xref: /linux/drivers/net/xen-netfront.c (revision 2ba9268dd603d23e17643437b2246acb6844953b)
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 
48 #include <asm/xen/page.h>
49 #include <xen/xen.h>
50 #include <xen/xenbus.h>
51 #include <xen/events.h>
52 #include <xen/page.h>
53 #include <xen/platform_pci.h>
54 #include <xen/grant_table.h>
55 
56 #include <xen/interface/io/netif.h>
57 #include <xen/interface/memory.h>
58 #include <xen/interface/grant_table.h>
59 
60 /* Module parameters */
61 static unsigned int xennet_max_queues;
62 module_param_named(max_queues, xennet_max_queues, uint, 0644);
63 MODULE_PARM_DESC(max_queues,
64 		 "Maximum number of queues per virtual interface");
65 
66 static const struct ethtool_ops xennet_ethtool_ops;
67 
68 struct netfront_cb {
69 	int pull_to;
70 };
71 
72 #define NETFRONT_SKB_CB(skb)	((struct netfront_cb *)((skb)->cb))
73 
74 #define RX_COPY_THRESHOLD 256
75 
76 #define GRANT_INVALID_REF	0
77 
78 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
79 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
80 
81 /* Minimum number of Rx slots (includes slot for GSO metadata). */
82 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
83 
84 /* Queue name is interface name with "-qNNN" appended */
85 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
86 
87 /* IRQ name is queue name with "-tx" or "-rx" appended */
88 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
89 
90 struct netfront_stats {
91 	u64			packets;
92 	u64			bytes;
93 	struct u64_stats_sync	syncp;
94 };
95 
96 struct netfront_info;
97 
98 struct netfront_queue {
99 	unsigned int id; /* Queue ID, 0-based */
100 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
101 	struct netfront_info *info;
102 
103 	struct napi_struct napi;
104 
105 	/* Split event channels support, tx_* == rx_* when using
106 	 * single event channel.
107 	 */
108 	unsigned int tx_evtchn, rx_evtchn;
109 	unsigned int tx_irq, rx_irq;
110 	/* Only used when split event channels support is enabled */
111 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
112 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
113 
114 	spinlock_t   tx_lock;
115 	struct xen_netif_tx_front_ring tx;
116 	int tx_ring_ref;
117 
118 	/*
119 	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
120 	 * are linked from tx_skb_freelist through skb_entry.link.
121 	 *
122 	 *  NB. Freelist index entries are always going to be less than
123 	 *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
124 	 *  greater than PAGE_OFFSET: we use this property to distinguish
125 	 *  them.
126 	 */
127 	union skb_entry {
128 		struct sk_buff *skb;
129 		unsigned long link;
130 	} tx_skbs[NET_TX_RING_SIZE];
131 	grant_ref_t gref_tx_head;
132 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
133 	struct page *grant_tx_page[NET_TX_RING_SIZE];
134 	unsigned tx_skb_freelist;
135 
136 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
137 	struct xen_netif_rx_front_ring rx;
138 	int rx_ring_ref;
139 
140 	struct timer_list rx_refill_timer;
141 
142 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
143 	grant_ref_t gref_rx_head;
144 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
145 };
146 
147 struct netfront_info {
148 	struct list_head list;
149 	struct net_device *netdev;
150 
151 	struct xenbus_device *xbdev;
152 
153 	/* Multi-queue support */
154 	struct netfront_queue *queues;
155 
156 	/* Statistics */
157 	struct netfront_stats __percpu *rx_stats;
158 	struct netfront_stats __percpu *tx_stats;
159 
160 	atomic_t rx_gso_checksum_fixup;
161 };
162 
163 struct netfront_rx_info {
164 	struct xen_netif_rx_response rx;
165 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
166 };
167 
168 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
169 {
170 	list->link = id;
171 }
172 
173 static int skb_entry_is_link(const union skb_entry *list)
174 {
175 	BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
176 	return (unsigned long)list->skb < PAGE_OFFSET;
177 }
178 
179 /*
180  * Access macros for acquiring freeing slots in tx_skbs[].
181  */
182 
183 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
184 			       unsigned short id)
185 {
186 	skb_entry_set_link(&list[id], *head);
187 	*head = id;
188 }
189 
190 static unsigned short get_id_from_freelist(unsigned *head,
191 					   union skb_entry *list)
192 {
193 	unsigned int id = *head;
194 	*head = list[id].link;
195 	return id;
196 }
197 
198 static int xennet_rxidx(RING_IDX idx)
199 {
200 	return idx & (NET_RX_RING_SIZE - 1);
201 }
202 
203 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
204 					 RING_IDX ri)
205 {
206 	int i = xennet_rxidx(ri);
207 	struct sk_buff *skb = queue->rx_skbs[i];
208 	queue->rx_skbs[i] = NULL;
209 	return skb;
210 }
211 
212 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
213 					    RING_IDX ri)
214 {
215 	int i = xennet_rxidx(ri);
216 	grant_ref_t ref = queue->grant_rx_ref[i];
217 	queue->grant_rx_ref[i] = GRANT_INVALID_REF;
218 	return ref;
219 }
220 
221 #ifdef CONFIG_SYSFS
222 static const struct attribute_group xennet_dev_group;
223 #endif
224 
225 static bool xennet_can_sg(struct net_device *dev)
226 {
227 	return dev->features & NETIF_F_SG;
228 }
229 
230 
231 static void rx_refill_timeout(unsigned long data)
232 {
233 	struct netfront_queue *queue = (struct netfront_queue *)data;
234 	napi_schedule(&queue->napi);
235 }
236 
237 static int netfront_tx_slot_available(struct netfront_queue *queue)
238 {
239 	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
240 		(NET_TX_RING_SIZE - MAX_SKB_FRAGS - 2);
241 }
242 
243 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
244 {
245 	struct net_device *dev = queue->info->netdev;
246 	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
247 
248 	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
249 	    netfront_tx_slot_available(queue) &&
250 	    likely(netif_running(dev)))
251 		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
252 }
253 
254 
255 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
256 {
257 	struct sk_buff *skb;
258 	struct page *page;
259 
260 	skb = __netdev_alloc_skb(queue->info->netdev,
261 				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
262 				 GFP_ATOMIC | __GFP_NOWARN);
263 	if (unlikely(!skb))
264 		return NULL;
265 
266 	page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
267 	if (!page) {
268 		kfree_skb(skb);
269 		return NULL;
270 	}
271 	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
272 
273 	/* Align ip header to a 16 bytes boundary */
274 	skb_reserve(skb, NET_IP_ALIGN);
275 	skb->dev = queue->info->netdev;
276 
277 	return skb;
278 }
279 
280 
281 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
282 {
283 	RING_IDX req_prod = queue->rx.req_prod_pvt;
284 	int notify;
285 
286 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
287 		return;
288 
289 	for (req_prod = queue->rx.req_prod_pvt;
290 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
291 	     req_prod++) {
292 		struct sk_buff *skb;
293 		unsigned short id;
294 		grant_ref_t ref;
295 		unsigned long pfn;
296 		struct xen_netif_rx_request *req;
297 
298 		skb = xennet_alloc_one_rx_buffer(queue);
299 		if (!skb)
300 			break;
301 
302 		id = xennet_rxidx(req_prod);
303 
304 		BUG_ON(queue->rx_skbs[id]);
305 		queue->rx_skbs[id] = skb;
306 
307 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
308 		BUG_ON((signed short)ref < 0);
309 		queue->grant_rx_ref[id] = ref;
310 
311 		pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
312 
313 		req = RING_GET_REQUEST(&queue->rx, req_prod);
314 		gnttab_grant_foreign_access_ref(ref,
315 						queue->info->xbdev->otherend_id,
316 						pfn_to_mfn(pfn),
317 						0);
318 
319 		req->id = id;
320 		req->gref = ref;
321 	}
322 
323 	queue->rx.req_prod_pvt = req_prod;
324 
325 	/* Not enough requests? Try again later. */
326 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN) {
327 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
328 		return;
329 	}
330 
331 	wmb();		/* barrier so backend seens requests */
332 
333 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
334 	if (notify)
335 		notify_remote_via_irq(queue->rx_irq);
336 }
337 
338 static int xennet_open(struct net_device *dev)
339 {
340 	struct netfront_info *np = netdev_priv(dev);
341 	unsigned int num_queues = dev->real_num_tx_queues;
342 	unsigned int i = 0;
343 	struct netfront_queue *queue = NULL;
344 
345 	for (i = 0; i < num_queues; ++i) {
346 		queue = &np->queues[i];
347 		napi_enable(&queue->napi);
348 
349 		spin_lock_bh(&queue->rx_lock);
350 		if (netif_carrier_ok(dev)) {
351 			xennet_alloc_rx_buffers(queue);
352 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
353 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
354 				napi_schedule(&queue->napi);
355 		}
356 		spin_unlock_bh(&queue->rx_lock);
357 	}
358 
359 	netif_tx_start_all_queues(dev);
360 
361 	return 0;
362 }
363 
364 static void xennet_tx_buf_gc(struct netfront_queue *queue)
365 {
366 	RING_IDX cons, prod;
367 	unsigned short id;
368 	struct sk_buff *skb;
369 
370 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
371 
372 	do {
373 		prod = queue->tx.sring->rsp_prod;
374 		rmb(); /* Ensure we see responses up to 'rp'. */
375 
376 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
377 			struct xen_netif_tx_response *txrsp;
378 
379 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
380 			if (txrsp->status == XEN_NETIF_RSP_NULL)
381 				continue;
382 
383 			id  = txrsp->id;
384 			skb = queue->tx_skbs[id].skb;
385 			if (unlikely(gnttab_query_foreign_access(
386 				queue->grant_tx_ref[id]) != 0)) {
387 				pr_alert("%s: warning -- grant still in use by backend domain\n",
388 					 __func__);
389 				BUG();
390 			}
391 			gnttab_end_foreign_access_ref(
392 				queue->grant_tx_ref[id], GNTMAP_readonly);
393 			gnttab_release_grant_reference(
394 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
395 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
396 			queue->grant_tx_page[id] = NULL;
397 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
398 			dev_kfree_skb_irq(skb);
399 		}
400 
401 		queue->tx.rsp_cons = prod;
402 
403 		/*
404 		 * Set a new event, then check for race with update of tx_cons.
405 		 * Note that it is essential to schedule a callback, no matter
406 		 * how few buffers are pending. Even if there is space in the
407 		 * transmit ring, higher layers may be blocked because too much
408 		 * data is outstanding: in such cases notification from Xen is
409 		 * likely to be the only kick that we'll get.
410 		 */
411 		queue->tx.sring->rsp_event =
412 			prod + ((queue->tx.sring->req_prod - prod) >> 1) + 1;
413 		mb();		/* update shared area */
414 	} while ((cons == prod) && (prod != queue->tx.sring->rsp_prod));
415 
416 	xennet_maybe_wake_tx(queue);
417 }
418 
419 static struct xen_netif_tx_request *xennet_make_one_txreq(
420 	struct netfront_queue *queue, struct sk_buff *skb,
421 	struct page *page, unsigned int offset, unsigned int len)
422 {
423 	unsigned int id;
424 	struct xen_netif_tx_request *tx;
425 	grant_ref_t ref;
426 
427 	len = min_t(unsigned int, PAGE_SIZE - offset, len);
428 
429 	id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
430 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
431 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
432 	BUG_ON((signed short)ref < 0);
433 
434 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
435 					page_to_mfn(page), GNTMAP_readonly);
436 
437 	queue->tx_skbs[id].skb = skb;
438 	queue->grant_tx_page[id] = page;
439 	queue->grant_tx_ref[id] = ref;
440 
441 	tx->id = id;
442 	tx->gref = ref;
443 	tx->offset = offset;
444 	tx->size = len;
445 	tx->flags = 0;
446 
447 	return tx;
448 }
449 
450 static struct xen_netif_tx_request *xennet_make_txreqs(
451 	struct netfront_queue *queue, struct xen_netif_tx_request *tx,
452 	struct sk_buff *skb, struct page *page,
453 	unsigned int offset, unsigned int len)
454 {
455 	/* Skip unused frames from start of page */
456 	page += offset >> PAGE_SHIFT;
457 	offset &= ~PAGE_MASK;
458 
459 	while (len) {
460 		tx->flags |= XEN_NETTXF_more_data;
461 		tx = xennet_make_one_txreq(queue, skb_get(skb),
462 					   page, offset, len);
463 		page++;
464 		offset = 0;
465 		len -= tx->size;
466 	}
467 
468 	return tx;
469 }
470 
471 /*
472  * Count how many ring slots are required to send this skb. Each frag
473  * might be a compound page.
474  */
475 static int xennet_count_skb_slots(struct sk_buff *skb)
476 {
477 	int i, frags = skb_shinfo(skb)->nr_frags;
478 	int pages;
479 
480 	pages = PFN_UP(offset_in_page(skb->data) + skb_headlen(skb));
481 
482 	for (i = 0; i < frags; i++) {
483 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
484 		unsigned long size = skb_frag_size(frag);
485 		unsigned long offset = frag->page_offset;
486 
487 		/* Skip unused frames from start of page */
488 		offset &= ~PAGE_MASK;
489 
490 		pages += PFN_UP(offset + size);
491 	}
492 
493 	return pages;
494 }
495 
496 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
497 			       void *accel_priv, select_queue_fallback_t fallback)
498 {
499 	unsigned int num_queues = dev->real_num_tx_queues;
500 	u32 hash;
501 	u16 queue_idx;
502 
503 	/* First, check if there is only one queue */
504 	if (num_queues == 1) {
505 		queue_idx = 0;
506 	} else {
507 		hash = skb_get_hash(skb);
508 		queue_idx = hash % num_queues;
509 	}
510 
511 	return queue_idx;
512 }
513 
514 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
515 {
516 	struct netfront_info *np = netdev_priv(dev);
517 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
518 	struct xen_netif_tx_request *tx, *first_tx;
519 	unsigned int i;
520 	int notify;
521 	int slots;
522 	struct page *page;
523 	unsigned int offset;
524 	unsigned int len;
525 	unsigned long flags;
526 	struct netfront_queue *queue = NULL;
527 	unsigned int num_queues = dev->real_num_tx_queues;
528 	u16 queue_index;
529 
530 	/* Drop the packet if no queues are set up */
531 	if (num_queues < 1)
532 		goto drop;
533 	/* Determine which queue to transmit this SKB on */
534 	queue_index = skb_get_queue_mapping(skb);
535 	queue = &np->queues[queue_index];
536 
537 	/* If skb->len is too big for wire format, drop skb and alert
538 	 * user about misconfiguration.
539 	 */
540 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
541 		net_alert_ratelimited(
542 			"xennet: skb->len = %u, too big for wire format\n",
543 			skb->len);
544 		goto drop;
545 	}
546 
547 	slots = xennet_count_skb_slots(skb);
548 	if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
549 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
550 				    slots, skb->len);
551 		if (skb_linearize(skb))
552 			goto drop;
553 	}
554 
555 	page = virt_to_page(skb->data);
556 	offset = offset_in_page(skb->data);
557 	len = skb_headlen(skb);
558 
559 	spin_lock_irqsave(&queue->tx_lock, flags);
560 
561 	if (unlikely(!netif_carrier_ok(dev) ||
562 		     (slots > 1 && !xennet_can_sg(dev)) ||
563 		     netif_needs_gso(dev, skb, netif_skb_features(skb)))) {
564 		spin_unlock_irqrestore(&queue->tx_lock, flags);
565 		goto drop;
566 	}
567 
568 	/* First request for the linear area. */
569 	first_tx = tx = xennet_make_one_txreq(queue, skb,
570 					      page, offset, len);
571 	page++;
572 	offset = 0;
573 	len -= tx->size;
574 
575 	if (skb->ip_summed == CHECKSUM_PARTIAL)
576 		/* local packet? */
577 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
578 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
579 		/* remote but checksummed. */
580 		tx->flags |= XEN_NETTXF_data_validated;
581 
582 	/* Optional extra info after the first request. */
583 	if (skb_shinfo(skb)->gso_size) {
584 		struct xen_netif_extra_info *gso;
585 
586 		gso = (struct xen_netif_extra_info *)
587 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
588 
589 		tx->flags |= XEN_NETTXF_extra_info;
590 
591 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
592 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
593 			XEN_NETIF_GSO_TYPE_TCPV6 :
594 			XEN_NETIF_GSO_TYPE_TCPV4;
595 		gso->u.gso.pad = 0;
596 		gso->u.gso.features = 0;
597 
598 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
599 		gso->flags = 0;
600 	}
601 
602 	/* Requests for the rest of the linear area. */
603 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
604 
605 	/* Requests for all the frags. */
606 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
607 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
608 		tx = xennet_make_txreqs(queue, tx, skb,
609 					skb_frag_page(frag), frag->page_offset,
610 					skb_frag_size(frag));
611 	}
612 
613 	/* First request has the packet length. */
614 	first_tx->size = skb->len;
615 
616 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
617 	if (notify)
618 		notify_remote_via_irq(queue->tx_irq);
619 
620 	u64_stats_update_begin(&tx_stats->syncp);
621 	tx_stats->bytes += skb->len;
622 	tx_stats->packets++;
623 	u64_stats_update_end(&tx_stats->syncp);
624 
625 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
626 	xennet_tx_buf_gc(queue);
627 
628 	if (!netfront_tx_slot_available(queue))
629 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
630 
631 	spin_unlock_irqrestore(&queue->tx_lock, flags);
632 
633 	return NETDEV_TX_OK;
634 
635  drop:
636 	dev->stats.tx_dropped++;
637 	dev_kfree_skb_any(skb);
638 	return NETDEV_TX_OK;
639 }
640 
641 static int xennet_close(struct net_device *dev)
642 {
643 	struct netfront_info *np = netdev_priv(dev);
644 	unsigned int num_queues = dev->real_num_tx_queues;
645 	unsigned int i;
646 	struct netfront_queue *queue;
647 	netif_tx_stop_all_queues(np->netdev);
648 	for (i = 0; i < num_queues; ++i) {
649 		queue = &np->queues[i];
650 		napi_disable(&queue->napi);
651 	}
652 	return 0;
653 }
654 
655 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
656 				grant_ref_t ref)
657 {
658 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
659 
660 	BUG_ON(queue->rx_skbs[new]);
661 	queue->rx_skbs[new] = skb;
662 	queue->grant_rx_ref[new] = ref;
663 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
664 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
665 	queue->rx.req_prod_pvt++;
666 }
667 
668 static int xennet_get_extras(struct netfront_queue *queue,
669 			     struct xen_netif_extra_info *extras,
670 			     RING_IDX rp)
671 
672 {
673 	struct xen_netif_extra_info *extra;
674 	struct device *dev = &queue->info->netdev->dev;
675 	RING_IDX cons = queue->rx.rsp_cons;
676 	int err = 0;
677 
678 	do {
679 		struct sk_buff *skb;
680 		grant_ref_t ref;
681 
682 		if (unlikely(cons + 1 == rp)) {
683 			if (net_ratelimit())
684 				dev_warn(dev, "Missing extra info\n");
685 			err = -EBADR;
686 			break;
687 		}
688 
689 		extra = (struct xen_netif_extra_info *)
690 			RING_GET_RESPONSE(&queue->rx, ++cons);
691 
692 		if (unlikely(!extra->type ||
693 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
694 			if (net_ratelimit())
695 				dev_warn(dev, "Invalid extra type: %d\n",
696 					extra->type);
697 			err = -EINVAL;
698 		} else {
699 			memcpy(&extras[extra->type - 1], extra,
700 			       sizeof(*extra));
701 		}
702 
703 		skb = xennet_get_rx_skb(queue, cons);
704 		ref = xennet_get_rx_ref(queue, cons);
705 		xennet_move_rx_slot(queue, skb, ref);
706 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
707 
708 	queue->rx.rsp_cons = cons;
709 	return err;
710 }
711 
712 static int xennet_get_responses(struct netfront_queue *queue,
713 				struct netfront_rx_info *rinfo, RING_IDX rp,
714 				struct sk_buff_head *list)
715 {
716 	struct xen_netif_rx_response *rx = &rinfo->rx;
717 	struct xen_netif_extra_info *extras = rinfo->extras;
718 	struct device *dev = &queue->info->netdev->dev;
719 	RING_IDX cons = queue->rx.rsp_cons;
720 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
721 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
722 	int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
723 	int slots = 1;
724 	int err = 0;
725 	unsigned long ret;
726 
727 	if (rx->flags & XEN_NETRXF_extra_info) {
728 		err = xennet_get_extras(queue, extras, rp);
729 		cons = queue->rx.rsp_cons;
730 	}
731 
732 	for (;;) {
733 		if (unlikely(rx->status < 0 ||
734 			     rx->offset + rx->status > PAGE_SIZE)) {
735 			if (net_ratelimit())
736 				dev_warn(dev, "rx->offset: %x, size: %u\n",
737 					 rx->offset, rx->status);
738 			xennet_move_rx_slot(queue, skb, ref);
739 			err = -EINVAL;
740 			goto next;
741 		}
742 
743 		/*
744 		 * This definitely indicates a bug, either in this driver or in
745 		 * the backend driver. In future this should flag the bad
746 		 * situation to the system controller to reboot the backend.
747 		 */
748 		if (ref == GRANT_INVALID_REF) {
749 			if (net_ratelimit())
750 				dev_warn(dev, "Bad rx response id %d.\n",
751 					 rx->id);
752 			err = -EINVAL;
753 			goto next;
754 		}
755 
756 		ret = gnttab_end_foreign_access_ref(ref, 0);
757 		BUG_ON(!ret);
758 
759 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
760 
761 		__skb_queue_tail(list, skb);
762 
763 next:
764 		if (!(rx->flags & XEN_NETRXF_more_data))
765 			break;
766 
767 		if (cons + slots == rp) {
768 			if (net_ratelimit())
769 				dev_warn(dev, "Need more slots\n");
770 			err = -ENOENT;
771 			break;
772 		}
773 
774 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
775 		skb = xennet_get_rx_skb(queue, cons + slots);
776 		ref = xennet_get_rx_ref(queue, cons + slots);
777 		slots++;
778 	}
779 
780 	if (unlikely(slots > max)) {
781 		if (net_ratelimit())
782 			dev_warn(dev, "Too many slots\n");
783 		err = -E2BIG;
784 	}
785 
786 	if (unlikely(err))
787 		queue->rx.rsp_cons = cons + slots;
788 
789 	return err;
790 }
791 
792 static int xennet_set_skb_gso(struct sk_buff *skb,
793 			      struct xen_netif_extra_info *gso)
794 {
795 	if (!gso->u.gso.size) {
796 		if (net_ratelimit())
797 			pr_warn("GSO size must not be zero\n");
798 		return -EINVAL;
799 	}
800 
801 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
802 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
803 		if (net_ratelimit())
804 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
805 		return -EINVAL;
806 	}
807 
808 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
809 	skb_shinfo(skb)->gso_type =
810 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
811 		SKB_GSO_TCPV4 :
812 		SKB_GSO_TCPV6;
813 
814 	/* Header must be checked, and gso_segs computed. */
815 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
816 	skb_shinfo(skb)->gso_segs = 0;
817 
818 	return 0;
819 }
820 
821 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
822 				  struct sk_buff *skb,
823 				  struct sk_buff_head *list)
824 {
825 	struct skb_shared_info *shinfo = skb_shinfo(skb);
826 	RING_IDX cons = queue->rx.rsp_cons;
827 	struct sk_buff *nskb;
828 
829 	while ((nskb = __skb_dequeue(list))) {
830 		struct xen_netif_rx_response *rx =
831 			RING_GET_RESPONSE(&queue->rx, ++cons);
832 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
833 
834 		if (shinfo->nr_frags == MAX_SKB_FRAGS) {
835 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
836 
837 			BUG_ON(pull_to <= skb_headlen(skb));
838 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
839 		}
840 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
841 
842 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
843 				rx->offset, rx->status, PAGE_SIZE);
844 
845 		skb_shinfo(nskb)->nr_frags = 0;
846 		kfree_skb(nskb);
847 	}
848 
849 	return cons;
850 }
851 
852 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
853 {
854 	bool recalculate_partial_csum = false;
855 
856 	/*
857 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
858 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
859 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
860 	 * recalculate the partial checksum.
861 	 */
862 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
863 		struct netfront_info *np = netdev_priv(dev);
864 		atomic_inc(&np->rx_gso_checksum_fixup);
865 		skb->ip_summed = CHECKSUM_PARTIAL;
866 		recalculate_partial_csum = true;
867 	}
868 
869 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
870 	if (skb->ip_summed != CHECKSUM_PARTIAL)
871 		return 0;
872 
873 	return skb_checksum_setup(skb, recalculate_partial_csum);
874 }
875 
876 static int handle_incoming_queue(struct netfront_queue *queue,
877 				 struct sk_buff_head *rxq)
878 {
879 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
880 	int packets_dropped = 0;
881 	struct sk_buff *skb;
882 
883 	while ((skb = __skb_dequeue(rxq)) != NULL) {
884 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
885 
886 		if (pull_to > skb_headlen(skb))
887 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
888 
889 		/* Ethernet work: Delayed to here as it peeks the header. */
890 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
891 		skb_reset_network_header(skb);
892 
893 		if (checksum_setup(queue->info->netdev, skb)) {
894 			kfree_skb(skb);
895 			packets_dropped++;
896 			queue->info->netdev->stats.rx_errors++;
897 			continue;
898 		}
899 
900 		u64_stats_update_begin(&rx_stats->syncp);
901 		rx_stats->packets++;
902 		rx_stats->bytes += skb->len;
903 		u64_stats_update_end(&rx_stats->syncp);
904 
905 		/* Pass it up. */
906 		napi_gro_receive(&queue->napi, skb);
907 	}
908 
909 	return packets_dropped;
910 }
911 
912 static int xennet_poll(struct napi_struct *napi, int budget)
913 {
914 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
915 	struct net_device *dev = queue->info->netdev;
916 	struct sk_buff *skb;
917 	struct netfront_rx_info rinfo;
918 	struct xen_netif_rx_response *rx = &rinfo.rx;
919 	struct xen_netif_extra_info *extras = rinfo.extras;
920 	RING_IDX i, rp;
921 	int work_done;
922 	struct sk_buff_head rxq;
923 	struct sk_buff_head errq;
924 	struct sk_buff_head tmpq;
925 	int err;
926 
927 	spin_lock(&queue->rx_lock);
928 
929 	skb_queue_head_init(&rxq);
930 	skb_queue_head_init(&errq);
931 	skb_queue_head_init(&tmpq);
932 
933 	rp = queue->rx.sring->rsp_prod;
934 	rmb(); /* Ensure we see queued responses up to 'rp'. */
935 
936 	i = queue->rx.rsp_cons;
937 	work_done = 0;
938 	while ((i != rp) && (work_done < budget)) {
939 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
940 		memset(extras, 0, sizeof(rinfo.extras));
941 
942 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
943 
944 		if (unlikely(err)) {
945 err:
946 			while ((skb = __skb_dequeue(&tmpq)))
947 				__skb_queue_tail(&errq, skb);
948 			dev->stats.rx_errors++;
949 			i = queue->rx.rsp_cons;
950 			continue;
951 		}
952 
953 		skb = __skb_dequeue(&tmpq);
954 
955 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
956 			struct xen_netif_extra_info *gso;
957 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
958 
959 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
960 				__skb_queue_head(&tmpq, skb);
961 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
962 				goto err;
963 			}
964 		}
965 
966 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
967 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
968 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
969 
970 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
971 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
972 		skb->data_len = rx->status;
973 		skb->len += rx->status;
974 
975 		i = xennet_fill_frags(queue, skb, &tmpq);
976 
977 		if (rx->flags & XEN_NETRXF_csum_blank)
978 			skb->ip_summed = CHECKSUM_PARTIAL;
979 		else if (rx->flags & XEN_NETRXF_data_validated)
980 			skb->ip_summed = CHECKSUM_UNNECESSARY;
981 
982 		__skb_queue_tail(&rxq, skb);
983 
984 		queue->rx.rsp_cons = ++i;
985 		work_done++;
986 	}
987 
988 	__skb_queue_purge(&errq);
989 
990 	work_done -= handle_incoming_queue(queue, &rxq);
991 
992 	xennet_alloc_rx_buffers(queue);
993 
994 	if (work_done < budget) {
995 		int more_to_do = 0;
996 
997 		napi_complete(napi);
998 
999 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1000 		if (more_to_do)
1001 			napi_schedule(napi);
1002 	}
1003 
1004 	spin_unlock(&queue->rx_lock);
1005 
1006 	return work_done;
1007 }
1008 
1009 static int xennet_change_mtu(struct net_device *dev, int mtu)
1010 {
1011 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1012 
1013 	if (mtu > max)
1014 		return -EINVAL;
1015 	dev->mtu = mtu;
1016 	return 0;
1017 }
1018 
1019 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1020 						    struct rtnl_link_stats64 *tot)
1021 {
1022 	struct netfront_info *np = netdev_priv(dev);
1023 	int cpu;
1024 
1025 	for_each_possible_cpu(cpu) {
1026 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1027 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1028 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1029 		unsigned int start;
1030 
1031 		do {
1032 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1033 			tx_packets = tx_stats->packets;
1034 			tx_bytes = tx_stats->bytes;
1035 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1036 
1037 		do {
1038 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1039 			rx_packets = rx_stats->packets;
1040 			rx_bytes = rx_stats->bytes;
1041 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1042 
1043 		tot->rx_packets += rx_packets;
1044 		tot->tx_packets += tx_packets;
1045 		tot->rx_bytes   += rx_bytes;
1046 		tot->tx_bytes   += tx_bytes;
1047 	}
1048 
1049 	tot->rx_errors  = dev->stats.rx_errors;
1050 	tot->tx_dropped = dev->stats.tx_dropped;
1051 
1052 	return tot;
1053 }
1054 
1055 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1056 {
1057 	struct sk_buff *skb;
1058 	int i;
1059 
1060 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1061 		/* Skip over entries which are actually freelist references */
1062 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1063 			continue;
1064 
1065 		skb = queue->tx_skbs[i].skb;
1066 		get_page(queue->grant_tx_page[i]);
1067 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1068 					  GNTMAP_readonly,
1069 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1070 		queue->grant_tx_page[i] = NULL;
1071 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1072 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1073 		dev_kfree_skb_irq(skb);
1074 	}
1075 }
1076 
1077 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1078 {
1079 	int id, ref;
1080 
1081 	spin_lock_bh(&queue->rx_lock);
1082 
1083 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1084 		struct sk_buff *skb;
1085 		struct page *page;
1086 
1087 		skb = queue->rx_skbs[id];
1088 		if (!skb)
1089 			continue;
1090 
1091 		ref = queue->grant_rx_ref[id];
1092 		if (ref == GRANT_INVALID_REF)
1093 			continue;
1094 
1095 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1096 
1097 		/* gnttab_end_foreign_access() needs a page ref until
1098 		 * foreign access is ended (which may be deferred).
1099 		 */
1100 		get_page(page);
1101 		gnttab_end_foreign_access(ref, 0,
1102 					  (unsigned long)page_address(page));
1103 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1104 
1105 		kfree_skb(skb);
1106 	}
1107 
1108 	spin_unlock_bh(&queue->rx_lock);
1109 }
1110 
1111 static netdev_features_t xennet_fix_features(struct net_device *dev,
1112 	netdev_features_t features)
1113 {
1114 	struct netfront_info *np = netdev_priv(dev);
1115 	int val;
1116 
1117 	if (features & NETIF_F_SG) {
1118 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1119 				 "%d", &val) < 0)
1120 			val = 0;
1121 
1122 		if (!val)
1123 			features &= ~NETIF_F_SG;
1124 	}
1125 
1126 	if (features & NETIF_F_IPV6_CSUM) {
1127 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1128 				 "feature-ipv6-csum-offload", "%d", &val) < 0)
1129 			val = 0;
1130 
1131 		if (!val)
1132 			features &= ~NETIF_F_IPV6_CSUM;
1133 	}
1134 
1135 	if (features & NETIF_F_TSO) {
1136 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1137 				 "feature-gso-tcpv4", "%d", &val) < 0)
1138 			val = 0;
1139 
1140 		if (!val)
1141 			features &= ~NETIF_F_TSO;
1142 	}
1143 
1144 	if (features & NETIF_F_TSO6) {
1145 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1146 				 "feature-gso-tcpv6", "%d", &val) < 0)
1147 			val = 0;
1148 
1149 		if (!val)
1150 			features &= ~NETIF_F_TSO6;
1151 	}
1152 
1153 	return features;
1154 }
1155 
1156 static int xennet_set_features(struct net_device *dev,
1157 	netdev_features_t features)
1158 {
1159 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1160 		netdev_info(dev, "Reducing MTU because no SG offload");
1161 		dev->mtu = ETH_DATA_LEN;
1162 	}
1163 
1164 	return 0;
1165 }
1166 
1167 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1168 {
1169 	struct netfront_queue *queue = dev_id;
1170 	unsigned long flags;
1171 
1172 	spin_lock_irqsave(&queue->tx_lock, flags);
1173 	xennet_tx_buf_gc(queue);
1174 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1175 
1176 	return IRQ_HANDLED;
1177 }
1178 
1179 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1180 {
1181 	struct netfront_queue *queue = dev_id;
1182 	struct net_device *dev = queue->info->netdev;
1183 
1184 	if (likely(netif_carrier_ok(dev) &&
1185 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1186 		napi_schedule(&queue->napi);
1187 
1188 	return IRQ_HANDLED;
1189 }
1190 
1191 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1192 {
1193 	xennet_tx_interrupt(irq, dev_id);
1194 	xennet_rx_interrupt(irq, dev_id);
1195 	return IRQ_HANDLED;
1196 }
1197 
1198 #ifdef CONFIG_NET_POLL_CONTROLLER
1199 static void xennet_poll_controller(struct net_device *dev)
1200 {
1201 	/* Poll each queue */
1202 	struct netfront_info *info = netdev_priv(dev);
1203 	unsigned int num_queues = dev->real_num_tx_queues;
1204 	unsigned int i;
1205 	for (i = 0; i < num_queues; ++i)
1206 		xennet_interrupt(0, &info->queues[i]);
1207 }
1208 #endif
1209 
1210 static const struct net_device_ops xennet_netdev_ops = {
1211 	.ndo_open            = xennet_open,
1212 	.ndo_stop            = xennet_close,
1213 	.ndo_start_xmit      = xennet_start_xmit,
1214 	.ndo_change_mtu	     = xennet_change_mtu,
1215 	.ndo_get_stats64     = xennet_get_stats64,
1216 	.ndo_set_mac_address = eth_mac_addr,
1217 	.ndo_validate_addr   = eth_validate_addr,
1218 	.ndo_fix_features    = xennet_fix_features,
1219 	.ndo_set_features    = xennet_set_features,
1220 	.ndo_select_queue    = xennet_select_queue,
1221 #ifdef CONFIG_NET_POLL_CONTROLLER
1222 	.ndo_poll_controller = xennet_poll_controller,
1223 #endif
1224 };
1225 
1226 static void xennet_free_netdev(struct net_device *netdev)
1227 {
1228 	struct netfront_info *np = netdev_priv(netdev);
1229 
1230 	free_percpu(np->rx_stats);
1231 	free_percpu(np->tx_stats);
1232 	free_netdev(netdev);
1233 }
1234 
1235 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1236 {
1237 	int err;
1238 	struct net_device *netdev;
1239 	struct netfront_info *np;
1240 
1241 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1242 	if (!netdev)
1243 		return ERR_PTR(-ENOMEM);
1244 
1245 	np                   = netdev_priv(netdev);
1246 	np->xbdev            = dev;
1247 
1248 	/* No need to use rtnl_lock() before the call below as it
1249 	 * happens before register_netdev().
1250 	 */
1251 	netif_set_real_num_tx_queues(netdev, 0);
1252 	np->queues = NULL;
1253 
1254 	err = -ENOMEM;
1255 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1256 	if (np->rx_stats == NULL)
1257 		goto exit;
1258 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1259 	if (np->tx_stats == NULL)
1260 		goto exit;
1261 
1262 	netdev->netdev_ops	= &xennet_netdev_ops;
1263 
1264 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1265 				  NETIF_F_GSO_ROBUST;
1266 	netdev->hw_features	= NETIF_F_SG |
1267 				  NETIF_F_IPV6_CSUM |
1268 				  NETIF_F_TSO | NETIF_F_TSO6;
1269 
1270 	/*
1271          * Assume that all hw features are available for now. This set
1272          * will be adjusted by the call to netdev_update_features() in
1273          * xennet_connect() which is the earliest point where we can
1274          * negotiate with the backend regarding supported features.
1275          */
1276 	netdev->features |= netdev->hw_features;
1277 
1278 	netdev->ethtool_ops = &xennet_ethtool_ops;
1279 	SET_NETDEV_DEV(netdev, &dev->dev);
1280 
1281 	np->netdev = netdev;
1282 
1283 	netif_carrier_off(netdev);
1284 
1285 	return netdev;
1286 
1287  exit:
1288 	xennet_free_netdev(netdev);
1289 	return ERR_PTR(err);
1290 }
1291 
1292 /**
1293  * Entry point to this code when a new device is created.  Allocate the basic
1294  * structures and the ring buffers for communication with the backend, and
1295  * inform the backend of the appropriate details for those.
1296  */
1297 static int netfront_probe(struct xenbus_device *dev,
1298 			  const struct xenbus_device_id *id)
1299 {
1300 	int err;
1301 	struct net_device *netdev;
1302 	struct netfront_info *info;
1303 
1304 	netdev = xennet_create_dev(dev);
1305 	if (IS_ERR(netdev)) {
1306 		err = PTR_ERR(netdev);
1307 		xenbus_dev_fatal(dev, err, "creating netdev");
1308 		return err;
1309 	}
1310 
1311 	info = netdev_priv(netdev);
1312 	dev_set_drvdata(&dev->dev, info);
1313 #ifdef CONFIG_SYSFS
1314 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1315 #endif
1316 	err = register_netdev(info->netdev);
1317 	if (err) {
1318 		pr_warn("%s: register_netdev err=%d\n", __func__, err);
1319 		goto fail;
1320 	}
1321 
1322 	return 0;
1323 
1324  fail:
1325 	xennet_free_netdev(netdev);
1326 	dev_set_drvdata(&dev->dev, NULL);
1327 	return err;
1328 }
1329 
1330 static void xennet_end_access(int ref, void *page)
1331 {
1332 	/* This frees the page as a side-effect */
1333 	if (ref != GRANT_INVALID_REF)
1334 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1335 }
1336 
1337 static void xennet_disconnect_backend(struct netfront_info *info)
1338 {
1339 	unsigned int i = 0;
1340 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1341 
1342 	netif_carrier_off(info->netdev);
1343 
1344 	for (i = 0; i < num_queues; ++i) {
1345 		struct netfront_queue *queue = &info->queues[i];
1346 
1347 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1348 			unbind_from_irqhandler(queue->tx_irq, queue);
1349 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1350 			unbind_from_irqhandler(queue->tx_irq, queue);
1351 			unbind_from_irqhandler(queue->rx_irq, queue);
1352 		}
1353 		queue->tx_evtchn = queue->rx_evtchn = 0;
1354 		queue->tx_irq = queue->rx_irq = 0;
1355 
1356 		napi_synchronize(&queue->napi);
1357 
1358 		xennet_release_tx_bufs(queue);
1359 		xennet_release_rx_bufs(queue);
1360 		gnttab_free_grant_references(queue->gref_tx_head);
1361 		gnttab_free_grant_references(queue->gref_rx_head);
1362 
1363 		/* End access and free the pages */
1364 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1365 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1366 
1367 		queue->tx_ring_ref = GRANT_INVALID_REF;
1368 		queue->rx_ring_ref = GRANT_INVALID_REF;
1369 		queue->tx.sring = NULL;
1370 		queue->rx.sring = NULL;
1371 	}
1372 }
1373 
1374 /**
1375  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1376  * driver restart.  We tear down our netif structure and recreate it, but
1377  * leave the device-layer structures intact so that this is transparent to the
1378  * rest of the kernel.
1379  */
1380 static int netfront_resume(struct xenbus_device *dev)
1381 {
1382 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1383 
1384 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1385 
1386 	xennet_disconnect_backend(info);
1387 	return 0;
1388 }
1389 
1390 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1391 {
1392 	char *s, *e, *macstr;
1393 	int i;
1394 
1395 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1396 	if (IS_ERR(macstr))
1397 		return PTR_ERR(macstr);
1398 
1399 	for (i = 0; i < ETH_ALEN; i++) {
1400 		mac[i] = simple_strtoul(s, &e, 16);
1401 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1402 			kfree(macstr);
1403 			return -ENOENT;
1404 		}
1405 		s = e+1;
1406 	}
1407 
1408 	kfree(macstr);
1409 	return 0;
1410 }
1411 
1412 static int setup_netfront_single(struct netfront_queue *queue)
1413 {
1414 	int err;
1415 
1416 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1417 	if (err < 0)
1418 		goto fail;
1419 
1420 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1421 					xennet_interrupt,
1422 					0, queue->info->netdev->name, queue);
1423 	if (err < 0)
1424 		goto bind_fail;
1425 	queue->rx_evtchn = queue->tx_evtchn;
1426 	queue->rx_irq = queue->tx_irq = err;
1427 
1428 	return 0;
1429 
1430 bind_fail:
1431 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1432 	queue->tx_evtchn = 0;
1433 fail:
1434 	return err;
1435 }
1436 
1437 static int setup_netfront_split(struct netfront_queue *queue)
1438 {
1439 	int err;
1440 
1441 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1442 	if (err < 0)
1443 		goto fail;
1444 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1445 	if (err < 0)
1446 		goto alloc_rx_evtchn_fail;
1447 
1448 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1449 		 "%s-tx", queue->name);
1450 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1451 					xennet_tx_interrupt,
1452 					0, queue->tx_irq_name, queue);
1453 	if (err < 0)
1454 		goto bind_tx_fail;
1455 	queue->tx_irq = err;
1456 
1457 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1458 		 "%s-rx", queue->name);
1459 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1460 					xennet_rx_interrupt,
1461 					0, queue->rx_irq_name, queue);
1462 	if (err < 0)
1463 		goto bind_rx_fail;
1464 	queue->rx_irq = err;
1465 
1466 	return 0;
1467 
1468 bind_rx_fail:
1469 	unbind_from_irqhandler(queue->tx_irq, queue);
1470 	queue->tx_irq = 0;
1471 bind_tx_fail:
1472 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1473 	queue->rx_evtchn = 0;
1474 alloc_rx_evtchn_fail:
1475 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1476 	queue->tx_evtchn = 0;
1477 fail:
1478 	return err;
1479 }
1480 
1481 static int setup_netfront(struct xenbus_device *dev,
1482 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1483 {
1484 	struct xen_netif_tx_sring *txs;
1485 	struct xen_netif_rx_sring *rxs;
1486 	int err;
1487 
1488 	queue->tx_ring_ref = GRANT_INVALID_REF;
1489 	queue->rx_ring_ref = GRANT_INVALID_REF;
1490 	queue->rx.sring = NULL;
1491 	queue->tx.sring = NULL;
1492 
1493 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1494 	if (!txs) {
1495 		err = -ENOMEM;
1496 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1497 		goto fail;
1498 	}
1499 	SHARED_RING_INIT(txs);
1500 	FRONT_RING_INIT(&queue->tx, txs, PAGE_SIZE);
1501 
1502 	err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1503 	if (err < 0)
1504 		goto grant_tx_ring_fail;
1505 	queue->tx_ring_ref = err;
1506 
1507 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1508 	if (!rxs) {
1509 		err = -ENOMEM;
1510 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1511 		goto alloc_rx_ring_fail;
1512 	}
1513 	SHARED_RING_INIT(rxs);
1514 	FRONT_RING_INIT(&queue->rx, rxs, PAGE_SIZE);
1515 
1516 	err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1517 	if (err < 0)
1518 		goto grant_rx_ring_fail;
1519 	queue->rx_ring_ref = err;
1520 
1521 	if (feature_split_evtchn)
1522 		err = setup_netfront_split(queue);
1523 	/* setup single event channel if
1524 	 *  a) feature-split-event-channels == 0
1525 	 *  b) feature-split-event-channels == 1 but failed to setup
1526 	 */
1527 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1528 		err = setup_netfront_single(queue);
1529 
1530 	if (err)
1531 		goto alloc_evtchn_fail;
1532 
1533 	return 0;
1534 
1535 	/* If we fail to setup netfront, it is safe to just revoke access to
1536 	 * granted pages because backend is not accessing it at this point.
1537 	 */
1538 alloc_evtchn_fail:
1539 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1540 grant_rx_ring_fail:
1541 	free_page((unsigned long)rxs);
1542 alloc_rx_ring_fail:
1543 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1544 grant_tx_ring_fail:
1545 	free_page((unsigned long)txs);
1546 fail:
1547 	return err;
1548 }
1549 
1550 /* Queue-specific initialisation
1551  * This used to be done in xennet_create_dev() but must now
1552  * be run per-queue.
1553  */
1554 static int xennet_init_queue(struct netfront_queue *queue)
1555 {
1556 	unsigned short i;
1557 	int err = 0;
1558 
1559 	spin_lock_init(&queue->tx_lock);
1560 	spin_lock_init(&queue->rx_lock);
1561 
1562 	init_timer(&queue->rx_refill_timer);
1563 	queue->rx_refill_timer.data = (unsigned long)queue;
1564 	queue->rx_refill_timer.function = rx_refill_timeout;
1565 
1566 	snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1567 		 queue->info->netdev->name, queue->id);
1568 
1569 	/* Initialise tx_skbs as a free chain containing every entry. */
1570 	queue->tx_skb_freelist = 0;
1571 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1572 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1573 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1574 		queue->grant_tx_page[i] = NULL;
1575 	}
1576 
1577 	/* Clear out rx_skbs */
1578 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1579 		queue->rx_skbs[i] = NULL;
1580 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1581 	}
1582 
1583 	/* A grant for every tx ring slot */
1584 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1585 					  &queue->gref_tx_head) < 0) {
1586 		pr_alert("can't alloc tx grant refs\n");
1587 		err = -ENOMEM;
1588 		goto exit;
1589 	}
1590 
1591 	/* A grant for every rx ring slot */
1592 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1593 					  &queue->gref_rx_head) < 0) {
1594 		pr_alert("can't alloc rx grant refs\n");
1595 		err = -ENOMEM;
1596 		goto exit_free_tx;
1597 	}
1598 
1599 	return 0;
1600 
1601  exit_free_tx:
1602 	gnttab_free_grant_references(queue->gref_tx_head);
1603  exit:
1604 	return err;
1605 }
1606 
1607 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1608 			   struct xenbus_transaction *xbt, int write_hierarchical)
1609 {
1610 	/* Write the queue-specific keys into XenStore in the traditional
1611 	 * way for a single queue, or in a queue subkeys for multiple
1612 	 * queues.
1613 	 */
1614 	struct xenbus_device *dev = queue->info->xbdev;
1615 	int err;
1616 	const char *message;
1617 	char *path;
1618 	size_t pathsize;
1619 
1620 	/* Choose the correct place to write the keys */
1621 	if (write_hierarchical) {
1622 		pathsize = strlen(dev->nodename) + 10;
1623 		path = kzalloc(pathsize, GFP_KERNEL);
1624 		if (!path) {
1625 			err = -ENOMEM;
1626 			message = "out of memory while writing ring references";
1627 			goto error;
1628 		}
1629 		snprintf(path, pathsize, "%s/queue-%u",
1630 				dev->nodename, queue->id);
1631 	} else {
1632 		path = (char *)dev->nodename;
1633 	}
1634 
1635 	/* Write ring references */
1636 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1637 			queue->tx_ring_ref);
1638 	if (err) {
1639 		message = "writing tx-ring-ref";
1640 		goto error;
1641 	}
1642 
1643 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1644 			queue->rx_ring_ref);
1645 	if (err) {
1646 		message = "writing rx-ring-ref";
1647 		goto error;
1648 	}
1649 
1650 	/* Write event channels; taking into account both shared
1651 	 * and split event channel scenarios.
1652 	 */
1653 	if (queue->tx_evtchn == queue->rx_evtchn) {
1654 		/* Shared event channel */
1655 		err = xenbus_printf(*xbt, path,
1656 				"event-channel", "%u", queue->tx_evtchn);
1657 		if (err) {
1658 			message = "writing event-channel";
1659 			goto error;
1660 		}
1661 	} else {
1662 		/* Split event channels */
1663 		err = xenbus_printf(*xbt, path,
1664 				"event-channel-tx", "%u", queue->tx_evtchn);
1665 		if (err) {
1666 			message = "writing event-channel-tx";
1667 			goto error;
1668 		}
1669 
1670 		err = xenbus_printf(*xbt, path,
1671 				"event-channel-rx", "%u", queue->rx_evtchn);
1672 		if (err) {
1673 			message = "writing event-channel-rx";
1674 			goto error;
1675 		}
1676 	}
1677 
1678 	if (write_hierarchical)
1679 		kfree(path);
1680 	return 0;
1681 
1682 error:
1683 	if (write_hierarchical)
1684 		kfree(path);
1685 	xenbus_dev_fatal(dev, err, "%s", message);
1686 	return err;
1687 }
1688 
1689 static void xennet_destroy_queues(struct netfront_info *info)
1690 {
1691 	unsigned int i;
1692 
1693 	rtnl_lock();
1694 
1695 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1696 		struct netfront_queue *queue = &info->queues[i];
1697 
1698 		if (netif_running(info->netdev))
1699 			napi_disable(&queue->napi);
1700 		netif_napi_del(&queue->napi);
1701 	}
1702 
1703 	rtnl_unlock();
1704 
1705 	kfree(info->queues);
1706 	info->queues = NULL;
1707 }
1708 
1709 static int xennet_create_queues(struct netfront_info *info,
1710 				unsigned int num_queues)
1711 {
1712 	unsigned int i;
1713 	int ret;
1714 
1715 	info->queues = kcalloc(num_queues, sizeof(struct netfront_queue),
1716 			       GFP_KERNEL);
1717 	if (!info->queues)
1718 		return -ENOMEM;
1719 
1720 	rtnl_lock();
1721 
1722 	for (i = 0; i < num_queues; i++) {
1723 		struct netfront_queue *queue = &info->queues[i];
1724 
1725 		queue->id = i;
1726 		queue->info = info;
1727 
1728 		ret = xennet_init_queue(queue);
1729 		if (ret < 0) {
1730 			dev_warn(&info->netdev->dev,
1731 				 "only created %d queues\n", i);
1732 			num_queues = i;
1733 			break;
1734 		}
1735 
1736 		netif_napi_add(queue->info->netdev, &queue->napi,
1737 			       xennet_poll, 64);
1738 		if (netif_running(info->netdev))
1739 			napi_enable(&queue->napi);
1740 	}
1741 
1742 	netif_set_real_num_tx_queues(info->netdev, num_queues);
1743 
1744 	rtnl_unlock();
1745 
1746 	if (num_queues == 0) {
1747 		dev_err(&info->netdev->dev, "no queues\n");
1748 		return -EINVAL;
1749 	}
1750 	return 0;
1751 }
1752 
1753 /* Common code used when first setting up, and when resuming. */
1754 static int talk_to_netback(struct xenbus_device *dev,
1755 			   struct netfront_info *info)
1756 {
1757 	const char *message;
1758 	struct xenbus_transaction xbt;
1759 	int err;
1760 	unsigned int feature_split_evtchn;
1761 	unsigned int i = 0;
1762 	unsigned int max_queues = 0;
1763 	struct netfront_queue *queue = NULL;
1764 	unsigned int num_queues = 1;
1765 
1766 	info->netdev->irq = 0;
1767 
1768 	/* Check if backend supports multiple queues */
1769 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1770 			   "multi-queue-max-queues", "%u", &max_queues);
1771 	if (err < 0)
1772 		max_queues = 1;
1773 	num_queues = min(max_queues, xennet_max_queues);
1774 
1775 	/* Check feature-split-event-channels */
1776 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1777 			   "feature-split-event-channels", "%u",
1778 			   &feature_split_evtchn);
1779 	if (err < 0)
1780 		feature_split_evtchn = 0;
1781 
1782 	/* Read mac addr. */
1783 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
1784 	if (err) {
1785 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1786 		goto out;
1787 	}
1788 
1789 	if (info->queues)
1790 		xennet_destroy_queues(info);
1791 
1792 	err = xennet_create_queues(info, num_queues);
1793 	if (err < 0)
1794 		goto destroy_ring;
1795 
1796 	/* Create shared ring, alloc event channel -- for each queue */
1797 	for (i = 0; i < num_queues; ++i) {
1798 		queue = &info->queues[i];
1799 		err = setup_netfront(dev, queue, feature_split_evtchn);
1800 		if (err) {
1801 			/* setup_netfront() will tidy up the current
1802 			 * queue on error, but we need to clean up
1803 			 * those already allocated.
1804 			 */
1805 			if (i > 0) {
1806 				rtnl_lock();
1807 				netif_set_real_num_tx_queues(info->netdev, i);
1808 				rtnl_unlock();
1809 				goto destroy_ring;
1810 			} else {
1811 				goto out;
1812 			}
1813 		}
1814 	}
1815 
1816 again:
1817 	err = xenbus_transaction_start(&xbt);
1818 	if (err) {
1819 		xenbus_dev_fatal(dev, err, "starting transaction");
1820 		goto destroy_ring;
1821 	}
1822 
1823 	if (num_queues == 1) {
1824 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1825 		if (err)
1826 			goto abort_transaction_no_dev_fatal;
1827 	} else {
1828 		/* Write the number of queues */
1829 		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues",
1830 				    "%u", num_queues);
1831 		if (err) {
1832 			message = "writing multi-queue-num-queues";
1833 			goto abort_transaction_no_dev_fatal;
1834 		}
1835 
1836 		/* Write the keys for each queue */
1837 		for (i = 0; i < num_queues; ++i) {
1838 			queue = &info->queues[i];
1839 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1840 			if (err)
1841 				goto abort_transaction_no_dev_fatal;
1842 		}
1843 	}
1844 
1845 	/* The remaining keys are not queue-specific */
1846 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1847 			    1);
1848 	if (err) {
1849 		message = "writing request-rx-copy";
1850 		goto abort_transaction;
1851 	}
1852 
1853 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1854 	if (err) {
1855 		message = "writing feature-rx-notify";
1856 		goto abort_transaction;
1857 	}
1858 
1859 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1860 	if (err) {
1861 		message = "writing feature-sg";
1862 		goto abort_transaction;
1863 	}
1864 
1865 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1866 	if (err) {
1867 		message = "writing feature-gso-tcpv4";
1868 		goto abort_transaction;
1869 	}
1870 
1871 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1872 	if (err) {
1873 		message = "writing feature-gso-tcpv6";
1874 		goto abort_transaction;
1875 	}
1876 
1877 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1878 			   "1");
1879 	if (err) {
1880 		message = "writing feature-ipv6-csum-offload";
1881 		goto abort_transaction;
1882 	}
1883 
1884 	err = xenbus_transaction_end(xbt, 0);
1885 	if (err) {
1886 		if (err == -EAGAIN)
1887 			goto again;
1888 		xenbus_dev_fatal(dev, err, "completing transaction");
1889 		goto destroy_ring;
1890 	}
1891 
1892 	return 0;
1893 
1894  abort_transaction:
1895 	xenbus_dev_fatal(dev, err, "%s", message);
1896 abort_transaction_no_dev_fatal:
1897 	xenbus_transaction_end(xbt, 1);
1898  destroy_ring:
1899 	xennet_disconnect_backend(info);
1900 	kfree(info->queues);
1901 	info->queues = NULL;
1902 	rtnl_lock();
1903 	netif_set_real_num_tx_queues(info->netdev, 0);
1904 	rtnl_unlock();
1905  out:
1906 	return err;
1907 }
1908 
1909 static int xennet_connect(struct net_device *dev)
1910 {
1911 	struct netfront_info *np = netdev_priv(dev);
1912 	unsigned int num_queues = 0;
1913 	int err;
1914 	unsigned int feature_rx_copy;
1915 	unsigned int j = 0;
1916 	struct netfront_queue *queue = NULL;
1917 
1918 	err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1919 			   "feature-rx-copy", "%u", &feature_rx_copy);
1920 	if (err != 1)
1921 		feature_rx_copy = 0;
1922 
1923 	if (!feature_rx_copy) {
1924 		dev_info(&dev->dev,
1925 			 "backend does not support copying receive path\n");
1926 		return -ENODEV;
1927 	}
1928 
1929 	err = talk_to_netback(np->xbdev, np);
1930 	if (err)
1931 		return err;
1932 
1933 	/* talk_to_netback() sets the correct number of queues */
1934 	num_queues = dev->real_num_tx_queues;
1935 
1936 	rtnl_lock();
1937 	netdev_update_features(dev);
1938 	rtnl_unlock();
1939 
1940 	/*
1941 	 * All public and private state should now be sane.  Get
1942 	 * ready to start sending and receiving packets and give the driver
1943 	 * domain a kick because we've probably just requeued some
1944 	 * packets.
1945 	 */
1946 	netif_carrier_on(np->netdev);
1947 	for (j = 0; j < num_queues; ++j) {
1948 		queue = &np->queues[j];
1949 
1950 		notify_remote_via_irq(queue->tx_irq);
1951 		if (queue->tx_irq != queue->rx_irq)
1952 			notify_remote_via_irq(queue->rx_irq);
1953 
1954 		spin_lock_irq(&queue->tx_lock);
1955 		xennet_tx_buf_gc(queue);
1956 		spin_unlock_irq(&queue->tx_lock);
1957 
1958 		spin_lock_bh(&queue->rx_lock);
1959 		xennet_alloc_rx_buffers(queue);
1960 		spin_unlock_bh(&queue->rx_lock);
1961 	}
1962 
1963 	return 0;
1964 }
1965 
1966 /**
1967  * Callback received when the backend's state changes.
1968  */
1969 static void netback_changed(struct xenbus_device *dev,
1970 			    enum xenbus_state backend_state)
1971 {
1972 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
1973 	struct net_device *netdev = np->netdev;
1974 
1975 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1976 
1977 	switch (backend_state) {
1978 	case XenbusStateInitialising:
1979 	case XenbusStateInitialised:
1980 	case XenbusStateReconfiguring:
1981 	case XenbusStateReconfigured:
1982 	case XenbusStateUnknown:
1983 		break;
1984 
1985 	case XenbusStateInitWait:
1986 		if (dev->state != XenbusStateInitialising)
1987 			break;
1988 		if (xennet_connect(netdev) != 0)
1989 			break;
1990 		xenbus_switch_state(dev, XenbusStateConnected);
1991 		break;
1992 
1993 	case XenbusStateConnected:
1994 		netdev_notify_peers(netdev);
1995 		break;
1996 
1997 	case XenbusStateClosed:
1998 		if (dev->state == XenbusStateClosed)
1999 			break;
2000 		/* Missed the backend's CLOSING state -- fallthrough */
2001 	case XenbusStateClosing:
2002 		xenbus_frontend_closed(dev);
2003 		break;
2004 	}
2005 }
2006 
2007 static const struct xennet_stat {
2008 	char name[ETH_GSTRING_LEN];
2009 	u16 offset;
2010 } xennet_stats[] = {
2011 	{
2012 		"rx_gso_checksum_fixup",
2013 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2014 	},
2015 };
2016 
2017 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2018 {
2019 	switch (string_set) {
2020 	case ETH_SS_STATS:
2021 		return ARRAY_SIZE(xennet_stats);
2022 	default:
2023 		return -EINVAL;
2024 	}
2025 }
2026 
2027 static void xennet_get_ethtool_stats(struct net_device *dev,
2028 				     struct ethtool_stats *stats, u64 * data)
2029 {
2030 	void *np = netdev_priv(dev);
2031 	int i;
2032 
2033 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2034 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2035 }
2036 
2037 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2038 {
2039 	int i;
2040 
2041 	switch (stringset) {
2042 	case ETH_SS_STATS:
2043 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2044 			memcpy(data + i * ETH_GSTRING_LEN,
2045 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2046 		break;
2047 	}
2048 }
2049 
2050 static const struct ethtool_ops xennet_ethtool_ops =
2051 {
2052 	.get_link = ethtool_op_get_link,
2053 
2054 	.get_sset_count = xennet_get_sset_count,
2055 	.get_ethtool_stats = xennet_get_ethtool_stats,
2056 	.get_strings = xennet_get_strings,
2057 };
2058 
2059 #ifdef CONFIG_SYSFS
2060 static ssize_t show_rxbuf(struct device *dev,
2061 			  struct device_attribute *attr, char *buf)
2062 {
2063 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2064 }
2065 
2066 static ssize_t store_rxbuf(struct device *dev,
2067 			   struct device_attribute *attr,
2068 			   const char *buf, size_t len)
2069 {
2070 	char *endp;
2071 	unsigned long target;
2072 
2073 	if (!capable(CAP_NET_ADMIN))
2074 		return -EPERM;
2075 
2076 	target = simple_strtoul(buf, &endp, 0);
2077 	if (endp == buf)
2078 		return -EBADMSG;
2079 
2080 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2081 
2082 	return len;
2083 }
2084 
2085 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2086 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2087 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2088 
2089 static struct attribute *xennet_dev_attrs[] = {
2090 	&dev_attr_rxbuf_min.attr,
2091 	&dev_attr_rxbuf_max.attr,
2092 	&dev_attr_rxbuf_cur.attr,
2093 	NULL
2094 };
2095 
2096 static const struct attribute_group xennet_dev_group = {
2097 	.attrs = xennet_dev_attrs
2098 };
2099 #endif /* CONFIG_SYSFS */
2100 
2101 static int xennet_remove(struct xenbus_device *dev)
2102 {
2103 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2104 	unsigned int num_queues = info->netdev->real_num_tx_queues;
2105 	struct netfront_queue *queue = NULL;
2106 	unsigned int i = 0;
2107 
2108 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2109 
2110 	xennet_disconnect_backend(info);
2111 
2112 	unregister_netdev(info->netdev);
2113 
2114 	for (i = 0; i < num_queues; ++i) {
2115 		queue = &info->queues[i];
2116 		del_timer_sync(&queue->rx_refill_timer);
2117 	}
2118 
2119 	if (num_queues) {
2120 		kfree(info->queues);
2121 		info->queues = NULL;
2122 	}
2123 
2124 	xennet_free_netdev(info->netdev);
2125 
2126 	return 0;
2127 }
2128 
2129 static const struct xenbus_device_id netfront_ids[] = {
2130 	{ "vif" },
2131 	{ "" }
2132 };
2133 
2134 static struct xenbus_driver netfront_driver = {
2135 	.ids = netfront_ids,
2136 	.probe = netfront_probe,
2137 	.remove = xennet_remove,
2138 	.resume = netfront_resume,
2139 	.otherend_changed = netback_changed,
2140 };
2141 
2142 static int __init netif_init(void)
2143 {
2144 	if (!xen_domain())
2145 		return -ENODEV;
2146 
2147 	if (!xen_has_pv_nic_devices())
2148 		return -ENODEV;
2149 
2150 	pr_info("Initialising Xen virtual ethernet driver\n");
2151 
2152 	/* Allow as many queues as there are CPUs, by default */
2153 	xennet_max_queues = num_online_cpus();
2154 
2155 	return xenbus_register_frontend(&netfront_driver);
2156 }
2157 module_init(netif_init);
2158 
2159 
2160 static void __exit netif_exit(void)
2161 {
2162 	xenbus_unregister_driver(&netfront_driver);
2163 }
2164 module_exit(netif_exit);
2165 
2166 MODULE_DESCRIPTION("Xen virtual network device frontend");
2167 MODULE_LICENSE("GPL");
2168 MODULE_ALIAS("xen:vif");
2169 MODULE_ALIAS("xennet");
2170