xref: /linux/drivers/net/xen-netfront.c (revision 49b3cd306e60b9d889c775cb2ebb709f80dd8ae9)
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 <xen/xen.h>
49 #include <xen/xenbus.h>
50 #include <xen/events.h>
51 #include <xen/page.h>
52 #include <xen/platform_pci.h>
53 #include <xen/grant_table.h>
54 
55 #include <xen/interface/io/netif.h>
56 #include <xen/interface/memory.h>
57 #include <xen/interface/grant_table.h>
58 
59 /* Module parameters */
60 static unsigned int xennet_max_queues;
61 module_param_named(max_queues, xennet_max_queues, uint, 0644);
62 MODULE_PARM_DESC(max_queues,
63 		 "Maximum number of queues per virtual interface");
64 
65 static const struct ethtool_ops xennet_ethtool_ops;
66 
67 struct netfront_cb {
68 	int pull_to;
69 };
70 
71 #define NETFRONT_SKB_CB(skb)	((struct netfront_cb *)((skb)->cb))
72 
73 #define RX_COPY_THRESHOLD 256
74 
75 #define GRANT_INVALID_REF	0
76 
77 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
78 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
79 
80 /* Minimum number of Rx slots (includes slot for GSO metadata). */
81 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
82 
83 /* Queue name is interface name with "-qNNN" appended */
84 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
85 
86 /* IRQ name is queue name with "-tx" or "-rx" appended */
87 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
88 
89 struct netfront_stats {
90 	u64			packets;
91 	u64			bytes;
92 	struct u64_stats_sync	syncp;
93 };
94 
95 struct netfront_info;
96 
97 struct netfront_queue {
98 	unsigned int id; /* Queue ID, 0-based */
99 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
100 	struct netfront_info *info;
101 
102 	struct napi_struct napi;
103 
104 	/* Split event channels support, tx_* == rx_* when using
105 	 * single event channel.
106 	 */
107 	unsigned int tx_evtchn, rx_evtchn;
108 	unsigned int tx_irq, rx_irq;
109 	/* Only used when split event channels support is enabled */
110 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
111 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
112 
113 	spinlock_t   tx_lock;
114 	struct xen_netif_tx_front_ring tx;
115 	int tx_ring_ref;
116 
117 	/*
118 	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
119 	 * are linked from tx_skb_freelist through skb_entry.link.
120 	 *
121 	 *  NB. Freelist index entries are always going to be less than
122 	 *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
123 	 *  greater than PAGE_OFFSET: we use this property to distinguish
124 	 *  them.
125 	 */
126 	union skb_entry {
127 		struct sk_buff *skb;
128 		unsigned long link;
129 	} tx_skbs[NET_TX_RING_SIZE];
130 	grant_ref_t gref_tx_head;
131 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
132 	struct page *grant_tx_page[NET_TX_RING_SIZE];
133 	unsigned tx_skb_freelist;
134 
135 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
136 	struct xen_netif_rx_front_ring rx;
137 	int rx_ring_ref;
138 
139 	struct timer_list rx_refill_timer;
140 
141 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
142 	grant_ref_t gref_rx_head;
143 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
144 };
145 
146 struct netfront_info {
147 	struct list_head list;
148 	struct net_device *netdev;
149 
150 	struct xenbus_device *xbdev;
151 
152 	/* Multi-queue support */
153 	struct netfront_queue *queues;
154 
155 	/* Statistics */
156 	struct netfront_stats __percpu *rx_stats;
157 	struct netfront_stats __percpu *tx_stats;
158 
159 	atomic_t rx_gso_checksum_fixup;
160 };
161 
162 struct netfront_rx_info {
163 	struct xen_netif_rx_response rx;
164 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
165 };
166 
167 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
168 {
169 	list->link = id;
170 }
171 
172 static int skb_entry_is_link(const union skb_entry *list)
173 {
174 	BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
175 	return (unsigned long)list->skb < PAGE_OFFSET;
176 }
177 
178 /*
179  * Access macros for acquiring freeing slots in tx_skbs[].
180  */
181 
182 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
183 			       unsigned short id)
184 {
185 	skb_entry_set_link(&list[id], *head);
186 	*head = id;
187 }
188 
189 static unsigned short get_id_from_freelist(unsigned *head,
190 					   union skb_entry *list)
191 {
192 	unsigned int id = *head;
193 	*head = list[id].link;
194 	return id;
195 }
196 
197 static int xennet_rxidx(RING_IDX idx)
198 {
199 	return idx & (NET_RX_RING_SIZE - 1);
200 }
201 
202 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
203 					 RING_IDX ri)
204 {
205 	int i = xennet_rxidx(ri);
206 	struct sk_buff *skb = queue->rx_skbs[i];
207 	queue->rx_skbs[i] = NULL;
208 	return skb;
209 }
210 
211 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
212 					    RING_IDX ri)
213 {
214 	int i = xennet_rxidx(ri);
215 	grant_ref_t ref = queue->grant_rx_ref[i];
216 	queue->grant_rx_ref[i] = GRANT_INVALID_REF;
217 	return ref;
218 }
219 
220 #ifdef CONFIG_SYSFS
221 static const struct attribute_group xennet_dev_group;
222 #endif
223 
224 static bool xennet_can_sg(struct net_device *dev)
225 {
226 	return dev->features & NETIF_F_SG;
227 }
228 
229 
230 static void rx_refill_timeout(unsigned long data)
231 {
232 	struct netfront_queue *queue = (struct netfront_queue *)data;
233 	napi_schedule(&queue->napi);
234 }
235 
236 static int netfront_tx_slot_available(struct netfront_queue *queue)
237 {
238 	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
239 		(NET_TX_RING_SIZE - MAX_SKB_FRAGS - 2);
240 }
241 
242 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
243 {
244 	struct net_device *dev = queue->info->netdev;
245 	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
246 
247 	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
248 	    netfront_tx_slot_available(queue) &&
249 	    likely(netif_running(dev)))
250 		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
251 }
252 
253 
254 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
255 {
256 	struct sk_buff *skb;
257 	struct page *page;
258 
259 	skb = __netdev_alloc_skb(queue->info->netdev,
260 				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
261 				 GFP_ATOMIC | __GFP_NOWARN);
262 	if (unlikely(!skb))
263 		return NULL;
264 
265 	page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
266 	if (!page) {
267 		kfree_skb(skb);
268 		return NULL;
269 	}
270 	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
271 
272 	/* Align ip header to a 16 bytes boundary */
273 	skb_reserve(skb, NET_IP_ALIGN);
274 	skb->dev = queue->info->netdev;
275 
276 	return skb;
277 }
278 
279 
280 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
281 {
282 	RING_IDX req_prod = queue->rx.req_prod_pvt;
283 	int notify;
284 	int err = 0;
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 		struct page *page;
296 		struct xen_netif_rx_request *req;
297 
298 		skb = xennet_alloc_one_rx_buffer(queue);
299 		if (!skb) {
300 			err = -ENOMEM;
301 			break;
302 		}
303 
304 		id = xennet_rxidx(req_prod);
305 
306 		BUG_ON(queue->rx_skbs[id]);
307 		queue->rx_skbs[id] = skb;
308 
309 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
310 		WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
311 		queue->grant_rx_ref[id] = ref;
312 
313 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
314 
315 		req = RING_GET_REQUEST(&queue->rx, req_prod);
316 		gnttab_page_grant_foreign_access_ref_one(ref,
317 							 queue->info->xbdev->otherend_id,
318 							 page,
319 							 0);
320 		req->id = id;
321 		req->gref = ref;
322 	}
323 
324 	queue->rx.req_prod_pvt = req_prod;
325 
326 	/* Try again later if there are not enough requests or skb allocation
327 	 * failed.
328 	 * Enough requests is quantified as the sum of newly created slots and
329 	 * the unconsumed slots at the backend.
330 	 */
331 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
332 	    unlikely(err)) {
333 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
334 		return;
335 	}
336 
337 	wmb();		/* barrier so backend seens requests */
338 
339 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
340 	if (notify)
341 		notify_remote_via_irq(queue->rx_irq);
342 }
343 
344 static int xennet_open(struct net_device *dev)
345 {
346 	struct netfront_info *np = netdev_priv(dev);
347 	unsigned int num_queues = dev->real_num_tx_queues;
348 	unsigned int i = 0;
349 	struct netfront_queue *queue = NULL;
350 
351 	for (i = 0; i < num_queues; ++i) {
352 		queue = &np->queues[i];
353 		napi_enable(&queue->napi);
354 
355 		spin_lock_bh(&queue->rx_lock);
356 		if (netif_carrier_ok(dev)) {
357 			xennet_alloc_rx_buffers(queue);
358 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
359 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
360 				napi_schedule(&queue->napi);
361 		}
362 		spin_unlock_bh(&queue->rx_lock);
363 	}
364 
365 	netif_tx_start_all_queues(dev);
366 
367 	return 0;
368 }
369 
370 static void xennet_tx_buf_gc(struct netfront_queue *queue)
371 {
372 	RING_IDX cons, prod;
373 	unsigned short id;
374 	struct sk_buff *skb;
375 	bool more_to_do;
376 
377 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
378 
379 	do {
380 		prod = queue->tx.sring->rsp_prod;
381 		rmb(); /* Ensure we see responses up to 'rp'. */
382 
383 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
384 			struct xen_netif_tx_response *txrsp;
385 
386 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
387 			if (txrsp->status == XEN_NETIF_RSP_NULL)
388 				continue;
389 
390 			id  = txrsp->id;
391 			skb = queue->tx_skbs[id].skb;
392 			if (unlikely(gnttab_query_foreign_access(
393 				queue->grant_tx_ref[id]) != 0)) {
394 				pr_alert("%s: warning -- grant still in use by backend domain\n",
395 					 __func__);
396 				BUG();
397 			}
398 			gnttab_end_foreign_access_ref(
399 				queue->grant_tx_ref[id], GNTMAP_readonly);
400 			gnttab_release_grant_reference(
401 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
402 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
403 			queue->grant_tx_page[id] = NULL;
404 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
405 			dev_kfree_skb_irq(skb);
406 		}
407 
408 		queue->tx.rsp_cons = prod;
409 
410 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
411 	} while (more_to_do);
412 
413 	xennet_maybe_wake_tx(queue);
414 }
415 
416 struct xennet_gnttab_make_txreq {
417 	struct netfront_queue *queue;
418 	struct sk_buff *skb;
419 	struct page *page;
420 	struct xen_netif_tx_request *tx; /* Last request */
421 	unsigned int size;
422 };
423 
424 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
425 				  unsigned int len, void *data)
426 {
427 	struct xennet_gnttab_make_txreq *info = data;
428 	unsigned int id;
429 	struct xen_netif_tx_request *tx;
430 	grant_ref_t ref;
431 	/* convenient aliases */
432 	struct page *page = info->page;
433 	struct netfront_queue *queue = info->queue;
434 	struct sk_buff *skb = info->skb;
435 
436 	id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
437 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
438 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
439 	WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
440 
441 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
442 					gfn, GNTMAP_readonly);
443 
444 	queue->tx_skbs[id].skb = skb;
445 	queue->grant_tx_page[id] = page;
446 	queue->grant_tx_ref[id] = ref;
447 
448 	tx->id = id;
449 	tx->gref = ref;
450 	tx->offset = offset;
451 	tx->size = len;
452 	tx->flags = 0;
453 
454 	info->tx = tx;
455 	info->size += tx->size;
456 }
457 
458 static struct xen_netif_tx_request *xennet_make_first_txreq(
459 	struct netfront_queue *queue, struct sk_buff *skb,
460 	struct page *page, unsigned int offset, unsigned int len)
461 {
462 	struct xennet_gnttab_make_txreq info = {
463 		.queue = queue,
464 		.skb = skb,
465 		.page = page,
466 		.size = 0,
467 	};
468 
469 	gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
470 
471 	return info.tx;
472 }
473 
474 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
475 				  unsigned int len, void *data)
476 {
477 	struct xennet_gnttab_make_txreq *info = data;
478 
479 	info->tx->flags |= XEN_NETTXF_more_data;
480 	skb_get(info->skb);
481 	xennet_tx_setup_grant(gfn, offset, len, data);
482 }
483 
484 static struct xen_netif_tx_request *xennet_make_txreqs(
485 	struct netfront_queue *queue, struct xen_netif_tx_request *tx,
486 	struct sk_buff *skb, struct page *page,
487 	unsigned int offset, unsigned int len)
488 {
489 	struct xennet_gnttab_make_txreq info = {
490 		.queue = queue,
491 		.skb = skb,
492 		.tx = tx,
493 	};
494 
495 	/* Skip unused frames from start of page */
496 	page += offset >> PAGE_SHIFT;
497 	offset &= ~PAGE_MASK;
498 
499 	while (len) {
500 		info.page = page;
501 		info.size = 0;
502 
503 		gnttab_foreach_grant_in_range(page, offset, len,
504 					      xennet_make_one_txreq,
505 					      &info);
506 
507 		page++;
508 		offset = 0;
509 		len -= info.size;
510 	}
511 
512 	return info.tx;
513 }
514 
515 /*
516  * Count how many ring slots are required to send this skb. Each frag
517  * might be a compound page.
518  */
519 static int xennet_count_skb_slots(struct sk_buff *skb)
520 {
521 	int i, frags = skb_shinfo(skb)->nr_frags;
522 	int slots;
523 
524 	slots = gnttab_count_grant(offset_in_page(skb->data),
525 				   skb_headlen(skb));
526 
527 	for (i = 0; i < frags; i++) {
528 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
529 		unsigned long size = skb_frag_size(frag);
530 		unsigned long offset = frag->page_offset;
531 
532 		/* Skip unused frames from start of page */
533 		offset &= ~PAGE_MASK;
534 
535 		slots += gnttab_count_grant(offset, size);
536 	}
537 
538 	return slots;
539 }
540 
541 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
542 			       void *accel_priv, select_queue_fallback_t fallback)
543 {
544 	unsigned int num_queues = dev->real_num_tx_queues;
545 	u32 hash;
546 	u16 queue_idx;
547 
548 	/* First, check if there is only one queue */
549 	if (num_queues == 1) {
550 		queue_idx = 0;
551 	} else {
552 		hash = skb_get_hash(skb);
553 		queue_idx = hash % num_queues;
554 	}
555 
556 	return queue_idx;
557 }
558 
559 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
560 
561 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
562 {
563 	struct netfront_info *np = netdev_priv(dev);
564 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
565 	struct xen_netif_tx_request *tx, *first_tx;
566 	unsigned int i;
567 	int notify;
568 	int slots;
569 	struct page *page;
570 	unsigned int offset;
571 	unsigned int len;
572 	unsigned long flags;
573 	struct netfront_queue *queue = NULL;
574 	unsigned int num_queues = dev->real_num_tx_queues;
575 	u16 queue_index;
576 	struct sk_buff *nskb;
577 
578 	/* Drop the packet if no queues are set up */
579 	if (num_queues < 1)
580 		goto drop;
581 	/* Determine which queue to transmit this SKB on */
582 	queue_index = skb_get_queue_mapping(skb);
583 	queue = &np->queues[queue_index];
584 
585 	/* If skb->len is too big for wire format, drop skb and alert
586 	 * user about misconfiguration.
587 	 */
588 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
589 		net_alert_ratelimited(
590 			"xennet: skb->len = %u, too big for wire format\n",
591 			skb->len);
592 		goto drop;
593 	}
594 
595 	slots = xennet_count_skb_slots(skb);
596 	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
597 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
598 				    slots, skb->len);
599 		if (skb_linearize(skb))
600 			goto drop;
601 	}
602 
603 	page = virt_to_page(skb->data);
604 	offset = offset_in_page(skb->data);
605 
606 	/* The first req should be at least ETH_HLEN size or the packet will be
607 	 * dropped by netback.
608 	 */
609 	if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
610 		nskb = skb_copy(skb, GFP_ATOMIC);
611 		if (!nskb)
612 			goto drop;
613 		dev_kfree_skb_any(skb);
614 		skb = nskb;
615 		page = virt_to_page(skb->data);
616 		offset = offset_in_page(skb->data);
617 	}
618 
619 	len = skb_headlen(skb);
620 
621 	spin_lock_irqsave(&queue->tx_lock, flags);
622 
623 	if (unlikely(!netif_carrier_ok(dev) ||
624 		     (slots > 1 && !xennet_can_sg(dev)) ||
625 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
626 		spin_unlock_irqrestore(&queue->tx_lock, flags);
627 		goto drop;
628 	}
629 
630 	/* First request for the linear area. */
631 	first_tx = tx = xennet_make_first_txreq(queue, skb,
632 						page, offset, len);
633 	offset += tx->size;
634 	if (offset == PAGE_SIZE) {
635 		page++;
636 		offset = 0;
637 	}
638 	len -= tx->size;
639 
640 	if (skb->ip_summed == CHECKSUM_PARTIAL)
641 		/* local packet? */
642 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
643 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
644 		/* remote but checksummed. */
645 		tx->flags |= XEN_NETTXF_data_validated;
646 
647 	/* Optional extra info after the first request. */
648 	if (skb_shinfo(skb)->gso_size) {
649 		struct xen_netif_extra_info *gso;
650 
651 		gso = (struct xen_netif_extra_info *)
652 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
653 
654 		tx->flags |= XEN_NETTXF_extra_info;
655 
656 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
657 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
658 			XEN_NETIF_GSO_TYPE_TCPV6 :
659 			XEN_NETIF_GSO_TYPE_TCPV4;
660 		gso->u.gso.pad = 0;
661 		gso->u.gso.features = 0;
662 
663 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
664 		gso->flags = 0;
665 	}
666 
667 	/* Requests for the rest of the linear area. */
668 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
669 
670 	/* Requests for all the frags. */
671 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
672 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
673 		tx = xennet_make_txreqs(queue, tx, skb,
674 					skb_frag_page(frag), frag->page_offset,
675 					skb_frag_size(frag));
676 	}
677 
678 	/* First request has the packet length. */
679 	first_tx->size = skb->len;
680 
681 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
682 	if (notify)
683 		notify_remote_via_irq(queue->tx_irq);
684 
685 	u64_stats_update_begin(&tx_stats->syncp);
686 	tx_stats->bytes += skb->len;
687 	tx_stats->packets++;
688 	u64_stats_update_end(&tx_stats->syncp);
689 
690 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
691 	xennet_tx_buf_gc(queue);
692 
693 	if (!netfront_tx_slot_available(queue))
694 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
695 
696 	spin_unlock_irqrestore(&queue->tx_lock, flags);
697 
698 	return NETDEV_TX_OK;
699 
700  drop:
701 	dev->stats.tx_dropped++;
702 	dev_kfree_skb_any(skb);
703 	return NETDEV_TX_OK;
704 }
705 
706 static int xennet_close(struct net_device *dev)
707 {
708 	struct netfront_info *np = netdev_priv(dev);
709 	unsigned int num_queues = dev->real_num_tx_queues;
710 	unsigned int i;
711 	struct netfront_queue *queue;
712 	netif_tx_stop_all_queues(np->netdev);
713 	for (i = 0; i < num_queues; ++i) {
714 		queue = &np->queues[i];
715 		napi_disable(&queue->napi);
716 	}
717 	return 0;
718 }
719 
720 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
721 				grant_ref_t ref)
722 {
723 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
724 
725 	BUG_ON(queue->rx_skbs[new]);
726 	queue->rx_skbs[new] = skb;
727 	queue->grant_rx_ref[new] = ref;
728 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
729 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
730 	queue->rx.req_prod_pvt++;
731 }
732 
733 static int xennet_get_extras(struct netfront_queue *queue,
734 			     struct xen_netif_extra_info *extras,
735 			     RING_IDX rp)
736 
737 {
738 	struct xen_netif_extra_info *extra;
739 	struct device *dev = &queue->info->netdev->dev;
740 	RING_IDX cons = queue->rx.rsp_cons;
741 	int err = 0;
742 
743 	do {
744 		struct sk_buff *skb;
745 		grant_ref_t ref;
746 
747 		if (unlikely(cons + 1 == rp)) {
748 			if (net_ratelimit())
749 				dev_warn(dev, "Missing extra info\n");
750 			err = -EBADR;
751 			break;
752 		}
753 
754 		extra = (struct xen_netif_extra_info *)
755 			RING_GET_RESPONSE(&queue->rx, ++cons);
756 
757 		if (unlikely(!extra->type ||
758 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
759 			if (net_ratelimit())
760 				dev_warn(dev, "Invalid extra type: %d\n",
761 					extra->type);
762 			err = -EINVAL;
763 		} else {
764 			memcpy(&extras[extra->type - 1], extra,
765 			       sizeof(*extra));
766 		}
767 
768 		skb = xennet_get_rx_skb(queue, cons);
769 		ref = xennet_get_rx_ref(queue, cons);
770 		xennet_move_rx_slot(queue, skb, ref);
771 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
772 
773 	queue->rx.rsp_cons = cons;
774 	return err;
775 }
776 
777 static int xennet_get_responses(struct netfront_queue *queue,
778 				struct netfront_rx_info *rinfo, RING_IDX rp,
779 				struct sk_buff_head *list)
780 {
781 	struct xen_netif_rx_response *rx = &rinfo->rx;
782 	struct xen_netif_extra_info *extras = rinfo->extras;
783 	struct device *dev = &queue->info->netdev->dev;
784 	RING_IDX cons = queue->rx.rsp_cons;
785 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
786 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
787 	int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
788 	int slots = 1;
789 	int err = 0;
790 	unsigned long ret;
791 
792 	if (rx->flags & XEN_NETRXF_extra_info) {
793 		err = xennet_get_extras(queue, extras, rp);
794 		cons = queue->rx.rsp_cons;
795 	}
796 
797 	for (;;) {
798 		if (unlikely(rx->status < 0 ||
799 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
800 			if (net_ratelimit())
801 				dev_warn(dev, "rx->offset: %u, size: %d\n",
802 					 rx->offset, rx->status);
803 			xennet_move_rx_slot(queue, skb, ref);
804 			err = -EINVAL;
805 			goto next;
806 		}
807 
808 		/*
809 		 * This definitely indicates a bug, either in this driver or in
810 		 * the backend driver. In future this should flag the bad
811 		 * situation to the system controller to reboot the backend.
812 		 */
813 		if (ref == GRANT_INVALID_REF) {
814 			if (net_ratelimit())
815 				dev_warn(dev, "Bad rx response id %d.\n",
816 					 rx->id);
817 			err = -EINVAL;
818 			goto next;
819 		}
820 
821 		ret = gnttab_end_foreign_access_ref(ref, 0);
822 		BUG_ON(!ret);
823 
824 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
825 
826 		__skb_queue_tail(list, skb);
827 
828 next:
829 		if (!(rx->flags & XEN_NETRXF_more_data))
830 			break;
831 
832 		if (cons + slots == rp) {
833 			if (net_ratelimit())
834 				dev_warn(dev, "Need more slots\n");
835 			err = -ENOENT;
836 			break;
837 		}
838 
839 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
840 		skb = xennet_get_rx_skb(queue, cons + slots);
841 		ref = xennet_get_rx_ref(queue, cons + slots);
842 		slots++;
843 	}
844 
845 	if (unlikely(slots > max)) {
846 		if (net_ratelimit())
847 			dev_warn(dev, "Too many slots\n");
848 		err = -E2BIG;
849 	}
850 
851 	if (unlikely(err))
852 		queue->rx.rsp_cons = cons + slots;
853 
854 	return err;
855 }
856 
857 static int xennet_set_skb_gso(struct sk_buff *skb,
858 			      struct xen_netif_extra_info *gso)
859 {
860 	if (!gso->u.gso.size) {
861 		if (net_ratelimit())
862 			pr_warn("GSO size must not be zero\n");
863 		return -EINVAL;
864 	}
865 
866 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
867 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
868 		if (net_ratelimit())
869 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
870 		return -EINVAL;
871 	}
872 
873 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
874 	skb_shinfo(skb)->gso_type =
875 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
876 		SKB_GSO_TCPV4 :
877 		SKB_GSO_TCPV6;
878 
879 	/* Header must be checked, and gso_segs computed. */
880 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
881 	skb_shinfo(skb)->gso_segs = 0;
882 
883 	return 0;
884 }
885 
886 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
887 				  struct sk_buff *skb,
888 				  struct sk_buff_head *list)
889 {
890 	struct skb_shared_info *shinfo = skb_shinfo(skb);
891 	RING_IDX cons = queue->rx.rsp_cons;
892 	struct sk_buff *nskb;
893 
894 	while ((nskb = __skb_dequeue(list))) {
895 		struct xen_netif_rx_response *rx =
896 			RING_GET_RESPONSE(&queue->rx, ++cons);
897 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
898 
899 		if (shinfo->nr_frags == MAX_SKB_FRAGS) {
900 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
901 
902 			BUG_ON(pull_to <= skb_headlen(skb));
903 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
904 		}
905 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
906 
907 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
908 				rx->offset, rx->status, PAGE_SIZE);
909 
910 		skb_shinfo(nskb)->nr_frags = 0;
911 		kfree_skb(nskb);
912 	}
913 
914 	return cons;
915 }
916 
917 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
918 {
919 	bool recalculate_partial_csum = false;
920 
921 	/*
922 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
923 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
924 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
925 	 * recalculate the partial checksum.
926 	 */
927 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
928 		struct netfront_info *np = netdev_priv(dev);
929 		atomic_inc(&np->rx_gso_checksum_fixup);
930 		skb->ip_summed = CHECKSUM_PARTIAL;
931 		recalculate_partial_csum = true;
932 	}
933 
934 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
935 	if (skb->ip_summed != CHECKSUM_PARTIAL)
936 		return 0;
937 
938 	return skb_checksum_setup(skb, recalculate_partial_csum);
939 }
940 
941 static int handle_incoming_queue(struct netfront_queue *queue,
942 				 struct sk_buff_head *rxq)
943 {
944 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
945 	int packets_dropped = 0;
946 	struct sk_buff *skb;
947 
948 	while ((skb = __skb_dequeue(rxq)) != NULL) {
949 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
950 
951 		if (pull_to > skb_headlen(skb))
952 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
953 
954 		/* Ethernet work: Delayed to here as it peeks the header. */
955 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
956 		skb_reset_network_header(skb);
957 
958 		if (checksum_setup(queue->info->netdev, skb)) {
959 			kfree_skb(skb);
960 			packets_dropped++;
961 			queue->info->netdev->stats.rx_errors++;
962 			continue;
963 		}
964 
965 		u64_stats_update_begin(&rx_stats->syncp);
966 		rx_stats->packets++;
967 		rx_stats->bytes += skb->len;
968 		u64_stats_update_end(&rx_stats->syncp);
969 
970 		/* Pass it up. */
971 		napi_gro_receive(&queue->napi, skb);
972 	}
973 
974 	return packets_dropped;
975 }
976 
977 static int xennet_poll(struct napi_struct *napi, int budget)
978 {
979 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
980 	struct net_device *dev = queue->info->netdev;
981 	struct sk_buff *skb;
982 	struct netfront_rx_info rinfo;
983 	struct xen_netif_rx_response *rx = &rinfo.rx;
984 	struct xen_netif_extra_info *extras = rinfo.extras;
985 	RING_IDX i, rp;
986 	int work_done;
987 	struct sk_buff_head rxq;
988 	struct sk_buff_head errq;
989 	struct sk_buff_head tmpq;
990 	int err;
991 
992 	spin_lock(&queue->rx_lock);
993 
994 	skb_queue_head_init(&rxq);
995 	skb_queue_head_init(&errq);
996 	skb_queue_head_init(&tmpq);
997 
998 	rp = queue->rx.sring->rsp_prod;
999 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1000 
1001 	i = queue->rx.rsp_cons;
1002 	work_done = 0;
1003 	while ((i != rp) && (work_done < budget)) {
1004 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1005 		memset(extras, 0, sizeof(rinfo.extras));
1006 
1007 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1008 
1009 		if (unlikely(err)) {
1010 err:
1011 			while ((skb = __skb_dequeue(&tmpq)))
1012 				__skb_queue_tail(&errq, skb);
1013 			dev->stats.rx_errors++;
1014 			i = queue->rx.rsp_cons;
1015 			continue;
1016 		}
1017 
1018 		skb = __skb_dequeue(&tmpq);
1019 
1020 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1021 			struct xen_netif_extra_info *gso;
1022 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1023 
1024 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1025 				__skb_queue_head(&tmpq, skb);
1026 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1027 				goto err;
1028 			}
1029 		}
1030 
1031 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1032 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1033 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1034 
1035 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1036 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1037 		skb->data_len = rx->status;
1038 		skb->len += rx->status;
1039 
1040 		i = xennet_fill_frags(queue, skb, &tmpq);
1041 
1042 		if (rx->flags & XEN_NETRXF_csum_blank)
1043 			skb->ip_summed = CHECKSUM_PARTIAL;
1044 		else if (rx->flags & XEN_NETRXF_data_validated)
1045 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1046 
1047 		__skb_queue_tail(&rxq, skb);
1048 
1049 		queue->rx.rsp_cons = ++i;
1050 		work_done++;
1051 	}
1052 
1053 	__skb_queue_purge(&errq);
1054 
1055 	work_done -= handle_incoming_queue(queue, &rxq);
1056 
1057 	xennet_alloc_rx_buffers(queue);
1058 
1059 	if (work_done < budget) {
1060 		int more_to_do = 0;
1061 
1062 		napi_complete(napi);
1063 
1064 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1065 		if (more_to_do)
1066 			napi_schedule(napi);
1067 	}
1068 
1069 	spin_unlock(&queue->rx_lock);
1070 
1071 	return work_done;
1072 }
1073 
1074 static int xennet_change_mtu(struct net_device *dev, int mtu)
1075 {
1076 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1077 
1078 	if (mtu > max)
1079 		return -EINVAL;
1080 	dev->mtu = mtu;
1081 	return 0;
1082 }
1083 
1084 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1085 						    struct rtnl_link_stats64 *tot)
1086 {
1087 	struct netfront_info *np = netdev_priv(dev);
1088 	int cpu;
1089 
1090 	for_each_possible_cpu(cpu) {
1091 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1092 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1093 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1094 		unsigned int start;
1095 
1096 		do {
1097 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1098 			tx_packets = tx_stats->packets;
1099 			tx_bytes = tx_stats->bytes;
1100 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1101 
1102 		do {
1103 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1104 			rx_packets = rx_stats->packets;
1105 			rx_bytes = rx_stats->bytes;
1106 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1107 
1108 		tot->rx_packets += rx_packets;
1109 		tot->tx_packets += tx_packets;
1110 		tot->rx_bytes   += rx_bytes;
1111 		tot->tx_bytes   += tx_bytes;
1112 	}
1113 
1114 	tot->rx_errors  = dev->stats.rx_errors;
1115 	tot->tx_dropped = dev->stats.tx_dropped;
1116 
1117 	return tot;
1118 }
1119 
1120 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1121 {
1122 	struct sk_buff *skb;
1123 	int i;
1124 
1125 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1126 		/* Skip over entries which are actually freelist references */
1127 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1128 			continue;
1129 
1130 		skb = queue->tx_skbs[i].skb;
1131 		get_page(queue->grant_tx_page[i]);
1132 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1133 					  GNTMAP_readonly,
1134 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1135 		queue->grant_tx_page[i] = NULL;
1136 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1137 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1138 		dev_kfree_skb_irq(skb);
1139 	}
1140 }
1141 
1142 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1143 {
1144 	int id, ref;
1145 
1146 	spin_lock_bh(&queue->rx_lock);
1147 
1148 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1149 		struct sk_buff *skb;
1150 		struct page *page;
1151 
1152 		skb = queue->rx_skbs[id];
1153 		if (!skb)
1154 			continue;
1155 
1156 		ref = queue->grant_rx_ref[id];
1157 		if (ref == GRANT_INVALID_REF)
1158 			continue;
1159 
1160 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1161 
1162 		/* gnttab_end_foreign_access() needs a page ref until
1163 		 * foreign access is ended (which may be deferred).
1164 		 */
1165 		get_page(page);
1166 		gnttab_end_foreign_access(ref, 0,
1167 					  (unsigned long)page_address(page));
1168 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1169 
1170 		kfree_skb(skb);
1171 	}
1172 
1173 	spin_unlock_bh(&queue->rx_lock);
1174 }
1175 
1176 static netdev_features_t xennet_fix_features(struct net_device *dev,
1177 	netdev_features_t features)
1178 {
1179 	struct netfront_info *np = netdev_priv(dev);
1180 
1181 	if (features & NETIF_F_SG &&
1182 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1183 		features &= ~NETIF_F_SG;
1184 
1185 	if (features & NETIF_F_IPV6_CSUM &&
1186 	    !xenbus_read_unsigned(np->xbdev->otherend,
1187 				  "feature-ipv6-csum-offload", 0))
1188 		features &= ~NETIF_F_IPV6_CSUM;
1189 
1190 	if (features & NETIF_F_TSO &&
1191 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1192 		features &= ~NETIF_F_TSO;
1193 
1194 	if (features & NETIF_F_TSO6 &&
1195 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1196 		features &= ~NETIF_F_TSO6;
1197 
1198 	return features;
1199 }
1200 
1201 static int xennet_set_features(struct net_device *dev,
1202 	netdev_features_t features)
1203 {
1204 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1205 		netdev_info(dev, "Reducing MTU because no SG offload");
1206 		dev->mtu = ETH_DATA_LEN;
1207 	}
1208 
1209 	return 0;
1210 }
1211 
1212 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1213 {
1214 	struct netfront_queue *queue = dev_id;
1215 	unsigned long flags;
1216 
1217 	spin_lock_irqsave(&queue->tx_lock, flags);
1218 	xennet_tx_buf_gc(queue);
1219 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1220 
1221 	return IRQ_HANDLED;
1222 }
1223 
1224 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1225 {
1226 	struct netfront_queue *queue = dev_id;
1227 	struct net_device *dev = queue->info->netdev;
1228 
1229 	if (likely(netif_carrier_ok(dev) &&
1230 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1231 		napi_schedule(&queue->napi);
1232 
1233 	return IRQ_HANDLED;
1234 }
1235 
1236 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1237 {
1238 	xennet_tx_interrupt(irq, dev_id);
1239 	xennet_rx_interrupt(irq, dev_id);
1240 	return IRQ_HANDLED;
1241 }
1242 
1243 #ifdef CONFIG_NET_POLL_CONTROLLER
1244 static void xennet_poll_controller(struct net_device *dev)
1245 {
1246 	/* Poll each queue */
1247 	struct netfront_info *info = netdev_priv(dev);
1248 	unsigned int num_queues = dev->real_num_tx_queues;
1249 	unsigned int i;
1250 	for (i = 0; i < num_queues; ++i)
1251 		xennet_interrupt(0, &info->queues[i]);
1252 }
1253 #endif
1254 
1255 static const struct net_device_ops xennet_netdev_ops = {
1256 	.ndo_open            = xennet_open,
1257 	.ndo_stop            = xennet_close,
1258 	.ndo_start_xmit      = xennet_start_xmit,
1259 	.ndo_change_mtu	     = xennet_change_mtu,
1260 	.ndo_get_stats64     = xennet_get_stats64,
1261 	.ndo_set_mac_address = eth_mac_addr,
1262 	.ndo_validate_addr   = eth_validate_addr,
1263 	.ndo_fix_features    = xennet_fix_features,
1264 	.ndo_set_features    = xennet_set_features,
1265 	.ndo_select_queue    = xennet_select_queue,
1266 #ifdef CONFIG_NET_POLL_CONTROLLER
1267 	.ndo_poll_controller = xennet_poll_controller,
1268 #endif
1269 };
1270 
1271 static void xennet_free_netdev(struct net_device *netdev)
1272 {
1273 	struct netfront_info *np = netdev_priv(netdev);
1274 
1275 	free_percpu(np->rx_stats);
1276 	free_percpu(np->tx_stats);
1277 	free_netdev(netdev);
1278 }
1279 
1280 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1281 {
1282 	int err;
1283 	struct net_device *netdev;
1284 	struct netfront_info *np;
1285 
1286 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1287 	if (!netdev)
1288 		return ERR_PTR(-ENOMEM);
1289 
1290 	np                   = netdev_priv(netdev);
1291 	np->xbdev            = dev;
1292 
1293 	np->queues = NULL;
1294 
1295 	err = -ENOMEM;
1296 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1297 	if (np->rx_stats == NULL)
1298 		goto exit;
1299 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1300 	if (np->tx_stats == NULL)
1301 		goto exit;
1302 
1303 	netdev->netdev_ops	= &xennet_netdev_ops;
1304 
1305 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1306 				  NETIF_F_GSO_ROBUST;
1307 	netdev->hw_features	= NETIF_F_SG |
1308 				  NETIF_F_IPV6_CSUM |
1309 				  NETIF_F_TSO | NETIF_F_TSO6;
1310 
1311 	/*
1312          * Assume that all hw features are available for now. This set
1313          * will be adjusted by the call to netdev_update_features() in
1314          * xennet_connect() which is the earliest point where we can
1315          * negotiate with the backend regarding supported features.
1316          */
1317 	netdev->features |= netdev->hw_features;
1318 
1319 	netdev->ethtool_ops = &xennet_ethtool_ops;
1320 	netdev->min_mtu = 0;
1321 	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1322 	SET_NETDEV_DEV(netdev, &dev->dev);
1323 
1324 	np->netdev = netdev;
1325 
1326 	netif_carrier_off(netdev);
1327 
1328 	return netdev;
1329 
1330  exit:
1331 	xennet_free_netdev(netdev);
1332 	return ERR_PTR(err);
1333 }
1334 
1335 /**
1336  * Entry point to this code when a new device is created.  Allocate the basic
1337  * structures and the ring buffers for communication with the backend, and
1338  * inform the backend of the appropriate details for those.
1339  */
1340 static int netfront_probe(struct xenbus_device *dev,
1341 			  const struct xenbus_device_id *id)
1342 {
1343 	int err;
1344 	struct net_device *netdev;
1345 	struct netfront_info *info;
1346 
1347 	netdev = xennet_create_dev(dev);
1348 	if (IS_ERR(netdev)) {
1349 		err = PTR_ERR(netdev);
1350 		xenbus_dev_fatal(dev, err, "creating netdev");
1351 		return err;
1352 	}
1353 
1354 	info = netdev_priv(netdev);
1355 	dev_set_drvdata(&dev->dev, info);
1356 #ifdef CONFIG_SYSFS
1357 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1358 #endif
1359 	err = register_netdev(info->netdev);
1360 	if (err) {
1361 		pr_warn("%s: register_netdev err=%d\n", __func__, err);
1362 		goto fail;
1363 	}
1364 
1365 	return 0;
1366 
1367  fail:
1368 	xennet_free_netdev(netdev);
1369 	dev_set_drvdata(&dev->dev, NULL);
1370 	return err;
1371 }
1372 
1373 static void xennet_end_access(int ref, void *page)
1374 {
1375 	/* This frees the page as a side-effect */
1376 	if (ref != GRANT_INVALID_REF)
1377 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1378 }
1379 
1380 static void xennet_disconnect_backend(struct netfront_info *info)
1381 {
1382 	unsigned int i = 0;
1383 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1384 
1385 	netif_carrier_off(info->netdev);
1386 
1387 	for (i = 0; i < num_queues && info->queues; ++i) {
1388 		struct netfront_queue *queue = &info->queues[i];
1389 
1390 		del_timer_sync(&queue->rx_refill_timer);
1391 
1392 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1393 			unbind_from_irqhandler(queue->tx_irq, queue);
1394 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1395 			unbind_from_irqhandler(queue->tx_irq, queue);
1396 			unbind_from_irqhandler(queue->rx_irq, queue);
1397 		}
1398 		queue->tx_evtchn = queue->rx_evtchn = 0;
1399 		queue->tx_irq = queue->rx_irq = 0;
1400 
1401 		if (netif_running(info->netdev))
1402 			napi_synchronize(&queue->napi);
1403 
1404 		xennet_release_tx_bufs(queue);
1405 		xennet_release_rx_bufs(queue);
1406 		gnttab_free_grant_references(queue->gref_tx_head);
1407 		gnttab_free_grant_references(queue->gref_rx_head);
1408 
1409 		/* End access and free the pages */
1410 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1411 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1412 
1413 		queue->tx_ring_ref = GRANT_INVALID_REF;
1414 		queue->rx_ring_ref = GRANT_INVALID_REF;
1415 		queue->tx.sring = NULL;
1416 		queue->rx.sring = NULL;
1417 	}
1418 }
1419 
1420 /**
1421  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1422  * driver restart.  We tear down our netif structure and recreate it, but
1423  * leave the device-layer structures intact so that this is transparent to the
1424  * rest of the kernel.
1425  */
1426 static int netfront_resume(struct xenbus_device *dev)
1427 {
1428 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1429 
1430 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1431 
1432 	xennet_disconnect_backend(info);
1433 	return 0;
1434 }
1435 
1436 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1437 {
1438 	char *s, *e, *macstr;
1439 	int i;
1440 
1441 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1442 	if (IS_ERR(macstr))
1443 		return PTR_ERR(macstr);
1444 
1445 	for (i = 0; i < ETH_ALEN; i++) {
1446 		mac[i] = simple_strtoul(s, &e, 16);
1447 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1448 			kfree(macstr);
1449 			return -ENOENT;
1450 		}
1451 		s = e+1;
1452 	}
1453 
1454 	kfree(macstr);
1455 	return 0;
1456 }
1457 
1458 static int setup_netfront_single(struct netfront_queue *queue)
1459 {
1460 	int err;
1461 
1462 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1463 	if (err < 0)
1464 		goto fail;
1465 
1466 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1467 					xennet_interrupt,
1468 					0, queue->info->netdev->name, queue);
1469 	if (err < 0)
1470 		goto bind_fail;
1471 	queue->rx_evtchn = queue->tx_evtchn;
1472 	queue->rx_irq = queue->tx_irq = err;
1473 
1474 	return 0;
1475 
1476 bind_fail:
1477 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1478 	queue->tx_evtchn = 0;
1479 fail:
1480 	return err;
1481 }
1482 
1483 static int setup_netfront_split(struct netfront_queue *queue)
1484 {
1485 	int err;
1486 
1487 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1488 	if (err < 0)
1489 		goto fail;
1490 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1491 	if (err < 0)
1492 		goto alloc_rx_evtchn_fail;
1493 
1494 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1495 		 "%s-tx", queue->name);
1496 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1497 					xennet_tx_interrupt,
1498 					0, queue->tx_irq_name, queue);
1499 	if (err < 0)
1500 		goto bind_tx_fail;
1501 	queue->tx_irq = err;
1502 
1503 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1504 		 "%s-rx", queue->name);
1505 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1506 					xennet_rx_interrupt,
1507 					0, queue->rx_irq_name, queue);
1508 	if (err < 0)
1509 		goto bind_rx_fail;
1510 	queue->rx_irq = err;
1511 
1512 	return 0;
1513 
1514 bind_rx_fail:
1515 	unbind_from_irqhandler(queue->tx_irq, queue);
1516 	queue->tx_irq = 0;
1517 bind_tx_fail:
1518 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1519 	queue->rx_evtchn = 0;
1520 alloc_rx_evtchn_fail:
1521 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1522 	queue->tx_evtchn = 0;
1523 fail:
1524 	return err;
1525 }
1526 
1527 static int setup_netfront(struct xenbus_device *dev,
1528 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1529 {
1530 	struct xen_netif_tx_sring *txs;
1531 	struct xen_netif_rx_sring *rxs;
1532 	grant_ref_t gref;
1533 	int err;
1534 
1535 	queue->tx_ring_ref = GRANT_INVALID_REF;
1536 	queue->rx_ring_ref = GRANT_INVALID_REF;
1537 	queue->rx.sring = NULL;
1538 	queue->tx.sring = NULL;
1539 
1540 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1541 	if (!txs) {
1542 		err = -ENOMEM;
1543 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1544 		goto fail;
1545 	}
1546 	SHARED_RING_INIT(txs);
1547 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1548 
1549 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1550 	if (err < 0)
1551 		goto grant_tx_ring_fail;
1552 	queue->tx_ring_ref = gref;
1553 
1554 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1555 	if (!rxs) {
1556 		err = -ENOMEM;
1557 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1558 		goto alloc_rx_ring_fail;
1559 	}
1560 	SHARED_RING_INIT(rxs);
1561 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1562 
1563 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1564 	if (err < 0)
1565 		goto grant_rx_ring_fail;
1566 	queue->rx_ring_ref = gref;
1567 
1568 	if (feature_split_evtchn)
1569 		err = setup_netfront_split(queue);
1570 	/* setup single event channel if
1571 	 *  a) feature-split-event-channels == 0
1572 	 *  b) feature-split-event-channels == 1 but failed to setup
1573 	 */
1574 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1575 		err = setup_netfront_single(queue);
1576 
1577 	if (err)
1578 		goto alloc_evtchn_fail;
1579 
1580 	return 0;
1581 
1582 	/* If we fail to setup netfront, it is safe to just revoke access to
1583 	 * granted pages because backend is not accessing it at this point.
1584 	 */
1585 alloc_evtchn_fail:
1586 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1587 grant_rx_ring_fail:
1588 	free_page((unsigned long)rxs);
1589 alloc_rx_ring_fail:
1590 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1591 grant_tx_ring_fail:
1592 	free_page((unsigned long)txs);
1593 fail:
1594 	return err;
1595 }
1596 
1597 /* Queue-specific initialisation
1598  * This used to be done in xennet_create_dev() but must now
1599  * be run per-queue.
1600  */
1601 static int xennet_init_queue(struct netfront_queue *queue)
1602 {
1603 	unsigned short i;
1604 	int err = 0;
1605 
1606 	spin_lock_init(&queue->tx_lock);
1607 	spin_lock_init(&queue->rx_lock);
1608 
1609 	setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1610 		    (unsigned long)queue);
1611 
1612 	snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1613 		 queue->info->netdev->name, queue->id);
1614 
1615 	/* Initialise tx_skbs as a free chain containing every entry. */
1616 	queue->tx_skb_freelist = 0;
1617 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1618 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1619 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1620 		queue->grant_tx_page[i] = NULL;
1621 	}
1622 
1623 	/* Clear out rx_skbs */
1624 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1625 		queue->rx_skbs[i] = NULL;
1626 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1627 	}
1628 
1629 	/* A grant for every tx ring slot */
1630 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1631 					  &queue->gref_tx_head) < 0) {
1632 		pr_alert("can't alloc tx grant refs\n");
1633 		err = -ENOMEM;
1634 		goto exit;
1635 	}
1636 
1637 	/* A grant for every rx ring slot */
1638 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1639 					  &queue->gref_rx_head) < 0) {
1640 		pr_alert("can't alloc rx grant refs\n");
1641 		err = -ENOMEM;
1642 		goto exit_free_tx;
1643 	}
1644 
1645 	return 0;
1646 
1647  exit_free_tx:
1648 	gnttab_free_grant_references(queue->gref_tx_head);
1649  exit:
1650 	return err;
1651 }
1652 
1653 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1654 			   struct xenbus_transaction *xbt, int write_hierarchical)
1655 {
1656 	/* Write the queue-specific keys into XenStore in the traditional
1657 	 * way for a single queue, or in a queue subkeys for multiple
1658 	 * queues.
1659 	 */
1660 	struct xenbus_device *dev = queue->info->xbdev;
1661 	int err;
1662 	const char *message;
1663 	char *path;
1664 	size_t pathsize;
1665 
1666 	/* Choose the correct place to write the keys */
1667 	if (write_hierarchical) {
1668 		pathsize = strlen(dev->nodename) + 10;
1669 		path = kzalloc(pathsize, GFP_KERNEL);
1670 		if (!path) {
1671 			err = -ENOMEM;
1672 			message = "out of memory while writing ring references";
1673 			goto error;
1674 		}
1675 		snprintf(path, pathsize, "%s/queue-%u",
1676 				dev->nodename, queue->id);
1677 	} else {
1678 		path = (char *)dev->nodename;
1679 	}
1680 
1681 	/* Write ring references */
1682 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1683 			queue->tx_ring_ref);
1684 	if (err) {
1685 		message = "writing tx-ring-ref";
1686 		goto error;
1687 	}
1688 
1689 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1690 			queue->rx_ring_ref);
1691 	if (err) {
1692 		message = "writing rx-ring-ref";
1693 		goto error;
1694 	}
1695 
1696 	/* Write event channels; taking into account both shared
1697 	 * and split event channel scenarios.
1698 	 */
1699 	if (queue->tx_evtchn == queue->rx_evtchn) {
1700 		/* Shared event channel */
1701 		err = xenbus_printf(*xbt, path,
1702 				"event-channel", "%u", queue->tx_evtchn);
1703 		if (err) {
1704 			message = "writing event-channel";
1705 			goto error;
1706 		}
1707 	} else {
1708 		/* Split event channels */
1709 		err = xenbus_printf(*xbt, path,
1710 				"event-channel-tx", "%u", queue->tx_evtchn);
1711 		if (err) {
1712 			message = "writing event-channel-tx";
1713 			goto error;
1714 		}
1715 
1716 		err = xenbus_printf(*xbt, path,
1717 				"event-channel-rx", "%u", queue->rx_evtchn);
1718 		if (err) {
1719 			message = "writing event-channel-rx";
1720 			goto error;
1721 		}
1722 	}
1723 
1724 	if (write_hierarchical)
1725 		kfree(path);
1726 	return 0;
1727 
1728 error:
1729 	if (write_hierarchical)
1730 		kfree(path);
1731 	xenbus_dev_fatal(dev, err, "%s", message);
1732 	return err;
1733 }
1734 
1735 static void xennet_destroy_queues(struct netfront_info *info)
1736 {
1737 	unsigned int i;
1738 
1739 	rtnl_lock();
1740 
1741 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1742 		struct netfront_queue *queue = &info->queues[i];
1743 
1744 		if (netif_running(info->netdev))
1745 			napi_disable(&queue->napi);
1746 		netif_napi_del(&queue->napi);
1747 	}
1748 
1749 	rtnl_unlock();
1750 
1751 	kfree(info->queues);
1752 	info->queues = NULL;
1753 }
1754 
1755 static int xennet_create_queues(struct netfront_info *info,
1756 				unsigned int *num_queues)
1757 {
1758 	unsigned int i;
1759 	int ret;
1760 
1761 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1762 			       GFP_KERNEL);
1763 	if (!info->queues)
1764 		return -ENOMEM;
1765 
1766 	rtnl_lock();
1767 
1768 	for (i = 0; i < *num_queues; i++) {
1769 		struct netfront_queue *queue = &info->queues[i];
1770 
1771 		queue->id = i;
1772 		queue->info = info;
1773 
1774 		ret = xennet_init_queue(queue);
1775 		if (ret < 0) {
1776 			dev_warn(&info->netdev->dev,
1777 				 "only created %d queues\n", i);
1778 			*num_queues = i;
1779 			break;
1780 		}
1781 
1782 		netif_napi_add(queue->info->netdev, &queue->napi,
1783 			       xennet_poll, 64);
1784 		if (netif_running(info->netdev))
1785 			napi_enable(&queue->napi);
1786 	}
1787 
1788 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
1789 
1790 	rtnl_unlock();
1791 
1792 	if (*num_queues == 0) {
1793 		dev_err(&info->netdev->dev, "no queues\n");
1794 		return -EINVAL;
1795 	}
1796 	return 0;
1797 }
1798 
1799 /* Common code used when first setting up, and when resuming. */
1800 static int talk_to_netback(struct xenbus_device *dev,
1801 			   struct netfront_info *info)
1802 {
1803 	const char *message;
1804 	struct xenbus_transaction xbt;
1805 	int err;
1806 	unsigned int feature_split_evtchn;
1807 	unsigned int i = 0;
1808 	unsigned int max_queues = 0;
1809 	struct netfront_queue *queue = NULL;
1810 	unsigned int num_queues = 1;
1811 
1812 	info->netdev->irq = 0;
1813 
1814 	/* Check if backend supports multiple queues */
1815 	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1816 					  "multi-queue-max-queues", 1);
1817 	num_queues = min(max_queues, xennet_max_queues);
1818 
1819 	/* Check feature-split-event-channels */
1820 	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1821 					"feature-split-event-channels", 0);
1822 
1823 	/* Read mac addr. */
1824 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
1825 	if (err) {
1826 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1827 		goto out;
1828 	}
1829 
1830 	if (info->queues)
1831 		xennet_destroy_queues(info);
1832 
1833 	err = xennet_create_queues(info, &num_queues);
1834 	if (err < 0) {
1835 		xenbus_dev_fatal(dev, err, "creating queues");
1836 		kfree(info->queues);
1837 		info->queues = NULL;
1838 		goto out;
1839 	}
1840 
1841 	/* Create shared ring, alloc event channel -- for each queue */
1842 	for (i = 0; i < num_queues; ++i) {
1843 		queue = &info->queues[i];
1844 		err = setup_netfront(dev, queue, feature_split_evtchn);
1845 		if (err)
1846 			goto destroy_ring;
1847 	}
1848 
1849 again:
1850 	err = xenbus_transaction_start(&xbt);
1851 	if (err) {
1852 		xenbus_dev_fatal(dev, err, "starting transaction");
1853 		goto destroy_ring;
1854 	}
1855 
1856 	if (xenbus_exists(XBT_NIL,
1857 			  info->xbdev->otherend, "multi-queue-max-queues")) {
1858 		/* Write the number of queues */
1859 		err = xenbus_printf(xbt, dev->nodename,
1860 				    "multi-queue-num-queues", "%u", num_queues);
1861 		if (err) {
1862 			message = "writing multi-queue-num-queues";
1863 			goto abort_transaction_no_dev_fatal;
1864 		}
1865 	}
1866 
1867 	if (num_queues == 1) {
1868 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1869 		if (err)
1870 			goto abort_transaction_no_dev_fatal;
1871 	} else {
1872 		/* Write the keys for each queue */
1873 		for (i = 0; i < num_queues; ++i) {
1874 			queue = &info->queues[i];
1875 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1876 			if (err)
1877 				goto abort_transaction_no_dev_fatal;
1878 		}
1879 	}
1880 
1881 	/* The remaining keys are not queue-specific */
1882 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1883 			    1);
1884 	if (err) {
1885 		message = "writing request-rx-copy";
1886 		goto abort_transaction;
1887 	}
1888 
1889 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1890 	if (err) {
1891 		message = "writing feature-rx-notify";
1892 		goto abort_transaction;
1893 	}
1894 
1895 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1896 	if (err) {
1897 		message = "writing feature-sg";
1898 		goto abort_transaction;
1899 	}
1900 
1901 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1902 	if (err) {
1903 		message = "writing feature-gso-tcpv4";
1904 		goto abort_transaction;
1905 	}
1906 
1907 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1908 	if (err) {
1909 		message = "writing feature-gso-tcpv6";
1910 		goto abort_transaction;
1911 	}
1912 
1913 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1914 			   "1");
1915 	if (err) {
1916 		message = "writing feature-ipv6-csum-offload";
1917 		goto abort_transaction;
1918 	}
1919 
1920 	err = xenbus_transaction_end(xbt, 0);
1921 	if (err) {
1922 		if (err == -EAGAIN)
1923 			goto again;
1924 		xenbus_dev_fatal(dev, err, "completing transaction");
1925 		goto destroy_ring;
1926 	}
1927 
1928 	return 0;
1929 
1930  abort_transaction:
1931 	xenbus_dev_fatal(dev, err, "%s", message);
1932 abort_transaction_no_dev_fatal:
1933 	xenbus_transaction_end(xbt, 1);
1934  destroy_ring:
1935 	xennet_disconnect_backend(info);
1936 	xennet_destroy_queues(info);
1937  out:
1938 	unregister_netdev(info->netdev);
1939 	xennet_free_netdev(info->netdev);
1940 	return err;
1941 }
1942 
1943 static int xennet_connect(struct net_device *dev)
1944 {
1945 	struct netfront_info *np = netdev_priv(dev);
1946 	unsigned int num_queues = 0;
1947 	int err;
1948 	unsigned int j = 0;
1949 	struct netfront_queue *queue = NULL;
1950 
1951 	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1952 		dev_info(&dev->dev,
1953 			 "backend does not support copying receive path\n");
1954 		return -ENODEV;
1955 	}
1956 
1957 	err = talk_to_netback(np->xbdev, np);
1958 	if (err)
1959 		return err;
1960 
1961 	/* talk_to_netback() sets the correct number of queues */
1962 	num_queues = dev->real_num_tx_queues;
1963 
1964 	rtnl_lock();
1965 	netdev_update_features(dev);
1966 	rtnl_unlock();
1967 
1968 	/*
1969 	 * All public and private state should now be sane.  Get
1970 	 * ready to start sending and receiving packets and give the driver
1971 	 * domain a kick because we've probably just requeued some
1972 	 * packets.
1973 	 */
1974 	netif_carrier_on(np->netdev);
1975 	for (j = 0; j < num_queues; ++j) {
1976 		queue = &np->queues[j];
1977 
1978 		notify_remote_via_irq(queue->tx_irq);
1979 		if (queue->tx_irq != queue->rx_irq)
1980 			notify_remote_via_irq(queue->rx_irq);
1981 
1982 		spin_lock_irq(&queue->tx_lock);
1983 		xennet_tx_buf_gc(queue);
1984 		spin_unlock_irq(&queue->tx_lock);
1985 
1986 		spin_lock_bh(&queue->rx_lock);
1987 		xennet_alloc_rx_buffers(queue);
1988 		spin_unlock_bh(&queue->rx_lock);
1989 	}
1990 
1991 	return 0;
1992 }
1993 
1994 /**
1995  * Callback received when the backend's state changes.
1996  */
1997 static void netback_changed(struct xenbus_device *dev,
1998 			    enum xenbus_state backend_state)
1999 {
2000 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2001 	struct net_device *netdev = np->netdev;
2002 
2003 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2004 
2005 	switch (backend_state) {
2006 	case XenbusStateInitialising:
2007 	case XenbusStateInitialised:
2008 	case XenbusStateReconfiguring:
2009 	case XenbusStateReconfigured:
2010 	case XenbusStateUnknown:
2011 		break;
2012 
2013 	case XenbusStateInitWait:
2014 		if (dev->state != XenbusStateInitialising)
2015 			break;
2016 		if (xennet_connect(netdev) != 0)
2017 			break;
2018 		xenbus_switch_state(dev, XenbusStateConnected);
2019 		break;
2020 
2021 	case XenbusStateConnected:
2022 		netdev_notify_peers(netdev);
2023 		break;
2024 
2025 	case XenbusStateClosed:
2026 		if (dev->state == XenbusStateClosed)
2027 			break;
2028 		/* Missed the backend's CLOSING state -- fallthrough */
2029 	case XenbusStateClosing:
2030 		xenbus_frontend_closed(dev);
2031 		break;
2032 	}
2033 }
2034 
2035 static const struct xennet_stat {
2036 	char name[ETH_GSTRING_LEN];
2037 	u16 offset;
2038 } xennet_stats[] = {
2039 	{
2040 		"rx_gso_checksum_fixup",
2041 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2042 	},
2043 };
2044 
2045 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2046 {
2047 	switch (string_set) {
2048 	case ETH_SS_STATS:
2049 		return ARRAY_SIZE(xennet_stats);
2050 	default:
2051 		return -EINVAL;
2052 	}
2053 }
2054 
2055 static void xennet_get_ethtool_stats(struct net_device *dev,
2056 				     struct ethtool_stats *stats, u64 * data)
2057 {
2058 	void *np = netdev_priv(dev);
2059 	int i;
2060 
2061 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2062 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2063 }
2064 
2065 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2066 {
2067 	int i;
2068 
2069 	switch (stringset) {
2070 	case ETH_SS_STATS:
2071 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2072 			memcpy(data + i * ETH_GSTRING_LEN,
2073 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2074 		break;
2075 	}
2076 }
2077 
2078 static const struct ethtool_ops xennet_ethtool_ops =
2079 {
2080 	.get_link = ethtool_op_get_link,
2081 
2082 	.get_sset_count = xennet_get_sset_count,
2083 	.get_ethtool_stats = xennet_get_ethtool_stats,
2084 	.get_strings = xennet_get_strings,
2085 };
2086 
2087 #ifdef CONFIG_SYSFS
2088 static ssize_t show_rxbuf(struct device *dev,
2089 			  struct device_attribute *attr, char *buf)
2090 {
2091 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2092 }
2093 
2094 static ssize_t store_rxbuf(struct device *dev,
2095 			   struct device_attribute *attr,
2096 			   const char *buf, size_t len)
2097 {
2098 	char *endp;
2099 	unsigned long target;
2100 
2101 	if (!capable(CAP_NET_ADMIN))
2102 		return -EPERM;
2103 
2104 	target = simple_strtoul(buf, &endp, 0);
2105 	if (endp == buf)
2106 		return -EBADMSG;
2107 
2108 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2109 
2110 	return len;
2111 }
2112 
2113 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2114 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2115 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2116 
2117 static struct attribute *xennet_dev_attrs[] = {
2118 	&dev_attr_rxbuf_min.attr,
2119 	&dev_attr_rxbuf_max.attr,
2120 	&dev_attr_rxbuf_cur.attr,
2121 	NULL
2122 };
2123 
2124 static const struct attribute_group xennet_dev_group = {
2125 	.attrs = xennet_dev_attrs
2126 };
2127 #endif /* CONFIG_SYSFS */
2128 
2129 static int xennet_remove(struct xenbus_device *dev)
2130 {
2131 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2132 
2133 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2134 
2135 	xennet_disconnect_backend(info);
2136 
2137 	unregister_netdev(info->netdev);
2138 
2139 	if (info->queues)
2140 		xennet_destroy_queues(info);
2141 	xennet_free_netdev(info->netdev);
2142 
2143 	return 0;
2144 }
2145 
2146 static const struct xenbus_device_id netfront_ids[] = {
2147 	{ "vif" },
2148 	{ "" }
2149 };
2150 
2151 static struct xenbus_driver netfront_driver = {
2152 	.ids = netfront_ids,
2153 	.probe = netfront_probe,
2154 	.remove = xennet_remove,
2155 	.resume = netfront_resume,
2156 	.otherend_changed = netback_changed,
2157 };
2158 
2159 static int __init netif_init(void)
2160 {
2161 	if (!xen_domain())
2162 		return -ENODEV;
2163 
2164 	if (!xen_has_pv_nic_devices())
2165 		return -ENODEV;
2166 
2167 	pr_info("Initialising Xen virtual ethernet driver\n");
2168 
2169 	/* Allow as many queues as there are CPUs if user has not
2170 	 * specified a value.
2171 	 */
2172 	if (xennet_max_queues == 0)
2173 		xennet_max_queues = num_online_cpus();
2174 
2175 	return xenbus_register_frontend(&netfront_driver);
2176 }
2177 module_init(netif_init);
2178 
2179 
2180 static void __exit netif_exit(void)
2181 {
2182 	xenbus_unregister_driver(&netfront_driver);
2183 }
2184 module_exit(netif_exit);
2185 
2186 MODULE_DESCRIPTION("Xen virtual network device frontend");
2187 MODULE_LICENSE("GPL");
2188 MODULE_ALIAS("xen:vif");
2189 MODULE_ALIAS("xennet");
2190