xref: /openbmc/linux/drivers/net/xen-netfront.c (revision 8c749ce9)
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 
285 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
286 		return;
287 
288 	for (req_prod = queue->rx.req_prod_pvt;
289 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
290 	     req_prod++) {
291 		struct sk_buff *skb;
292 		unsigned short id;
293 		grant_ref_t ref;
294 		struct page *page;
295 		struct xen_netif_rx_request *req;
296 
297 		skb = xennet_alloc_one_rx_buffer(queue);
298 		if (!skb)
299 			break;
300 
301 		id = xennet_rxidx(req_prod);
302 
303 		BUG_ON(queue->rx_skbs[id]);
304 		queue->rx_skbs[id] = skb;
305 
306 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
307 		BUG_ON((signed short)ref < 0);
308 		queue->grant_rx_ref[id] = ref;
309 
310 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
311 
312 		req = RING_GET_REQUEST(&queue->rx, req_prod);
313 		gnttab_page_grant_foreign_access_ref_one(ref,
314 							 queue->info->xbdev->otherend_id,
315 							 page,
316 							 0);
317 		req->id = id;
318 		req->gref = ref;
319 	}
320 
321 	queue->rx.req_prod_pvt = req_prod;
322 
323 	/* Not enough requests? Try again later. */
324 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN) {
325 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
326 		return;
327 	}
328 
329 	wmb();		/* barrier so backend seens requests */
330 
331 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
332 	if (notify)
333 		notify_remote_via_irq(queue->rx_irq);
334 }
335 
336 static int xennet_open(struct net_device *dev)
337 {
338 	struct netfront_info *np = netdev_priv(dev);
339 	unsigned int num_queues = dev->real_num_tx_queues;
340 	unsigned int i = 0;
341 	struct netfront_queue *queue = NULL;
342 
343 	for (i = 0; i < num_queues; ++i) {
344 		queue = &np->queues[i];
345 		napi_enable(&queue->napi);
346 
347 		spin_lock_bh(&queue->rx_lock);
348 		if (netif_carrier_ok(dev)) {
349 			xennet_alloc_rx_buffers(queue);
350 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
351 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
352 				napi_schedule(&queue->napi);
353 		}
354 		spin_unlock_bh(&queue->rx_lock);
355 	}
356 
357 	netif_tx_start_all_queues(dev);
358 
359 	return 0;
360 }
361 
362 static void xennet_tx_buf_gc(struct netfront_queue *queue)
363 {
364 	RING_IDX cons, prod;
365 	unsigned short id;
366 	struct sk_buff *skb;
367 
368 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
369 
370 	do {
371 		prod = queue->tx.sring->rsp_prod;
372 		rmb(); /* Ensure we see responses up to 'rp'. */
373 
374 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
375 			struct xen_netif_tx_response *txrsp;
376 
377 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
378 			if (txrsp->status == XEN_NETIF_RSP_NULL)
379 				continue;
380 
381 			id  = txrsp->id;
382 			skb = queue->tx_skbs[id].skb;
383 			if (unlikely(gnttab_query_foreign_access(
384 				queue->grant_tx_ref[id]) != 0)) {
385 				pr_alert("%s: warning -- grant still in use by backend domain\n",
386 					 __func__);
387 				BUG();
388 			}
389 			gnttab_end_foreign_access_ref(
390 				queue->grant_tx_ref[id], GNTMAP_readonly);
391 			gnttab_release_grant_reference(
392 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
393 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
394 			queue->grant_tx_page[id] = NULL;
395 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
396 			dev_kfree_skb_irq(skb);
397 		}
398 
399 		queue->tx.rsp_cons = prod;
400 
401 		/*
402 		 * Set a new event, then check for race with update of tx_cons.
403 		 * Note that it is essential to schedule a callback, no matter
404 		 * how few buffers are pending. Even if there is space in the
405 		 * transmit ring, higher layers may be blocked because too much
406 		 * data is outstanding: in such cases notification from Xen is
407 		 * likely to be the only kick that we'll get.
408 		 */
409 		queue->tx.sring->rsp_event =
410 			prod + ((queue->tx.sring->req_prod - prod) >> 1) + 1;
411 		mb();		/* update shared area */
412 	} while ((cons == prod) && (prod != queue->tx.sring->rsp_prod));
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 	BUG_ON((signed short)ref < 0);
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 
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 	len = skb_headlen(skb);
606 
607 	spin_lock_irqsave(&queue->tx_lock, flags);
608 
609 	if (unlikely(!netif_carrier_ok(dev) ||
610 		     (slots > 1 && !xennet_can_sg(dev)) ||
611 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
612 		spin_unlock_irqrestore(&queue->tx_lock, flags);
613 		goto drop;
614 	}
615 
616 	/* First request for the linear area. */
617 	first_tx = tx = xennet_make_first_txreq(queue, skb,
618 						page, offset, len);
619 	offset += tx->size;
620 	if (offset == PAGE_SIZE) {
621 		page++;
622 		offset = 0;
623 	}
624 	len -= tx->size;
625 
626 	if (skb->ip_summed == CHECKSUM_PARTIAL)
627 		/* local packet? */
628 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
629 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
630 		/* remote but checksummed. */
631 		tx->flags |= XEN_NETTXF_data_validated;
632 
633 	/* Optional extra info after the first request. */
634 	if (skb_shinfo(skb)->gso_size) {
635 		struct xen_netif_extra_info *gso;
636 
637 		gso = (struct xen_netif_extra_info *)
638 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
639 
640 		tx->flags |= XEN_NETTXF_extra_info;
641 
642 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
643 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
644 			XEN_NETIF_GSO_TYPE_TCPV6 :
645 			XEN_NETIF_GSO_TYPE_TCPV4;
646 		gso->u.gso.pad = 0;
647 		gso->u.gso.features = 0;
648 
649 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
650 		gso->flags = 0;
651 	}
652 
653 	/* Requests for the rest of the linear area. */
654 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
655 
656 	/* Requests for all the frags. */
657 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
658 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
659 		tx = xennet_make_txreqs(queue, tx, skb,
660 					skb_frag_page(frag), frag->page_offset,
661 					skb_frag_size(frag));
662 	}
663 
664 	/* First request has the packet length. */
665 	first_tx->size = skb->len;
666 
667 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
668 	if (notify)
669 		notify_remote_via_irq(queue->tx_irq);
670 
671 	u64_stats_update_begin(&tx_stats->syncp);
672 	tx_stats->bytes += skb->len;
673 	tx_stats->packets++;
674 	u64_stats_update_end(&tx_stats->syncp);
675 
676 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
677 	xennet_tx_buf_gc(queue);
678 
679 	if (!netfront_tx_slot_available(queue))
680 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
681 
682 	spin_unlock_irqrestore(&queue->tx_lock, flags);
683 
684 	return NETDEV_TX_OK;
685 
686  drop:
687 	dev->stats.tx_dropped++;
688 	dev_kfree_skb_any(skb);
689 	return NETDEV_TX_OK;
690 }
691 
692 static int xennet_close(struct net_device *dev)
693 {
694 	struct netfront_info *np = netdev_priv(dev);
695 	unsigned int num_queues = dev->real_num_tx_queues;
696 	unsigned int i;
697 	struct netfront_queue *queue;
698 	netif_tx_stop_all_queues(np->netdev);
699 	for (i = 0; i < num_queues; ++i) {
700 		queue = &np->queues[i];
701 		napi_disable(&queue->napi);
702 	}
703 	return 0;
704 }
705 
706 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
707 				grant_ref_t ref)
708 {
709 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
710 
711 	BUG_ON(queue->rx_skbs[new]);
712 	queue->rx_skbs[new] = skb;
713 	queue->grant_rx_ref[new] = ref;
714 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
715 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
716 	queue->rx.req_prod_pvt++;
717 }
718 
719 static int xennet_get_extras(struct netfront_queue *queue,
720 			     struct xen_netif_extra_info *extras,
721 			     RING_IDX rp)
722 
723 {
724 	struct xen_netif_extra_info *extra;
725 	struct device *dev = &queue->info->netdev->dev;
726 	RING_IDX cons = queue->rx.rsp_cons;
727 	int err = 0;
728 
729 	do {
730 		struct sk_buff *skb;
731 		grant_ref_t ref;
732 
733 		if (unlikely(cons + 1 == rp)) {
734 			if (net_ratelimit())
735 				dev_warn(dev, "Missing extra info\n");
736 			err = -EBADR;
737 			break;
738 		}
739 
740 		extra = (struct xen_netif_extra_info *)
741 			RING_GET_RESPONSE(&queue->rx, ++cons);
742 
743 		if (unlikely(!extra->type ||
744 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
745 			if (net_ratelimit())
746 				dev_warn(dev, "Invalid extra type: %d\n",
747 					extra->type);
748 			err = -EINVAL;
749 		} else {
750 			memcpy(&extras[extra->type - 1], extra,
751 			       sizeof(*extra));
752 		}
753 
754 		skb = xennet_get_rx_skb(queue, cons);
755 		ref = xennet_get_rx_ref(queue, cons);
756 		xennet_move_rx_slot(queue, skb, ref);
757 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
758 
759 	queue->rx.rsp_cons = cons;
760 	return err;
761 }
762 
763 static int xennet_get_responses(struct netfront_queue *queue,
764 				struct netfront_rx_info *rinfo, RING_IDX rp,
765 				struct sk_buff_head *list)
766 {
767 	struct xen_netif_rx_response *rx = &rinfo->rx;
768 	struct xen_netif_extra_info *extras = rinfo->extras;
769 	struct device *dev = &queue->info->netdev->dev;
770 	RING_IDX cons = queue->rx.rsp_cons;
771 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
772 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
773 	int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
774 	int slots = 1;
775 	int err = 0;
776 	unsigned long ret;
777 
778 	if (rx->flags & XEN_NETRXF_extra_info) {
779 		err = xennet_get_extras(queue, extras, rp);
780 		cons = queue->rx.rsp_cons;
781 	}
782 
783 	for (;;) {
784 		if (unlikely(rx->status < 0 ||
785 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
786 			if (net_ratelimit())
787 				dev_warn(dev, "rx->offset: %u, size: %d\n",
788 					 rx->offset, rx->status);
789 			xennet_move_rx_slot(queue, skb, ref);
790 			err = -EINVAL;
791 			goto next;
792 		}
793 
794 		/*
795 		 * This definitely indicates a bug, either in this driver or in
796 		 * the backend driver. In future this should flag the bad
797 		 * situation to the system controller to reboot the backend.
798 		 */
799 		if (ref == GRANT_INVALID_REF) {
800 			if (net_ratelimit())
801 				dev_warn(dev, "Bad rx response id %d.\n",
802 					 rx->id);
803 			err = -EINVAL;
804 			goto next;
805 		}
806 
807 		ret = gnttab_end_foreign_access_ref(ref, 0);
808 		BUG_ON(!ret);
809 
810 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
811 
812 		__skb_queue_tail(list, skb);
813 
814 next:
815 		if (!(rx->flags & XEN_NETRXF_more_data))
816 			break;
817 
818 		if (cons + slots == rp) {
819 			if (net_ratelimit())
820 				dev_warn(dev, "Need more slots\n");
821 			err = -ENOENT;
822 			break;
823 		}
824 
825 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
826 		skb = xennet_get_rx_skb(queue, cons + slots);
827 		ref = xennet_get_rx_ref(queue, cons + slots);
828 		slots++;
829 	}
830 
831 	if (unlikely(slots > max)) {
832 		if (net_ratelimit())
833 			dev_warn(dev, "Too many slots\n");
834 		err = -E2BIG;
835 	}
836 
837 	if (unlikely(err))
838 		queue->rx.rsp_cons = cons + slots;
839 
840 	return err;
841 }
842 
843 static int xennet_set_skb_gso(struct sk_buff *skb,
844 			      struct xen_netif_extra_info *gso)
845 {
846 	if (!gso->u.gso.size) {
847 		if (net_ratelimit())
848 			pr_warn("GSO size must not be zero\n");
849 		return -EINVAL;
850 	}
851 
852 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
853 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
854 		if (net_ratelimit())
855 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
856 		return -EINVAL;
857 	}
858 
859 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
860 	skb_shinfo(skb)->gso_type =
861 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
862 		SKB_GSO_TCPV4 :
863 		SKB_GSO_TCPV6;
864 
865 	/* Header must be checked, and gso_segs computed. */
866 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
867 	skb_shinfo(skb)->gso_segs = 0;
868 
869 	return 0;
870 }
871 
872 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
873 				  struct sk_buff *skb,
874 				  struct sk_buff_head *list)
875 {
876 	struct skb_shared_info *shinfo = skb_shinfo(skb);
877 	RING_IDX cons = queue->rx.rsp_cons;
878 	struct sk_buff *nskb;
879 
880 	while ((nskb = __skb_dequeue(list))) {
881 		struct xen_netif_rx_response *rx =
882 			RING_GET_RESPONSE(&queue->rx, ++cons);
883 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
884 
885 		if (shinfo->nr_frags == MAX_SKB_FRAGS) {
886 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
887 
888 			BUG_ON(pull_to <= skb_headlen(skb));
889 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
890 		}
891 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
892 
893 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
894 				rx->offset, rx->status, PAGE_SIZE);
895 
896 		skb_shinfo(nskb)->nr_frags = 0;
897 		kfree_skb(nskb);
898 	}
899 
900 	return cons;
901 }
902 
903 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
904 {
905 	bool recalculate_partial_csum = false;
906 
907 	/*
908 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
909 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
910 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
911 	 * recalculate the partial checksum.
912 	 */
913 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
914 		struct netfront_info *np = netdev_priv(dev);
915 		atomic_inc(&np->rx_gso_checksum_fixup);
916 		skb->ip_summed = CHECKSUM_PARTIAL;
917 		recalculate_partial_csum = true;
918 	}
919 
920 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
921 	if (skb->ip_summed != CHECKSUM_PARTIAL)
922 		return 0;
923 
924 	return skb_checksum_setup(skb, recalculate_partial_csum);
925 }
926 
927 static int handle_incoming_queue(struct netfront_queue *queue,
928 				 struct sk_buff_head *rxq)
929 {
930 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
931 	int packets_dropped = 0;
932 	struct sk_buff *skb;
933 
934 	while ((skb = __skb_dequeue(rxq)) != NULL) {
935 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
936 
937 		if (pull_to > skb_headlen(skb))
938 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
939 
940 		/* Ethernet work: Delayed to here as it peeks the header. */
941 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
942 		skb_reset_network_header(skb);
943 
944 		if (checksum_setup(queue->info->netdev, skb)) {
945 			kfree_skb(skb);
946 			packets_dropped++;
947 			queue->info->netdev->stats.rx_errors++;
948 			continue;
949 		}
950 
951 		u64_stats_update_begin(&rx_stats->syncp);
952 		rx_stats->packets++;
953 		rx_stats->bytes += skb->len;
954 		u64_stats_update_end(&rx_stats->syncp);
955 
956 		/* Pass it up. */
957 		napi_gro_receive(&queue->napi, skb);
958 	}
959 
960 	return packets_dropped;
961 }
962 
963 static int xennet_poll(struct napi_struct *napi, int budget)
964 {
965 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
966 	struct net_device *dev = queue->info->netdev;
967 	struct sk_buff *skb;
968 	struct netfront_rx_info rinfo;
969 	struct xen_netif_rx_response *rx = &rinfo.rx;
970 	struct xen_netif_extra_info *extras = rinfo.extras;
971 	RING_IDX i, rp;
972 	int work_done;
973 	struct sk_buff_head rxq;
974 	struct sk_buff_head errq;
975 	struct sk_buff_head tmpq;
976 	int err;
977 
978 	spin_lock(&queue->rx_lock);
979 
980 	skb_queue_head_init(&rxq);
981 	skb_queue_head_init(&errq);
982 	skb_queue_head_init(&tmpq);
983 
984 	rp = queue->rx.sring->rsp_prod;
985 	rmb(); /* Ensure we see queued responses up to 'rp'. */
986 
987 	i = queue->rx.rsp_cons;
988 	work_done = 0;
989 	while ((i != rp) && (work_done < budget)) {
990 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
991 		memset(extras, 0, sizeof(rinfo.extras));
992 
993 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
994 
995 		if (unlikely(err)) {
996 err:
997 			while ((skb = __skb_dequeue(&tmpq)))
998 				__skb_queue_tail(&errq, skb);
999 			dev->stats.rx_errors++;
1000 			i = queue->rx.rsp_cons;
1001 			continue;
1002 		}
1003 
1004 		skb = __skb_dequeue(&tmpq);
1005 
1006 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1007 			struct xen_netif_extra_info *gso;
1008 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1009 
1010 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1011 				__skb_queue_head(&tmpq, skb);
1012 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1013 				goto err;
1014 			}
1015 		}
1016 
1017 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1018 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1019 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1020 
1021 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1022 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1023 		skb->data_len = rx->status;
1024 		skb->len += rx->status;
1025 
1026 		i = xennet_fill_frags(queue, skb, &tmpq);
1027 
1028 		if (rx->flags & XEN_NETRXF_csum_blank)
1029 			skb->ip_summed = CHECKSUM_PARTIAL;
1030 		else if (rx->flags & XEN_NETRXF_data_validated)
1031 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1032 
1033 		__skb_queue_tail(&rxq, skb);
1034 
1035 		queue->rx.rsp_cons = ++i;
1036 		work_done++;
1037 	}
1038 
1039 	__skb_queue_purge(&errq);
1040 
1041 	work_done -= handle_incoming_queue(queue, &rxq);
1042 
1043 	xennet_alloc_rx_buffers(queue);
1044 
1045 	if (work_done < budget) {
1046 		int more_to_do = 0;
1047 
1048 		napi_complete(napi);
1049 
1050 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1051 		if (more_to_do)
1052 			napi_schedule(napi);
1053 	}
1054 
1055 	spin_unlock(&queue->rx_lock);
1056 
1057 	return work_done;
1058 }
1059 
1060 static int xennet_change_mtu(struct net_device *dev, int mtu)
1061 {
1062 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1063 
1064 	if (mtu > max)
1065 		return -EINVAL;
1066 	dev->mtu = mtu;
1067 	return 0;
1068 }
1069 
1070 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1071 						    struct rtnl_link_stats64 *tot)
1072 {
1073 	struct netfront_info *np = netdev_priv(dev);
1074 	int cpu;
1075 
1076 	for_each_possible_cpu(cpu) {
1077 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1078 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1079 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1080 		unsigned int start;
1081 
1082 		do {
1083 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1084 			tx_packets = tx_stats->packets;
1085 			tx_bytes = tx_stats->bytes;
1086 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1087 
1088 		do {
1089 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1090 			rx_packets = rx_stats->packets;
1091 			rx_bytes = rx_stats->bytes;
1092 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1093 
1094 		tot->rx_packets += rx_packets;
1095 		tot->tx_packets += tx_packets;
1096 		tot->rx_bytes   += rx_bytes;
1097 		tot->tx_bytes   += tx_bytes;
1098 	}
1099 
1100 	tot->rx_errors  = dev->stats.rx_errors;
1101 	tot->tx_dropped = dev->stats.tx_dropped;
1102 
1103 	return tot;
1104 }
1105 
1106 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1107 {
1108 	struct sk_buff *skb;
1109 	int i;
1110 
1111 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1112 		/* Skip over entries which are actually freelist references */
1113 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1114 			continue;
1115 
1116 		skb = queue->tx_skbs[i].skb;
1117 		get_page(queue->grant_tx_page[i]);
1118 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1119 					  GNTMAP_readonly,
1120 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1121 		queue->grant_tx_page[i] = NULL;
1122 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1123 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1124 		dev_kfree_skb_irq(skb);
1125 	}
1126 }
1127 
1128 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1129 {
1130 	int id, ref;
1131 
1132 	spin_lock_bh(&queue->rx_lock);
1133 
1134 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1135 		struct sk_buff *skb;
1136 		struct page *page;
1137 
1138 		skb = queue->rx_skbs[id];
1139 		if (!skb)
1140 			continue;
1141 
1142 		ref = queue->grant_rx_ref[id];
1143 		if (ref == GRANT_INVALID_REF)
1144 			continue;
1145 
1146 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1147 
1148 		/* gnttab_end_foreign_access() needs a page ref until
1149 		 * foreign access is ended (which may be deferred).
1150 		 */
1151 		get_page(page);
1152 		gnttab_end_foreign_access(ref, 0,
1153 					  (unsigned long)page_address(page));
1154 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1155 
1156 		kfree_skb(skb);
1157 	}
1158 
1159 	spin_unlock_bh(&queue->rx_lock);
1160 }
1161 
1162 static netdev_features_t xennet_fix_features(struct net_device *dev,
1163 	netdev_features_t features)
1164 {
1165 	struct netfront_info *np = netdev_priv(dev);
1166 	int val;
1167 
1168 	if (features & NETIF_F_SG) {
1169 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1170 				 "%d", &val) < 0)
1171 			val = 0;
1172 
1173 		if (!val)
1174 			features &= ~NETIF_F_SG;
1175 	}
1176 
1177 	if (features & NETIF_F_IPV6_CSUM) {
1178 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1179 				 "feature-ipv6-csum-offload", "%d", &val) < 0)
1180 			val = 0;
1181 
1182 		if (!val)
1183 			features &= ~NETIF_F_IPV6_CSUM;
1184 	}
1185 
1186 	if (features & NETIF_F_TSO) {
1187 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1188 				 "feature-gso-tcpv4", "%d", &val) < 0)
1189 			val = 0;
1190 
1191 		if (!val)
1192 			features &= ~NETIF_F_TSO;
1193 	}
1194 
1195 	if (features & NETIF_F_TSO6) {
1196 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1197 				 "feature-gso-tcpv6", "%d", &val) < 0)
1198 			val = 0;
1199 
1200 		if (!val)
1201 			features &= ~NETIF_F_TSO6;
1202 	}
1203 
1204 	return features;
1205 }
1206 
1207 static int xennet_set_features(struct net_device *dev,
1208 	netdev_features_t features)
1209 {
1210 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1211 		netdev_info(dev, "Reducing MTU because no SG offload");
1212 		dev->mtu = ETH_DATA_LEN;
1213 	}
1214 
1215 	return 0;
1216 }
1217 
1218 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1219 {
1220 	struct netfront_queue *queue = dev_id;
1221 	unsigned long flags;
1222 
1223 	spin_lock_irqsave(&queue->tx_lock, flags);
1224 	xennet_tx_buf_gc(queue);
1225 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1226 
1227 	return IRQ_HANDLED;
1228 }
1229 
1230 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1231 {
1232 	struct netfront_queue *queue = dev_id;
1233 	struct net_device *dev = queue->info->netdev;
1234 
1235 	if (likely(netif_carrier_ok(dev) &&
1236 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1237 		napi_schedule(&queue->napi);
1238 
1239 	return IRQ_HANDLED;
1240 }
1241 
1242 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1243 {
1244 	xennet_tx_interrupt(irq, dev_id);
1245 	xennet_rx_interrupt(irq, dev_id);
1246 	return IRQ_HANDLED;
1247 }
1248 
1249 #ifdef CONFIG_NET_POLL_CONTROLLER
1250 static void xennet_poll_controller(struct net_device *dev)
1251 {
1252 	/* Poll each queue */
1253 	struct netfront_info *info = netdev_priv(dev);
1254 	unsigned int num_queues = dev->real_num_tx_queues;
1255 	unsigned int i;
1256 	for (i = 0; i < num_queues; ++i)
1257 		xennet_interrupt(0, &info->queues[i]);
1258 }
1259 #endif
1260 
1261 static const struct net_device_ops xennet_netdev_ops = {
1262 	.ndo_open            = xennet_open,
1263 	.ndo_stop            = xennet_close,
1264 	.ndo_start_xmit      = xennet_start_xmit,
1265 	.ndo_change_mtu	     = xennet_change_mtu,
1266 	.ndo_get_stats64     = xennet_get_stats64,
1267 	.ndo_set_mac_address = eth_mac_addr,
1268 	.ndo_validate_addr   = eth_validate_addr,
1269 	.ndo_fix_features    = xennet_fix_features,
1270 	.ndo_set_features    = xennet_set_features,
1271 	.ndo_select_queue    = xennet_select_queue,
1272 #ifdef CONFIG_NET_POLL_CONTROLLER
1273 	.ndo_poll_controller = xennet_poll_controller,
1274 #endif
1275 };
1276 
1277 static void xennet_free_netdev(struct net_device *netdev)
1278 {
1279 	struct netfront_info *np = netdev_priv(netdev);
1280 
1281 	free_percpu(np->rx_stats);
1282 	free_percpu(np->tx_stats);
1283 	free_netdev(netdev);
1284 }
1285 
1286 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1287 {
1288 	int err;
1289 	struct net_device *netdev;
1290 	struct netfront_info *np;
1291 
1292 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1293 	if (!netdev)
1294 		return ERR_PTR(-ENOMEM);
1295 
1296 	np                   = netdev_priv(netdev);
1297 	np->xbdev            = dev;
1298 
1299 	np->queues = NULL;
1300 
1301 	err = -ENOMEM;
1302 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1303 	if (np->rx_stats == NULL)
1304 		goto exit;
1305 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1306 	if (np->tx_stats == NULL)
1307 		goto exit;
1308 
1309 	netdev->netdev_ops	= &xennet_netdev_ops;
1310 
1311 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1312 				  NETIF_F_GSO_ROBUST;
1313 	netdev->hw_features	= NETIF_F_SG |
1314 				  NETIF_F_IPV6_CSUM |
1315 				  NETIF_F_TSO | NETIF_F_TSO6;
1316 
1317 	/*
1318          * Assume that all hw features are available for now. This set
1319          * will be adjusted by the call to netdev_update_features() in
1320          * xennet_connect() which is the earliest point where we can
1321          * negotiate with the backend regarding supported features.
1322          */
1323 	netdev->features |= netdev->hw_features;
1324 
1325 	netdev->ethtool_ops = &xennet_ethtool_ops;
1326 	SET_NETDEV_DEV(netdev, &dev->dev);
1327 
1328 	np->netdev = netdev;
1329 
1330 	netif_carrier_off(netdev);
1331 
1332 	return netdev;
1333 
1334  exit:
1335 	xennet_free_netdev(netdev);
1336 	return ERR_PTR(err);
1337 }
1338 
1339 /**
1340  * Entry point to this code when a new device is created.  Allocate the basic
1341  * structures and the ring buffers for communication with the backend, and
1342  * inform the backend of the appropriate details for those.
1343  */
1344 static int netfront_probe(struct xenbus_device *dev,
1345 			  const struct xenbus_device_id *id)
1346 {
1347 	int err;
1348 	struct net_device *netdev;
1349 	struct netfront_info *info;
1350 
1351 	netdev = xennet_create_dev(dev);
1352 	if (IS_ERR(netdev)) {
1353 		err = PTR_ERR(netdev);
1354 		xenbus_dev_fatal(dev, err, "creating netdev");
1355 		return err;
1356 	}
1357 
1358 	info = netdev_priv(netdev);
1359 	dev_set_drvdata(&dev->dev, info);
1360 #ifdef CONFIG_SYSFS
1361 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1362 #endif
1363 	err = register_netdev(info->netdev);
1364 	if (err) {
1365 		pr_warn("%s: register_netdev err=%d\n", __func__, err);
1366 		goto fail;
1367 	}
1368 
1369 	return 0;
1370 
1371  fail:
1372 	xennet_free_netdev(netdev);
1373 	dev_set_drvdata(&dev->dev, NULL);
1374 	return err;
1375 }
1376 
1377 static void xennet_end_access(int ref, void *page)
1378 {
1379 	/* This frees the page as a side-effect */
1380 	if (ref != GRANT_INVALID_REF)
1381 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1382 }
1383 
1384 static void xennet_disconnect_backend(struct netfront_info *info)
1385 {
1386 	unsigned int i = 0;
1387 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1388 
1389 	netif_carrier_off(info->netdev);
1390 
1391 	for (i = 0; i < num_queues && info->queues; ++i) {
1392 		struct netfront_queue *queue = &info->queues[i];
1393 
1394 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1395 			unbind_from_irqhandler(queue->tx_irq, queue);
1396 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1397 			unbind_from_irqhandler(queue->tx_irq, queue);
1398 			unbind_from_irqhandler(queue->rx_irq, queue);
1399 		}
1400 		queue->tx_evtchn = queue->rx_evtchn = 0;
1401 		queue->tx_irq = queue->rx_irq = 0;
1402 
1403 		if (netif_running(info->netdev))
1404 			napi_synchronize(&queue->napi);
1405 
1406 		xennet_release_tx_bufs(queue);
1407 		xennet_release_rx_bufs(queue);
1408 		gnttab_free_grant_references(queue->gref_tx_head);
1409 		gnttab_free_grant_references(queue->gref_rx_head);
1410 
1411 		/* End access and free the pages */
1412 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1413 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1414 
1415 		queue->tx_ring_ref = GRANT_INVALID_REF;
1416 		queue->rx_ring_ref = GRANT_INVALID_REF;
1417 		queue->tx.sring = NULL;
1418 		queue->rx.sring = NULL;
1419 	}
1420 }
1421 
1422 /**
1423  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1424  * driver restart.  We tear down our netif structure and recreate it, but
1425  * leave the device-layer structures intact so that this is transparent to the
1426  * rest of the kernel.
1427  */
1428 static int netfront_resume(struct xenbus_device *dev)
1429 {
1430 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1431 
1432 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1433 
1434 	xennet_disconnect_backend(info);
1435 	return 0;
1436 }
1437 
1438 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1439 {
1440 	char *s, *e, *macstr;
1441 	int i;
1442 
1443 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1444 	if (IS_ERR(macstr))
1445 		return PTR_ERR(macstr);
1446 
1447 	for (i = 0; i < ETH_ALEN; i++) {
1448 		mac[i] = simple_strtoul(s, &e, 16);
1449 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1450 			kfree(macstr);
1451 			return -ENOENT;
1452 		}
1453 		s = e+1;
1454 	}
1455 
1456 	kfree(macstr);
1457 	return 0;
1458 }
1459 
1460 static int setup_netfront_single(struct netfront_queue *queue)
1461 {
1462 	int err;
1463 
1464 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1465 	if (err < 0)
1466 		goto fail;
1467 
1468 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1469 					xennet_interrupt,
1470 					0, queue->info->netdev->name, queue);
1471 	if (err < 0)
1472 		goto bind_fail;
1473 	queue->rx_evtchn = queue->tx_evtchn;
1474 	queue->rx_irq = queue->tx_irq = err;
1475 
1476 	return 0;
1477 
1478 bind_fail:
1479 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1480 	queue->tx_evtchn = 0;
1481 fail:
1482 	return err;
1483 }
1484 
1485 static int setup_netfront_split(struct netfront_queue *queue)
1486 {
1487 	int err;
1488 
1489 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1490 	if (err < 0)
1491 		goto fail;
1492 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1493 	if (err < 0)
1494 		goto alloc_rx_evtchn_fail;
1495 
1496 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1497 		 "%s-tx", queue->name);
1498 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1499 					xennet_tx_interrupt,
1500 					0, queue->tx_irq_name, queue);
1501 	if (err < 0)
1502 		goto bind_tx_fail;
1503 	queue->tx_irq = err;
1504 
1505 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1506 		 "%s-rx", queue->name);
1507 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1508 					xennet_rx_interrupt,
1509 					0, queue->rx_irq_name, queue);
1510 	if (err < 0)
1511 		goto bind_rx_fail;
1512 	queue->rx_irq = err;
1513 
1514 	return 0;
1515 
1516 bind_rx_fail:
1517 	unbind_from_irqhandler(queue->tx_irq, queue);
1518 	queue->tx_irq = 0;
1519 bind_tx_fail:
1520 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1521 	queue->rx_evtchn = 0;
1522 alloc_rx_evtchn_fail:
1523 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1524 	queue->tx_evtchn = 0;
1525 fail:
1526 	return err;
1527 }
1528 
1529 static int setup_netfront(struct xenbus_device *dev,
1530 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1531 {
1532 	struct xen_netif_tx_sring *txs;
1533 	struct xen_netif_rx_sring *rxs;
1534 	grant_ref_t gref;
1535 	int err;
1536 
1537 	queue->tx_ring_ref = GRANT_INVALID_REF;
1538 	queue->rx_ring_ref = GRANT_INVALID_REF;
1539 	queue->rx.sring = NULL;
1540 	queue->tx.sring = NULL;
1541 
1542 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1543 	if (!txs) {
1544 		err = -ENOMEM;
1545 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1546 		goto fail;
1547 	}
1548 	SHARED_RING_INIT(txs);
1549 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1550 
1551 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1552 	if (err < 0)
1553 		goto grant_tx_ring_fail;
1554 	queue->tx_ring_ref = gref;
1555 
1556 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1557 	if (!rxs) {
1558 		err = -ENOMEM;
1559 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1560 		goto alloc_rx_ring_fail;
1561 	}
1562 	SHARED_RING_INIT(rxs);
1563 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1564 
1565 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1566 	if (err < 0)
1567 		goto grant_rx_ring_fail;
1568 	queue->rx_ring_ref = gref;
1569 
1570 	if (feature_split_evtchn)
1571 		err = setup_netfront_split(queue);
1572 	/* setup single event channel if
1573 	 *  a) feature-split-event-channels == 0
1574 	 *  b) feature-split-event-channels == 1 but failed to setup
1575 	 */
1576 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1577 		err = setup_netfront_single(queue);
1578 
1579 	if (err)
1580 		goto alloc_evtchn_fail;
1581 
1582 	return 0;
1583 
1584 	/* If we fail to setup netfront, it is safe to just revoke access to
1585 	 * granted pages because backend is not accessing it at this point.
1586 	 */
1587 alloc_evtchn_fail:
1588 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1589 grant_rx_ring_fail:
1590 	free_page((unsigned long)rxs);
1591 alloc_rx_ring_fail:
1592 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1593 grant_tx_ring_fail:
1594 	free_page((unsigned long)txs);
1595 fail:
1596 	return err;
1597 }
1598 
1599 /* Queue-specific initialisation
1600  * This used to be done in xennet_create_dev() but must now
1601  * be run per-queue.
1602  */
1603 static int xennet_init_queue(struct netfront_queue *queue)
1604 {
1605 	unsigned short i;
1606 	int err = 0;
1607 
1608 	spin_lock_init(&queue->tx_lock);
1609 	spin_lock_init(&queue->rx_lock);
1610 
1611 	setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1612 		    (unsigned long)queue);
1613 
1614 	snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1615 		 queue->info->netdev->name, queue->id);
1616 
1617 	/* Initialise tx_skbs as a free chain containing every entry. */
1618 	queue->tx_skb_freelist = 0;
1619 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1620 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1621 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1622 		queue->grant_tx_page[i] = NULL;
1623 	}
1624 
1625 	/* Clear out rx_skbs */
1626 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1627 		queue->rx_skbs[i] = NULL;
1628 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1629 	}
1630 
1631 	/* A grant for every tx ring slot */
1632 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1633 					  &queue->gref_tx_head) < 0) {
1634 		pr_alert("can't alloc tx grant refs\n");
1635 		err = -ENOMEM;
1636 		goto exit;
1637 	}
1638 
1639 	/* A grant for every rx ring slot */
1640 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1641 					  &queue->gref_rx_head) < 0) {
1642 		pr_alert("can't alloc rx grant refs\n");
1643 		err = -ENOMEM;
1644 		goto exit_free_tx;
1645 	}
1646 
1647 	return 0;
1648 
1649  exit_free_tx:
1650 	gnttab_free_grant_references(queue->gref_tx_head);
1651  exit:
1652 	return err;
1653 }
1654 
1655 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1656 			   struct xenbus_transaction *xbt, int write_hierarchical)
1657 {
1658 	/* Write the queue-specific keys into XenStore in the traditional
1659 	 * way for a single queue, or in a queue subkeys for multiple
1660 	 * queues.
1661 	 */
1662 	struct xenbus_device *dev = queue->info->xbdev;
1663 	int err;
1664 	const char *message;
1665 	char *path;
1666 	size_t pathsize;
1667 
1668 	/* Choose the correct place to write the keys */
1669 	if (write_hierarchical) {
1670 		pathsize = strlen(dev->nodename) + 10;
1671 		path = kzalloc(pathsize, GFP_KERNEL);
1672 		if (!path) {
1673 			err = -ENOMEM;
1674 			message = "out of memory while writing ring references";
1675 			goto error;
1676 		}
1677 		snprintf(path, pathsize, "%s/queue-%u",
1678 				dev->nodename, queue->id);
1679 	} else {
1680 		path = (char *)dev->nodename;
1681 	}
1682 
1683 	/* Write ring references */
1684 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1685 			queue->tx_ring_ref);
1686 	if (err) {
1687 		message = "writing tx-ring-ref";
1688 		goto error;
1689 	}
1690 
1691 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1692 			queue->rx_ring_ref);
1693 	if (err) {
1694 		message = "writing rx-ring-ref";
1695 		goto error;
1696 	}
1697 
1698 	/* Write event channels; taking into account both shared
1699 	 * and split event channel scenarios.
1700 	 */
1701 	if (queue->tx_evtchn == queue->rx_evtchn) {
1702 		/* Shared event channel */
1703 		err = xenbus_printf(*xbt, path,
1704 				"event-channel", "%u", queue->tx_evtchn);
1705 		if (err) {
1706 			message = "writing event-channel";
1707 			goto error;
1708 		}
1709 	} else {
1710 		/* Split event channels */
1711 		err = xenbus_printf(*xbt, path,
1712 				"event-channel-tx", "%u", queue->tx_evtchn);
1713 		if (err) {
1714 			message = "writing event-channel-tx";
1715 			goto error;
1716 		}
1717 
1718 		err = xenbus_printf(*xbt, path,
1719 				"event-channel-rx", "%u", queue->rx_evtchn);
1720 		if (err) {
1721 			message = "writing event-channel-rx";
1722 			goto error;
1723 		}
1724 	}
1725 
1726 	if (write_hierarchical)
1727 		kfree(path);
1728 	return 0;
1729 
1730 error:
1731 	if (write_hierarchical)
1732 		kfree(path);
1733 	xenbus_dev_fatal(dev, err, "%s", message);
1734 	return err;
1735 }
1736 
1737 static void xennet_destroy_queues(struct netfront_info *info)
1738 {
1739 	unsigned int i;
1740 
1741 	rtnl_lock();
1742 
1743 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1744 		struct netfront_queue *queue = &info->queues[i];
1745 
1746 		if (netif_running(info->netdev))
1747 			napi_disable(&queue->napi);
1748 		del_timer_sync(&queue->rx_refill_timer);
1749 		netif_napi_del(&queue->napi);
1750 	}
1751 
1752 	rtnl_unlock();
1753 
1754 	kfree(info->queues);
1755 	info->queues = NULL;
1756 }
1757 
1758 static int xennet_create_queues(struct netfront_info *info,
1759 				unsigned int *num_queues)
1760 {
1761 	unsigned int i;
1762 	int ret;
1763 
1764 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1765 			       GFP_KERNEL);
1766 	if (!info->queues)
1767 		return -ENOMEM;
1768 
1769 	rtnl_lock();
1770 
1771 	for (i = 0; i < *num_queues; i++) {
1772 		struct netfront_queue *queue = &info->queues[i];
1773 
1774 		queue->id = i;
1775 		queue->info = info;
1776 
1777 		ret = xennet_init_queue(queue);
1778 		if (ret < 0) {
1779 			dev_warn(&info->netdev->dev,
1780 				 "only created %d queues\n", i);
1781 			*num_queues = i;
1782 			break;
1783 		}
1784 
1785 		netif_napi_add(queue->info->netdev, &queue->napi,
1786 			       xennet_poll, 64);
1787 		if (netif_running(info->netdev))
1788 			napi_enable(&queue->napi);
1789 	}
1790 
1791 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
1792 
1793 	rtnl_unlock();
1794 
1795 	if (*num_queues == 0) {
1796 		dev_err(&info->netdev->dev, "no queues\n");
1797 		return -EINVAL;
1798 	}
1799 	return 0;
1800 }
1801 
1802 /* Common code used when first setting up, and when resuming. */
1803 static int talk_to_netback(struct xenbus_device *dev,
1804 			   struct netfront_info *info)
1805 {
1806 	const char *message;
1807 	struct xenbus_transaction xbt;
1808 	int err;
1809 	unsigned int feature_split_evtchn;
1810 	unsigned int i = 0;
1811 	unsigned int max_queues = 0;
1812 	struct netfront_queue *queue = NULL;
1813 	unsigned int num_queues = 1;
1814 
1815 	info->netdev->irq = 0;
1816 
1817 	/* Check if backend supports multiple queues */
1818 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1819 			   "multi-queue-max-queues", "%u", &max_queues);
1820 	if (err < 0)
1821 		max_queues = 1;
1822 	num_queues = min(max_queues, xennet_max_queues);
1823 
1824 	/* Check feature-split-event-channels */
1825 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1826 			   "feature-split-event-channels", "%u",
1827 			   &feature_split_evtchn);
1828 	if (err < 0)
1829 		feature_split_evtchn = 0;
1830 
1831 	/* Read mac addr. */
1832 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
1833 	if (err) {
1834 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1835 		goto out;
1836 	}
1837 
1838 	if (info->queues)
1839 		xennet_destroy_queues(info);
1840 
1841 	err = xennet_create_queues(info, &num_queues);
1842 	if (err < 0)
1843 		goto destroy_ring;
1844 
1845 	/* Create shared ring, alloc event channel -- for each queue */
1846 	for (i = 0; i < num_queues; ++i) {
1847 		queue = &info->queues[i];
1848 		err = setup_netfront(dev, queue, feature_split_evtchn);
1849 		if (err) {
1850 			/* setup_netfront() will tidy up the current
1851 			 * queue on error, but we need to clean up
1852 			 * those already allocated.
1853 			 */
1854 			if (i > 0) {
1855 				rtnl_lock();
1856 				netif_set_real_num_tx_queues(info->netdev, i);
1857 				rtnl_unlock();
1858 				goto destroy_ring;
1859 			} else {
1860 				goto out;
1861 			}
1862 		}
1863 	}
1864 
1865 again:
1866 	err = xenbus_transaction_start(&xbt);
1867 	if (err) {
1868 		xenbus_dev_fatal(dev, err, "starting transaction");
1869 		goto destroy_ring;
1870 	}
1871 
1872 	if (xenbus_exists(XBT_NIL,
1873 			  info->xbdev->otherend, "multi-queue-max-queues")) {
1874 		/* Write the number of queues */
1875 		err = xenbus_printf(xbt, dev->nodename,
1876 				    "multi-queue-num-queues", "%u", num_queues);
1877 		if (err) {
1878 			message = "writing multi-queue-num-queues";
1879 			goto abort_transaction_no_dev_fatal;
1880 		}
1881 	}
1882 
1883 	if (num_queues == 1) {
1884 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1885 		if (err)
1886 			goto abort_transaction_no_dev_fatal;
1887 	} else {
1888 		/* Write the keys for each queue */
1889 		for (i = 0; i < num_queues; ++i) {
1890 			queue = &info->queues[i];
1891 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1892 			if (err)
1893 				goto abort_transaction_no_dev_fatal;
1894 		}
1895 	}
1896 
1897 	/* The remaining keys are not queue-specific */
1898 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1899 			    1);
1900 	if (err) {
1901 		message = "writing request-rx-copy";
1902 		goto abort_transaction;
1903 	}
1904 
1905 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1906 	if (err) {
1907 		message = "writing feature-rx-notify";
1908 		goto abort_transaction;
1909 	}
1910 
1911 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1912 	if (err) {
1913 		message = "writing feature-sg";
1914 		goto abort_transaction;
1915 	}
1916 
1917 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1918 	if (err) {
1919 		message = "writing feature-gso-tcpv4";
1920 		goto abort_transaction;
1921 	}
1922 
1923 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1924 	if (err) {
1925 		message = "writing feature-gso-tcpv6";
1926 		goto abort_transaction;
1927 	}
1928 
1929 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1930 			   "1");
1931 	if (err) {
1932 		message = "writing feature-ipv6-csum-offload";
1933 		goto abort_transaction;
1934 	}
1935 
1936 	err = xenbus_transaction_end(xbt, 0);
1937 	if (err) {
1938 		if (err == -EAGAIN)
1939 			goto again;
1940 		xenbus_dev_fatal(dev, err, "completing transaction");
1941 		goto destroy_ring;
1942 	}
1943 
1944 	return 0;
1945 
1946  abort_transaction:
1947 	xenbus_dev_fatal(dev, err, "%s", message);
1948 abort_transaction_no_dev_fatal:
1949 	xenbus_transaction_end(xbt, 1);
1950  destroy_ring:
1951 	xennet_disconnect_backend(info);
1952 	kfree(info->queues);
1953 	info->queues = NULL;
1954  out:
1955 	return err;
1956 }
1957 
1958 static int xennet_connect(struct net_device *dev)
1959 {
1960 	struct netfront_info *np = netdev_priv(dev);
1961 	unsigned int num_queues = 0;
1962 	int err;
1963 	unsigned int feature_rx_copy;
1964 	unsigned int j = 0;
1965 	struct netfront_queue *queue = NULL;
1966 
1967 	err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1968 			   "feature-rx-copy", "%u", &feature_rx_copy);
1969 	if (err != 1)
1970 		feature_rx_copy = 0;
1971 
1972 	if (!feature_rx_copy) {
1973 		dev_info(&dev->dev,
1974 			 "backend does not support copying receive path\n");
1975 		return -ENODEV;
1976 	}
1977 
1978 	err = talk_to_netback(np->xbdev, np);
1979 	if (err)
1980 		return err;
1981 
1982 	/* talk_to_netback() sets the correct number of queues */
1983 	num_queues = dev->real_num_tx_queues;
1984 
1985 	rtnl_lock();
1986 	netdev_update_features(dev);
1987 	rtnl_unlock();
1988 
1989 	/*
1990 	 * All public and private state should now be sane.  Get
1991 	 * ready to start sending and receiving packets and give the driver
1992 	 * domain a kick because we've probably just requeued some
1993 	 * packets.
1994 	 */
1995 	netif_carrier_on(np->netdev);
1996 	for (j = 0; j < num_queues; ++j) {
1997 		queue = &np->queues[j];
1998 
1999 		notify_remote_via_irq(queue->tx_irq);
2000 		if (queue->tx_irq != queue->rx_irq)
2001 			notify_remote_via_irq(queue->rx_irq);
2002 
2003 		spin_lock_irq(&queue->tx_lock);
2004 		xennet_tx_buf_gc(queue);
2005 		spin_unlock_irq(&queue->tx_lock);
2006 
2007 		spin_lock_bh(&queue->rx_lock);
2008 		xennet_alloc_rx_buffers(queue);
2009 		spin_unlock_bh(&queue->rx_lock);
2010 	}
2011 
2012 	return 0;
2013 }
2014 
2015 /**
2016  * Callback received when the backend's state changes.
2017  */
2018 static void netback_changed(struct xenbus_device *dev,
2019 			    enum xenbus_state backend_state)
2020 {
2021 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2022 	struct net_device *netdev = np->netdev;
2023 
2024 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2025 
2026 	switch (backend_state) {
2027 	case XenbusStateInitialising:
2028 	case XenbusStateInitialised:
2029 	case XenbusStateReconfiguring:
2030 	case XenbusStateReconfigured:
2031 	case XenbusStateUnknown:
2032 		break;
2033 
2034 	case XenbusStateInitWait:
2035 		if (dev->state != XenbusStateInitialising)
2036 			break;
2037 		if (xennet_connect(netdev) != 0)
2038 			break;
2039 		xenbus_switch_state(dev, XenbusStateConnected);
2040 		break;
2041 
2042 	case XenbusStateConnected:
2043 		netdev_notify_peers(netdev);
2044 		break;
2045 
2046 	case XenbusStateClosed:
2047 		if (dev->state == XenbusStateClosed)
2048 			break;
2049 		/* Missed the backend's CLOSING state -- fallthrough */
2050 	case XenbusStateClosing:
2051 		xenbus_frontend_closed(dev);
2052 		break;
2053 	}
2054 }
2055 
2056 static const struct xennet_stat {
2057 	char name[ETH_GSTRING_LEN];
2058 	u16 offset;
2059 } xennet_stats[] = {
2060 	{
2061 		"rx_gso_checksum_fixup",
2062 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2063 	},
2064 };
2065 
2066 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2067 {
2068 	switch (string_set) {
2069 	case ETH_SS_STATS:
2070 		return ARRAY_SIZE(xennet_stats);
2071 	default:
2072 		return -EINVAL;
2073 	}
2074 }
2075 
2076 static void xennet_get_ethtool_stats(struct net_device *dev,
2077 				     struct ethtool_stats *stats, u64 * data)
2078 {
2079 	void *np = netdev_priv(dev);
2080 	int i;
2081 
2082 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2083 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2084 }
2085 
2086 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2087 {
2088 	int i;
2089 
2090 	switch (stringset) {
2091 	case ETH_SS_STATS:
2092 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2093 			memcpy(data + i * ETH_GSTRING_LEN,
2094 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2095 		break;
2096 	}
2097 }
2098 
2099 static const struct ethtool_ops xennet_ethtool_ops =
2100 {
2101 	.get_link = ethtool_op_get_link,
2102 
2103 	.get_sset_count = xennet_get_sset_count,
2104 	.get_ethtool_stats = xennet_get_ethtool_stats,
2105 	.get_strings = xennet_get_strings,
2106 };
2107 
2108 #ifdef CONFIG_SYSFS
2109 static ssize_t show_rxbuf(struct device *dev,
2110 			  struct device_attribute *attr, char *buf)
2111 {
2112 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2113 }
2114 
2115 static ssize_t store_rxbuf(struct device *dev,
2116 			   struct device_attribute *attr,
2117 			   const char *buf, size_t len)
2118 {
2119 	char *endp;
2120 	unsigned long target;
2121 
2122 	if (!capable(CAP_NET_ADMIN))
2123 		return -EPERM;
2124 
2125 	target = simple_strtoul(buf, &endp, 0);
2126 	if (endp == buf)
2127 		return -EBADMSG;
2128 
2129 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2130 
2131 	return len;
2132 }
2133 
2134 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2135 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2136 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2137 
2138 static struct attribute *xennet_dev_attrs[] = {
2139 	&dev_attr_rxbuf_min.attr,
2140 	&dev_attr_rxbuf_max.attr,
2141 	&dev_attr_rxbuf_cur.attr,
2142 	NULL
2143 };
2144 
2145 static const struct attribute_group xennet_dev_group = {
2146 	.attrs = xennet_dev_attrs
2147 };
2148 #endif /* CONFIG_SYSFS */
2149 
2150 static int xennet_remove(struct xenbus_device *dev)
2151 {
2152 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2153 
2154 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2155 
2156 	xennet_disconnect_backend(info);
2157 
2158 	unregister_netdev(info->netdev);
2159 
2160 	if (info->queues)
2161 		xennet_destroy_queues(info);
2162 	xennet_free_netdev(info->netdev);
2163 
2164 	return 0;
2165 }
2166 
2167 static const struct xenbus_device_id netfront_ids[] = {
2168 	{ "vif" },
2169 	{ "" }
2170 };
2171 
2172 static struct xenbus_driver netfront_driver = {
2173 	.ids = netfront_ids,
2174 	.probe = netfront_probe,
2175 	.remove = xennet_remove,
2176 	.resume = netfront_resume,
2177 	.otherend_changed = netback_changed,
2178 };
2179 
2180 static int __init netif_init(void)
2181 {
2182 	if (!xen_domain())
2183 		return -ENODEV;
2184 
2185 	if (!xen_has_pv_nic_devices())
2186 		return -ENODEV;
2187 
2188 	pr_info("Initialising Xen virtual ethernet driver\n");
2189 
2190 	/* Allow as many queues as there are CPUs if user has not
2191 	 * specified a value.
2192 	 */
2193 	if (xennet_max_queues == 0)
2194 		xennet_max_queues = num_online_cpus();
2195 
2196 	return xenbus_register_frontend(&netfront_driver);
2197 }
2198 module_init(netif_init);
2199 
2200 
2201 static void __exit netif_exit(void)
2202 {
2203 	xenbus_unregister_driver(&netfront_driver);
2204 }
2205 module_exit(netif_exit);
2206 
2207 MODULE_DESCRIPTION("Xen virtual network device frontend");
2208 MODULE_LICENSE("GPL");
2209 MODULE_ALIAS("xen:vif");
2210 MODULE_ALIAS("xennet");
2211