLL_CONTROLLER
1526 	.ndo_poll_controller = xennet_poll_controller,
1527 #endif
1528 };
1529 
1530 static void xennet_free_netdev(struct net_device *netdev)
1531 {
1532 	struct netfront_info *np = netdev_priv(netdev);
1533 
1534 	free_percpu(np->rx_stats);
1535 	free_percpu(np->tx_stats);
1536 	free_netdev(netdev);
1537 }
1538 
1539 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1540 {
1541 	int err;
1542 	struct net_device *netdev;
1543 	struct netfront_info *np;
1544 
1545 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1546 	if (!netdev)
1547 		return ERR_PTR(-ENOMEM);
1548 
1549 	np                   = netdev_priv(netdev);
1550 	np->xbdev            = dev;
1551 
1552 	np->queues = NULL;
1553 
1554 	err = -ENOMEM;
1555 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1556 	if (np->rx_stats == NULL)
1557 		goto exit;
1558 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1559 	if (np->tx_stats == NULL)
1560 		goto exit;
1561 
1562 	netdev->netdev_ops	= &xennet_netdev_ops;
1563 
1564 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1565 				  NETIF_F_GSO_ROBUST;
1566 	netdev->hw_features	= NETIF_F_SG |
1567 				  NETIF_F_IPV6_CSUM |
1568 				  NETIF_F_TSO | NETIF_F_TSO6;
1569 
1570 	/*
1571          * Assume that all hw features are available for now. This set
1572          * will be adjusted by the call to netdev_update_features() in
1573          * xennet_connect() which is the earliest point where we can
1574          * negotiate with the backend regarding supported features.
1575          */
1576 	netdev->features |= netdev->hw_features;
1577 
1578 	netdev->ethtool_ops = &xennet_ethtool_ops;
1579 	netdev->min_mtu = ETH_MIN_MTU;
1580 	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1581 	SET_NETDEV_DEV(netdev, &dev->dev);
1582 
1583 	np->netdev = netdev;
1584 	np->netfront_xdp_enabled = false;
1585 
1586 	netif_carrier_off(netdev);
1587 
1588 	xenbus_switch_state(dev, XenbusStateInitialising);
1589 	wait_event(module_wq,
1590 		   xenbus_read_driver_state(dev->otherend) !=
1591 		   XenbusStateClosed &&
1592 		   xenbus_read_driver_state(dev->otherend) !=
1593 		   XenbusStateUnknown);
1594 	return netdev;
1595 
1596  exit:
1597 	xennet_free_netdev(netdev);
1598 	return ERR_PTR(err);
1599 }
1600 
1601 /**
1602  * Entry point to this code when a new device is created.  Allocate the basic
1603  * structures and the ring buffers for communication with the backend, and
1604  * inform the backend of the appropriate details for those.
1605  */
1606 static int netfront_probe(struct xenbus_device *dev,
1607 			  const struct xenbus_device_id *id)
1608 {
1609 	int err;
1610 	struct net_device *netdev;
1611 	struct netfront_info *info;
1612 
1613 	netdev = xennet_create_dev(dev);
1614 	if (IS_ERR(netdev)) {
1615 		err = PTR_ERR(netdev);
1616 		xenbus_dev_fatal(dev, err, "creating netdev");
1617 		return err;
1618 	}
1619 
1620 	info = netdev_priv(netdev);
1621 	dev_set_drvdata(&dev->dev, info);
1622 #ifdef CONFIG_SYSFS
1623 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1624 #endif
1625 
1626 	return 0;
1627 }
1628 
1629 static void xennet_end_access(int ref, void *page)
1630 {
1631 	/* This frees the page as a side-effect */
1632 	if (ref != GRANT_INVALID_REF)
1633 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1634 }
1635 
1636 static void xennet_disconnect_backend(struct netfront_info *info)
1637 {
1638 	unsigned int i = 0;
1639 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1640 
1641 	netif_carrier_off(info->netdev);
1642 
1643 	for (i = 0; i < num_queues && info->queues; ++i) {
1644 		struct netfront_queue *queue = &info->queues[i];
1645 
1646 		del_timer_sync(&queue->rx_refill_timer);
1647 
1648 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1649 			unbind_from_irqhandler(queue->tx_irq, queue);
1650 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1651 			unbind_from_irqhandler(queue->tx_irq, queue);
1652 			unbind_from_irqhandler(queue->rx_irq, queue);
1653 		}
1654 		queue->tx_evtchn = queue->rx_evtchn = 0;
1655 		queue->tx_irq = queue->rx_irq = 0;
1656 
1657 		if (netif_running(info->netdev))
1658 			napi_synchronize(&queue->napi);
1659 
1660 		xennet_release_tx_bufs(queue);
1661 		xennet_release_rx_bufs(queue);
1662 		gnttab_free_grant_references(queue->gref_tx_head);
1663 		gnttab_free_grant_references(queue->gref_rx_head);
1664 
1665 		/* End access and free the pages */
1666 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1667 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1668 
1669 		queue->tx_ring_ref = GRANT_INVALID_REF;
1670 		queue->rx_ring_ref = GRANT_INVALID_REF;
1671 		queue->tx.sring = NULL;
1672 		queue->rx.sring = NULL;
1673 
1674 		page_pool_destroy(queue->page_pool);
1675 	}
1676 }
1677 
1678 /**
1679  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1680  * driver restart.  We tear down our netif structure and recreate it, but
1681  * leave the device-layer structures intact so that this is transparent to the
1682  * rest of the kernel.
1683  */
1684 static int netfront_resume(struct xenbus_device *dev)
1685 {
1686 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1687 
1688 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1689 
1690 	xennet_disconnect_backend(info);
1691 	return 0;
1692 }
1693 
1694 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1695 {
1696 	char *s, *e, *macstr;
1697 	int i;
1698 
1699 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1700 	if (IS_ERR(macstr))
1701 		return PTR_ERR(macstr);
1702 
1703 	for (i = 0; i < ETH_ALEN; i++) {
1704 		mac[i] = simple_strtoul(s, &e, 16);
1705 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1706 			kfree(macstr);
1707 			return -ENOENT;
1708 		}
1709 		s = e+1;
1710 	}
1711 
1712 	kfree(macstr);
1713 	return 0;
1714 }
1715 
1716 static int setup_netfront_single(struct netfront_queue *queue)
1717 {
1718 	int err;
1719 
1720 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1721 	if (err < 0)
1722 		goto fail;
1723 
1724 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1725 					xennet_interrupt,
1726 					0, queue->info->netdev->name, queue);
1727 	if (err < 0)
1728 		goto bind_fail;
1729 	queue->rx_evtchn = queue->tx_evtchn;
1730 	queue->rx_irq = queue->tx_irq = err;
1731 
1732 	return 0;
1733 
1734 bind_fail:
1735 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1736 	queue->tx_evtchn = 0;
1737 fail:
1738 	return err;
1739 }
1740 
1741 static int setup_netfront_split(struct netfront_queue *queue)
1742 {
1743 	int err;
1744 
1745 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1746 	if (err < 0)
1747 		goto fail;
1748 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1749 	if (err < 0)
1750 		goto alloc_rx_evtchn_fail;
1751 
1752 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1753 		 "%s-tx", queue->name);
1754 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1755 					xennet_tx_interrupt,
1756 					0, queue->tx_irq_name, queue);
1757 	if (err < 0)
1758 		goto bind_tx_fail;
1759 	queue->tx_irq = err;
1760 
1761 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1762 		 "%s-rx", queue->name);
1763 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1764 					xennet_rx_interrupt,
1765 					0, queue->rx_irq_name, queue);
1766 	if (err < 0)
1767 		goto bind_rx_fail;
1768 	queue->rx_irq = err;
1769 
1770 	return 0;
1771 
1772 bind_rx_fail:
1773 	unbind_from_irqhandler(queue->tx_irq, queue);
1774 	queue->tx_irq = 0;
1775 bind_tx_fail:
1776 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1777 	queue->rx_evtchn = 0;
1778 alloc_rx_evtchn_fail:
1779 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1780 	queue->tx_evtchn = 0;
1781 fail:
1782 	return err;
1783 }
1784 
1785 static int setup_netfront(struct xenbus_device *dev,
1786 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1787 {
1788 	struct xen_netif_tx_sring *txs;
1789 	struct xen_netif_rx_sring *rxs;
1790 	grant_ref_t gref;
1791 	int err;
1792 
1793 	queue->tx_ring_ref = GRANT_INVALID_REF;
1794 	queue->rx_ring_ref = GRANT_INVALID_REF;
1795 	queue->rx.sring = NULL;
1796 	queue->tx.sring = NULL;
1797 
1798 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1799 	if (!txs) {
1800 		err = -ENOMEM;
1801 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1802 		goto fail;
1803 	}
1804 	SHARED_RING_INIT(txs);
1805 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1806 
1807 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1808 	if (err < 0)
1809 		goto grant_tx_ring_fail;
1810 	queue->tx_ring_ref = gref;
1811 
1812 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1813 	if (!rxs) {
1814 		err = -ENOMEM;
1815 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1816 		goto alloc_rx_ring_fail;
1817 	}
1818 	SHARED_RING_INIT(rxs);
1819 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1820 
1821 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1822 	if (err < 0)
1823 		goto grant_rx_ring_fail;
1824 	queue->rx_ring_ref = gref;
1825 
1826 	if (feature_split_evtchn)
1827 		err = setup_netfront_split(queue);
1828 	/* setup single event channel if
1829 	 *  a) feature-split-event-channels == 0
1830 	 *  b) feature-split-event-channels == 1 but failed to setup
1831 	 */
1832 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1833 		err = setup_netfront_single(queue);
1834 
1835 	if (err)
1836 		goto alloc_evtchn_fail;
1837 
1838 	return 0;
1839 
1840 	/* If we fail to setup netfront, it is safe to just revoke access to
1841 	 * granted pages because backend is not accessing it at this point.
1842 	 */
1843 alloc_evtchn_fail:
1844 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1845 grant_rx_ring_fail:
1846 	free_page((unsigned long)rxs);
1847 alloc_rx_ring_fail:
1848 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1849 grant_tx_ring_fail:
1850 	free_page((unsigned long)txs);
1851 fail:
1852 	return err;
1853 }
1854 
1855 /* Queue-specific initialisation
1856  * This used to be done in xennet_create_dev() but must now
1857  * be run per-queue.
1858  */
1859 static int xennet_init_queue(struct netfront_queue *queue)
1860 {
1861 	unsigned short i;
1862 	int err = 0;
1863 	char *devid;
1864 
1865 	spin_lock_init(&queue->tx_lock);
1866 	spin_lock_init(&queue->rx_lock);
1867 
1868 	timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1869 
1870 	devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1871 	snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1872 		 devid, queue->id);
1873 
1874 	/* Initialise tx_skbs as a free chain containing every entry. */
1875 	queue->tx_skb_freelist = 0;
1876 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1877 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1878 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1879 		queue->grant_tx_page[i] = NULL;
1880 	}
1881 
1882 	/* Clear out rx_skbs */
1883 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1884 		queue->rx_skbs[i] = NULL;
1885 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1886 	}
1887 
1888 	/* A grant for every tx ring slot */
1889 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1890 					  &queue->gref_tx_head) < 0) {
1891 		pr_alert("can't alloc tx grant refs\n");
1892 		err = -ENOMEM;
1893 		goto exit;
1894 	}
1895 
1896 	/* A grant for every rx ring slot */
1897 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1898 					  &queue->gref_rx_head) < 0) {
1899 		pr_alert("can't alloc rx grant refs\n");
1900 		err = -ENOMEM;
1901 		goto exit_free_tx;
1902 	}
1903 
1904 	return 0;
1905 
1906  exit_free_tx:
1907 	gnttab_free_grant_references(queue->gref_tx_head);
1908  exit:
1909 	return err;
1910 }
1911 
1912 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1913 			   struct xenbus_transaction *xbt, int write_hierarchical)
1914 {
1915 	/* Write the queue-specific keys into XenStore in the traditional
1916 	 * way for a single queue, or in a queue subkeys for multiple
1917 	 * queues.
1918 	 */
1919 	struct xenbus_device *dev = queue->info->xbdev;
1920 	int err;
1921 	const char *message;
1922 	char *path;
1923 	size_t pathsize;
1924 
1925 	/* Choose the correct place to write the keys */
1926 	if (write_hierarchical) {
1927 		pathsize = strlen(dev->nodename) + 10;
1928 		path = kzalloc(pathsize, GFP_KERNEL);
1929 		if (!path) {
1930 			err = -ENOMEM;
1931 			message = "out of memory while writing ring references";
1932 			goto error;
1933 		}
1934 		snprintf(path, pathsize, "%s/queue-%u",
1935 				dev->nodename, queue->id);
1936 	} else {
1937 		path = (char *)dev->nodename;
1938 	}
1939 
1940 	/* Write ring references */
1941 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1942 			queue->tx_ring_ref);
1943 	if (err) {
1944 		message = "writing tx-ring-ref";
1945 		goto error;
1946 	}
1947 
1948 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1949 			queue->rx_ring_ref);
1950 	if (err) {
1951 		message = "writing rx-ring-ref";
1952 		goto error;
1953 	}
1954 
1955 	/* Write event channels; taking into account both shared
1956 	 * and split event channel scenarios.
1957 	 */
1958 	if (queue->tx_evtchn == queue->rx_evtchn) {
1959 		/* Shared event channel */
1960 		err = xenbus_printf(*xbt, path,
1961 				"event-channel", "%u", queue->tx_evtchn);
1962 		if (err) {
1963 			message = "writing event-channel";
1964 			goto error;
1965 		}
1966 	} else {
1967 		/* Split event channels */
1968 		err = xenbus_printf(*xbt, path,
1969 				"event-channel-tx", "%u", queue->tx_evtchn);
1970 		if (err) {
1971 			message = "writing event-channel-tx";
1972 			goto error;
1973 		}
1974 
1975 		err = xenbus_printf(*xbt, path,
1976 				"event-channel-rx", "%u", queue->rx_evtchn);
1977 		if (err) {
1978 			message = "writing event-channel-rx";
1979 			goto error;
1980 		}
1981 	}
1982 
1983 	if (write_hierarchical)
1984 		kfree(path);
1985 	return 0;
1986 
1987 error:
1988 	if (write_hierarchical)
1989 		kfree(path);
1990 	xenbus_dev_fatal(dev, err, "%s", message);
1991 	return err;
1992 }
1993 
1994 static void xennet_destroy_queues(struct netfront_info *info)
1995 {
1996 	unsigned int i;
1997 
1998 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1999 		struct netfront_queue *queue = &info->queues[i];
2000 
2001 		if (netif_running(info->netdev))
2002 			napi_disable(&queue->napi);
2003 		netif_napi_del(&queue->napi);
2004 	}
2005 
2006 	kfree(info->queues);
2007 	info->queues = NULL;
2008 }
2009 
2010 
2011 
2012 static int xennet_create_page_pool(struct netfront_queue *queue)
2013 {
2014 	int err;
2015 	struct page_pool_params pp_params = {
2016 		.order = 0,
2017 		.flags = 0,
2018 		.pool_size = NET_RX_RING_SIZE,
2019 		.nid = NUMA_NO_NODE,
2020 		.dev = &queue->info->netdev->dev,
2021 		.offset = XDP_PACKET_HEADROOM,
2022 		.max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2023 	};
2024 
2025 	queue->page_pool = page_pool_create(&pp_params);
2026 	if (IS_ERR(queue->page_pool)) {
2027 		err = PTR_ERR(queue->page_pool);
2028 		queue->page_pool = NULL;
2029 		return err;
2030 	}
2031 
2032 	err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2033 			       queue->id);
2034 	if (err) {
2035 		netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2036 		goto err_free_pp;
2037 	}
2038 
2039 	err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2040 					 MEM_TYPE_PAGE_POOL, queue->page_pool);
2041 	if (err) {
2042 		netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2043 		goto err_unregister_rxq;
2044 	}
2045 	return 0;
2046 
2047 err_unregister_rxq:
2048 	xdp_rxq_info_unreg(&queue->xdp_rxq);
2049 err_free_pp:
2050 	page_pool_destroy(queue->page_pool);
2051 	queue->page_pool = NULL;
2052 	return err;
2053 }
2054 
2055 static int xennet_create_queues(struct netfront_info *info,
2056 				unsigned int *num_queues)
2057 {
2058 	unsigned int i;
2059 	int ret;
2060 
2061 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2062 			       GFP_KERNEL);
2063 	if (!info->queues)
2064 		return -ENOMEM;
2065 
2066 	for (i = 0; i < *num_queues; i++) {
2067 		struct netfront_queue *queue = &info->queues[i];
2068 
2069 		queue->id = i;
2070 		queue->info = info;
2071 
2072 		ret = xennet_init_queue(queue);
2073 		if (ret < 0) {
2074 			dev_warn(&info->xbdev->dev,
2075 				 "only created %d queues\n", i);
2076 			*num_queues = i;
2077 			break;
2078 		}
2079 
2080 		/* use page pool recycling instead of buddy allocator */
2081 		ret = xennet_create_page_pool(queue);
2082 		if (ret < 0) {
2083 			dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2084 			*num_queues = i;
2085 			return ret;
2086 		}
2087 
2088 		netif_napi_add(queue->info->netdev, &queue->napi,
2089 			       xennet_poll, 64);
2090 		if (netif_running(info->netdev))
2091 			napi_enable(&queue->napi);
2092 	}
2093 
2094 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
2095 
2096 	if (*num_queues == 0) {
2097 		dev_err(&info->xbdev->dev, "no queues\n");
2098 		return -EINVAL;
2099 	}
2100 	return 0;
2101 }
2102 
2103 /* Common code used when first setting up, and when resuming. */
2104 static int talk_to_netback(struct xenbus_device *dev,
2105 			   struct netfront_info *info)
2106 {
2107 	const char *message;
2108 	struct xenbus_transaction xbt;
2109 	int err;
2110 	unsigned int feature_split_evtchn;
2111 	unsigned int i = 0;
2112 	unsigned int max_queues = 0;
2113 	struct netfront_queue *queue = NULL;
2114 	unsigned int num_queues = 1;
2115 
2116 	info->netdev->irq = 0;
2117 
2118 	/* Check if backend supports multiple queues */
2119 	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2120 					  "multi-queue-max-queues", 1);
2121 	num_queues = min(max_queues, xennet_max_queues);
2122 
2123 	/* Check feature-split-event-channels */
2124 	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2125 					"feature-split-event-channels", 0);
2126 
2127 	/* Read mac addr. */
2128 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
2129 	if (err) {
2130 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2131 		goto out_unlocked;
2132 	}
2133 
2134 	info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2135 							      "feature-xdp-headroom", 0);
2136 	if (info->netback_has_xdp_headroom) {
2137 		/* set the current xen-netfront xdp state */
2138 		err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2139 					  NETBACK_XDP_HEADROOM_ENABLE :
2140 					  NETBACK_XDP_HEADROOM_DISABLE);
2141 		if (err)
2142 			goto out_unlocked;
2143 	}
2144 
2145 	rtnl_lock();
2146 	if (info->queues)
2147 		xennet_destroy_queues(info);
2148 
2149 	err = xennet_create_queues(info, &num_queues);
2150 	if (err < 0) {
2151 		xenbus_dev_fatal(dev, err, "creating queues");
2152 		kfree(info->queues);
2153 		info->queues = NULL;
2154 		goto out;
2155 	}
2156 	rtnl_unlock();
2157 
2158 	/* Create shared ring, alloc event channel -- for each queue */
2159 	for (i = 0; i < num_queues; ++i) {
2160 		queue = &info->queues[i];
2161 		err = setup_netfront(dev, queue, feature_split_evtchn);
2162 		if (err)
2163 			goto destroy_ring;
2164 	}
2165 
2166 again:
2167 	err = xenbus_transaction_start(&xbt);
2168 	if (err) {
2169 		xenbus_dev_fatal(dev, err, "starting transaction");
2170 		goto destroy_ring;
2171 	}
2172 
2173 	if (xenbus_exists(XBT_NIL,
2174 			  info->xbdev->otherend, "multi-queue-max-queues")) {
2175 		/* Write the number of queues */
2176 		err = xenbus_printf(xbt, dev->nodename,
2177 				    "multi-queue-num-queues", "%u", num_queues);
2178 		if (err) {
2179 			message = "writing multi-queue-num-queues";
2180 			goto abort_transaction_no_dev_fatal;
2181 		}
2182 	}
2183 
2184 	if (num_queues == 1) {
2185 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2186 		if (err)
2187 			goto abort_transaction_no_dev_fatal;
2188 	} else {
2189 		/* Write the keys for each queue */
2190 		for (i = 0; i < num_queues; ++i) {
2191 			queue = &info->queues[i];
2192 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2193 			if (err)
2194 				goto abort_transaction_no_dev_fatal;
2195 		}
2196 	}
2197 
2198 	/* The remaining keys are not queue-specific */
2199 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2200 			    1);
2201 	if (err) {
2202 		message = "writing request-rx-copy";
2203 		goto abort_transaction;
2204 	}
2205 
2206 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2207 	if (err) {
2208 		message = "writing feature-rx-notify";
2209 		goto abort_transaction;
2210 	}
2211 
2212 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2213 	if (err) {
2214 		message = "writing feature-sg";
2215 		goto abort_transaction;
2216 	}
2217 
2218 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2219 	if (err) {
2220 		message = "writing feature-gso-tcpv4";
2221 		goto abort_transaction;
2222 	}
2223 
2224 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2225 	if (err) {
2226 		message = "writing feature-gso-tcpv6";
2227 		goto abort_transaction;
2228 	}
2229 
2230 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2231 			   "1");
2232 	if (err) {
2233 		message = "writing feature-ipv6-csum-offload";
2234 		goto abort_transaction;
2235 	}
2236 
2237 	err = xenbus_transaction_end(xbt, 0);
2238 	if (err) {
2239 		if (err == -EAGAIN)
2240 			goto again;
2241 		xenbus_dev_fatal(dev, err, "completing transaction");
2242 		goto destroy_ring;
2243 	}
2244 
2245 	return 0;
2246 
2247  abort_transaction:
2248 	xenbus_dev_fatal(dev, err, "%s", message);
2249 abort_transaction_no_dev_fatal:
2250 	xenbus_transaction_end(xbt, 1);
2251  destroy_ring:
2252 	xennet_disconnect_backend(info);
2253 	rtnl_lock();
2254 	xennet_destroy_queues(info);
2255  out:
2256 	rtnl_unlock();
2257 out_unlocked:
2258 	device_unregister(&dev->dev);
2259 	return err;
2260 }
2261 
2262 static int xennet_connect(struct net_device *dev)
2263 {
2264 	struct netfront_info *np = netdev_priv(dev);
2265 	unsigned int num_queues = 0;
2266 	int err;
2267 	unsigned int j = 0;
2268 	struct netfront_queue *queue = NULL;
2269 
2270 	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2271 		dev_info(&dev->dev,
2272 			 "backend does not support copying receive path\n");
2273 		return -ENODEV;
2274 	}
2275 
2276 	err = talk_to_netback(np->xbdev, np);
2277 	if (err)
2278 		return err;
2279 	if (np->netback_has_xdp_headroom)
2280 		pr_info("backend supports XDP headroom\n");
2281 
2282 	/* talk_to_netback() sets the correct number of queues */
2283 	num_queues = dev->real_num_tx_queues;
2284 
2285 	if (dev->reg_state == NETREG_UNINITIALIZED) {
2286 		err = register_netdev(dev);
2287 		if (err) {
2288 			pr_warn("%s: register_netdev err=%d\n", __func__, err);
2289 			device_unregister(&np->xbdev->dev);
2290 			return err;
2291 		}
2292 	}
2293 
2294 	rtnl_lock();
2295 	netdev_update_features(dev);
2296 	rtnl_unlock();
2297 
2298 	/*
2299 	 * All public and private state should now be sane.  Get
2300 	 * ready to start sending and receiving packets and give the driver
2301 	 * domain a kick because we've probably just requeued some
2302 	 * packets.
2303 	 */
2304 	netif_carrier_on(np->netdev);
2305 	for (j = 0; j < num_queues; ++j) {
2306 		queue = &np->queues[j];
2307 
2308 		notify_remote_via_irq(queue->tx_irq);
2309 		if (queue->tx_irq != queue->rx_irq)
2310 			notify_remote_via_irq(queue->rx_irq);
2311 
2312 		spin_lock_irq(&queue->tx_lock);
2313 		xennet_tx_buf_gc(queue);
2314 		spin_unlock_irq(&queue->tx_lock);
2315 
2316 		spin_lock_bh(&queue->rx_lock);
2317 		xennet_alloc_rx_buffers(queue);
2318 		spin_unlock_bh(&queue->rx_lock);
2319 	}
2320 
2321 	return 0;
2322 }
2323 
2324 /**
2325  * Callback received when the backend's state changes.
2326  */
2327 static void netback_changed(struct xenbus_device *dev,
2328 			    enum xenbus_state backend_state)
2329 {
2330 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2331 	struct net_device *netdev = np->netdev;
2332 
2333 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2334 
2335 	wake_up_all(&module_wq);
2336 
2337 	switch (backend_state) {
2338 	case XenbusStateInitialising:
2339 	case XenbusStateInitialised:
2340 	case XenbusStateReconfiguring:
2341 	case XenbusStateReconfigured:
2342 	case XenbusStateUnknown:
2343 		break;
2344 
2345 	case XenbusStateInitWait:
2346 		if (dev->state != XenbusStateInitialising)
2347 			break;
2348 		if (xennet_connect(netdev) != 0)
2349 			break;
2350 		xenbus_switch_state(dev, XenbusStateConnected);
2351 		break;
2352 
2353 	case XenbusStateConnected:
2354 		netdev_notify_peers(netdev);
2355 		break;
2356 
2357 	case XenbusStateClosed:
2358 		if (dev->state == XenbusStateClosed)
2359 			break;
2360 		/* Fall through - Missed the backend's CLOSING state. */
2361 	case XenbusStateClosing:
2362 		xenbus_frontend_closed(dev);
2363 		break;
2364 	}
2365 }
2366 
2367 static const struct xennet_stat {
2368 	char name[ETH_GSTRING_LEN];
2369 	u16 offset;
2370 } xennet_stats[] = {
2371 	{
2372 		"rx_gso_checksum_fixup",
2373 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2374 	},
2375 };
2376 
2377 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2378 {
2379 	switch (string_set) {
2380 	case ETH_SS_STATS:
2381 		return ARRAY_SIZE(xennet_stats);
2382 	default:
2383 		return -EINVAL;
2384 	}
2385 }
2386 
2387 static void xennet_get_ethtool_stats(struct net_device *dev,
2388 				     struct ethtool_stats *stats, u64 * data)
2389 {
2390 	void *np = netdev_priv(dev);
2391 	int i;
2392 
2393 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2394 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2395 }
2396 
2397 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2398 {
2399 	int i;
2400 
2401 	switch (stringset) {
2402 	case ETH_SS_STATS:
2403 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2404 			memcpy(data + i * ETH_GSTRING_LEN,
2405 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2406 		break;
2407 	}
2408 }
2409 
2410 static const struct ethtool_ops xennet_ethtool_ops =
2411 {
2412 	.get_link = ethtool_op_get_link,
2413 
2414 	.get_sset_count = xennet_get_sset_count,
2415 	.get_ethtool_stats = xennet_get_ethtool_stats,
2416 	.get_strings = xennet_get_strings,
2417 	.get_ts_info = ethtool_op_get_ts_info,
2418 };
2419 
2420 #ifdef CONFIG_SYSFS
2421 static ssize_t show_rxbuf(struct device *dev,
2422 			  struct device_attribute *attr, char *buf)
2423 {
2424 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2425 }
2426 
2427 static ssize_t store_rxbuf(struct device *dev,
2428 			   struct device_attribute *attr,
2429 			   const char *buf, size_t len)
2430 {
2431 	char *endp;
2432 	unsigned long target;
2433 
2434 	if (!capable(CAP_NET_ADMIN))
2435 		return -EPERM;
2436 
2437 	target = simple_strtoul(buf, &endp, 0);
2438 	if (endp == buf)
2439 		return -EBADMSG;
2440 
2441 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2442 
2443 	return len;
2444 }
2445 
2446 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2447 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2448 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2449 
2450 static struct attribute *xennet_dev_attrs[] = {
2451 	&dev_attr_rxbuf_min.attr,
2452 	&dev_attr_rxbuf_max.attr,
2453 	&dev_attr_rxbuf_cur.attr,
2454 	NULL
2455 };
2456 
2457 static const struct attribute_group xennet_dev_group = {
2458 	.attrs = xennet_dev_attrs
2459 };
2460 #endif /* CONFIG_SYSFS */
2461 
2462 static int xennet_remove(struct xenbus_device *dev)
2463 {
2464 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2465 
2466 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2467 
2468 	if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) {
2469 		xenbus_switch_state(dev, XenbusStateClosing);
2470 		wait_event(module_wq,
2471 			   xenbus_read_driver_state(dev->otherend) ==
2472 			   XenbusStateClosing ||
2473 			   xenbus_read_driver_state(dev->otherend) ==
2474 			   XenbusStateUnknown);
2475 
2476 		xenbus_switch_state(dev, XenbusStateClosed);
2477 		wait_event(module_wq,
2478 			   xenbus_read_driver_state(dev->otherend) ==
2479 			   XenbusStateClosed ||
2480 			   xenbus_read_driver_state(dev->otherend) ==
2481 			   XenbusStateUnknown);
2482 	}
2483 
2484 	xennet_disconnect_backend(info);
2485 
2486 	if (info->netdev->reg_state == NETREG_REGISTERED)
2487 		unregister_netdev(info->netdev);
2488 
2489 	if (info->queues) {
2490 		rtnl_lock();
2491 		xennet_destroy_queues(info);
2492 		rtnl_unlock();
2493 	}
2494 	xennet_free_netdev(info->netdev);
2495 
2496 	return 0;
2497 }
2498 
2499 static const struct xenbus_device_id netfront_ids[] = {
2500 	{ "vif" },
2501 	{ "" }
2502 };
2503 
2504 static struct xenbus_driver netfront_driver = {
2505 	.ids = netfront_ids,
2506 	.probe = netfront_probe,
2507 	.remove = xennet_remove,
2508 	.resume = netfront_resume,
2509 	.otherend_changed = netback_changed,
2510 };
2511 
2512 static int __init netif_init(void)
2513 {
2514 	if (!xen_domain())
2515 		return -ENODEV;
2516 
2517 	if (!xen_has_pv_nic_devices())
2518 		return -ENODEV;
2519 
2520 	pr_info("Initialising Xen virtual ethernet driver\n");
2521 
2522 	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2523 	 * specified a value.
2524 	 */
2525 	if (xennet_max_queues == 0)
2526 		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2527 					  num_online_cpus());
2528 
2529 	return xenbus_register_frontend(&netfront_driver);
2530 }
2531 module_init(netif_init);
2532 
2533 
2534 static void __exit netif_exit(void)
2535 {
2536 	xenbus_unregister_driver(&netfront_driver);
2537 }
2538 module_exit(netif_exit);
2539 
2540 MODULE_DESCRIPTION("Xen virtual network device frontend");
2541 MODULE_LICENSE("GPL");
2542 MODULE_ALIAS("xen:vif");
2543 MODULE_ALIAS("xennet");
2544