xref: /openbmc/linux/drivers/net/xen-netfront.c (revision 97e6ea6d)
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 tx_link.
130 	 */
131 	struct sk_buff *tx_skbs[NET_TX_RING_SIZE];
132 	unsigned short tx_link[NET_TX_RING_SIZE];
133 #define TX_LINK_NONE 0xffff
134 #define TX_PENDING   0xfffe
135 	grant_ref_t gref_tx_head;
136 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
137 	struct page *grant_tx_page[NET_TX_RING_SIZE];
138 	unsigned tx_skb_freelist;
139 	unsigned int tx_pend_queue;
140 
141 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
142 	struct xen_netif_rx_front_ring rx;
143 	int rx_ring_ref;
144 
145 	struct timer_list rx_refill_timer;
146 
147 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
148 	grant_ref_t gref_rx_head;
149 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
150 
151 	struct page_pool *page_pool;
152 	struct xdp_rxq_info xdp_rxq;
153 };
154 
155 struct netfront_info {
156 	struct list_head list;
157 	struct net_device *netdev;
158 
159 	struct xenbus_device *xbdev;
160 
161 	/* Multi-queue support */
162 	struct netfront_queue *queues;
163 
164 	/* Statistics */
165 	struct netfront_stats __percpu *rx_stats;
166 	struct netfront_stats __percpu *tx_stats;
167 
168 	/* XDP state */
169 	bool netback_has_xdp_headroom;
170 	bool netfront_xdp_enabled;
171 
172 	/* Is device behaving sane? */
173 	bool broken;
174 
175 	atomic_t rx_gso_checksum_fixup;
176 };
177 
178 struct netfront_rx_info {
179 	struct xen_netif_rx_response rx;
180 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
181 };
182 
183 /*
184  * Access macros for acquiring freeing slots in tx_skbs[].
185  */
186 
187 static void add_id_to_list(unsigned *head, unsigned short *list,
188 			   unsigned short id)
189 {
190 	list[id] = *head;
191 	*head = id;
192 }
193 
194 static unsigned short get_id_from_list(unsigned *head, unsigned short *list)
195 {
196 	unsigned int id = *head;
197 
198 	if (id != TX_LINK_NONE) {
199 		*head = list[id];
200 		list[id] = TX_LINK_NONE;
201 	}
202 	return id;
203 }
204 
205 static int xennet_rxidx(RING_IDX idx)
206 {
207 	return idx & (NET_RX_RING_SIZE - 1);
208 }
209 
210 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
211 					 RING_IDX ri)
212 {
213 	int i = xennet_rxidx(ri);
214 	struct sk_buff *skb = queue->rx_skbs[i];
215 	queue->rx_skbs[i] = NULL;
216 	return skb;
217 }
218 
219 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
220 					    RING_IDX ri)
221 {
222 	int i = xennet_rxidx(ri);
223 	grant_ref_t ref = queue->grant_rx_ref[i];
224 	queue->grant_rx_ref[i] = GRANT_INVALID_REF;
225 	return ref;
226 }
227 
228 #ifdef CONFIG_SYSFS
229 static const struct attribute_group xennet_dev_group;
230 #endif
231 
232 static bool xennet_can_sg(struct net_device *dev)
233 {
234 	return dev->features & NETIF_F_SG;
235 }
236 
237 
238 static void rx_refill_timeout(struct timer_list *t)
239 {
240 	struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
241 	napi_schedule(&queue->napi);
242 }
243 
244 static int netfront_tx_slot_available(struct netfront_queue *queue)
245 {
246 	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
247 		(NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
248 }
249 
250 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
251 {
252 	struct net_device *dev = queue->info->netdev;
253 	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
254 
255 	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
256 	    netfront_tx_slot_available(queue) &&
257 	    likely(netif_running(dev)))
258 		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
259 }
260 
261 
262 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
263 {
264 	struct sk_buff *skb;
265 	struct page *page;
266 
267 	skb = __netdev_alloc_skb(queue->info->netdev,
268 				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
269 				 GFP_ATOMIC | __GFP_NOWARN);
270 	if (unlikely(!skb))
271 		return NULL;
272 
273 	page = page_pool_dev_alloc_pages(queue->page_pool);
274 	if (unlikely(!page)) {
275 		kfree_skb(skb);
276 		return NULL;
277 	}
278 	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
279 
280 	/* Align ip header to a 16 bytes boundary */
281 	skb_reserve(skb, NET_IP_ALIGN);
282 	skb->dev = queue->info->netdev;
283 
284 	return skb;
285 }
286 
287 
288 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
289 {
290 	RING_IDX req_prod = queue->rx.req_prod_pvt;
291 	int notify;
292 	int err = 0;
293 
294 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
295 		return;
296 
297 	for (req_prod = queue->rx.req_prod_pvt;
298 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
299 	     req_prod++) {
300 		struct sk_buff *skb;
301 		unsigned short id;
302 		grant_ref_t ref;
303 		struct page *page;
304 		struct xen_netif_rx_request *req;
305 
306 		skb = xennet_alloc_one_rx_buffer(queue);
307 		if (!skb) {
308 			err = -ENOMEM;
309 			break;
310 		}
311 
312 		id = xennet_rxidx(req_prod);
313 
314 		BUG_ON(queue->rx_skbs[id]);
315 		queue->rx_skbs[id] = skb;
316 
317 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
318 		WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
319 		queue->grant_rx_ref[id] = ref;
320 
321 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
322 
323 		req = RING_GET_REQUEST(&queue->rx, req_prod);
324 		gnttab_page_grant_foreign_access_ref_one(ref,
325 							 queue->info->xbdev->otherend_id,
326 							 page,
327 							 0);
328 		req->id = id;
329 		req->gref = ref;
330 	}
331 
332 	queue->rx.req_prod_pvt = req_prod;
333 
334 	/* Try again later if there are not enough requests or skb allocation
335 	 * failed.
336 	 * Enough requests is quantified as the sum of newly created slots and
337 	 * the unconsumed slots at the backend.
338 	 */
339 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
340 	    unlikely(err)) {
341 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
342 		return;
343 	}
344 
345 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
346 	if (notify)
347 		notify_remote_via_irq(queue->rx_irq);
348 }
349 
350 static int xennet_open(struct net_device *dev)
351 {
352 	struct netfront_info *np = netdev_priv(dev);
353 	unsigned int num_queues = dev->real_num_tx_queues;
354 	unsigned int i = 0;
355 	struct netfront_queue *queue = NULL;
356 
357 	if (!np->queues || np->broken)
358 		return -ENODEV;
359 
360 	for (i = 0; i < num_queues; ++i) {
361 		queue = &np->queues[i];
362 		napi_enable(&queue->napi);
363 
364 		spin_lock_bh(&queue->rx_lock);
365 		if (netif_carrier_ok(dev)) {
366 			xennet_alloc_rx_buffers(queue);
367 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
368 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
369 				napi_schedule(&queue->napi);
370 		}
371 		spin_unlock_bh(&queue->rx_lock);
372 	}
373 
374 	netif_tx_start_all_queues(dev);
375 
376 	return 0;
377 }
378 
379 static void xennet_tx_buf_gc(struct netfront_queue *queue)
380 {
381 	RING_IDX cons, prod;
382 	unsigned short id;
383 	struct sk_buff *skb;
384 	bool more_to_do;
385 	const struct device *dev = &queue->info->netdev->dev;
386 
387 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
388 
389 	do {
390 		prod = queue->tx.sring->rsp_prod;
391 		if (RING_RESPONSE_PROD_OVERFLOW(&queue->tx, prod)) {
392 			dev_alert(dev, "Illegal number of responses %u\n",
393 				  prod - queue->tx.rsp_cons);
394 			goto err;
395 		}
396 		rmb(); /* Ensure we see responses up to 'rp'. */
397 
398 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
399 			struct xen_netif_tx_response txrsp;
400 
401 			RING_COPY_RESPONSE(&queue->tx, cons, &txrsp);
402 			if (txrsp.status == XEN_NETIF_RSP_NULL)
403 				continue;
404 
405 			id = txrsp.id;
406 			if (id >= RING_SIZE(&queue->tx)) {
407 				dev_alert(dev,
408 					  "Response has incorrect id (%u)\n",
409 					  id);
410 				goto err;
411 			}
412 			if (queue->tx_link[id] != TX_PENDING) {
413 				dev_alert(dev,
414 					  "Response for inactive request\n");
415 				goto err;
416 			}
417 
418 			queue->tx_link[id] = TX_LINK_NONE;
419 			skb = queue->tx_skbs[id];
420 			queue->tx_skbs[id] = NULL;
421 			if (unlikely(gnttab_query_foreign_access(
422 				queue->grant_tx_ref[id]) != 0)) {
423 				dev_alert(dev,
424 					  "Grant still in use by backend domain\n");
425 				goto err;
426 			}
427 			gnttab_end_foreign_access_ref(
428 				queue->grant_tx_ref[id], GNTMAP_readonly);
429 			gnttab_release_grant_reference(
430 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
431 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
432 			queue->grant_tx_page[id] = NULL;
433 			add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, id);
434 			dev_kfree_skb_irq(skb);
435 		}
436 
437 		queue->tx.rsp_cons = prod;
438 
439 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
440 	} while (more_to_do);
441 
442 	xennet_maybe_wake_tx(queue);
443 
444 	return;
445 
446  err:
447 	queue->info->broken = true;
448 	dev_alert(dev, "Disabled for further use\n");
449 }
450 
451 struct xennet_gnttab_make_txreq {
452 	struct netfront_queue *queue;
453 	struct sk_buff *skb;
454 	struct page *page;
455 	struct xen_netif_tx_request *tx;      /* Last request on ring page */
456 	struct xen_netif_tx_request tx_local; /* Last request local copy*/
457 	unsigned int size;
458 };
459 
460 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
461 				  unsigned int len, void *data)
462 {
463 	struct xennet_gnttab_make_txreq *info = data;
464 	unsigned int id;
465 	struct xen_netif_tx_request *tx;
466 	grant_ref_t ref;
467 	/* convenient aliases */
468 	struct page *page = info->page;
469 	struct netfront_queue *queue = info->queue;
470 	struct sk_buff *skb = info->skb;
471 
472 	id = get_id_from_list(&queue->tx_skb_freelist, queue->tx_link);
473 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
474 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
475 	WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
476 
477 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
478 					gfn, GNTMAP_readonly);
479 
480 	queue->tx_skbs[id] = skb;
481 	queue->grant_tx_page[id] = page;
482 	queue->grant_tx_ref[id] = ref;
483 
484 	info->tx_local.id = id;
485 	info->tx_local.gref = ref;
486 	info->tx_local.offset = offset;
487 	info->tx_local.size = len;
488 	info->tx_local.flags = 0;
489 
490 	*tx = info->tx_local;
491 
492 	/*
493 	 * Put the request in the pending queue, it will be set to be pending
494 	 * when the producer index is about to be raised.
495 	 */
496 	add_id_to_list(&queue->tx_pend_queue, queue->tx_link, id);
497 
498 	info->tx = tx;
499 	info->size += info->tx_local.size;
500 }
501 
502 static struct xen_netif_tx_request *xennet_make_first_txreq(
503 	struct xennet_gnttab_make_txreq *info,
504 	unsigned int offset, unsigned int len)
505 {
506 	info->size = 0;
507 
508 	gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info);
509 
510 	return info->tx;
511 }
512 
513 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
514 				  unsigned int len, void *data)
515 {
516 	struct xennet_gnttab_make_txreq *info = data;
517 
518 	info->tx->flags |= XEN_NETTXF_more_data;
519 	skb_get(info->skb);
520 	xennet_tx_setup_grant(gfn, offset, len, data);
521 }
522 
523 static void xennet_make_txreqs(
524 	struct xennet_gnttab_make_txreq *info,
525 	struct page *page,
526 	unsigned int offset, unsigned int len)
527 {
528 	/* Skip unused frames from start of page */
529 	page += offset >> PAGE_SHIFT;
530 	offset &= ~PAGE_MASK;
531 
532 	while (len) {
533 		info->page = page;
534 		info->size = 0;
535 
536 		gnttab_foreach_grant_in_range(page, offset, len,
537 					      xennet_make_one_txreq,
538 					      info);
539 
540 		page++;
541 		offset = 0;
542 		len -= info->size;
543 	}
544 }
545 
546 /*
547  * Count how many ring slots are required to send this skb. Each frag
548  * might be a compound page.
549  */
550 static int xennet_count_skb_slots(struct sk_buff *skb)
551 {
552 	int i, frags = skb_shinfo(skb)->nr_frags;
553 	int slots;
554 
555 	slots = gnttab_count_grant(offset_in_page(skb->data),
556 				   skb_headlen(skb));
557 
558 	for (i = 0; i < frags; i++) {
559 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
560 		unsigned long size = skb_frag_size(frag);
561 		unsigned long offset = skb_frag_off(frag);
562 
563 		/* Skip unused frames from start of page */
564 		offset &= ~PAGE_MASK;
565 
566 		slots += gnttab_count_grant(offset, size);
567 	}
568 
569 	return slots;
570 }
571 
572 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
573 			       struct net_device *sb_dev)
574 {
575 	unsigned int num_queues = dev->real_num_tx_queues;
576 	u32 hash;
577 	u16 queue_idx;
578 
579 	/* First, check if there is only one queue */
580 	if (num_queues == 1) {
581 		queue_idx = 0;
582 	} else {
583 		hash = skb_get_hash(skb);
584 		queue_idx = hash % num_queues;
585 	}
586 
587 	return queue_idx;
588 }
589 
590 static void xennet_mark_tx_pending(struct netfront_queue *queue)
591 {
592 	unsigned int i;
593 
594 	while ((i = get_id_from_list(&queue->tx_pend_queue, queue->tx_link)) !=
595 	       TX_LINK_NONE)
596 		queue->tx_link[i] = TX_PENDING;
597 }
598 
599 static int xennet_xdp_xmit_one(struct net_device *dev,
600 			       struct netfront_queue *queue,
601 			       struct xdp_frame *xdpf)
602 {
603 	struct netfront_info *np = netdev_priv(dev);
604 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
605 	struct xennet_gnttab_make_txreq info = {
606 		.queue = queue,
607 		.skb = NULL,
608 		.page = virt_to_page(xdpf->data),
609 	};
610 	int notify;
611 
612 	xennet_make_first_txreq(&info,
613 				offset_in_page(xdpf->data),
614 				xdpf->len);
615 
616 	xennet_mark_tx_pending(queue);
617 
618 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
619 	if (notify)
620 		notify_remote_via_irq(queue->tx_irq);
621 
622 	u64_stats_update_begin(&tx_stats->syncp);
623 	tx_stats->bytes += xdpf->len;
624 	tx_stats->packets++;
625 	u64_stats_update_end(&tx_stats->syncp);
626 
627 	xennet_tx_buf_gc(queue);
628 
629 	return 0;
630 }
631 
632 static int xennet_xdp_xmit(struct net_device *dev, int n,
633 			   struct xdp_frame **frames, u32 flags)
634 {
635 	unsigned int num_queues = dev->real_num_tx_queues;
636 	struct netfront_info *np = netdev_priv(dev);
637 	struct netfront_queue *queue = NULL;
638 	unsigned long irq_flags;
639 	int nxmit = 0;
640 	int i;
641 
642 	if (unlikely(np->broken))
643 		return -ENODEV;
644 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
645 		return -EINVAL;
646 
647 	queue = &np->queues[smp_processor_id() % num_queues];
648 
649 	spin_lock_irqsave(&queue->tx_lock, irq_flags);
650 	for (i = 0; i < n; i++) {
651 		struct xdp_frame *xdpf = frames[i];
652 
653 		if (!xdpf)
654 			continue;
655 		if (xennet_xdp_xmit_one(dev, queue, xdpf))
656 			break;
657 		nxmit++;
658 	}
659 	spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
660 
661 	return nxmit;
662 }
663 
664 
665 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
666 
667 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
668 {
669 	struct netfront_info *np = netdev_priv(dev);
670 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
671 	struct xen_netif_tx_request *first_tx;
672 	unsigned int i;
673 	int notify;
674 	int slots;
675 	struct page *page;
676 	unsigned int offset;
677 	unsigned int len;
678 	unsigned long flags;
679 	struct netfront_queue *queue = NULL;
680 	struct xennet_gnttab_make_txreq info = { };
681 	unsigned int num_queues = dev->real_num_tx_queues;
682 	u16 queue_index;
683 	struct sk_buff *nskb;
684 
685 	/* Drop the packet if no queues are set up */
686 	if (num_queues < 1)
687 		goto drop;
688 	if (unlikely(np->broken))
689 		goto drop;
690 	/* Determine which queue to transmit this SKB on */
691 	queue_index = skb_get_queue_mapping(skb);
692 	queue = &np->queues[queue_index];
693 
694 	/* If skb->len is too big for wire format, drop skb and alert
695 	 * user about misconfiguration.
696 	 */
697 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
698 		net_alert_ratelimited(
699 			"xennet: skb->len = %u, too big for wire format\n",
700 			skb->len);
701 		goto drop;
702 	}
703 
704 	slots = xennet_count_skb_slots(skb);
705 	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
706 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
707 				    slots, skb->len);
708 		if (skb_linearize(skb))
709 			goto drop;
710 	}
711 
712 	page = virt_to_page(skb->data);
713 	offset = offset_in_page(skb->data);
714 
715 	/* The first req should be at least ETH_HLEN size or the packet will be
716 	 * dropped by netback.
717 	 */
718 	if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
719 		nskb = skb_copy(skb, GFP_ATOMIC);
720 		if (!nskb)
721 			goto drop;
722 		dev_consume_skb_any(skb);
723 		skb = nskb;
724 		page = virt_to_page(skb->data);
725 		offset = offset_in_page(skb->data);
726 	}
727 
728 	len = skb_headlen(skb);
729 
730 	spin_lock_irqsave(&queue->tx_lock, flags);
731 
732 	if (unlikely(!netif_carrier_ok(dev) ||
733 		     (slots > 1 && !xennet_can_sg(dev)) ||
734 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
735 		spin_unlock_irqrestore(&queue->tx_lock, flags);
736 		goto drop;
737 	}
738 
739 	/* First request for the linear area. */
740 	info.queue = queue;
741 	info.skb = skb;
742 	info.page = page;
743 	first_tx = xennet_make_first_txreq(&info, offset, len);
744 	offset += info.tx_local.size;
745 	if (offset == PAGE_SIZE) {
746 		page++;
747 		offset = 0;
748 	}
749 	len -= info.tx_local.size;
750 
751 	if (skb->ip_summed == CHECKSUM_PARTIAL)
752 		/* local packet? */
753 		first_tx->flags |= XEN_NETTXF_csum_blank |
754 				   XEN_NETTXF_data_validated;
755 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
756 		/* remote but checksummed. */
757 		first_tx->flags |= XEN_NETTXF_data_validated;
758 
759 	/* Optional extra info after the first request. */
760 	if (skb_shinfo(skb)->gso_size) {
761 		struct xen_netif_extra_info *gso;
762 
763 		gso = (struct xen_netif_extra_info *)
764 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
765 
766 		first_tx->flags |= XEN_NETTXF_extra_info;
767 
768 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
769 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
770 			XEN_NETIF_GSO_TYPE_TCPV6 :
771 			XEN_NETIF_GSO_TYPE_TCPV4;
772 		gso->u.gso.pad = 0;
773 		gso->u.gso.features = 0;
774 
775 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
776 		gso->flags = 0;
777 	}
778 
779 	/* Requests for the rest of the linear area. */
780 	xennet_make_txreqs(&info, page, offset, len);
781 
782 	/* Requests for all the frags. */
783 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
784 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
785 		xennet_make_txreqs(&info, skb_frag_page(frag),
786 					skb_frag_off(frag),
787 					skb_frag_size(frag));
788 	}
789 
790 	/* First request has the packet length. */
791 	first_tx->size = skb->len;
792 
793 	/* timestamp packet in software */
794 	skb_tx_timestamp(skb);
795 
796 	xennet_mark_tx_pending(queue);
797 
798 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
799 	if (notify)
800 		notify_remote_via_irq(queue->tx_irq);
801 
802 	u64_stats_update_begin(&tx_stats->syncp);
803 	tx_stats->bytes += skb->len;
804 	tx_stats->packets++;
805 	u64_stats_update_end(&tx_stats->syncp);
806 
807 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
808 	xennet_tx_buf_gc(queue);
809 
810 	if (!netfront_tx_slot_available(queue))
811 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
812 
813 	spin_unlock_irqrestore(&queue->tx_lock, flags);
814 
815 	return NETDEV_TX_OK;
816 
817  drop:
818 	dev->stats.tx_dropped++;
819 	dev_kfree_skb_any(skb);
820 	return NETDEV_TX_OK;
821 }
822 
823 static int xennet_close(struct net_device *dev)
824 {
825 	struct netfront_info *np = netdev_priv(dev);
826 	unsigned int num_queues = dev->real_num_tx_queues;
827 	unsigned int i;
828 	struct netfront_queue *queue;
829 	netif_tx_stop_all_queues(np->netdev);
830 	for (i = 0; i < num_queues; ++i) {
831 		queue = &np->queues[i];
832 		napi_disable(&queue->napi);
833 	}
834 	return 0;
835 }
836 
837 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
838 				grant_ref_t ref)
839 {
840 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
841 
842 	BUG_ON(queue->rx_skbs[new]);
843 	queue->rx_skbs[new] = skb;
844 	queue->grant_rx_ref[new] = ref;
845 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
846 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
847 	queue->rx.req_prod_pvt++;
848 }
849 
850 static int xennet_get_extras(struct netfront_queue *queue,
851 			     struct xen_netif_extra_info *extras,
852 			     RING_IDX rp)
853 
854 {
855 	struct xen_netif_extra_info extra;
856 	struct device *dev = &queue->info->netdev->dev;
857 	RING_IDX cons = queue->rx.rsp_cons;
858 	int err = 0;
859 
860 	do {
861 		struct sk_buff *skb;
862 		grant_ref_t ref;
863 
864 		if (unlikely(cons + 1 == rp)) {
865 			if (net_ratelimit())
866 				dev_warn(dev, "Missing extra info\n");
867 			err = -EBADR;
868 			break;
869 		}
870 
871 		RING_COPY_RESPONSE(&queue->rx, ++cons, &extra);
872 
873 		if (unlikely(!extra.type ||
874 			     extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
875 			if (net_ratelimit())
876 				dev_warn(dev, "Invalid extra type: %d\n",
877 					 extra.type);
878 			err = -EINVAL;
879 		} else {
880 			extras[extra.type - 1] = extra;
881 		}
882 
883 		skb = xennet_get_rx_skb(queue, cons);
884 		ref = xennet_get_rx_ref(queue, cons);
885 		xennet_move_rx_slot(queue, skb, ref);
886 	} while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
887 
888 	queue->rx.rsp_cons = cons;
889 	return err;
890 }
891 
892 static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
893 		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
894 		   struct xdp_buff *xdp, bool *need_xdp_flush)
895 {
896 	struct xdp_frame *xdpf;
897 	u32 len = rx->status;
898 	u32 act;
899 	int err;
900 
901 	xdp_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
902 		      &queue->xdp_rxq);
903 	xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM,
904 			 len, false);
905 
906 	act = bpf_prog_run_xdp(prog, xdp);
907 	switch (act) {
908 	case XDP_TX:
909 		get_page(pdata);
910 		xdpf = xdp_convert_buff_to_frame(xdp);
911 		err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
912 		if (unlikely(!err))
913 			xdp_return_frame_rx_napi(xdpf);
914 		else if (unlikely(err < 0))
915 			trace_xdp_exception(queue->info->netdev, prog, act);
916 		break;
917 	case XDP_REDIRECT:
918 		get_page(pdata);
919 		err = xdp_do_redirect(queue->info->netdev, xdp, prog);
920 		*need_xdp_flush = true;
921 		if (unlikely(err))
922 			trace_xdp_exception(queue->info->netdev, prog, act);
923 		break;
924 	case XDP_PASS:
925 	case XDP_DROP:
926 		break;
927 
928 	case XDP_ABORTED:
929 		trace_xdp_exception(queue->info->netdev, prog, act);
930 		break;
931 
932 	default:
933 		bpf_warn_invalid_xdp_action(act);
934 	}
935 
936 	return act;
937 }
938 
939 static int xennet_get_responses(struct netfront_queue *queue,
940 				struct netfront_rx_info *rinfo, RING_IDX rp,
941 				struct sk_buff_head *list,
942 				bool *need_xdp_flush)
943 {
944 	struct xen_netif_rx_response *rx = &rinfo->rx, rx_local;
945 	int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
946 	RING_IDX cons = queue->rx.rsp_cons;
947 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
948 	struct xen_netif_extra_info *extras = rinfo->extras;
949 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
950 	struct device *dev = &queue->info->netdev->dev;
951 	struct bpf_prog *xdp_prog;
952 	struct xdp_buff xdp;
953 	unsigned long ret;
954 	int slots = 1;
955 	int err = 0;
956 	u32 verdict;
957 
958 	if (rx->flags & XEN_NETRXF_extra_info) {
959 		err = xennet_get_extras(queue, extras, rp);
960 		if (!err) {
961 			if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
962 				struct xen_netif_extra_info *xdp;
963 
964 				xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
965 				rx->offset = xdp->u.xdp.headroom;
966 			}
967 		}
968 		cons = queue->rx.rsp_cons;
969 	}
970 
971 	for (;;) {
972 		if (unlikely(rx->status < 0 ||
973 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
974 			if (net_ratelimit())
975 				dev_warn(dev, "rx->offset: %u, size: %d\n",
976 					 rx->offset, rx->status);
977 			xennet_move_rx_slot(queue, skb, ref);
978 			err = -EINVAL;
979 			goto next;
980 		}
981 
982 		/*
983 		 * This definitely indicates a bug, either in this driver or in
984 		 * the backend driver. In future this should flag the bad
985 		 * situation to the system controller to reboot the backend.
986 		 */
987 		if (ref == GRANT_INVALID_REF) {
988 			if (net_ratelimit())
989 				dev_warn(dev, "Bad rx response id %d.\n",
990 					 rx->id);
991 			err = -EINVAL;
992 			goto next;
993 		}
994 
995 		ret = gnttab_end_foreign_access_ref(ref, 0);
996 		BUG_ON(!ret);
997 
998 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
999 
1000 		rcu_read_lock();
1001 		xdp_prog = rcu_dereference(queue->xdp_prog);
1002 		if (xdp_prog) {
1003 			if (!(rx->flags & XEN_NETRXF_more_data)) {
1004 				/* currently only a single page contains data */
1005 				verdict = xennet_run_xdp(queue,
1006 							 skb_frag_page(&skb_shinfo(skb)->frags[0]),
1007 							 rx, xdp_prog, &xdp, need_xdp_flush);
1008 				if (verdict != XDP_PASS)
1009 					err = -EINVAL;
1010 			} else {
1011 				/* drop the frame */
1012 				err = -EINVAL;
1013 			}
1014 		}
1015 		rcu_read_unlock();
1016 next:
1017 		__skb_queue_tail(list, skb);
1018 		if (!(rx->flags & XEN_NETRXF_more_data))
1019 			break;
1020 
1021 		if (cons + slots == rp) {
1022 			if (net_ratelimit())
1023 				dev_warn(dev, "Need more slots\n");
1024 			err = -ENOENT;
1025 			break;
1026 		}
1027 
1028 		RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local);
1029 		rx = &rx_local;
1030 		skb = xennet_get_rx_skb(queue, cons + slots);
1031 		ref = xennet_get_rx_ref(queue, cons + slots);
1032 		slots++;
1033 	}
1034 
1035 	if (unlikely(slots > max)) {
1036 		if (net_ratelimit())
1037 			dev_warn(dev, "Too many slots\n");
1038 		err = -E2BIG;
1039 	}
1040 
1041 	if (unlikely(err))
1042 		queue->rx.rsp_cons = cons + slots;
1043 
1044 	return err;
1045 }
1046 
1047 static int xennet_set_skb_gso(struct sk_buff *skb,
1048 			      struct xen_netif_extra_info *gso)
1049 {
1050 	if (!gso->u.gso.size) {
1051 		if (net_ratelimit())
1052 			pr_warn("GSO size must not be zero\n");
1053 		return -EINVAL;
1054 	}
1055 
1056 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1057 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
1058 		if (net_ratelimit())
1059 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
1060 		return -EINVAL;
1061 	}
1062 
1063 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
1064 	skb_shinfo(skb)->gso_type =
1065 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1066 		SKB_GSO_TCPV4 :
1067 		SKB_GSO_TCPV6;
1068 
1069 	/* Header must be checked, and gso_segs computed. */
1070 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1071 	skb_shinfo(skb)->gso_segs = 0;
1072 
1073 	return 0;
1074 }
1075 
1076 static int xennet_fill_frags(struct netfront_queue *queue,
1077 			     struct sk_buff *skb,
1078 			     struct sk_buff_head *list)
1079 {
1080 	RING_IDX cons = queue->rx.rsp_cons;
1081 	struct sk_buff *nskb;
1082 
1083 	while ((nskb = __skb_dequeue(list))) {
1084 		struct xen_netif_rx_response rx;
1085 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
1086 
1087 		RING_COPY_RESPONSE(&queue->rx, ++cons, &rx);
1088 
1089 		if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
1090 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1091 
1092 			BUG_ON(pull_to < skb_headlen(skb));
1093 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1094 		}
1095 		if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
1096 			queue->rx.rsp_cons = ++cons + skb_queue_len(list);
1097 			kfree_skb(nskb);
1098 			return -ENOENT;
1099 		}
1100 
1101 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1102 				skb_frag_page(nfrag),
1103 				rx.offset, rx.status, PAGE_SIZE);
1104 
1105 		skb_shinfo(nskb)->nr_frags = 0;
1106 		kfree_skb(nskb);
1107 	}
1108 
1109 	queue->rx.rsp_cons = cons;
1110 
1111 	return 0;
1112 }
1113 
1114 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
1115 {
1116 	bool recalculate_partial_csum = false;
1117 
1118 	/*
1119 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1120 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
1121 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1122 	 * recalculate the partial checksum.
1123 	 */
1124 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1125 		struct netfront_info *np = netdev_priv(dev);
1126 		atomic_inc(&np->rx_gso_checksum_fixup);
1127 		skb->ip_summed = CHECKSUM_PARTIAL;
1128 		recalculate_partial_csum = true;
1129 	}
1130 
1131 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1132 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1133 		return 0;
1134 
1135 	return skb_checksum_setup(skb, recalculate_partial_csum);
1136 }
1137 
1138 static int handle_incoming_queue(struct netfront_queue *queue,
1139 				 struct sk_buff_head *rxq)
1140 {
1141 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
1142 	int packets_dropped = 0;
1143 	struct sk_buff *skb;
1144 
1145 	while ((skb = __skb_dequeue(rxq)) != NULL) {
1146 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1147 
1148 		if (pull_to > skb_headlen(skb))
1149 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1150 
1151 		/* Ethernet work: Delayed to here as it peeks the header. */
1152 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
1153 		skb_reset_network_header(skb);
1154 
1155 		if (checksum_setup(queue->info->netdev, skb)) {
1156 			kfree_skb(skb);
1157 			packets_dropped++;
1158 			queue->info->netdev->stats.rx_errors++;
1159 			continue;
1160 		}
1161 
1162 		u64_stats_update_begin(&rx_stats->syncp);
1163 		rx_stats->packets++;
1164 		rx_stats->bytes += skb->len;
1165 		u64_stats_update_end(&rx_stats->syncp);
1166 
1167 		/* Pass it up. */
1168 		napi_gro_receive(&queue->napi, skb);
1169 	}
1170 
1171 	return packets_dropped;
1172 }
1173 
1174 static int xennet_poll(struct napi_struct *napi, int budget)
1175 {
1176 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1177 	struct net_device *dev = queue->info->netdev;
1178 	struct sk_buff *skb;
1179 	struct netfront_rx_info rinfo;
1180 	struct xen_netif_rx_response *rx = &rinfo.rx;
1181 	struct xen_netif_extra_info *extras = rinfo.extras;
1182 	RING_IDX i, rp;
1183 	int work_done;
1184 	struct sk_buff_head rxq;
1185 	struct sk_buff_head errq;
1186 	struct sk_buff_head tmpq;
1187 	int err;
1188 	bool need_xdp_flush = false;
1189 
1190 	spin_lock(&queue->rx_lock);
1191 
1192 	skb_queue_head_init(&rxq);
1193 	skb_queue_head_init(&errq);
1194 	skb_queue_head_init(&tmpq);
1195 
1196 	rp = queue->rx.sring->rsp_prod;
1197 	if (RING_RESPONSE_PROD_OVERFLOW(&queue->rx, rp)) {
1198 		dev_alert(&dev->dev, "Illegal number of responses %u\n",
1199 			  rp - queue->rx.rsp_cons);
1200 		queue->info->broken = true;
1201 		spin_unlock(&queue->rx_lock);
1202 		return 0;
1203 	}
1204 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1205 
1206 	i = queue->rx.rsp_cons;
1207 	work_done = 0;
1208 	while ((i != rp) && (work_done < budget)) {
1209 		RING_COPY_RESPONSE(&queue->rx, i, rx);
1210 		memset(extras, 0, sizeof(rinfo.extras));
1211 
1212 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1213 					   &need_xdp_flush);
1214 
1215 		if (unlikely(err)) {
1216 err:
1217 			while ((skb = __skb_dequeue(&tmpq)))
1218 				__skb_queue_tail(&errq, skb);
1219 			dev->stats.rx_errors++;
1220 			i = queue->rx.rsp_cons;
1221 			continue;
1222 		}
1223 
1224 		skb = __skb_dequeue(&tmpq);
1225 
1226 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1227 			struct xen_netif_extra_info *gso;
1228 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1229 
1230 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1231 				__skb_queue_head(&tmpq, skb);
1232 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1233 				goto err;
1234 			}
1235 		}
1236 
1237 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1238 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1239 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1240 
1241 		skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1242 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1243 		skb->data_len = rx->status;
1244 		skb->len += rx->status;
1245 
1246 		if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1247 			goto err;
1248 
1249 		if (rx->flags & XEN_NETRXF_csum_blank)
1250 			skb->ip_summed = CHECKSUM_PARTIAL;
1251 		else if (rx->flags & XEN_NETRXF_data_validated)
1252 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1253 
1254 		__skb_queue_tail(&rxq, skb);
1255 
1256 		i = ++queue->rx.rsp_cons;
1257 		work_done++;
1258 	}
1259 	if (need_xdp_flush)
1260 		xdp_do_flush();
1261 
1262 	__skb_queue_purge(&errq);
1263 
1264 	work_done -= handle_incoming_queue(queue, &rxq);
1265 
1266 	xennet_alloc_rx_buffers(queue);
1267 
1268 	if (work_done < budget) {
1269 		int more_to_do = 0;
1270 
1271 		napi_complete_done(napi, work_done);
1272 
1273 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1274 		if (more_to_do)
1275 			napi_schedule(napi);
1276 	}
1277 
1278 	spin_unlock(&queue->rx_lock);
1279 
1280 	return work_done;
1281 }
1282 
1283 static int xennet_change_mtu(struct net_device *dev, int mtu)
1284 {
1285 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1286 
1287 	if (mtu > max)
1288 		return -EINVAL;
1289 	dev->mtu = mtu;
1290 	return 0;
1291 }
1292 
1293 static void xennet_get_stats64(struct net_device *dev,
1294 			       struct rtnl_link_stats64 *tot)
1295 {
1296 	struct netfront_info *np = netdev_priv(dev);
1297 	int cpu;
1298 
1299 	for_each_possible_cpu(cpu) {
1300 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1301 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1302 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1303 		unsigned int start;
1304 
1305 		do {
1306 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1307 			tx_packets = tx_stats->packets;
1308 			tx_bytes = tx_stats->bytes;
1309 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1310 
1311 		do {
1312 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1313 			rx_packets = rx_stats->packets;
1314 			rx_bytes = rx_stats->bytes;
1315 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1316 
1317 		tot->rx_packets += rx_packets;
1318 		tot->tx_packets += tx_packets;
1319 		tot->rx_bytes   += rx_bytes;
1320 		tot->tx_bytes   += tx_bytes;
1321 	}
1322 
1323 	tot->rx_errors  = dev->stats.rx_errors;
1324 	tot->tx_dropped = dev->stats.tx_dropped;
1325 }
1326 
1327 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1328 {
1329 	struct sk_buff *skb;
1330 	int i;
1331 
1332 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1333 		/* Skip over entries which are actually freelist references */
1334 		if (!queue->tx_skbs[i])
1335 			continue;
1336 
1337 		skb = queue->tx_skbs[i];
1338 		queue->tx_skbs[i] = NULL;
1339 		get_page(queue->grant_tx_page[i]);
1340 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1341 					  GNTMAP_readonly,
1342 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1343 		queue->grant_tx_page[i] = NULL;
1344 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1345 		add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, i);
1346 		dev_kfree_skb_irq(skb);
1347 	}
1348 }
1349 
1350 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1351 {
1352 	int id, ref;
1353 
1354 	spin_lock_bh(&queue->rx_lock);
1355 
1356 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1357 		struct sk_buff *skb;
1358 		struct page *page;
1359 
1360 		skb = queue->rx_skbs[id];
1361 		if (!skb)
1362 			continue;
1363 
1364 		ref = queue->grant_rx_ref[id];
1365 		if (ref == GRANT_INVALID_REF)
1366 			continue;
1367 
1368 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1369 
1370 		/* gnttab_end_foreign_access() needs a page ref until
1371 		 * foreign access is ended (which may be deferred).
1372 		 */
1373 		get_page(page);
1374 		gnttab_end_foreign_access(ref, 0,
1375 					  (unsigned long)page_address(page));
1376 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1377 
1378 		kfree_skb(skb);
1379 	}
1380 
1381 	spin_unlock_bh(&queue->rx_lock);
1382 }
1383 
1384 static netdev_features_t xennet_fix_features(struct net_device *dev,
1385 	netdev_features_t features)
1386 {
1387 	struct netfront_info *np = netdev_priv(dev);
1388 
1389 	if (features & NETIF_F_SG &&
1390 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1391 		features &= ~NETIF_F_SG;
1392 
1393 	if (features & NETIF_F_IPV6_CSUM &&
1394 	    !xenbus_read_unsigned(np->xbdev->otherend,
1395 				  "feature-ipv6-csum-offload", 0))
1396 		features &= ~NETIF_F_IPV6_CSUM;
1397 
1398 	if (features & NETIF_F_TSO &&
1399 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1400 		features &= ~NETIF_F_TSO;
1401 
1402 	if (features & NETIF_F_TSO6 &&
1403 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1404 		features &= ~NETIF_F_TSO6;
1405 
1406 	return features;
1407 }
1408 
1409 static int xennet_set_features(struct net_device *dev,
1410 	netdev_features_t features)
1411 {
1412 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1413 		netdev_info(dev, "Reducing MTU because no SG offload");
1414 		dev->mtu = ETH_DATA_LEN;
1415 	}
1416 
1417 	return 0;
1418 }
1419 
1420 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1421 {
1422 	struct netfront_queue *queue = dev_id;
1423 	unsigned long flags;
1424 
1425 	if (queue->info->broken)
1426 		return IRQ_HANDLED;
1427 
1428 	spin_lock_irqsave(&queue->tx_lock, flags);
1429 	xennet_tx_buf_gc(queue);
1430 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1431 
1432 	return IRQ_HANDLED;
1433 }
1434 
1435 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1436 {
1437 	struct netfront_queue *queue = dev_id;
1438 	struct net_device *dev = queue->info->netdev;
1439 
1440 	if (queue->info->broken)
1441 		return IRQ_HANDLED;
1442 
1443 	if (likely(netif_carrier_ok(dev) &&
1444 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1445 		napi_schedule(&queue->napi);
1446 
1447 	return IRQ_HANDLED;
1448 }
1449 
1450 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1451 {
1452 	xennet_tx_interrupt(irq, dev_id);
1453 	xennet_rx_interrupt(irq, dev_id);
1454 	return IRQ_HANDLED;
1455 }
1456 
1457 #ifdef CONFIG_NET_POLL_CONTROLLER
1458 static void xennet_poll_controller(struct net_device *dev)
1459 {
1460 	/* Poll each queue */
1461 	struct netfront_info *info = netdev_priv(dev);
1462 	unsigned int num_queues = dev->real_num_tx_queues;
1463 	unsigned int i;
1464 
1465 	if (info->broken)
1466 		return;
1467 
1468 	for (i = 0; i < num_queues; ++i)
1469 		xennet_interrupt(0, &info->queues[i]);
1470 }
1471 #endif
1472 
1473 #define NETBACK_XDP_HEADROOM_DISABLE	0
1474 #define NETBACK_XDP_HEADROOM_ENABLE	1
1475 
1476 static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1477 {
1478 	int err;
1479 	unsigned short headroom;
1480 
1481 	headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1482 	err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1483 			    "xdp-headroom", "%hu",
1484 			    headroom);
1485 	if (err)
1486 		pr_warn("Error writing xdp-headroom\n");
1487 
1488 	return err;
1489 }
1490 
1491 static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1492 			  struct netlink_ext_ack *extack)
1493 {
1494 	unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1495 	struct netfront_info *np = netdev_priv(dev);
1496 	struct bpf_prog *old_prog;
1497 	unsigned int i, err;
1498 
1499 	if (dev->mtu > max_mtu) {
1500 		netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1501 		return -EINVAL;
1502 	}
1503 
1504 	if (!np->netback_has_xdp_headroom)
1505 		return 0;
1506 
1507 	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1508 
1509 	err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1510 				  NETBACK_XDP_HEADROOM_DISABLE);
1511 	if (err)
1512 		return err;
1513 
1514 	/* avoid the race with XDP headroom adjustment */
1515 	wait_event(module_wq,
1516 		   xenbus_read_driver_state(np->xbdev->otherend) ==
1517 		   XenbusStateReconfigured);
1518 	np->netfront_xdp_enabled = true;
1519 
1520 	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1521 
1522 	if (prog)
1523 		bpf_prog_add(prog, dev->real_num_tx_queues);
1524 
1525 	for (i = 0; i < dev->real_num_tx_queues; ++i)
1526 		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1527 
1528 	if (old_prog)
1529 		for (i = 0; i < dev->real_num_tx_queues; ++i)
1530 			bpf_prog_put(old_prog);
1531 
1532 	xenbus_switch_state(np->xbdev, XenbusStateConnected);
1533 
1534 	return 0;
1535 }
1536 
1537 static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1538 {
1539 	struct netfront_info *np = netdev_priv(dev);
1540 
1541 	if (np->broken)
1542 		return -ENODEV;
1543 
1544 	switch (xdp->command) {
1545 	case XDP_SETUP_PROG:
1546 		return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1547 	default:
1548 		return -EINVAL;
1549 	}
1550 }
1551 
1552 static const struct net_device_ops xennet_netdev_ops = {
1553 	.ndo_open            = xennet_open,
1554 	.ndo_stop            = xennet_close,
1555 	.ndo_start_xmit      = xennet_start_xmit,
1556 	.ndo_change_mtu	     = xennet_change_mtu,
1557 	.ndo_get_stats64     = xennet_get_stats64,
1558 	.ndo_set_mac_address = eth_mac_addr,
1559 	.ndo_validate_addr   = eth_validate_addr,
1560 	.ndo_fix_features    = xennet_fix_features,
1561 	.ndo_set_features    = xennet_set_features,
1562 	.ndo_select_queue    = xennet_select_queue,
1563 	.ndo_bpf            = xennet_xdp,
1564 	.ndo_xdp_xmit	    = xennet_xdp_xmit,
1565 #ifdef CONFIG_NET_POLL_CONTROLLER
1566 	.ndo_poll_controller = xennet_poll_controller,
1567 #endif
1568 };
1569 
1570 static void xennet_free_netdev(struct net_device *netdev)
1571 {
1572 	struct netfront_info *np = netdev_priv(netdev);
1573 
1574 	free_percpu(np->rx_stats);
1575 	free_percpu(np->tx_stats);
1576 	free_netdev(netdev);
1577 }
1578 
1579 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1580 {
1581 	int err;
1582 	struct net_device *netdev;
1583 	struct netfront_info *np;
1584 
1585 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1586 	if (!netdev)
1587 		return ERR_PTR(-ENOMEM);
1588 
1589 	np                   = netdev_priv(netdev);
1590 	np->xbdev            = dev;
1591 
1592 	np->queues = NULL;
1593 
1594 	err = -ENOMEM;
1595 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1596 	if (np->rx_stats == NULL)
1597 		goto exit;
1598 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1599 	if (np->tx_stats == NULL)
1600 		goto exit;
1601 
1602 	netdev->netdev_ops	= &xennet_netdev_ops;
1603 
1604 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1605 				  NETIF_F_GSO_ROBUST;
1606 	netdev->hw_features	= NETIF_F_SG |
1607 				  NETIF_F_IPV6_CSUM |
1608 				  NETIF_F_TSO | NETIF_F_TSO6;
1609 
1610 	/*
1611          * Assume that all hw features are available for now. This set
1612          * will be adjusted by the call to netdev_update_features() in
1613          * xennet_connect() which is the earliest point where we can
1614          * negotiate with the backend regarding supported features.
1615          */
1616 	netdev->features |= netdev->hw_features;
1617 
1618 	netdev->ethtool_ops = &xennet_ethtool_ops;
1619 	netdev->min_mtu = ETH_MIN_MTU;
1620 	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1621 	SET_NETDEV_DEV(netdev, &dev->dev);
1622 
1623 	np->netdev = netdev;
1624 	np->netfront_xdp_enabled = false;
1625 
1626 	netif_carrier_off(netdev);
1627 
1628 	do {
1629 		xenbus_switch_state(dev, XenbusStateInitialising);
1630 		err = wait_event_timeout(module_wq,
1631 				 xenbus_read_driver_state(dev->otherend) !=
1632 				 XenbusStateClosed &&
1633 				 xenbus_read_driver_state(dev->otherend) !=
1634 				 XenbusStateUnknown, XENNET_TIMEOUT);
1635 	} while (!err);
1636 
1637 	return netdev;
1638 
1639  exit:
1640 	xennet_free_netdev(netdev);
1641 	return ERR_PTR(err);
1642 }
1643 
1644 /*
1645  * Entry point to this code when a new device is created.  Allocate the basic
1646  * structures and the ring buffers for communication with the backend, and
1647  * inform the backend of the appropriate details for those.
1648  */
1649 static int netfront_probe(struct xenbus_device *dev,
1650 			  const struct xenbus_device_id *id)
1651 {
1652 	int err;
1653 	struct net_device *netdev;
1654 	struct netfront_info *info;
1655 
1656 	netdev = xennet_create_dev(dev);
1657 	if (IS_ERR(netdev)) {
1658 		err = PTR_ERR(netdev);
1659 		xenbus_dev_fatal(dev, err, "creating netdev");
1660 		return err;
1661 	}
1662 
1663 	info = netdev_priv(netdev);
1664 	dev_set_drvdata(&dev->dev, info);
1665 #ifdef CONFIG_SYSFS
1666 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1667 #endif
1668 
1669 	return 0;
1670 }
1671 
1672 static void xennet_end_access(int ref, void *page)
1673 {
1674 	/* This frees the page as a side-effect */
1675 	if (ref != GRANT_INVALID_REF)
1676 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1677 }
1678 
1679 static void xennet_disconnect_backend(struct netfront_info *info)
1680 {
1681 	unsigned int i = 0;
1682 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1683 
1684 	netif_carrier_off(info->netdev);
1685 
1686 	for (i = 0; i < num_queues && info->queues; ++i) {
1687 		struct netfront_queue *queue = &info->queues[i];
1688 
1689 		del_timer_sync(&queue->rx_refill_timer);
1690 
1691 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1692 			unbind_from_irqhandler(queue->tx_irq, queue);
1693 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1694 			unbind_from_irqhandler(queue->tx_irq, queue);
1695 			unbind_from_irqhandler(queue->rx_irq, queue);
1696 		}
1697 		queue->tx_evtchn = queue->rx_evtchn = 0;
1698 		queue->tx_irq = queue->rx_irq = 0;
1699 
1700 		if (netif_running(info->netdev))
1701 			napi_synchronize(&queue->napi);
1702 
1703 		xennet_release_tx_bufs(queue);
1704 		xennet_release_rx_bufs(queue);
1705 		gnttab_free_grant_references(queue->gref_tx_head);
1706 		gnttab_free_grant_references(queue->gref_rx_head);
1707 
1708 		/* End access and free the pages */
1709 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1710 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1711 
1712 		queue->tx_ring_ref = GRANT_INVALID_REF;
1713 		queue->rx_ring_ref = GRANT_INVALID_REF;
1714 		queue->tx.sring = NULL;
1715 		queue->rx.sring = NULL;
1716 
1717 		page_pool_destroy(queue->page_pool);
1718 	}
1719 }
1720 
1721 /*
1722  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1723  * driver restart.  We tear down our netif structure and recreate it, but
1724  * leave the device-layer structures intact so that this is transparent to the
1725  * rest of the kernel.
1726  */
1727 static int netfront_resume(struct xenbus_device *dev)
1728 {
1729 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1730 
1731 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1732 
1733 	xennet_disconnect_backend(info);
1734 	return 0;
1735 }
1736 
1737 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1738 {
1739 	char *s, *e, *macstr;
1740 	int i;
1741 
1742 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1743 	if (IS_ERR(macstr))
1744 		return PTR_ERR(macstr);
1745 
1746 	for (i = 0; i < ETH_ALEN; i++) {
1747 		mac[i] = simple_strtoul(s, &e, 16);
1748 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1749 			kfree(macstr);
1750 			return -ENOENT;
1751 		}
1752 		s = e+1;
1753 	}
1754 
1755 	kfree(macstr);
1756 	return 0;
1757 }
1758 
1759 static int setup_netfront_single(struct netfront_queue *queue)
1760 {
1761 	int err;
1762 
1763 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1764 	if (err < 0)
1765 		goto fail;
1766 
1767 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1768 					xennet_interrupt,
1769 					0, queue->info->netdev->name, queue);
1770 	if (err < 0)
1771 		goto bind_fail;
1772 	queue->rx_evtchn = queue->tx_evtchn;
1773 	queue->rx_irq = queue->tx_irq = err;
1774 
1775 	return 0;
1776 
1777 bind_fail:
1778 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1779 	queue->tx_evtchn = 0;
1780 fail:
1781 	return err;
1782 }
1783 
1784 static int setup_netfront_split(struct netfront_queue *queue)
1785 {
1786 	int err;
1787 
1788 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1789 	if (err < 0)
1790 		goto fail;
1791 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1792 	if (err < 0)
1793 		goto alloc_rx_evtchn_fail;
1794 
1795 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1796 		 "%s-tx", queue->name);
1797 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1798 					xennet_tx_interrupt,
1799 					0, queue->tx_irq_name, queue);
1800 	if (err < 0)
1801 		goto bind_tx_fail;
1802 	queue->tx_irq = err;
1803 
1804 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1805 		 "%s-rx", queue->name);
1806 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1807 					xennet_rx_interrupt,
1808 					0, queue->rx_irq_name, queue);
1809 	if (err < 0)
1810 		goto bind_rx_fail;
1811 	queue->rx_irq = err;
1812 
1813 	return 0;
1814 
1815 bind_rx_fail:
1816 	unbind_from_irqhandler(queue->tx_irq, queue);
1817 	queue->tx_irq = 0;
1818 bind_tx_fail:
1819 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1820 	queue->rx_evtchn = 0;
1821 alloc_rx_evtchn_fail:
1822 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1823 	queue->tx_evtchn = 0;
1824 fail:
1825 	return err;
1826 }
1827 
1828 static int setup_netfront(struct xenbus_device *dev,
1829 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1830 {
1831 	struct xen_netif_tx_sring *txs;
1832 	struct xen_netif_rx_sring *rxs;
1833 	grant_ref_t gref;
1834 	int err;
1835 
1836 	queue->tx_ring_ref = GRANT_INVALID_REF;
1837 	queue->rx_ring_ref = GRANT_INVALID_REF;
1838 	queue->rx.sring = NULL;
1839 	queue->tx.sring = NULL;
1840 
1841 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1842 	if (!txs) {
1843 		err = -ENOMEM;
1844 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1845 		goto fail;
1846 	}
1847 	SHARED_RING_INIT(txs);
1848 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1849 
1850 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1851 	if (err < 0)
1852 		goto grant_tx_ring_fail;
1853 	queue->tx_ring_ref = gref;
1854 
1855 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1856 	if (!rxs) {
1857 		err = -ENOMEM;
1858 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1859 		goto alloc_rx_ring_fail;
1860 	}
1861 	SHARED_RING_INIT(rxs);
1862 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1863 
1864 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1865 	if (err < 0)
1866 		goto grant_rx_ring_fail;
1867 	queue->rx_ring_ref = gref;
1868 
1869 	if (feature_split_evtchn)
1870 		err = setup_netfront_split(queue);
1871 	/* setup single event channel if
1872 	 *  a) feature-split-event-channels == 0
1873 	 *  b) feature-split-event-channels == 1 but failed to setup
1874 	 */
1875 	if (!feature_split_evtchn || err)
1876 		err = setup_netfront_single(queue);
1877 
1878 	if (err)
1879 		goto alloc_evtchn_fail;
1880 
1881 	return 0;
1882 
1883 	/* If we fail to setup netfront, it is safe to just revoke access to
1884 	 * granted pages because backend is not accessing it at this point.
1885 	 */
1886 alloc_evtchn_fail:
1887 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1888 grant_rx_ring_fail:
1889 	free_page((unsigned long)rxs);
1890 alloc_rx_ring_fail:
1891 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1892 grant_tx_ring_fail:
1893 	free_page((unsigned long)txs);
1894 fail:
1895 	return err;
1896 }
1897 
1898 /* Queue-specific initialisation
1899  * This used to be done in xennet_create_dev() but must now
1900  * be run per-queue.
1901  */
1902 static int xennet_init_queue(struct netfront_queue *queue)
1903 {
1904 	unsigned short i;
1905 	int err = 0;
1906 	char *devid;
1907 
1908 	spin_lock_init(&queue->tx_lock);
1909 	spin_lock_init(&queue->rx_lock);
1910 
1911 	timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1912 
1913 	devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1914 	snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1915 		 devid, queue->id);
1916 
1917 	/* Initialise tx_skb_freelist as a free chain containing every entry. */
1918 	queue->tx_skb_freelist = 0;
1919 	queue->tx_pend_queue = TX_LINK_NONE;
1920 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1921 		queue->tx_link[i] = i + 1;
1922 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1923 		queue->grant_tx_page[i] = NULL;
1924 	}
1925 	queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE;
1926 
1927 	/* Clear out rx_skbs */
1928 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1929 		queue->rx_skbs[i] = NULL;
1930 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1931 	}
1932 
1933 	/* A grant for every tx ring slot */
1934 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1935 					  &queue->gref_tx_head) < 0) {
1936 		pr_alert("can't alloc tx grant refs\n");
1937 		err = -ENOMEM;
1938 		goto exit;
1939 	}
1940 
1941 	/* A grant for every rx ring slot */
1942 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1943 					  &queue->gref_rx_head) < 0) {
1944 		pr_alert("can't alloc rx grant refs\n");
1945 		err = -ENOMEM;
1946 		goto exit_free_tx;
1947 	}
1948 
1949 	return 0;
1950 
1951  exit_free_tx:
1952 	gnttab_free_grant_references(queue->gref_tx_head);
1953  exit:
1954 	return err;
1955 }
1956 
1957 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1958 			   struct xenbus_transaction *xbt, int write_hierarchical)
1959 {
1960 	/* Write the queue-specific keys into XenStore in the traditional
1961 	 * way for a single queue, or in a queue subkeys for multiple
1962 	 * queues.
1963 	 */
1964 	struct xenbus_device *dev = queue->info->xbdev;
1965 	int err;
1966 	const char *message;
1967 	char *path;
1968 	size_t pathsize;
1969 
1970 	/* Choose the correct place to write the keys */
1971 	if (write_hierarchical) {
1972 		pathsize = strlen(dev->nodename) + 10;
1973 		path = kzalloc(pathsize, GFP_KERNEL);
1974 		if (!path) {
1975 			err = -ENOMEM;
1976 			message = "out of memory while writing ring references";
1977 			goto error;
1978 		}
1979 		snprintf(path, pathsize, "%s/queue-%u",
1980 				dev->nodename, queue->id);
1981 	} else {
1982 		path = (char *)dev->nodename;
1983 	}
1984 
1985 	/* Write ring references */
1986 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1987 			queue->tx_ring_ref);
1988 	if (err) {
1989 		message = "writing tx-ring-ref";
1990 		goto error;
1991 	}
1992 
1993 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1994 			queue->rx_ring_ref);
1995 	if (err) {
1996 		message = "writing rx-ring-ref";
1997 		goto error;
1998 	}
1999 
2000 	/* Write event channels; taking into account both shared
2001 	 * and split event channel scenarios.
2002 	 */
2003 	if (queue->tx_evtchn == queue->rx_evtchn) {
2004 		/* Shared event channel */
2005 		err = xenbus_printf(*xbt, path,
2006 				"event-channel", "%u", queue->tx_evtchn);
2007 		if (err) {
2008 			message = "writing event-channel";
2009 			goto error;
2010 		}
2011 	} else {
2012 		/* Split event channels */
2013 		err = xenbus_printf(*xbt, path,
2014 				"event-channel-tx", "%u", queue->tx_evtchn);
2015 		if (err) {
2016 			message = "writing event-channel-tx";
2017 			goto error;
2018 		}
2019 
2020 		err = xenbus_printf(*xbt, path,
2021 				"event-channel-rx", "%u", queue->rx_evtchn);
2022 		if (err) {
2023 			message = "writing event-channel-rx";
2024 			goto error;
2025 		}
2026 	}
2027 
2028 	if (write_hierarchical)
2029 		kfree(path);
2030 	return 0;
2031 
2032 error:
2033 	if (write_hierarchical)
2034 		kfree(path);
2035 	xenbus_dev_fatal(dev, err, "%s", message);
2036 	return err;
2037 }
2038 
2039 static void xennet_destroy_queues(struct netfront_info *info)
2040 {
2041 	unsigned int i;
2042 
2043 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
2044 		struct netfront_queue *queue = &info->queues[i];
2045 
2046 		if (netif_running(info->netdev))
2047 			napi_disable(&queue->napi);
2048 		netif_napi_del(&queue->napi);
2049 	}
2050 
2051 	kfree(info->queues);
2052 	info->queues = NULL;
2053 }
2054 
2055 
2056 
2057 static int xennet_create_page_pool(struct netfront_queue *queue)
2058 {
2059 	int err;
2060 	struct page_pool_params pp_params = {
2061 		.order = 0,
2062 		.flags = 0,
2063 		.pool_size = NET_RX_RING_SIZE,
2064 		.nid = NUMA_NO_NODE,
2065 		.dev = &queue->info->netdev->dev,
2066 		.offset = XDP_PACKET_HEADROOM,
2067 		.max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2068 	};
2069 
2070 	queue->page_pool = page_pool_create(&pp_params);
2071 	if (IS_ERR(queue->page_pool)) {
2072 		err = PTR_ERR(queue->page_pool);
2073 		queue->page_pool = NULL;
2074 		return err;
2075 	}
2076 
2077 	err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2078 			       queue->id, 0);
2079 	if (err) {
2080 		netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2081 		goto err_free_pp;
2082 	}
2083 
2084 	err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2085 					 MEM_TYPE_PAGE_POOL, queue->page_pool);
2086 	if (err) {
2087 		netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2088 		goto err_unregister_rxq;
2089 	}
2090 	return 0;
2091 
2092 err_unregister_rxq:
2093 	xdp_rxq_info_unreg(&queue->xdp_rxq);
2094 err_free_pp:
2095 	page_pool_destroy(queue->page_pool);
2096 	queue->page_pool = NULL;
2097 	return err;
2098 }
2099 
2100 static int xennet_create_queues(struct netfront_info *info,
2101 				unsigned int *num_queues)
2102 {
2103 	unsigned int i;
2104 	int ret;
2105 
2106 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2107 			       GFP_KERNEL);
2108 	if (!info->queues)
2109 		return -ENOMEM;
2110 
2111 	for (i = 0; i < *num_queues; i++) {
2112 		struct netfront_queue *queue = &info->queues[i];
2113 
2114 		queue->id = i;
2115 		queue->info = info;
2116 
2117 		ret = xennet_init_queue(queue);
2118 		if (ret < 0) {
2119 			dev_warn(&info->xbdev->dev,
2120 				 "only created %d queues\n", i);
2121 			*num_queues = i;
2122 			break;
2123 		}
2124 
2125 		/* use page pool recycling instead of buddy allocator */
2126 		ret = xennet_create_page_pool(queue);
2127 		if (ret < 0) {
2128 			dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2129 			*num_queues = i;
2130 			return ret;
2131 		}
2132 
2133 		netif_napi_add(queue->info->netdev, &queue->napi,
2134 			       xennet_poll, 64);
2135 		if (netif_running(info->netdev))
2136 			napi_enable(&queue->napi);
2137 	}
2138 
2139 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
2140 
2141 	if (*num_queues == 0) {
2142 		dev_err(&info->xbdev->dev, "no queues\n");
2143 		return -EINVAL;
2144 	}
2145 	return 0;
2146 }
2147 
2148 /* Common code used when first setting up, and when resuming. */
2149 static int talk_to_netback(struct xenbus_device *dev,
2150 			   struct netfront_info *info)
2151 {
2152 	const char *message;
2153 	struct xenbus_transaction xbt;
2154 	int err;
2155 	unsigned int feature_split_evtchn;
2156 	unsigned int i = 0;
2157 	unsigned int max_queues = 0;
2158 	struct netfront_queue *queue = NULL;
2159 	unsigned int num_queues = 1;
2160 
2161 	info->netdev->irq = 0;
2162 
2163 	/* Check if backend supports multiple queues */
2164 	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2165 					  "multi-queue-max-queues", 1);
2166 	num_queues = min(max_queues, xennet_max_queues);
2167 
2168 	/* Check feature-split-event-channels */
2169 	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2170 					"feature-split-event-channels", 0);
2171 
2172 	/* Read mac addr. */
2173 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
2174 	if (err) {
2175 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2176 		goto out_unlocked;
2177 	}
2178 
2179 	info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2180 							      "feature-xdp-headroom", 0);
2181 	if (info->netback_has_xdp_headroom) {
2182 		/* set the current xen-netfront xdp state */
2183 		err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2184 					  NETBACK_XDP_HEADROOM_ENABLE :
2185 					  NETBACK_XDP_HEADROOM_DISABLE);
2186 		if (err)
2187 			goto out_unlocked;
2188 	}
2189 
2190 	rtnl_lock();
2191 	if (info->queues)
2192 		xennet_destroy_queues(info);
2193 
2194 	/* For the case of a reconnect reset the "broken" indicator. */
2195 	info->broken = false;
2196 
2197 	err = xennet_create_queues(info, &num_queues);
2198 	if (err < 0) {
2199 		xenbus_dev_fatal(dev, err, "creating queues");
2200 		kfree(info->queues);
2201 		info->queues = NULL;
2202 		goto out;
2203 	}
2204 	rtnl_unlock();
2205 
2206 	/* Create shared ring, alloc event channel -- for each queue */
2207 	for (i = 0; i < num_queues; ++i) {
2208 		queue = &info->queues[i];
2209 		err = setup_netfront(dev, queue, feature_split_evtchn);
2210 		if (err)
2211 			goto destroy_ring;
2212 	}
2213 
2214 again:
2215 	err = xenbus_transaction_start(&xbt);
2216 	if (err) {
2217 		xenbus_dev_fatal(dev, err, "starting transaction");
2218 		goto destroy_ring;
2219 	}
2220 
2221 	if (xenbus_exists(XBT_NIL,
2222 			  info->xbdev->otherend, "multi-queue-max-queues")) {
2223 		/* Write the number of queues */
2224 		err = xenbus_printf(xbt, dev->nodename,
2225 				    "multi-queue-num-queues", "%u", num_queues);
2226 		if (err) {
2227 			message = "writing multi-queue-num-queues";
2228 			goto abort_transaction_no_dev_fatal;
2229 		}
2230 	}
2231 
2232 	if (num_queues == 1) {
2233 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2234 		if (err)
2235 			goto abort_transaction_no_dev_fatal;
2236 	} else {
2237 		/* Write the keys for each queue */
2238 		for (i = 0; i < num_queues; ++i) {
2239 			queue = &info->queues[i];
2240 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2241 			if (err)
2242 				goto abort_transaction_no_dev_fatal;
2243 		}
2244 	}
2245 
2246 	/* The remaining keys are not queue-specific */
2247 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2248 			    1);
2249 	if (err) {
2250 		message = "writing request-rx-copy";
2251 		goto abort_transaction;
2252 	}
2253 
2254 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2255 	if (err) {
2256 		message = "writing feature-rx-notify";
2257 		goto abort_transaction;
2258 	}
2259 
2260 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2261 	if (err) {
2262 		message = "writing feature-sg";
2263 		goto abort_transaction;
2264 	}
2265 
2266 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2267 	if (err) {
2268 		message = "writing feature-gso-tcpv4";
2269 		goto abort_transaction;
2270 	}
2271 
2272 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2273 	if (err) {
2274 		message = "writing feature-gso-tcpv6";
2275 		goto abort_transaction;
2276 	}
2277 
2278 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2279 			   "1");
2280 	if (err) {
2281 		message = "writing feature-ipv6-csum-offload";
2282 		goto abort_transaction;
2283 	}
2284 
2285 	err = xenbus_transaction_end(xbt, 0);
2286 	if (err) {
2287 		if (err == -EAGAIN)
2288 			goto again;
2289 		xenbus_dev_fatal(dev, err, "completing transaction");
2290 		goto destroy_ring;
2291 	}
2292 
2293 	return 0;
2294 
2295  abort_transaction:
2296 	xenbus_dev_fatal(dev, err, "%s", message);
2297 abort_transaction_no_dev_fatal:
2298 	xenbus_transaction_end(xbt, 1);
2299  destroy_ring:
2300 	xennet_disconnect_backend(info);
2301 	rtnl_lock();
2302 	xennet_destroy_queues(info);
2303  out:
2304 	rtnl_unlock();
2305 out_unlocked:
2306 	device_unregister(&dev->dev);
2307 	return err;
2308 }
2309 
2310 static int xennet_connect(struct net_device *dev)
2311 {
2312 	struct netfront_info *np = netdev_priv(dev);
2313 	unsigned int num_queues = 0;
2314 	int err;
2315 	unsigned int j = 0;
2316 	struct netfront_queue *queue = NULL;
2317 
2318 	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2319 		dev_info(&dev->dev,
2320 			 "backend does not support copying receive path\n");
2321 		return -ENODEV;
2322 	}
2323 
2324 	err = talk_to_netback(np->xbdev, np);
2325 	if (err)
2326 		return err;
2327 	if (np->netback_has_xdp_headroom)
2328 		pr_info("backend supports XDP headroom\n");
2329 
2330 	/* talk_to_netback() sets the correct number of queues */
2331 	num_queues = dev->real_num_tx_queues;
2332 
2333 	if (dev->reg_state == NETREG_UNINITIALIZED) {
2334 		err = register_netdev(dev);
2335 		if (err) {
2336 			pr_warn("%s: register_netdev err=%d\n", __func__, err);
2337 			device_unregister(&np->xbdev->dev);
2338 			return err;
2339 		}
2340 	}
2341 
2342 	rtnl_lock();
2343 	netdev_update_features(dev);
2344 	rtnl_unlock();
2345 
2346 	/*
2347 	 * All public and private state should now be sane.  Get
2348 	 * ready to start sending and receiving packets and give the driver
2349 	 * domain a kick because we've probably just requeued some
2350 	 * packets.
2351 	 */
2352 	netif_carrier_on(np->netdev);
2353 	for (j = 0; j < num_queues; ++j) {
2354 		queue = &np->queues[j];
2355 
2356 		notify_remote_via_irq(queue->tx_irq);
2357 		if (queue->tx_irq != queue->rx_irq)
2358 			notify_remote_via_irq(queue->rx_irq);
2359 
2360 		spin_lock_irq(&queue->tx_lock);
2361 		xennet_tx_buf_gc(queue);
2362 		spin_unlock_irq(&queue->tx_lock);
2363 
2364 		spin_lock_bh(&queue->rx_lock);
2365 		xennet_alloc_rx_buffers(queue);
2366 		spin_unlock_bh(&queue->rx_lock);
2367 	}
2368 
2369 	return 0;
2370 }
2371 
2372 /*
2373  * Callback received when the backend's state changes.
2374  */
2375 static void netback_changed(struct xenbus_device *dev,
2376 			    enum xenbus_state backend_state)
2377 {
2378 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2379 	struct net_device *netdev = np->netdev;
2380 
2381 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2382 
2383 	wake_up_all(&module_wq);
2384 
2385 	switch (backend_state) {
2386 	case XenbusStateInitialising:
2387 	case XenbusStateInitialised:
2388 	case XenbusStateReconfiguring:
2389 	case XenbusStateReconfigured:
2390 	case XenbusStateUnknown:
2391 		break;
2392 
2393 	case XenbusStateInitWait:
2394 		if (dev->state != XenbusStateInitialising)
2395 			break;
2396 		if (xennet_connect(netdev) != 0)
2397 			break;
2398 		xenbus_switch_state(dev, XenbusStateConnected);
2399 		break;
2400 
2401 	case XenbusStateConnected:
2402 		netdev_notify_peers(netdev);
2403 		break;
2404 
2405 	case XenbusStateClosed:
2406 		if (dev->state == XenbusStateClosed)
2407 			break;
2408 		fallthrough;	/* Missed the backend's CLOSING state */
2409 	case XenbusStateClosing:
2410 		xenbus_frontend_closed(dev);
2411 		break;
2412 	}
2413 }
2414 
2415 static const struct xennet_stat {
2416 	char name[ETH_GSTRING_LEN];
2417 	u16 offset;
2418 } xennet_stats[] = {
2419 	{
2420 		"rx_gso_checksum_fixup",
2421 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2422 	},
2423 };
2424 
2425 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2426 {
2427 	switch (string_set) {
2428 	case ETH_SS_STATS:
2429 		return ARRAY_SIZE(xennet_stats);
2430 	default:
2431 		return -EINVAL;
2432 	}
2433 }
2434 
2435 static void xennet_get_ethtool_stats(struct net_device *dev,
2436 				     struct ethtool_stats *stats, u64 * data)
2437 {
2438 	void *np = netdev_priv(dev);
2439 	int i;
2440 
2441 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2442 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2443 }
2444 
2445 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2446 {
2447 	int i;
2448 
2449 	switch (stringset) {
2450 	case ETH_SS_STATS:
2451 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2452 			memcpy(data + i * ETH_GSTRING_LEN,
2453 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2454 		break;
2455 	}
2456 }
2457 
2458 static const struct ethtool_ops xennet_ethtool_ops =
2459 {
2460 	.get_link = ethtool_op_get_link,
2461 
2462 	.get_sset_count = xennet_get_sset_count,
2463 	.get_ethtool_stats = xennet_get_ethtool_stats,
2464 	.get_strings = xennet_get_strings,
2465 	.get_ts_info = ethtool_op_get_ts_info,
2466 };
2467 
2468 #ifdef CONFIG_SYSFS
2469 static ssize_t show_rxbuf(struct device *dev,
2470 			  struct device_attribute *attr, char *buf)
2471 {
2472 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2473 }
2474 
2475 static ssize_t store_rxbuf(struct device *dev,
2476 			   struct device_attribute *attr,
2477 			   const char *buf, size_t len)
2478 {
2479 	char *endp;
2480 
2481 	if (!capable(CAP_NET_ADMIN))
2482 		return -EPERM;
2483 
2484 	simple_strtoul(buf, &endp, 0);
2485 	if (endp == buf)
2486 		return -EBADMSG;
2487 
2488 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2489 
2490 	return len;
2491 }
2492 
2493 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2494 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2495 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2496 
2497 static struct attribute *xennet_dev_attrs[] = {
2498 	&dev_attr_rxbuf_min.attr,
2499 	&dev_attr_rxbuf_max.attr,
2500 	&dev_attr_rxbuf_cur.attr,
2501 	NULL
2502 };
2503 
2504 static const struct attribute_group xennet_dev_group = {
2505 	.attrs = xennet_dev_attrs
2506 };
2507 #endif /* CONFIG_SYSFS */
2508 
2509 static void xennet_bus_close(struct xenbus_device *dev)
2510 {
2511 	int ret;
2512 
2513 	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2514 		return;
2515 	do {
2516 		xenbus_switch_state(dev, XenbusStateClosing);
2517 		ret = wait_event_timeout(module_wq,
2518 				   xenbus_read_driver_state(dev->otherend) ==
2519 				   XenbusStateClosing ||
2520 				   xenbus_read_driver_state(dev->otherend) ==
2521 				   XenbusStateClosed ||
2522 				   xenbus_read_driver_state(dev->otherend) ==
2523 				   XenbusStateUnknown,
2524 				   XENNET_TIMEOUT);
2525 	} while (!ret);
2526 
2527 	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2528 		return;
2529 
2530 	do {
2531 		xenbus_switch_state(dev, XenbusStateClosed);
2532 		ret = wait_event_timeout(module_wq,
2533 				   xenbus_read_driver_state(dev->otherend) ==
2534 				   XenbusStateClosed ||
2535 				   xenbus_read_driver_state(dev->otherend) ==
2536 				   XenbusStateUnknown,
2537 				   XENNET_TIMEOUT);
2538 	} while (!ret);
2539 }
2540 
2541 static int xennet_remove(struct xenbus_device *dev)
2542 {
2543 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2544 
2545 	xennet_bus_close(dev);
2546 	xennet_disconnect_backend(info);
2547 
2548 	if (info->netdev->reg_state == NETREG_REGISTERED)
2549 		unregister_netdev(info->netdev);
2550 
2551 	if (info->queues) {
2552 		rtnl_lock();
2553 		xennet_destroy_queues(info);
2554 		rtnl_unlock();
2555 	}
2556 	xennet_free_netdev(info->netdev);
2557 
2558 	return 0;
2559 }
2560 
2561 static const struct xenbus_device_id netfront_ids[] = {
2562 	{ "vif" },
2563 	{ "" }
2564 };
2565 
2566 static struct xenbus_driver netfront_driver = {
2567 	.ids = netfront_ids,
2568 	.probe = netfront_probe,
2569 	.remove = xennet_remove,
2570 	.resume = netfront_resume,
2571 	.otherend_changed = netback_changed,
2572 };
2573 
2574 static int __init netif_init(void)
2575 {
2576 	if (!xen_domain())
2577 		return -ENODEV;
2578 
2579 	if (!xen_has_pv_nic_devices())
2580 		return -ENODEV;
2581 
2582 	pr_info("Initialising Xen virtual ethernet driver\n");
2583 
2584 	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2585 	 * specified a value.
2586 	 */
2587 	if (xennet_max_queues == 0)
2588 		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2589 					  num_online_cpus());
2590 
2591 	return xenbus_register_frontend(&netfront_driver);
2592 }
2593 module_init(netif_init);
2594 
2595 
2596 static void __exit netif_exit(void)
2597 {
2598 	xenbus_unregister_driver(&netfront_driver);
2599 }
2600 module_exit(netif_exit);
2601 
2602 MODULE_DESCRIPTION("Xen virtual network device frontend");
2603 MODULE_LICENSE("GPL");
2604 MODULE_ALIAS("xen:vif");
2605 MODULE_ALIAS("xennet");
2606