xref: /linux/drivers/net/xen-netfront.c (revision 905e46acd3272d04566fec49afbd7ad9e2ed9ae3)
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 #define MAX_QUEUES_DEFAULT 8
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, XEN_PAGE_SIZE)
79 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_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 	int err = 0;
286 
287 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
288 		return;
289 
290 	for (req_prod = queue->rx.req_prod_pvt;
291 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
292 	     req_prod++) {
293 		struct sk_buff *skb;
294 		unsigned short id;
295 		grant_ref_t ref;
296 		struct page *page;
297 		struct xen_netif_rx_request *req;
298 
299 		skb = xennet_alloc_one_rx_buffer(queue);
300 		if (!skb) {
301 			err = -ENOMEM;
302 			break;
303 		}
304 
305 		id = xennet_rxidx(req_prod);
306 
307 		BUG_ON(queue->rx_skbs[id]);
308 		queue->rx_skbs[id] = skb;
309 
310 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
311 		WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
312 		queue->grant_rx_ref[id] = ref;
313 
314 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
315 
316 		req = RING_GET_REQUEST(&queue->rx, req_prod);
317 		gnttab_page_grant_foreign_access_ref_one(ref,
318 							 queue->info->xbdev->otherend_id,
319 							 page,
320 							 0);
321 		req->id = id;
322 		req->gref = ref;
323 	}
324 
325 	queue->rx.req_prod_pvt = req_prod;
326 
327 	/* Try again later if there are not enough requests or skb allocation
328 	 * failed.
329 	 * Enough requests is quantified as the sum of newly created slots and
330 	 * the unconsumed slots at the backend.
331 	 */
332 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
333 	    unlikely(err)) {
334 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
335 		return;
336 	}
337 
338 	wmb();		/* barrier so backend seens requests */
339 
340 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
341 	if (notify)
342 		notify_remote_via_irq(queue->rx_irq);
343 }
344 
345 static int xennet_open(struct net_device *dev)
346 {
347 	struct netfront_info *np = netdev_priv(dev);
348 	unsigned int num_queues = dev->real_num_tx_queues;
349 	unsigned int i = 0;
350 	struct netfront_queue *queue = NULL;
351 
352 	for (i = 0; i < num_queues; ++i) {
353 		queue = &np->queues[i];
354 		napi_enable(&queue->napi);
355 
356 		spin_lock_bh(&queue->rx_lock);
357 		if (netif_carrier_ok(dev)) {
358 			xennet_alloc_rx_buffers(queue);
359 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
360 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
361 				napi_schedule(&queue->napi);
362 		}
363 		spin_unlock_bh(&queue->rx_lock);
364 	}
365 
366 	netif_tx_start_all_queues(dev);
367 
368 	return 0;
369 }
370 
371 static void xennet_tx_buf_gc(struct netfront_queue *queue)
372 {
373 	RING_IDX cons, prod;
374 	unsigned short id;
375 	struct sk_buff *skb;
376 	bool more_to_do;
377 
378 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
379 
380 	do {
381 		prod = queue->tx.sring->rsp_prod;
382 		rmb(); /* Ensure we see responses up to 'rp'. */
383 
384 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
385 			struct xen_netif_tx_response *txrsp;
386 
387 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
388 			if (txrsp->status == XEN_NETIF_RSP_NULL)
389 				continue;
390 
391 			id  = txrsp->id;
392 			skb = queue->tx_skbs[id].skb;
393 			if (unlikely(gnttab_query_foreign_access(
394 				queue->grant_tx_ref[id]) != 0)) {
395 				pr_alert("%s: warning -- grant still in use by backend domain\n",
396 					 __func__);
397 				BUG();
398 			}
399 			gnttab_end_foreign_access_ref(
400 				queue->grant_tx_ref[id], GNTMAP_readonly);
401 			gnttab_release_grant_reference(
402 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
403 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
404 			queue->grant_tx_page[id] = NULL;
405 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
406 			dev_kfree_skb_irq(skb);
407 		}
408 
409 		queue->tx.rsp_cons = prod;
410 
411 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
412 	} while (more_to_do);
413 
414 	xennet_maybe_wake_tx(queue);
415 }
416 
417 struct xennet_gnttab_make_txreq {
418 	struct netfront_queue *queue;
419 	struct sk_buff *skb;
420 	struct page *page;
421 	struct xen_netif_tx_request *tx; /* Last request */
422 	unsigned int size;
423 };
424 
425 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
426 				  unsigned int len, void *data)
427 {
428 	struct xennet_gnttab_make_txreq *info = data;
429 	unsigned int id;
430 	struct xen_netif_tx_request *tx;
431 	grant_ref_t ref;
432 	/* convenient aliases */
433 	struct page *page = info->page;
434 	struct netfront_queue *queue = info->queue;
435 	struct sk_buff *skb = info->skb;
436 
437 	id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
438 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
439 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
440 	WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
441 
442 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
443 					gfn, GNTMAP_readonly);
444 
445 	queue->tx_skbs[id].skb = skb;
446 	queue->grant_tx_page[id] = page;
447 	queue->grant_tx_ref[id] = ref;
448 
449 	tx->id = id;
450 	tx->gref = ref;
451 	tx->offset = offset;
452 	tx->size = len;
453 	tx->flags = 0;
454 
455 	info->tx = tx;
456 	info->size += tx->size;
457 }
458 
459 static struct xen_netif_tx_request *xennet_make_first_txreq(
460 	struct netfront_queue *queue, struct sk_buff *skb,
461 	struct page *page, unsigned int offset, unsigned int len)
462 {
463 	struct xennet_gnttab_make_txreq info = {
464 		.queue = queue,
465 		.skb = skb,
466 		.page = page,
467 		.size = 0,
468 	};
469 
470 	gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
471 
472 	return info.tx;
473 }
474 
475 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
476 				  unsigned int len, void *data)
477 {
478 	struct xennet_gnttab_make_txreq *info = data;
479 
480 	info->tx->flags |= XEN_NETTXF_more_data;
481 	skb_get(info->skb);
482 	xennet_tx_setup_grant(gfn, offset, len, data);
483 }
484 
485 static struct xen_netif_tx_request *xennet_make_txreqs(
486 	struct netfront_queue *queue, struct xen_netif_tx_request *tx,
487 	struct sk_buff *skb, struct page *page,
488 	unsigned int offset, unsigned int len)
489 {
490 	struct xennet_gnttab_make_txreq info = {
491 		.queue = queue,
492 		.skb = skb,
493 		.tx = tx,
494 	};
495 
496 	/* Skip unused frames from start of page */
497 	page += offset >> PAGE_SHIFT;
498 	offset &= ~PAGE_MASK;
499 
500 	while (len) {
501 		info.page = page;
502 		info.size = 0;
503 
504 		gnttab_foreach_grant_in_range(page, offset, len,
505 					      xennet_make_one_txreq,
506 					      &info);
507 
508 		page++;
509 		offset = 0;
510 		len -= info.size;
511 	}
512 
513 	return info.tx;
514 }
515 
516 /*
517  * Count how many ring slots are required to send this skb. Each frag
518  * might be a compound page.
519  */
520 static int xennet_count_skb_slots(struct sk_buff *skb)
521 {
522 	int i, frags = skb_shinfo(skb)->nr_frags;
523 	int slots;
524 
525 	slots = gnttab_count_grant(offset_in_page(skb->data),
526 				   skb_headlen(skb));
527 
528 	for (i = 0; i < frags; i++) {
529 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
530 		unsigned long size = skb_frag_size(frag);
531 		unsigned long offset = frag->page_offset;
532 
533 		/* Skip unused frames from start of page */
534 		offset &= ~PAGE_MASK;
535 
536 		slots += gnttab_count_grant(offset, size);
537 	}
538 
539 	return slots;
540 }
541 
542 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
543 			       void *accel_priv, select_queue_fallback_t fallback)
544 {
545 	unsigned int num_queues = dev->real_num_tx_queues;
546 	u32 hash;
547 	u16 queue_idx;
548 
549 	/* First, check if there is only one queue */
550 	if (num_queues == 1) {
551 		queue_idx = 0;
552 	} else {
553 		hash = skb_get_hash(skb);
554 		queue_idx = hash % num_queues;
555 	}
556 
557 	return queue_idx;
558 }
559 
560 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
561 
562 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
563 {
564 	struct netfront_info *np = netdev_priv(dev);
565 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
566 	struct xen_netif_tx_request *tx, *first_tx;
567 	unsigned int i;
568 	int notify;
569 	int slots;
570 	struct page *page;
571 	unsigned int offset;
572 	unsigned int len;
573 	unsigned long flags;
574 	struct netfront_queue *queue = NULL;
575 	unsigned int num_queues = dev->real_num_tx_queues;
576 	u16 queue_index;
577 	struct sk_buff *nskb;
578 
579 	/* Drop the packet if no queues are set up */
580 	if (num_queues < 1)
581 		goto drop;
582 	/* Determine which queue to transmit this SKB on */
583 	queue_index = skb_get_queue_mapping(skb);
584 	queue = &np->queues[queue_index];
585 
586 	/* If skb->len is too big for wire format, drop skb and alert
587 	 * user about misconfiguration.
588 	 */
589 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
590 		net_alert_ratelimited(
591 			"xennet: skb->len = %u, too big for wire format\n",
592 			skb->len);
593 		goto drop;
594 	}
595 
596 	slots = xennet_count_skb_slots(skb);
597 	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
598 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
599 				    slots, skb->len);
600 		if (skb_linearize(skb))
601 			goto drop;
602 	}
603 
604 	page = virt_to_page(skb->data);
605 	offset = offset_in_page(skb->data);
606 
607 	/* The first req should be at least ETH_HLEN size or the packet will be
608 	 * dropped by netback.
609 	 */
610 	if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
611 		nskb = skb_copy(skb, GFP_ATOMIC);
612 		if (!nskb)
613 			goto drop;
614 		dev_kfree_skb_any(skb);
615 		skb = nskb;
616 		page = virt_to_page(skb->data);
617 		offset = offset_in_page(skb->data);
618 	}
619 
620 	len = skb_headlen(skb);
621 
622 	spin_lock_irqsave(&queue->tx_lock, flags);
623 
624 	if (unlikely(!netif_carrier_ok(dev) ||
625 		     (slots > 1 && !xennet_can_sg(dev)) ||
626 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
627 		spin_unlock_irqrestore(&queue->tx_lock, flags);
628 		goto drop;
629 	}
630 
631 	/* First request for the linear area. */
632 	first_tx = tx = xennet_make_first_txreq(queue, skb,
633 						page, offset, len);
634 	offset += tx->size;
635 	if (offset == PAGE_SIZE) {
636 		page++;
637 		offset = 0;
638 	}
639 	len -= tx->size;
640 
641 	if (skb->ip_summed == CHECKSUM_PARTIAL)
642 		/* local packet? */
643 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
644 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
645 		/* remote but checksummed. */
646 		tx->flags |= XEN_NETTXF_data_validated;
647 
648 	/* Optional extra info after the first request. */
649 	if (skb_shinfo(skb)->gso_size) {
650 		struct xen_netif_extra_info *gso;
651 
652 		gso = (struct xen_netif_extra_info *)
653 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
654 
655 		tx->flags |= XEN_NETTXF_extra_info;
656 
657 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
658 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
659 			XEN_NETIF_GSO_TYPE_TCPV6 :
660 			XEN_NETIF_GSO_TYPE_TCPV4;
661 		gso->u.gso.pad = 0;
662 		gso->u.gso.features = 0;
663 
664 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
665 		gso->flags = 0;
666 	}
667 
668 	/* Requests for the rest of the linear area. */
669 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
670 
671 	/* Requests for all the frags. */
672 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
673 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
674 		tx = xennet_make_txreqs(queue, tx, skb,
675 					skb_frag_page(frag), frag->page_offset,
676 					skb_frag_size(frag));
677 	}
678 
679 	/* First request has the packet length. */
680 	first_tx->size = skb->len;
681 
682 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
683 	if (notify)
684 		notify_remote_via_irq(queue->tx_irq);
685 
686 	u64_stats_update_begin(&tx_stats->syncp);
687 	tx_stats->bytes += skb->len;
688 	tx_stats->packets++;
689 	u64_stats_update_end(&tx_stats->syncp);
690 
691 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
692 	xennet_tx_buf_gc(queue);
693 
694 	if (!netfront_tx_slot_available(queue))
695 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
696 
697 	spin_unlock_irqrestore(&queue->tx_lock, flags);
698 
699 	return NETDEV_TX_OK;
700 
701  drop:
702 	dev->stats.tx_dropped++;
703 	dev_kfree_skb_any(skb);
704 	return NETDEV_TX_OK;
705 }
706 
707 static int xennet_close(struct net_device *dev)
708 {
709 	struct netfront_info *np = netdev_priv(dev);
710 	unsigned int num_queues = dev->real_num_tx_queues;
711 	unsigned int i;
712 	struct netfront_queue *queue;
713 	netif_tx_stop_all_queues(np->netdev);
714 	for (i = 0; i < num_queues; ++i) {
715 		queue = &np->queues[i];
716 		napi_disable(&queue->napi);
717 	}
718 	return 0;
719 }
720 
721 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
722 				grant_ref_t ref)
723 {
724 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
725 
726 	BUG_ON(queue->rx_skbs[new]);
727 	queue->rx_skbs[new] = skb;
728 	queue->grant_rx_ref[new] = ref;
729 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
730 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
731 	queue->rx.req_prod_pvt++;
732 }
733 
734 static int xennet_get_extras(struct netfront_queue *queue,
735 			     struct xen_netif_extra_info *extras,
736 			     RING_IDX rp)
737 
738 {
739 	struct xen_netif_extra_info *extra;
740 	struct device *dev = &queue->info->netdev->dev;
741 	RING_IDX cons = queue->rx.rsp_cons;
742 	int err = 0;
743 
744 	do {
745 		struct sk_buff *skb;
746 		grant_ref_t ref;
747 
748 		if (unlikely(cons + 1 == rp)) {
749 			if (net_ratelimit())
750 				dev_warn(dev, "Missing extra info\n");
751 			err = -EBADR;
752 			break;
753 		}
754 
755 		extra = (struct xen_netif_extra_info *)
756 			RING_GET_RESPONSE(&queue->rx, ++cons);
757 
758 		if (unlikely(!extra->type ||
759 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
760 			if (net_ratelimit())
761 				dev_warn(dev, "Invalid extra type: %d\n",
762 					extra->type);
763 			err = -EINVAL;
764 		} else {
765 			memcpy(&extras[extra->type - 1], extra,
766 			       sizeof(*extra));
767 		}
768 
769 		skb = xennet_get_rx_skb(queue, cons);
770 		ref = xennet_get_rx_ref(queue, cons);
771 		xennet_move_rx_slot(queue, skb, ref);
772 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
773 
774 	queue->rx.rsp_cons = cons;
775 	return err;
776 }
777 
778 static int xennet_get_responses(struct netfront_queue *queue,
779 				struct netfront_rx_info *rinfo, RING_IDX rp,
780 				struct sk_buff_head *list)
781 {
782 	struct xen_netif_rx_response *rx = &rinfo->rx;
783 	struct xen_netif_extra_info *extras = rinfo->extras;
784 	struct device *dev = &queue->info->netdev->dev;
785 	RING_IDX cons = queue->rx.rsp_cons;
786 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
787 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
788 	int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
789 	int slots = 1;
790 	int err = 0;
791 	unsigned long ret;
792 
793 	if (rx->flags & XEN_NETRXF_extra_info) {
794 		err = xennet_get_extras(queue, extras, rp);
795 		cons = queue->rx.rsp_cons;
796 	}
797 
798 	for (;;) {
799 		if (unlikely(rx->status < 0 ||
800 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
801 			if (net_ratelimit())
802 				dev_warn(dev, "rx->offset: %u, size: %d\n",
803 					 rx->offset, rx->status);
804 			xennet_move_rx_slot(queue, skb, ref);
805 			err = -EINVAL;
806 			goto next;
807 		}
808 
809 		/*
810 		 * This definitely indicates a bug, either in this driver or in
811 		 * the backend driver. In future this should flag the bad
812 		 * situation to the system controller to reboot the backend.
813 		 */
814 		if (ref == GRANT_INVALID_REF) {
815 			if (net_ratelimit())
816 				dev_warn(dev, "Bad rx response id %d.\n",
817 					 rx->id);
818 			err = -EINVAL;
819 			goto next;
820 		}
821 
822 		ret = gnttab_end_foreign_access_ref(ref, 0);
823 		BUG_ON(!ret);
824 
825 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
826 
827 		__skb_queue_tail(list, skb);
828 
829 next:
830 		if (!(rx->flags & XEN_NETRXF_more_data))
831 			break;
832 
833 		if (cons + slots == rp) {
834 			if (net_ratelimit())
835 				dev_warn(dev, "Need more slots\n");
836 			err = -ENOENT;
837 			break;
838 		}
839 
840 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
841 		skb = xennet_get_rx_skb(queue, cons + slots);
842 		ref = xennet_get_rx_ref(queue, cons + slots);
843 		slots++;
844 	}
845 
846 	if (unlikely(slots > max)) {
847 		if (net_ratelimit())
848 			dev_warn(dev, "Too many slots\n");
849 		err = -E2BIG;
850 	}
851 
852 	if (unlikely(err))
853 		queue->rx.rsp_cons = cons + slots;
854 
855 	return err;
856 }
857 
858 static int xennet_set_skb_gso(struct sk_buff *skb,
859 			      struct xen_netif_extra_info *gso)
860 {
861 	if (!gso->u.gso.size) {
862 		if (net_ratelimit())
863 			pr_warn("GSO size must not be zero\n");
864 		return -EINVAL;
865 	}
866 
867 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
868 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
869 		if (net_ratelimit())
870 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
871 		return -EINVAL;
872 	}
873 
874 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
875 	skb_shinfo(skb)->gso_type =
876 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
877 		SKB_GSO_TCPV4 :
878 		SKB_GSO_TCPV6;
879 
880 	/* Header must be checked, and gso_segs computed. */
881 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
882 	skb_shinfo(skb)->gso_segs = 0;
883 
884 	return 0;
885 }
886 
887 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
888 				  struct sk_buff *skb,
889 				  struct sk_buff_head *list)
890 {
891 	struct skb_shared_info *shinfo = skb_shinfo(skb);
892 	RING_IDX cons = queue->rx.rsp_cons;
893 	struct sk_buff *nskb;
894 
895 	while ((nskb = __skb_dequeue(list))) {
896 		struct xen_netif_rx_response *rx =
897 			RING_GET_RESPONSE(&queue->rx, ++cons);
898 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
899 
900 		if (shinfo->nr_frags == MAX_SKB_FRAGS) {
901 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
902 
903 			BUG_ON(pull_to <= skb_headlen(skb));
904 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
905 		}
906 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
907 
908 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
909 				rx->offset, rx->status, PAGE_SIZE);
910 
911 		skb_shinfo(nskb)->nr_frags = 0;
912 		kfree_skb(nskb);
913 	}
914 
915 	return cons;
916 }
917 
918 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
919 {
920 	bool recalculate_partial_csum = false;
921 
922 	/*
923 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
924 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
925 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
926 	 * recalculate the partial checksum.
927 	 */
928 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
929 		struct netfront_info *np = netdev_priv(dev);
930 		atomic_inc(&np->rx_gso_checksum_fixup);
931 		skb->ip_summed = CHECKSUM_PARTIAL;
932 		recalculate_partial_csum = true;
933 	}
934 
935 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
936 	if (skb->ip_summed != CHECKSUM_PARTIAL)
937 		return 0;
938 
939 	return skb_checksum_setup(skb, recalculate_partial_csum);
940 }
941 
942 static int handle_incoming_queue(struct netfront_queue *queue,
943 				 struct sk_buff_head *rxq)
944 {
945 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
946 	int packets_dropped = 0;
947 	struct sk_buff *skb;
948 
949 	while ((skb = __skb_dequeue(rxq)) != NULL) {
950 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
951 
952 		if (pull_to > skb_headlen(skb))
953 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
954 
955 		/* Ethernet work: Delayed to here as it peeks the header. */
956 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
957 		skb_reset_network_header(skb);
958 
959 		if (checksum_setup(queue->info->netdev, skb)) {
960 			kfree_skb(skb);
961 			packets_dropped++;
962 			queue->info->netdev->stats.rx_errors++;
963 			continue;
964 		}
965 
966 		u64_stats_update_begin(&rx_stats->syncp);
967 		rx_stats->packets++;
968 		rx_stats->bytes += skb->len;
969 		u64_stats_update_end(&rx_stats->syncp);
970 
971 		/* Pass it up. */
972 		napi_gro_receive(&queue->napi, skb);
973 	}
974 
975 	return packets_dropped;
976 }
977 
978 static int xennet_poll(struct napi_struct *napi, int budget)
979 {
980 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
981 	struct net_device *dev = queue->info->netdev;
982 	struct sk_buff *skb;
983 	struct netfront_rx_info rinfo;
984 	struct xen_netif_rx_response *rx = &rinfo.rx;
985 	struct xen_netif_extra_info *extras = rinfo.extras;
986 	RING_IDX i, rp;
987 	int work_done;
988 	struct sk_buff_head rxq;
989 	struct sk_buff_head errq;
990 	struct sk_buff_head tmpq;
991 	int err;
992 
993 	spin_lock(&queue->rx_lock);
994 
995 	skb_queue_head_init(&rxq);
996 	skb_queue_head_init(&errq);
997 	skb_queue_head_init(&tmpq);
998 
999 	rp = queue->rx.sring->rsp_prod;
1000 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1001 
1002 	i = queue->rx.rsp_cons;
1003 	work_done = 0;
1004 	while ((i != rp) && (work_done < budget)) {
1005 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1006 		memset(extras, 0, sizeof(rinfo.extras));
1007 
1008 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1009 
1010 		if (unlikely(err)) {
1011 err:
1012 			while ((skb = __skb_dequeue(&tmpq)))
1013 				__skb_queue_tail(&errq, skb);
1014 			dev->stats.rx_errors++;
1015 			i = queue->rx.rsp_cons;
1016 			continue;
1017 		}
1018 
1019 		skb = __skb_dequeue(&tmpq);
1020 
1021 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1022 			struct xen_netif_extra_info *gso;
1023 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1024 
1025 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1026 				__skb_queue_head(&tmpq, skb);
1027 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1028 				goto err;
1029 			}
1030 		}
1031 
1032 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1033 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1034 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1035 
1036 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1037 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1038 		skb->data_len = rx->status;
1039 		skb->len += rx->status;
1040 
1041 		i = xennet_fill_frags(queue, skb, &tmpq);
1042 
1043 		if (rx->flags & XEN_NETRXF_csum_blank)
1044 			skb->ip_summed = CHECKSUM_PARTIAL;
1045 		else if (rx->flags & XEN_NETRXF_data_validated)
1046 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1047 
1048 		__skb_queue_tail(&rxq, skb);
1049 
1050 		queue->rx.rsp_cons = ++i;
1051 		work_done++;
1052 	}
1053 
1054 	__skb_queue_purge(&errq);
1055 
1056 	work_done -= handle_incoming_queue(queue, &rxq);
1057 
1058 	xennet_alloc_rx_buffers(queue);
1059 
1060 	if (work_done < budget) {
1061 		int more_to_do = 0;
1062 
1063 		napi_complete_done(napi, work_done);
1064 
1065 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1066 		if (more_to_do)
1067 			napi_schedule(napi);
1068 	}
1069 
1070 	spin_unlock(&queue->rx_lock);
1071 
1072 	return work_done;
1073 }
1074 
1075 static int xennet_change_mtu(struct net_device *dev, int mtu)
1076 {
1077 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1078 
1079 	if (mtu > max)
1080 		return -EINVAL;
1081 	dev->mtu = mtu;
1082 	return 0;
1083 }
1084 
1085 static void xennet_get_stats64(struct net_device *dev,
1086 			       struct rtnl_link_stats64 *tot)
1087 {
1088 	struct netfront_info *np = netdev_priv(dev);
1089 	int cpu;
1090 
1091 	for_each_possible_cpu(cpu) {
1092 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1093 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1094 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1095 		unsigned int start;
1096 
1097 		do {
1098 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1099 			tx_packets = tx_stats->packets;
1100 			tx_bytes = tx_stats->bytes;
1101 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1102 
1103 		do {
1104 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1105 			rx_packets = rx_stats->packets;
1106 			rx_bytes = rx_stats->bytes;
1107 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1108 
1109 		tot->rx_packets += rx_packets;
1110 		tot->tx_packets += tx_packets;
1111 		tot->rx_bytes   += rx_bytes;
1112 		tot->tx_bytes   += tx_bytes;
1113 	}
1114 
1115 	tot->rx_errors  = dev->stats.rx_errors;
1116 	tot->tx_dropped = dev->stats.tx_dropped;
1117 }
1118 
1119 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1120 {
1121 	struct sk_buff *skb;
1122 	int i;
1123 
1124 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1125 		/* Skip over entries which are actually freelist references */
1126 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1127 			continue;
1128 
1129 		skb = queue->tx_skbs[i].skb;
1130 		get_page(queue->grant_tx_page[i]);
1131 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1132 					  GNTMAP_readonly,
1133 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1134 		queue->grant_tx_page[i] = NULL;
1135 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1136 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1137 		dev_kfree_skb_irq(skb);
1138 	}
1139 }
1140 
1141 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1142 {
1143 	int id, ref;
1144 
1145 	spin_lock_bh(&queue->rx_lock);
1146 
1147 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1148 		struct sk_buff *skb;
1149 		struct page *page;
1150 
1151 		skb = queue->rx_skbs[id];
1152 		if (!skb)
1153 			continue;
1154 
1155 		ref = queue->grant_rx_ref[id];
1156 		if (ref == GRANT_INVALID_REF)
1157 			continue;
1158 
1159 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1160 
1161 		/* gnttab_end_foreign_access() needs a page ref until
1162 		 * foreign access is ended (which may be deferred).
1163 		 */
1164 		get_page(page);
1165 		gnttab_end_foreign_access(ref, 0,
1166 					  (unsigned long)page_address(page));
1167 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1168 
1169 		kfree_skb(skb);
1170 	}
1171 
1172 	spin_unlock_bh(&queue->rx_lock);
1173 }
1174 
1175 static netdev_features_t xennet_fix_features(struct net_device *dev,
1176 	netdev_features_t features)
1177 {
1178 	struct netfront_info *np = netdev_priv(dev);
1179 
1180 	if (features & NETIF_F_SG &&
1181 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1182 		features &= ~NETIF_F_SG;
1183 
1184 	if (features & NETIF_F_IPV6_CSUM &&
1185 	    !xenbus_read_unsigned(np->xbdev->otherend,
1186 				  "feature-ipv6-csum-offload", 0))
1187 		features &= ~NETIF_F_IPV6_CSUM;
1188 
1189 	if (features & NETIF_F_TSO &&
1190 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1191 		features &= ~NETIF_F_TSO;
1192 
1193 	if (features & NETIF_F_TSO6 &&
1194 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1195 		features &= ~NETIF_F_TSO6;
1196 
1197 	return features;
1198 }
1199 
1200 static int xennet_set_features(struct net_device *dev,
1201 	netdev_features_t features)
1202 {
1203 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1204 		netdev_info(dev, "Reducing MTU because no SG offload");
1205 		dev->mtu = ETH_DATA_LEN;
1206 	}
1207 
1208 	return 0;
1209 }
1210 
1211 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1212 {
1213 	struct netfront_queue *queue = dev_id;
1214 	unsigned long flags;
1215 
1216 	spin_lock_irqsave(&queue->tx_lock, flags);
1217 	xennet_tx_buf_gc(queue);
1218 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1219 
1220 	return IRQ_HANDLED;
1221 }
1222 
1223 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1224 {
1225 	struct netfront_queue *queue = dev_id;
1226 	struct net_device *dev = queue->info->netdev;
1227 
1228 	if (likely(netif_carrier_ok(dev) &&
1229 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1230 		napi_schedule(&queue->napi);
1231 
1232 	return IRQ_HANDLED;
1233 }
1234 
1235 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1236 {
1237 	xennet_tx_interrupt(irq, dev_id);
1238 	xennet_rx_interrupt(irq, dev_id);
1239 	return IRQ_HANDLED;
1240 }
1241 
1242 #ifdef CONFIG_NET_POLL_CONTROLLER
1243 static void xennet_poll_controller(struct net_device *dev)
1244 {
1245 	/* Poll each queue */
1246 	struct netfront_info *info = netdev_priv(dev);
1247 	unsigned int num_queues = dev->real_num_tx_queues;
1248 	unsigned int i;
1249 	for (i = 0; i < num_queues; ++i)
1250 		xennet_interrupt(0, &info->queues[i]);
1251 }
1252 #endif
1253 
1254 static const struct net_device_ops xennet_netdev_ops = {
1255 	.ndo_open            = xennet_open,
1256 	.ndo_stop            = xennet_close,
1257 	.ndo_start_xmit      = xennet_start_xmit,
1258 	.ndo_change_mtu	     = xennet_change_mtu,
1259 	.ndo_get_stats64     = xennet_get_stats64,
1260 	.ndo_set_mac_address = eth_mac_addr,
1261 	.ndo_validate_addr   = eth_validate_addr,
1262 	.ndo_fix_features    = xennet_fix_features,
1263 	.ndo_set_features    = xennet_set_features,
1264 	.ndo_select_queue    = xennet_select_queue,
1265 #ifdef CONFIG_NET_POLL_CONTROLLER
1266 	.ndo_poll_controller = xennet_poll_controller,
1267 #endif
1268 };
1269 
1270 static void xennet_free_netdev(struct net_device *netdev)
1271 {
1272 	struct netfront_info *np = netdev_priv(netdev);
1273 
1274 	free_percpu(np->rx_stats);
1275 	free_percpu(np->tx_stats);
1276 	free_netdev(netdev);
1277 }
1278 
1279 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1280 {
1281 	int err;
1282 	struct net_device *netdev;
1283 	struct netfront_info *np;
1284 
1285 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1286 	if (!netdev)
1287 		return ERR_PTR(-ENOMEM);
1288 
1289 	np                   = netdev_priv(netdev);
1290 	np->xbdev            = dev;
1291 
1292 	np->queues = NULL;
1293 
1294 	err = -ENOMEM;
1295 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1296 	if (np->rx_stats == NULL)
1297 		goto exit;
1298 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1299 	if (np->tx_stats == NULL)
1300 		goto exit;
1301 
1302 	netdev->netdev_ops	= &xennet_netdev_ops;
1303 
1304 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1305 				  NETIF_F_GSO_ROBUST;
1306 	netdev->hw_features	= NETIF_F_SG |
1307 				  NETIF_F_IPV6_CSUM |
1308 				  NETIF_F_TSO | NETIF_F_TSO6;
1309 
1310 	/*
1311          * Assume that all hw features are available for now. This set
1312          * will be adjusted by the call to netdev_update_features() in
1313          * xennet_connect() which is the earliest point where we can
1314          * negotiate with the backend regarding supported features.
1315          */
1316 	netdev->features |= netdev->hw_features;
1317 
1318 	netdev->ethtool_ops = &xennet_ethtool_ops;
1319 	netdev->min_mtu = 0;
1320 	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1321 	SET_NETDEV_DEV(netdev, &dev->dev);
1322 
1323 	np->netdev = netdev;
1324 
1325 	netif_carrier_off(netdev);
1326 
1327 	return netdev;
1328 
1329  exit:
1330 	xennet_free_netdev(netdev);
1331 	return ERR_PTR(err);
1332 }
1333 
1334 /**
1335  * Entry point to this code when a new device is created.  Allocate the basic
1336  * structures and the ring buffers for communication with the backend, and
1337  * inform the backend of the appropriate details for those.
1338  */
1339 static int netfront_probe(struct xenbus_device *dev,
1340 			  const struct xenbus_device_id *id)
1341 {
1342 	int err;
1343 	struct net_device *netdev;
1344 	struct netfront_info *info;
1345 
1346 	netdev = xennet_create_dev(dev);
1347 	if (IS_ERR(netdev)) {
1348 		err = PTR_ERR(netdev);
1349 		xenbus_dev_fatal(dev, err, "creating netdev");
1350 		return err;
1351 	}
1352 
1353 	info = netdev_priv(netdev);
1354 	dev_set_drvdata(&dev->dev, info);
1355 #ifdef CONFIG_SYSFS
1356 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1357 #endif
1358 	err = register_netdev(info->netdev);
1359 	if (err) {
1360 		pr_warn("%s: register_netdev err=%d\n", __func__, err);
1361 		goto fail;
1362 	}
1363 
1364 	return 0;
1365 
1366  fail:
1367 	xennet_free_netdev(netdev);
1368 	dev_set_drvdata(&dev->dev, NULL);
1369 	return err;
1370 }
1371 
1372 static void xennet_end_access(int ref, void *page)
1373 {
1374 	/* This frees the page as a side-effect */
1375 	if (ref != GRANT_INVALID_REF)
1376 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1377 }
1378 
1379 static void xennet_disconnect_backend(struct netfront_info *info)
1380 {
1381 	unsigned int i = 0;
1382 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1383 
1384 	netif_carrier_off(info->netdev);
1385 
1386 	for (i = 0; i < num_queues && info->queues; ++i) {
1387 		struct netfront_queue *queue = &info->queues[i];
1388 
1389 		del_timer_sync(&queue->rx_refill_timer);
1390 
1391 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1392 			unbind_from_irqhandler(queue->tx_irq, queue);
1393 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1394 			unbind_from_irqhandler(queue->tx_irq, queue);
1395 			unbind_from_irqhandler(queue->rx_irq, queue);
1396 		}
1397 		queue->tx_evtchn = queue->rx_evtchn = 0;
1398 		queue->tx_irq = queue->rx_irq = 0;
1399 
1400 		if (netif_running(info->netdev))
1401 			napi_synchronize(&queue->napi);
1402 
1403 		xennet_release_tx_bufs(queue);
1404 		xennet_release_rx_bufs(queue);
1405 		gnttab_free_grant_references(queue->gref_tx_head);
1406 		gnttab_free_grant_references(queue->gref_rx_head);
1407 
1408 		/* End access and free the pages */
1409 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1410 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1411 
1412 		queue->tx_ring_ref = GRANT_INVALID_REF;
1413 		queue->rx_ring_ref = GRANT_INVALID_REF;
1414 		queue->tx.sring = NULL;
1415 		queue->rx.sring = NULL;
1416 	}
1417 }
1418 
1419 /**
1420  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1421  * driver restart.  We tear down our netif structure and recreate it, but
1422  * leave the device-layer structures intact so that this is transparent to the
1423  * rest of the kernel.
1424  */
1425 static int netfront_resume(struct xenbus_device *dev)
1426 {
1427 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1428 
1429 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1430 
1431 	xennet_disconnect_backend(info);
1432 	return 0;
1433 }
1434 
1435 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1436 {
1437 	char *s, *e, *macstr;
1438 	int i;
1439 
1440 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1441 	if (IS_ERR(macstr))
1442 		return PTR_ERR(macstr);
1443 
1444 	for (i = 0; i < ETH_ALEN; i++) {
1445 		mac[i] = simple_strtoul(s, &e, 16);
1446 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1447 			kfree(macstr);
1448 			return -ENOENT;
1449 		}
1450 		s = e+1;
1451 	}
1452 
1453 	kfree(macstr);
1454 	return 0;
1455 }
1456 
1457 static int setup_netfront_single(struct netfront_queue *queue)
1458 {
1459 	int err;
1460 
1461 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1462 	if (err < 0)
1463 		goto fail;
1464 
1465 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1466 					xennet_interrupt,
1467 					0, queue->info->netdev->name, queue);
1468 	if (err < 0)
1469 		goto bind_fail;
1470 	queue->rx_evtchn = queue->tx_evtchn;
1471 	queue->rx_irq = queue->tx_irq = err;
1472 
1473 	return 0;
1474 
1475 bind_fail:
1476 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1477 	queue->tx_evtchn = 0;
1478 fail:
1479 	return err;
1480 }
1481 
1482 static int setup_netfront_split(struct netfront_queue *queue)
1483 {
1484 	int err;
1485 
1486 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1487 	if (err < 0)
1488 		goto fail;
1489 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1490 	if (err < 0)
1491 		goto alloc_rx_evtchn_fail;
1492 
1493 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1494 		 "%s-tx", queue->name);
1495 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1496 					xennet_tx_interrupt,
1497 					0, queue->tx_irq_name, queue);
1498 	if (err < 0)
1499 		goto bind_tx_fail;
1500 	queue->tx_irq = err;
1501 
1502 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1503 		 "%s-rx", queue->name);
1504 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1505 					xennet_rx_interrupt,
1506 					0, queue->rx_irq_name, queue);
1507 	if (err < 0)
1508 		goto bind_rx_fail;
1509 	queue->rx_irq = err;
1510 
1511 	return 0;
1512 
1513 bind_rx_fail:
1514 	unbind_from_irqhandler(queue->tx_irq, queue);
1515 	queue->tx_irq = 0;
1516 bind_tx_fail:
1517 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1518 	queue->rx_evtchn = 0;
1519 alloc_rx_evtchn_fail:
1520 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1521 	queue->tx_evtchn = 0;
1522 fail:
1523 	return err;
1524 }
1525 
1526 static int setup_netfront(struct xenbus_device *dev,
1527 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1528 {
1529 	struct xen_netif_tx_sring *txs;
1530 	struct xen_netif_rx_sring *rxs;
1531 	grant_ref_t gref;
1532 	int err;
1533 
1534 	queue->tx_ring_ref = GRANT_INVALID_REF;
1535 	queue->rx_ring_ref = GRANT_INVALID_REF;
1536 	queue->rx.sring = NULL;
1537 	queue->tx.sring = NULL;
1538 
1539 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1540 	if (!txs) {
1541 		err = -ENOMEM;
1542 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1543 		goto fail;
1544 	}
1545 	SHARED_RING_INIT(txs);
1546 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1547 
1548 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1549 	if (err < 0)
1550 		goto grant_tx_ring_fail;
1551 	queue->tx_ring_ref = gref;
1552 
1553 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1554 	if (!rxs) {
1555 		err = -ENOMEM;
1556 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1557 		goto alloc_rx_ring_fail;
1558 	}
1559 	SHARED_RING_INIT(rxs);
1560 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1561 
1562 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1563 	if (err < 0)
1564 		goto grant_rx_ring_fail;
1565 	queue->rx_ring_ref = gref;
1566 
1567 	if (feature_split_evtchn)
1568 		err = setup_netfront_split(queue);
1569 	/* setup single event channel if
1570 	 *  a) feature-split-event-channels == 0
1571 	 *  b) feature-split-event-channels == 1 but failed to setup
1572 	 */
1573 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1574 		err = setup_netfront_single(queue);
1575 
1576 	if (err)
1577 		goto alloc_evtchn_fail;
1578 
1579 	return 0;
1580 
1581 	/* If we fail to setup netfront, it is safe to just revoke access to
1582 	 * granted pages because backend is not accessing it at this point.
1583 	 */
1584 alloc_evtchn_fail:
1585 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1586 grant_rx_ring_fail:
1587 	free_page((unsigned long)rxs);
1588 alloc_rx_ring_fail:
1589 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1590 grant_tx_ring_fail:
1591 	free_page((unsigned long)txs);
1592 fail:
1593 	return err;
1594 }
1595 
1596 /* Queue-specific initialisation
1597  * This used to be done in xennet_create_dev() but must now
1598  * be run per-queue.
1599  */
1600 static int xennet_init_queue(struct netfront_queue *queue)
1601 {
1602 	unsigned short i;
1603 	int err = 0;
1604 
1605 	spin_lock_init(&queue->tx_lock);
1606 	spin_lock_init(&queue->rx_lock);
1607 
1608 	setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1609 		    (unsigned long)queue);
1610 
1611 	snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1612 		 queue->info->netdev->name, queue->id);
1613 
1614 	/* Initialise tx_skbs as a free chain containing every entry. */
1615 	queue->tx_skb_freelist = 0;
1616 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1617 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1618 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1619 		queue->grant_tx_page[i] = NULL;
1620 	}
1621 
1622 	/* Clear out rx_skbs */
1623 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1624 		queue->rx_skbs[i] = NULL;
1625 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1626 	}
1627 
1628 	/* A grant for every tx ring slot */
1629 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1630 					  &queue->gref_tx_head) < 0) {
1631 		pr_alert("can't alloc tx grant refs\n");
1632 		err = -ENOMEM;
1633 		goto exit;
1634 	}
1635 
1636 	/* A grant for every rx ring slot */
1637 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1638 					  &queue->gref_rx_head) < 0) {
1639 		pr_alert("can't alloc rx grant refs\n");
1640 		err = -ENOMEM;
1641 		goto exit_free_tx;
1642 	}
1643 
1644 	return 0;
1645 
1646  exit_free_tx:
1647 	gnttab_free_grant_references(queue->gref_tx_head);
1648  exit:
1649 	return err;
1650 }
1651 
1652 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1653 			   struct xenbus_transaction *xbt, int write_hierarchical)
1654 {
1655 	/* Write the queue-specific keys into XenStore in the traditional
1656 	 * way for a single queue, or in a queue subkeys for multiple
1657 	 * queues.
1658 	 */
1659 	struct xenbus_device *dev = queue->info->xbdev;
1660 	int err;
1661 	const char *message;
1662 	char *path;
1663 	size_t pathsize;
1664 
1665 	/* Choose the correct place to write the keys */
1666 	if (write_hierarchical) {
1667 		pathsize = strlen(dev->nodename) + 10;
1668 		path = kzalloc(pathsize, GFP_KERNEL);
1669 		if (!path) {
1670 			err = -ENOMEM;
1671 			message = "out of memory while writing ring references";
1672 			goto error;
1673 		}
1674 		snprintf(path, pathsize, "%s/queue-%u",
1675 				dev->nodename, queue->id);
1676 	} else {
1677 		path = (char *)dev->nodename;
1678 	}
1679 
1680 	/* Write ring references */
1681 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1682 			queue->tx_ring_ref);
1683 	if (err) {
1684 		message = "writing tx-ring-ref";
1685 		goto error;
1686 	}
1687 
1688 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1689 			queue->rx_ring_ref);
1690 	if (err) {
1691 		message = "writing rx-ring-ref";
1692 		goto error;
1693 	}
1694 
1695 	/* Write event channels; taking into account both shared
1696 	 * and split event channel scenarios.
1697 	 */
1698 	if (queue->tx_evtchn == queue->rx_evtchn) {
1699 		/* Shared event channel */
1700 		err = xenbus_printf(*xbt, path,
1701 				"event-channel", "%u", queue->tx_evtchn);
1702 		if (err) {
1703 			message = "writing event-channel";
1704 			goto error;
1705 		}
1706 	} else {
1707 		/* Split event channels */
1708 		err = xenbus_printf(*xbt, path,
1709 				"event-channel-tx", "%u", queue->tx_evtchn);
1710 		if (err) {
1711 			message = "writing event-channel-tx";
1712 			goto error;
1713 		}
1714 
1715 		err = xenbus_printf(*xbt, path,
1716 				"event-channel-rx", "%u", queue->rx_evtchn);
1717 		if (err) {
1718 			message = "writing event-channel-rx";
1719 			goto error;
1720 		}
1721 	}
1722 
1723 	if (write_hierarchical)
1724 		kfree(path);
1725 	return 0;
1726 
1727 error:
1728 	if (write_hierarchical)
1729 		kfree(path);
1730 	xenbus_dev_fatal(dev, err, "%s", message);
1731 	return err;
1732 }
1733 
1734 static void xennet_destroy_queues(struct netfront_info *info)
1735 {
1736 	unsigned int i;
1737 
1738 	rtnl_lock();
1739 
1740 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1741 		struct netfront_queue *queue = &info->queues[i];
1742 
1743 		if (netif_running(info->netdev))
1744 			napi_disable(&queue->napi);
1745 		netif_napi_del(&queue->napi);
1746 	}
1747 
1748 	rtnl_unlock();
1749 
1750 	kfree(info->queues);
1751 	info->queues = NULL;
1752 }
1753 
1754 static int xennet_create_queues(struct netfront_info *info,
1755 				unsigned int *num_queues)
1756 {
1757 	unsigned int i;
1758 	int ret;
1759 
1760 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1761 			       GFP_KERNEL);
1762 	if (!info->queues)
1763 		return -ENOMEM;
1764 
1765 	rtnl_lock();
1766 
1767 	for (i = 0; i < *num_queues; i++) {
1768 		struct netfront_queue *queue = &info->queues[i];
1769 
1770 		queue->id = i;
1771 		queue->info = info;
1772 
1773 		ret = xennet_init_queue(queue);
1774 		if (ret < 0) {
1775 			dev_warn(&info->netdev->dev,
1776 				 "only created %d queues\n", i);
1777 			*num_queues = i;
1778 			break;
1779 		}
1780 
1781 		netif_napi_add(queue->info->netdev, &queue->napi,
1782 			       xennet_poll, 64);
1783 		if (netif_running(info->netdev))
1784 			napi_enable(&queue->napi);
1785 	}
1786 
1787 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
1788 
1789 	rtnl_unlock();
1790 
1791 	if (*num_queues == 0) {
1792 		dev_err(&info->netdev->dev, "no queues\n");
1793 		return -EINVAL;
1794 	}
1795 	return 0;
1796 }
1797 
1798 /* Common code used when first setting up, and when resuming. */
1799 static int talk_to_netback(struct xenbus_device *dev,
1800 			   struct netfront_info *info)
1801 {
1802 	const char *message;
1803 	struct xenbus_transaction xbt;
1804 	int err;
1805 	unsigned int feature_split_evtchn;
1806 	unsigned int i = 0;
1807 	unsigned int max_queues = 0;
1808 	struct netfront_queue *queue = NULL;
1809 	unsigned int num_queues = 1;
1810 
1811 	info->netdev->irq = 0;
1812 
1813 	/* Check if backend supports multiple queues */
1814 	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1815 					  "multi-queue-max-queues", 1);
1816 	num_queues = min(max_queues, xennet_max_queues);
1817 
1818 	/* Check feature-split-event-channels */
1819 	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1820 					"feature-split-event-channels", 0);
1821 
1822 	/* Read mac addr. */
1823 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
1824 	if (err) {
1825 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1826 		goto out;
1827 	}
1828 
1829 	if (info->queues)
1830 		xennet_destroy_queues(info);
1831 
1832 	err = xennet_create_queues(info, &num_queues);
1833 	if (err < 0) {
1834 		xenbus_dev_fatal(dev, err, "creating queues");
1835 		kfree(info->queues);
1836 		info->queues = NULL;
1837 		goto out;
1838 	}
1839 
1840 	/* Create shared ring, alloc event channel -- for each queue */
1841 	for (i = 0; i < num_queues; ++i) {
1842 		queue = &info->queues[i];
1843 		err = setup_netfront(dev, queue, feature_split_evtchn);
1844 		if (err)
1845 			goto destroy_ring;
1846 	}
1847 
1848 again:
1849 	err = xenbus_transaction_start(&xbt);
1850 	if (err) {
1851 		xenbus_dev_fatal(dev, err, "starting transaction");
1852 		goto destroy_ring;
1853 	}
1854 
1855 	if (xenbus_exists(XBT_NIL,
1856 			  info->xbdev->otherend, "multi-queue-max-queues")) {
1857 		/* Write the number of queues */
1858 		err = xenbus_printf(xbt, dev->nodename,
1859 				    "multi-queue-num-queues", "%u", num_queues);
1860 		if (err) {
1861 			message = "writing multi-queue-num-queues";
1862 			goto abort_transaction_no_dev_fatal;
1863 		}
1864 	}
1865 
1866 	if (num_queues == 1) {
1867 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1868 		if (err)
1869 			goto abort_transaction_no_dev_fatal;
1870 	} else {
1871 		/* Write the keys for each queue */
1872 		for (i = 0; i < num_queues; ++i) {
1873 			queue = &info->queues[i];
1874 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1875 			if (err)
1876 				goto abort_transaction_no_dev_fatal;
1877 		}
1878 	}
1879 
1880 	/* The remaining keys are not queue-specific */
1881 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1882 			    1);
1883 	if (err) {
1884 		message = "writing request-rx-copy";
1885 		goto abort_transaction;
1886 	}
1887 
1888 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1889 	if (err) {
1890 		message = "writing feature-rx-notify";
1891 		goto abort_transaction;
1892 	}
1893 
1894 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1895 	if (err) {
1896 		message = "writing feature-sg";
1897 		goto abort_transaction;
1898 	}
1899 
1900 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1901 	if (err) {
1902 		message = "writing feature-gso-tcpv4";
1903 		goto abort_transaction;
1904 	}
1905 
1906 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1907 	if (err) {
1908 		message = "writing feature-gso-tcpv6";
1909 		goto abort_transaction;
1910 	}
1911 
1912 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1913 			   "1");
1914 	if (err) {
1915 		message = "writing feature-ipv6-csum-offload";
1916 		goto abort_transaction;
1917 	}
1918 
1919 	err = xenbus_transaction_end(xbt, 0);
1920 	if (err) {
1921 		if (err == -EAGAIN)
1922 			goto again;
1923 		xenbus_dev_fatal(dev, err, "completing transaction");
1924 		goto destroy_ring;
1925 	}
1926 
1927 	return 0;
1928 
1929  abort_transaction:
1930 	xenbus_dev_fatal(dev, err, "%s", message);
1931 abort_transaction_no_dev_fatal:
1932 	xenbus_transaction_end(xbt, 1);
1933  destroy_ring:
1934 	xennet_disconnect_backend(info);
1935 	xennet_destroy_queues(info);
1936  out:
1937 	unregister_netdev(info->netdev);
1938 	xennet_free_netdev(info->netdev);
1939 	return err;
1940 }
1941 
1942 static int xennet_connect(struct net_device *dev)
1943 {
1944 	struct netfront_info *np = netdev_priv(dev);
1945 	unsigned int num_queues = 0;
1946 	int err;
1947 	unsigned int j = 0;
1948 	struct netfront_queue *queue = NULL;
1949 
1950 	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1951 		dev_info(&dev->dev,
1952 			 "backend does not support copying receive path\n");
1953 		return -ENODEV;
1954 	}
1955 
1956 	err = talk_to_netback(np->xbdev, np);
1957 	if (err)
1958 		return err;
1959 
1960 	/* talk_to_netback() sets the correct number of queues */
1961 	num_queues = dev->real_num_tx_queues;
1962 
1963 	rtnl_lock();
1964 	netdev_update_features(dev);
1965 	rtnl_unlock();
1966 
1967 	/*
1968 	 * All public and private state should now be sane.  Get
1969 	 * ready to start sending and receiving packets and give the driver
1970 	 * domain a kick because we've probably just requeued some
1971 	 * packets.
1972 	 */
1973 	netif_carrier_on(np->netdev);
1974 	for (j = 0; j < num_queues; ++j) {
1975 		queue = &np->queues[j];
1976 
1977 		notify_remote_via_irq(queue->tx_irq);
1978 		if (queue->tx_irq != queue->rx_irq)
1979 			notify_remote_via_irq(queue->rx_irq);
1980 
1981 		spin_lock_irq(&queue->tx_lock);
1982 		xennet_tx_buf_gc(queue);
1983 		spin_unlock_irq(&queue->tx_lock);
1984 
1985 		spin_lock_bh(&queue->rx_lock);
1986 		xennet_alloc_rx_buffers(queue);
1987 		spin_unlock_bh(&queue->rx_lock);
1988 	}
1989 
1990 	return 0;
1991 }
1992 
1993 /**
1994  * Callback received when the backend's state changes.
1995  */
1996 static void netback_changed(struct xenbus_device *dev,
1997 			    enum xenbus_state backend_state)
1998 {
1999 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2000 	struct net_device *netdev = np->netdev;
2001 
2002 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2003 
2004 	switch (backend_state) {
2005 	case XenbusStateInitialising:
2006 	case XenbusStateInitialised:
2007 	case XenbusStateReconfiguring:
2008 	case XenbusStateReconfigured:
2009 	case XenbusStateUnknown:
2010 		break;
2011 
2012 	case XenbusStateInitWait:
2013 		if (dev->state != XenbusStateInitialising)
2014 			break;
2015 		if (xennet_connect(netdev) != 0)
2016 			break;
2017 		xenbus_switch_state(dev, XenbusStateConnected);
2018 		break;
2019 
2020 	case XenbusStateConnected:
2021 		netdev_notify_peers(netdev);
2022 		break;
2023 
2024 	case XenbusStateClosed:
2025 		if (dev->state == XenbusStateClosed)
2026 			break;
2027 		/* Missed the backend's CLOSING state -- fallthrough */
2028 	case XenbusStateClosing:
2029 		xenbus_frontend_closed(dev);
2030 		break;
2031 	}
2032 }
2033 
2034 static const struct xennet_stat {
2035 	char name[ETH_GSTRING_LEN];
2036 	u16 offset;
2037 } xennet_stats[] = {
2038 	{
2039 		"rx_gso_checksum_fixup",
2040 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2041 	},
2042 };
2043 
2044 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2045 {
2046 	switch (string_set) {
2047 	case ETH_SS_STATS:
2048 		return ARRAY_SIZE(xennet_stats);
2049 	default:
2050 		return -EINVAL;
2051 	}
2052 }
2053 
2054 static void xennet_get_ethtool_stats(struct net_device *dev,
2055 				     struct ethtool_stats *stats, u64 * data)
2056 {
2057 	void *np = netdev_priv(dev);
2058 	int i;
2059 
2060 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2061 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2062 }
2063 
2064 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2065 {
2066 	int i;
2067 
2068 	switch (stringset) {
2069 	case ETH_SS_STATS:
2070 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2071 			memcpy(data + i * ETH_GSTRING_LEN,
2072 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2073 		break;
2074 	}
2075 }
2076 
2077 static const struct ethtool_ops xennet_ethtool_ops =
2078 {
2079 	.get_link = ethtool_op_get_link,
2080 
2081 	.get_sset_count = xennet_get_sset_count,
2082 	.get_ethtool_stats = xennet_get_ethtool_stats,
2083 	.get_strings = xennet_get_strings,
2084 };
2085 
2086 #ifdef CONFIG_SYSFS
2087 static ssize_t show_rxbuf(struct device *dev,
2088 			  struct device_attribute *attr, char *buf)
2089 {
2090 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2091 }
2092 
2093 static ssize_t store_rxbuf(struct device *dev,
2094 			   struct device_attribute *attr,
2095 			   const char *buf, size_t len)
2096 {
2097 	char *endp;
2098 	unsigned long target;
2099 
2100 	if (!capable(CAP_NET_ADMIN))
2101 		return -EPERM;
2102 
2103 	target = simple_strtoul(buf, &endp, 0);
2104 	if (endp == buf)
2105 		return -EBADMSG;
2106 
2107 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2108 
2109 	return len;
2110 }
2111 
2112 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2113 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2114 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2115 
2116 static struct attribute *xennet_dev_attrs[] = {
2117 	&dev_attr_rxbuf_min.attr,
2118 	&dev_attr_rxbuf_max.attr,
2119 	&dev_attr_rxbuf_cur.attr,
2120 	NULL
2121 };
2122 
2123 static const struct attribute_group xennet_dev_group = {
2124 	.attrs = xennet_dev_attrs
2125 };
2126 #endif /* CONFIG_SYSFS */
2127 
2128 static int xennet_remove(struct xenbus_device *dev)
2129 {
2130 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2131 
2132 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2133 
2134 	xennet_disconnect_backend(info);
2135 
2136 	unregister_netdev(info->netdev);
2137 
2138 	if (info->queues)
2139 		xennet_destroy_queues(info);
2140 	xennet_free_netdev(info->netdev);
2141 
2142 	return 0;
2143 }
2144 
2145 static const struct xenbus_device_id netfront_ids[] = {
2146 	{ "vif" },
2147 	{ "" }
2148 };
2149 
2150 static struct xenbus_driver netfront_driver = {
2151 	.ids = netfront_ids,
2152 	.probe = netfront_probe,
2153 	.remove = xennet_remove,
2154 	.resume = netfront_resume,
2155 	.otherend_changed = netback_changed,
2156 };
2157 
2158 static int __init netif_init(void)
2159 {
2160 	if (!xen_domain())
2161 		return -ENODEV;
2162 
2163 	if (!xen_has_pv_nic_devices())
2164 		return -ENODEV;
2165 
2166 	pr_info("Initialising Xen virtual ethernet driver\n");
2167 
2168 	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2169 	 * specified a value.
2170 	 */
2171 	if (xennet_max_queues == 0)
2172 		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2173 					  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