xref: /openbmc/linux/drivers/net/xen-netfront.c (revision 78700c0a)
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 	bool more_to_do;
368 
369 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
370 
371 	do {
372 		prod = queue->tx.sring->rsp_prod;
373 		rmb(); /* Ensure we see responses up to 'rp'. */
374 
375 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
376 			struct xen_netif_tx_response *txrsp;
377 
378 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
379 			if (txrsp->status == XEN_NETIF_RSP_NULL)
380 				continue;
381 
382 			id  = txrsp->id;
383 			skb = queue->tx_skbs[id].skb;
384 			if (unlikely(gnttab_query_foreign_access(
385 				queue->grant_tx_ref[id]) != 0)) {
386 				pr_alert("%s: warning -- grant still in use by backend domain\n",
387 					 __func__);
388 				BUG();
389 			}
390 			gnttab_end_foreign_access_ref(
391 				queue->grant_tx_ref[id], GNTMAP_readonly);
392 			gnttab_release_grant_reference(
393 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
394 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
395 			queue->grant_tx_page[id] = NULL;
396 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
397 			dev_kfree_skb_irq(skb);
398 		}
399 
400 		queue->tx.rsp_cons = prod;
401 
402 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
403 	} while (more_to_do);
404 
405 	xennet_maybe_wake_tx(queue);
406 }
407 
408 struct xennet_gnttab_make_txreq {
409 	struct netfront_queue *queue;
410 	struct sk_buff *skb;
411 	struct page *page;
412 	struct xen_netif_tx_request *tx; /* Last request */
413 	unsigned int size;
414 };
415 
416 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
417 				  unsigned int len, void *data)
418 {
419 	struct xennet_gnttab_make_txreq *info = data;
420 	unsigned int id;
421 	struct xen_netif_tx_request *tx;
422 	grant_ref_t ref;
423 	/* convenient aliases */
424 	struct page *page = info->page;
425 	struct netfront_queue *queue = info->queue;
426 	struct sk_buff *skb = info->skb;
427 
428 	id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
429 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
430 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
431 	BUG_ON((signed short)ref < 0);
432 
433 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
434 					gfn, GNTMAP_readonly);
435 
436 	queue->tx_skbs[id].skb = skb;
437 	queue->grant_tx_page[id] = page;
438 	queue->grant_tx_ref[id] = ref;
439 
440 	tx->id = id;
441 	tx->gref = ref;
442 	tx->offset = offset;
443 	tx->size = len;
444 	tx->flags = 0;
445 
446 	info->tx = tx;
447 	info->size += tx->size;
448 }
449 
450 static struct xen_netif_tx_request *xennet_make_first_txreq(
451 	struct netfront_queue *queue, struct sk_buff *skb,
452 	struct page *page, unsigned int offset, unsigned int len)
453 {
454 	struct xennet_gnttab_make_txreq info = {
455 		.queue = queue,
456 		.skb = skb,
457 		.page = page,
458 		.size = 0,
459 	};
460 
461 	gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
462 
463 	return info.tx;
464 }
465 
466 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
467 				  unsigned int len, void *data)
468 {
469 	struct xennet_gnttab_make_txreq *info = data;
470 
471 	info->tx->flags |= XEN_NETTXF_more_data;
472 	skb_get(info->skb);
473 	xennet_tx_setup_grant(gfn, offset, len, data);
474 }
475 
476 static struct xen_netif_tx_request *xennet_make_txreqs(
477 	struct netfront_queue *queue, struct xen_netif_tx_request *tx,
478 	struct sk_buff *skb, struct page *page,
479 	unsigned int offset, unsigned int len)
480 {
481 	struct xennet_gnttab_make_txreq info = {
482 		.queue = queue,
483 		.skb = skb,
484 		.tx = tx,
485 	};
486 
487 	/* Skip unused frames from start of page */
488 	page += offset >> PAGE_SHIFT;
489 	offset &= ~PAGE_MASK;
490 
491 	while (len) {
492 		info.page = page;
493 		info.size = 0;
494 
495 		gnttab_foreach_grant_in_range(page, offset, len,
496 					      xennet_make_one_txreq,
497 					      &info);
498 
499 		page++;
500 		offset = 0;
501 		len -= info.size;
502 	}
503 
504 	return info.tx;
505 }
506 
507 /*
508  * Count how many ring slots are required to send this skb. Each frag
509  * might be a compound page.
510  */
511 static int xennet_count_skb_slots(struct sk_buff *skb)
512 {
513 	int i, frags = skb_shinfo(skb)->nr_frags;
514 	int slots;
515 
516 	slots = gnttab_count_grant(offset_in_page(skb->data),
517 				   skb_headlen(skb));
518 
519 	for (i = 0; i < frags; i++) {
520 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
521 		unsigned long size = skb_frag_size(frag);
522 		unsigned long offset = frag->page_offset;
523 
524 		/* Skip unused frames from start of page */
525 		offset &= ~PAGE_MASK;
526 
527 		slots += gnttab_count_grant(offset, size);
528 	}
529 
530 	return slots;
531 }
532 
533 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
534 			       void *accel_priv, select_queue_fallback_t fallback)
535 {
536 	unsigned int num_queues = dev->real_num_tx_queues;
537 	u32 hash;
538 	u16 queue_idx;
539 
540 	/* First, check if there is only one queue */
541 	if (num_queues == 1) {
542 		queue_idx = 0;
543 	} else {
544 		hash = skb_get_hash(skb);
545 		queue_idx = hash % num_queues;
546 	}
547 
548 	return queue_idx;
549 }
550 
551 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
552 
553 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
554 {
555 	struct netfront_info *np = netdev_priv(dev);
556 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
557 	struct xen_netif_tx_request *tx, *first_tx;
558 	unsigned int i;
559 	int notify;
560 	int slots;
561 	struct page *page;
562 	unsigned int offset;
563 	unsigned int len;
564 	unsigned long flags;
565 	struct netfront_queue *queue = NULL;
566 	unsigned int num_queues = dev->real_num_tx_queues;
567 	u16 queue_index;
568 
569 	/* Drop the packet if no queues are set up */
570 	if (num_queues < 1)
571 		goto drop;
572 	/* Determine which queue to transmit this SKB on */
573 	queue_index = skb_get_queue_mapping(skb);
574 	queue = &np->queues[queue_index];
575 
576 	/* If skb->len is too big for wire format, drop skb and alert
577 	 * user about misconfiguration.
578 	 */
579 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
580 		net_alert_ratelimited(
581 			"xennet: skb->len = %u, too big for wire format\n",
582 			skb->len);
583 		goto drop;
584 	}
585 
586 	slots = xennet_count_skb_slots(skb);
587 	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
588 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
589 				    slots, skb->len);
590 		if (skb_linearize(skb))
591 			goto drop;
592 	}
593 
594 	page = virt_to_page(skb->data);
595 	offset = offset_in_page(skb->data);
596 	len = skb_headlen(skb);
597 
598 	spin_lock_irqsave(&queue->tx_lock, flags);
599 
600 	if (unlikely(!netif_carrier_ok(dev) ||
601 		     (slots > 1 && !xennet_can_sg(dev)) ||
602 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
603 		spin_unlock_irqrestore(&queue->tx_lock, flags);
604 		goto drop;
605 	}
606 
607 	/* First request for the linear area. */
608 	first_tx = tx = xennet_make_first_txreq(queue, skb,
609 						page, offset, len);
610 	offset += tx->size;
611 	if (offset == PAGE_SIZE) {
612 		page++;
613 		offset = 0;
614 	}
615 	len -= tx->size;
616 
617 	if (skb->ip_summed == CHECKSUM_PARTIAL)
618 		/* local packet? */
619 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
620 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
621 		/* remote but checksummed. */
622 		tx->flags |= XEN_NETTXF_data_validated;
623 
624 	/* Optional extra info after the first request. */
625 	if (skb_shinfo(skb)->gso_size) {
626 		struct xen_netif_extra_info *gso;
627 
628 		gso = (struct xen_netif_extra_info *)
629 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
630 
631 		tx->flags |= XEN_NETTXF_extra_info;
632 
633 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
634 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
635 			XEN_NETIF_GSO_TYPE_TCPV6 :
636 			XEN_NETIF_GSO_TYPE_TCPV4;
637 		gso->u.gso.pad = 0;
638 		gso->u.gso.features = 0;
639 
640 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
641 		gso->flags = 0;
642 	}
643 
644 	/* Requests for the rest of the linear area. */
645 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
646 
647 	/* Requests for all the frags. */
648 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
649 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
650 		tx = xennet_make_txreqs(queue, tx, skb,
651 					skb_frag_page(frag), frag->page_offset,
652 					skb_frag_size(frag));
653 	}
654 
655 	/* First request has the packet length. */
656 	first_tx->size = skb->len;
657 
658 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
659 	if (notify)
660 		notify_remote_via_irq(queue->tx_irq);
661 
662 	u64_stats_update_begin(&tx_stats->syncp);
663 	tx_stats->bytes += skb->len;
664 	tx_stats->packets++;
665 	u64_stats_update_end(&tx_stats->syncp);
666 
667 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
668 	xennet_tx_buf_gc(queue);
669 
670 	if (!netfront_tx_slot_available(queue))
671 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
672 
673 	spin_unlock_irqrestore(&queue->tx_lock, flags);
674 
675 	return NETDEV_TX_OK;
676 
677  drop:
678 	dev->stats.tx_dropped++;
679 	dev_kfree_skb_any(skb);
680 	return NETDEV_TX_OK;
681 }
682 
683 static int xennet_close(struct net_device *dev)
684 {
685 	struct netfront_info *np = netdev_priv(dev);
686 	unsigned int num_queues = dev->real_num_tx_queues;
687 	unsigned int i;
688 	struct netfront_queue *queue;
689 	netif_tx_stop_all_queues(np->netdev);
690 	for (i = 0; i < num_queues; ++i) {
691 		queue = &np->queues[i];
692 		napi_disable(&queue->napi);
693 	}
694 	return 0;
695 }
696 
697 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
698 				grant_ref_t ref)
699 {
700 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
701 
702 	BUG_ON(queue->rx_skbs[new]);
703 	queue->rx_skbs[new] = skb;
704 	queue->grant_rx_ref[new] = ref;
705 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
706 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
707 	queue->rx.req_prod_pvt++;
708 }
709 
710 static int xennet_get_extras(struct netfront_queue *queue,
711 			     struct xen_netif_extra_info *extras,
712 			     RING_IDX rp)
713 
714 {
715 	struct xen_netif_extra_info *extra;
716 	struct device *dev = &queue->info->netdev->dev;
717 	RING_IDX cons = queue->rx.rsp_cons;
718 	int err = 0;
719 
720 	do {
721 		struct sk_buff *skb;
722 		grant_ref_t ref;
723 
724 		if (unlikely(cons + 1 == rp)) {
725 			if (net_ratelimit())
726 				dev_warn(dev, "Missing extra info\n");
727 			err = -EBADR;
728 			break;
729 		}
730 
731 		extra = (struct xen_netif_extra_info *)
732 			RING_GET_RESPONSE(&queue->rx, ++cons);
733 
734 		if (unlikely(!extra->type ||
735 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
736 			if (net_ratelimit())
737 				dev_warn(dev, "Invalid extra type: %d\n",
738 					extra->type);
739 			err = -EINVAL;
740 		} else {
741 			memcpy(&extras[extra->type - 1], extra,
742 			       sizeof(*extra));
743 		}
744 
745 		skb = xennet_get_rx_skb(queue, cons);
746 		ref = xennet_get_rx_ref(queue, cons);
747 		xennet_move_rx_slot(queue, skb, ref);
748 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
749 
750 	queue->rx.rsp_cons = cons;
751 	return err;
752 }
753 
754 static int xennet_get_responses(struct netfront_queue *queue,
755 				struct netfront_rx_info *rinfo, RING_IDX rp,
756 				struct sk_buff_head *list)
757 {
758 	struct xen_netif_rx_response *rx = &rinfo->rx;
759 	struct xen_netif_extra_info *extras = rinfo->extras;
760 	struct device *dev = &queue->info->netdev->dev;
761 	RING_IDX cons = queue->rx.rsp_cons;
762 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
763 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
764 	int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
765 	int slots = 1;
766 	int err = 0;
767 	unsigned long ret;
768 
769 	if (rx->flags & XEN_NETRXF_extra_info) {
770 		err = xennet_get_extras(queue, extras, rp);
771 		cons = queue->rx.rsp_cons;
772 	}
773 
774 	for (;;) {
775 		if (unlikely(rx->status < 0 ||
776 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
777 			if (net_ratelimit())
778 				dev_warn(dev, "rx->offset: %u, size: %d\n",
779 					 rx->offset, rx->status);
780 			xennet_move_rx_slot(queue, skb, ref);
781 			err = -EINVAL;
782 			goto next;
783 		}
784 
785 		/*
786 		 * This definitely indicates a bug, either in this driver or in
787 		 * the backend driver. In future this should flag the bad
788 		 * situation to the system controller to reboot the backend.
789 		 */
790 		if (ref == GRANT_INVALID_REF) {
791 			if (net_ratelimit())
792 				dev_warn(dev, "Bad rx response id %d.\n",
793 					 rx->id);
794 			err = -EINVAL;
795 			goto next;
796 		}
797 
798 		ret = gnttab_end_foreign_access_ref(ref, 0);
799 		BUG_ON(!ret);
800 
801 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
802 
803 		__skb_queue_tail(list, skb);
804 
805 next:
806 		if (!(rx->flags & XEN_NETRXF_more_data))
807 			break;
808 
809 		if (cons + slots == rp) {
810 			if (net_ratelimit())
811 				dev_warn(dev, "Need more slots\n");
812 			err = -ENOENT;
813 			break;
814 		}
815 
816 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
817 		skb = xennet_get_rx_skb(queue, cons + slots);
818 		ref = xennet_get_rx_ref(queue, cons + slots);
819 		slots++;
820 	}
821 
822 	if (unlikely(slots > max)) {
823 		if (net_ratelimit())
824 			dev_warn(dev, "Too many slots\n");
825 		err = -E2BIG;
826 	}
827 
828 	if (unlikely(err))
829 		queue->rx.rsp_cons = cons + slots;
830 
831 	return err;
832 }
833 
834 static int xennet_set_skb_gso(struct sk_buff *skb,
835 			      struct xen_netif_extra_info *gso)
836 {
837 	if (!gso->u.gso.size) {
838 		if (net_ratelimit())
839 			pr_warn("GSO size must not be zero\n");
840 		return -EINVAL;
841 	}
842 
843 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
844 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
845 		if (net_ratelimit())
846 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
847 		return -EINVAL;
848 	}
849 
850 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
851 	skb_shinfo(skb)->gso_type =
852 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
853 		SKB_GSO_TCPV4 :
854 		SKB_GSO_TCPV6;
855 
856 	/* Header must be checked, and gso_segs computed. */
857 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
858 	skb_shinfo(skb)->gso_segs = 0;
859 
860 	return 0;
861 }
862 
863 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
864 				  struct sk_buff *skb,
865 				  struct sk_buff_head *list)
866 {
867 	struct skb_shared_info *shinfo = skb_shinfo(skb);
868 	RING_IDX cons = queue->rx.rsp_cons;
869 	struct sk_buff *nskb;
870 
871 	while ((nskb = __skb_dequeue(list))) {
872 		struct xen_netif_rx_response *rx =
873 			RING_GET_RESPONSE(&queue->rx, ++cons);
874 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
875 
876 		if (shinfo->nr_frags == MAX_SKB_FRAGS) {
877 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
878 
879 			BUG_ON(pull_to <= skb_headlen(skb));
880 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
881 		}
882 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
883 
884 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
885 				rx->offset, rx->status, PAGE_SIZE);
886 
887 		skb_shinfo(nskb)->nr_frags = 0;
888 		kfree_skb(nskb);
889 	}
890 
891 	return cons;
892 }
893 
894 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
895 {
896 	bool recalculate_partial_csum = false;
897 
898 	/*
899 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
900 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
901 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
902 	 * recalculate the partial checksum.
903 	 */
904 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
905 		struct netfront_info *np = netdev_priv(dev);
906 		atomic_inc(&np->rx_gso_checksum_fixup);
907 		skb->ip_summed = CHECKSUM_PARTIAL;
908 		recalculate_partial_csum = true;
909 	}
910 
911 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
912 	if (skb->ip_summed != CHECKSUM_PARTIAL)
913 		return 0;
914 
915 	return skb_checksum_setup(skb, recalculate_partial_csum);
916 }
917 
918 static int handle_incoming_queue(struct netfront_queue *queue,
919 				 struct sk_buff_head *rxq)
920 {
921 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
922 	int packets_dropped = 0;
923 	struct sk_buff *skb;
924 
925 	while ((skb = __skb_dequeue(rxq)) != NULL) {
926 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
927 
928 		if (pull_to > skb_headlen(skb))
929 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
930 
931 		/* Ethernet work: Delayed to here as it peeks the header. */
932 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
933 		skb_reset_network_header(skb);
934 
935 		if (checksum_setup(queue->info->netdev, skb)) {
936 			kfree_skb(skb);
937 			packets_dropped++;
938 			queue->info->netdev->stats.rx_errors++;
939 			continue;
940 		}
941 
942 		u64_stats_update_begin(&rx_stats->syncp);
943 		rx_stats->packets++;
944 		rx_stats->bytes += skb->len;
945 		u64_stats_update_end(&rx_stats->syncp);
946 
947 		/* Pass it up. */
948 		napi_gro_receive(&queue->napi, skb);
949 	}
950 
951 	return packets_dropped;
952 }
953 
954 static int xennet_poll(struct napi_struct *napi, int budget)
955 {
956 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
957 	struct net_device *dev = queue->info->netdev;
958 	struct sk_buff *skb;
959 	struct netfront_rx_info rinfo;
960 	struct xen_netif_rx_response *rx = &rinfo.rx;
961 	struct xen_netif_extra_info *extras = rinfo.extras;
962 	RING_IDX i, rp;
963 	int work_done;
964 	struct sk_buff_head rxq;
965 	struct sk_buff_head errq;
966 	struct sk_buff_head tmpq;
967 	int err;
968 
969 	spin_lock(&queue->rx_lock);
970 
971 	skb_queue_head_init(&rxq);
972 	skb_queue_head_init(&errq);
973 	skb_queue_head_init(&tmpq);
974 
975 	rp = queue->rx.sring->rsp_prod;
976 	rmb(); /* Ensure we see queued responses up to 'rp'. */
977 
978 	i = queue->rx.rsp_cons;
979 	work_done = 0;
980 	while ((i != rp) && (work_done < budget)) {
981 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
982 		memset(extras, 0, sizeof(rinfo.extras));
983 
984 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
985 
986 		if (unlikely(err)) {
987 err:
988 			while ((skb = __skb_dequeue(&tmpq)))
989 				__skb_queue_tail(&errq, skb);
990 			dev->stats.rx_errors++;
991 			i = queue->rx.rsp_cons;
992 			continue;
993 		}
994 
995 		skb = __skb_dequeue(&tmpq);
996 
997 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
998 			struct xen_netif_extra_info *gso;
999 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1000 
1001 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1002 				__skb_queue_head(&tmpq, skb);
1003 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1004 				goto err;
1005 			}
1006 		}
1007 
1008 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1009 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1010 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1011 
1012 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1013 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1014 		skb->data_len = rx->status;
1015 		skb->len += rx->status;
1016 
1017 		i = xennet_fill_frags(queue, skb, &tmpq);
1018 
1019 		if (rx->flags & XEN_NETRXF_csum_blank)
1020 			skb->ip_summed = CHECKSUM_PARTIAL;
1021 		else if (rx->flags & XEN_NETRXF_data_validated)
1022 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1023 
1024 		__skb_queue_tail(&rxq, skb);
1025 
1026 		queue->rx.rsp_cons = ++i;
1027 		work_done++;
1028 	}
1029 
1030 	__skb_queue_purge(&errq);
1031 
1032 	work_done -= handle_incoming_queue(queue, &rxq);
1033 
1034 	xennet_alloc_rx_buffers(queue);
1035 
1036 	if (work_done < budget) {
1037 		int more_to_do = 0;
1038 
1039 		napi_complete(napi);
1040 
1041 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1042 		if (more_to_do)
1043 			napi_schedule(napi);
1044 	}
1045 
1046 	spin_unlock(&queue->rx_lock);
1047 
1048 	return work_done;
1049 }
1050 
1051 static int xennet_change_mtu(struct net_device *dev, int mtu)
1052 {
1053 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1054 
1055 	if (mtu > max)
1056 		return -EINVAL;
1057 	dev->mtu = mtu;
1058 	return 0;
1059 }
1060 
1061 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1062 						    struct rtnl_link_stats64 *tot)
1063 {
1064 	struct netfront_info *np = netdev_priv(dev);
1065 	int cpu;
1066 
1067 	for_each_possible_cpu(cpu) {
1068 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1069 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1070 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1071 		unsigned int start;
1072 
1073 		do {
1074 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1075 			tx_packets = tx_stats->packets;
1076 			tx_bytes = tx_stats->bytes;
1077 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1078 
1079 		do {
1080 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1081 			rx_packets = rx_stats->packets;
1082 			rx_bytes = rx_stats->bytes;
1083 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1084 
1085 		tot->rx_packets += rx_packets;
1086 		tot->tx_packets += tx_packets;
1087 		tot->rx_bytes   += rx_bytes;
1088 		tot->tx_bytes   += tx_bytes;
1089 	}
1090 
1091 	tot->rx_errors  = dev->stats.rx_errors;
1092 	tot->tx_dropped = dev->stats.tx_dropped;
1093 
1094 	return tot;
1095 }
1096 
1097 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1098 {
1099 	struct sk_buff *skb;
1100 	int i;
1101 
1102 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1103 		/* Skip over entries which are actually freelist references */
1104 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1105 			continue;
1106 
1107 		skb = queue->tx_skbs[i].skb;
1108 		get_page(queue->grant_tx_page[i]);
1109 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1110 					  GNTMAP_readonly,
1111 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1112 		queue->grant_tx_page[i] = NULL;
1113 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1114 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1115 		dev_kfree_skb_irq(skb);
1116 	}
1117 }
1118 
1119 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1120 {
1121 	int id, ref;
1122 
1123 	spin_lock_bh(&queue->rx_lock);
1124 
1125 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1126 		struct sk_buff *skb;
1127 		struct page *page;
1128 
1129 		skb = queue->rx_skbs[id];
1130 		if (!skb)
1131 			continue;
1132 
1133 		ref = queue->grant_rx_ref[id];
1134 		if (ref == GRANT_INVALID_REF)
1135 			continue;
1136 
1137 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1138 
1139 		/* gnttab_end_foreign_access() needs a page ref until
1140 		 * foreign access is ended (which may be deferred).
1141 		 */
1142 		get_page(page);
1143 		gnttab_end_foreign_access(ref, 0,
1144 					  (unsigned long)page_address(page));
1145 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1146 
1147 		kfree_skb(skb);
1148 	}
1149 
1150 	spin_unlock_bh(&queue->rx_lock);
1151 }
1152 
1153 static netdev_features_t xennet_fix_features(struct net_device *dev,
1154 	netdev_features_t features)
1155 {
1156 	struct netfront_info *np = netdev_priv(dev);
1157 	int val;
1158 
1159 	if (features & NETIF_F_SG) {
1160 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1161 				 "%d", &val) < 0)
1162 			val = 0;
1163 
1164 		if (!val)
1165 			features &= ~NETIF_F_SG;
1166 	}
1167 
1168 	if (features & NETIF_F_IPV6_CSUM) {
1169 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1170 				 "feature-ipv6-csum-offload", "%d", &val) < 0)
1171 			val = 0;
1172 
1173 		if (!val)
1174 			features &= ~NETIF_F_IPV6_CSUM;
1175 	}
1176 
1177 	if (features & NETIF_F_TSO) {
1178 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1179 				 "feature-gso-tcpv4", "%d", &val) < 0)
1180 			val = 0;
1181 
1182 		if (!val)
1183 			features &= ~NETIF_F_TSO;
1184 	}
1185 
1186 	if (features & NETIF_F_TSO6) {
1187 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1188 				 "feature-gso-tcpv6", "%d", &val) < 0)
1189 			val = 0;
1190 
1191 		if (!val)
1192 			features &= ~NETIF_F_TSO6;
1193 	}
1194 
1195 	return features;
1196 }
1197 
1198 static int xennet_set_features(struct net_device *dev,
1199 	netdev_features_t features)
1200 {
1201 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1202 		netdev_info(dev, "Reducing MTU because no SG offload");
1203 		dev->mtu = ETH_DATA_LEN;
1204 	}
1205 
1206 	return 0;
1207 }
1208 
1209 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1210 {
1211 	struct netfront_queue *queue = dev_id;
1212 	unsigned long flags;
1213 
1214 	spin_lock_irqsave(&queue->tx_lock, flags);
1215 	xennet_tx_buf_gc(queue);
1216 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1217 
1218 	return IRQ_HANDLED;
1219 }
1220 
1221 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1222 {
1223 	struct netfront_queue *queue = dev_id;
1224 	struct net_device *dev = queue->info->netdev;
1225 
1226 	if (likely(netif_carrier_ok(dev) &&
1227 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1228 		napi_schedule(&queue->napi);
1229 
1230 	return IRQ_HANDLED;
1231 }
1232 
1233 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1234 {
1235 	xennet_tx_interrupt(irq, dev_id);
1236 	xennet_rx_interrupt(irq, dev_id);
1237 	return IRQ_HANDLED;
1238 }
1239 
1240 #ifdef CONFIG_NET_POLL_CONTROLLER
1241 static void xennet_poll_controller(struct net_device *dev)
1242 {
1243 	/* Poll each queue */
1244 	struct netfront_info *info = netdev_priv(dev);
1245 	unsigned int num_queues = dev->real_num_tx_queues;
1246 	unsigned int i;
1247 	for (i = 0; i < num_queues; ++i)
1248 		xennet_interrupt(0, &info->queues[i]);
1249 }
1250 #endif
1251 
1252 static const struct net_device_ops xennet_netdev_ops = {
1253 	.ndo_open            = xennet_open,
1254 	.ndo_stop            = xennet_close,
1255 	.ndo_start_xmit      = xennet_start_xmit,
1256 	.ndo_change_mtu	     = xennet_change_mtu,
1257 	.ndo_get_stats64     = xennet_get_stats64,
1258 	.ndo_set_mac_address = eth_mac_addr,
1259 	.ndo_validate_addr   = eth_validate_addr,
1260 	.ndo_fix_features    = xennet_fix_features,
1261 	.ndo_set_features    = xennet_set_features,
1262 	.ndo_select_queue    = xennet_select_queue,
1263 #ifdef CONFIG_NET_POLL_CONTROLLER
1264 	.ndo_poll_controller = xennet_poll_controller,
1265 #endif
1266 };
1267 
1268 static void xennet_free_netdev(struct net_device *netdev)
1269 {
1270 	struct netfront_info *np = netdev_priv(netdev);
1271 
1272 	free_percpu(np->rx_stats);
1273 	free_percpu(np->tx_stats);
1274 	free_netdev(netdev);
1275 }
1276 
1277 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1278 {
1279 	int err;
1280 	struct net_device *netdev;
1281 	struct netfront_info *np;
1282 
1283 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1284 	if (!netdev)
1285 		return ERR_PTR(-ENOMEM);
1286 
1287 	np                   = netdev_priv(netdev);
1288 	np->xbdev            = dev;
1289 
1290 	np->queues = NULL;
1291 
1292 	err = -ENOMEM;
1293 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1294 	if (np->rx_stats == NULL)
1295 		goto exit;
1296 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1297 	if (np->tx_stats == NULL)
1298 		goto exit;
1299 
1300 	netdev->netdev_ops	= &xennet_netdev_ops;
1301 
1302 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1303 				  NETIF_F_GSO_ROBUST;
1304 	netdev->hw_features	= NETIF_F_SG |
1305 				  NETIF_F_IPV6_CSUM |
1306 				  NETIF_F_TSO | NETIF_F_TSO6;
1307 
1308 	/*
1309          * Assume that all hw features are available for now. This set
1310          * will be adjusted by the call to netdev_update_features() in
1311          * xennet_connect() which is the earliest point where we can
1312          * negotiate with the backend regarding supported features.
1313          */
1314 	netdev->features |= netdev->hw_features;
1315 
1316 	netdev->ethtool_ops = &xennet_ethtool_ops;
1317 	SET_NETDEV_DEV(netdev, &dev->dev);
1318 
1319 	np->netdev = netdev;
1320 
1321 	netif_carrier_off(netdev);
1322 
1323 	return netdev;
1324 
1325  exit:
1326 	xennet_free_netdev(netdev);
1327 	return ERR_PTR(err);
1328 }
1329 
1330 /**
1331  * Entry point to this code when a new device is created.  Allocate the basic
1332  * structures and the ring buffers for communication with the backend, and
1333  * inform the backend of the appropriate details for those.
1334  */
1335 static int netfront_probe(struct xenbus_device *dev,
1336 			  const struct xenbus_device_id *id)
1337 {
1338 	int err;
1339 	struct net_device *netdev;
1340 	struct netfront_info *info;
1341 
1342 	netdev = xennet_create_dev(dev);
1343 	if (IS_ERR(netdev)) {
1344 		err = PTR_ERR(netdev);
1345 		xenbus_dev_fatal(dev, err, "creating netdev");
1346 		return err;
1347 	}
1348 
1349 	info = netdev_priv(netdev);
1350 	dev_set_drvdata(&dev->dev, info);
1351 #ifdef CONFIG_SYSFS
1352 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1353 #endif
1354 	err = register_netdev(info->netdev);
1355 	if (err) {
1356 		pr_warn("%s: register_netdev err=%d\n", __func__, err);
1357 		goto fail;
1358 	}
1359 
1360 	return 0;
1361 
1362  fail:
1363 	xennet_free_netdev(netdev);
1364 	dev_set_drvdata(&dev->dev, NULL);
1365 	return err;
1366 }
1367 
1368 static void xennet_end_access(int ref, void *page)
1369 {
1370 	/* This frees the page as a side-effect */
1371 	if (ref != GRANT_INVALID_REF)
1372 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1373 }
1374 
1375 static void xennet_disconnect_backend(struct netfront_info *info)
1376 {
1377 	unsigned int i = 0;
1378 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1379 
1380 	netif_carrier_off(info->netdev);
1381 
1382 	for (i = 0; i < num_queues && info->queues; ++i) {
1383 		struct netfront_queue *queue = &info->queues[i];
1384 
1385 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1386 			unbind_from_irqhandler(queue->tx_irq, queue);
1387 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1388 			unbind_from_irqhandler(queue->tx_irq, queue);
1389 			unbind_from_irqhandler(queue->rx_irq, queue);
1390 		}
1391 		queue->tx_evtchn = queue->rx_evtchn = 0;
1392 		queue->tx_irq = queue->rx_irq = 0;
1393 
1394 		if (netif_running(info->netdev))
1395 			napi_synchronize(&queue->napi);
1396 
1397 		xennet_release_tx_bufs(queue);
1398 		xennet_release_rx_bufs(queue);
1399 		gnttab_free_grant_references(queue->gref_tx_head);
1400 		gnttab_free_grant_references(queue->gref_rx_head);
1401 
1402 		/* End access and free the pages */
1403 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1404 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1405 
1406 		queue->tx_ring_ref = GRANT_INVALID_REF;
1407 		queue->rx_ring_ref = GRANT_INVALID_REF;
1408 		queue->tx.sring = NULL;
1409 		queue->rx.sring = NULL;
1410 	}
1411 }
1412 
1413 /**
1414  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1415  * driver restart.  We tear down our netif structure and recreate it, but
1416  * leave the device-layer structures intact so that this is transparent to the
1417  * rest of the kernel.
1418  */
1419 static int netfront_resume(struct xenbus_device *dev)
1420 {
1421 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1422 
1423 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1424 
1425 	xennet_disconnect_backend(info);
1426 	return 0;
1427 }
1428 
1429 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1430 {
1431 	char *s, *e, *macstr;
1432 	int i;
1433 
1434 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1435 	if (IS_ERR(macstr))
1436 		return PTR_ERR(macstr);
1437 
1438 	for (i = 0; i < ETH_ALEN; i++) {
1439 		mac[i] = simple_strtoul(s, &e, 16);
1440 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1441 			kfree(macstr);
1442 			return -ENOENT;
1443 		}
1444 		s = e+1;
1445 	}
1446 
1447 	kfree(macstr);
1448 	return 0;
1449 }
1450 
1451 static int setup_netfront_single(struct netfront_queue *queue)
1452 {
1453 	int err;
1454 
1455 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1456 	if (err < 0)
1457 		goto fail;
1458 
1459 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1460 					xennet_interrupt,
1461 					0, queue->info->netdev->name, queue);
1462 	if (err < 0)
1463 		goto bind_fail;
1464 	queue->rx_evtchn = queue->tx_evtchn;
1465 	queue->rx_irq = queue->tx_irq = err;
1466 
1467 	return 0;
1468 
1469 bind_fail:
1470 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1471 	queue->tx_evtchn = 0;
1472 fail:
1473 	return err;
1474 }
1475 
1476 static int setup_netfront_split(struct netfront_queue *queue)
1477 {
1478 	int err;
1479 
1480 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1481 	if (err < 0)
1482 		goto fail;
1483 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1484 	if (err < 0)
1485 		goto alloc_rx_evtchn_fail;
1486 
1487 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1488 		 "%s-tx", queue->name);
1489 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1490 					xennet_tx_interrupt,
1491 					0, queue->tx_irq_name, queue);
1492 	if (err < 0)
1493 		goto bind_tx_fail;
1494 	queue->tx_irq = err;
1495 
1496 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1497 		 "%s-rx", queue->name);
1498 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1499 					xennet_rx_interrupt,
1500 					0, queue->rx_irq_name, queue);
1501 	if (err < 0)
1502 		goto bind_rx_fail;
1503 	queue->rx_irq = err;
1504 
1505 	return 0;
1506 
1507 bind_rx_fail:
1508 	unbind_from_irqhandler(queue->tx_irq, queue);
1509 	queue->tx_irq = 0;
1510 bind_tx_fail:
1511 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1512 	queue->rx_evtchn = 0;
1513 alloc_rx_evtchn_fail:
1514 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1515 	queue->tx_evtchn = 0;
1516 fail:
1517 	return err;
1518 }
1519 
1520 static int setup_netfront(struct xenbus_device *dev,
1521 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1522 {
1523 	struct xen_netif_tx_sring *txs;
1524 	struct xen_netif_rx_sring *rxs;
1525 	grant_ref_t gref;
1526 	int err;
1527 
1528 	queue->tx_ring_ref = GRANT_INVALID_REF;
1529 	queue->rx_ring_ref = GRANT_INVALID_REF;
1530 	queue->rx.sring = NULL;
1531 	queue->tx.sring = NULL;
1532 
1533 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1534 	if (!txs) {
1535 		err = -ENOMEM;
1536 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1537 		goto fail;
1538 	}
1539 	SHARED_RING_INIT(txs);
1540 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1541 
1542 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1543 	if (err < 0)
1544 		goto grant_tx_ring_fail;
1545 	queue->tx_ring_ref = gref;
1546 
1547 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1548 	if (!rxs) {
1549 		err = -ENOMEM;
1550 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1551 		goto alloc_rx_ring_fail;
1552 	}
1553 	SHARED_RING_INIT(rxs);
1554 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1555 
1556 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1557 	if (err < 0)
1558 		goto grant_rx_ring_fail;
1559 	queue->rx_ring_ref = gref;
1560 
1561 	if (feature_split_evtchn)
1562 		err = setup_netfront_split(queue);
1563 	/* setup single event channel if
1564 	 *  a) feature-split-event-channels == 0
1565 	 *  b) feature-split-event-channels == 1 but failed to setup
1566 	 */
1567 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1568 		err = setup_netfront_single(queue);
1569 
1570 	if (err)
1571 		goto alloc_evtchn_fail;
1572 
1573 	return 0;
1574 
1575 	/* If we fail to setup netfront, it is safe to just revoke access to
1576 	 * granted pages because backend is not accessing it at this point.
1577 	 */
1578 alloc_evtchn_fail:
1579 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1580 grant_rx_ring_fail:
1581 	free_page((unsigned long)rxs);
1582 alloc_rx_ring_fail:
1583 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1584 grant_tx_ring_fail:
1585 	free_page((unsigned long)txs);
1586 fail:
1587 	return err;
1588 }
1589 
1590 /* Queue-specific initialisation
1591  * This used to be done in xennet_create_dev() but must now
1592  * be run per-queue.
1593  */
1594 static int xennet_init_queue(struct netfront_queue *queue)
1595 {
1596 	unsigned short i;
1597 	int err = 0;
1598 
1599 	spin_lock_init(&queue->tx_lock);
1600 	spin_lock_init(&queue->rx_lock);
1601 
1602 	setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1603 		    (unsigned long)queue);
1604 
1605 	snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1606 		 queue->info->netdev->name, queue->id);
1607 
1608 	/* Initialise tx_skbs as a free chain containing every entry. */
1609 	queue->tx_skb_freelist = 0;
1610 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1611 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1612 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1613 		queue->grant_tx_page[i] = NULL;
1614 	}
1615 
1616 	/* Clear out rx_skbs */
1617 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1618 		queue->rx_skbs[i] = NULL;
1619 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1620 	}
1621 
1622 	/* A grant for every tx ring slot */
1623 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1624 					  &queue->gref_tx_head) < 0) {
1625 		pr_alert("can't alloc tx grant refs\n");
1626 		err = -ENOMEM;
1627 		goto exit;
1628 	}
1629 
1630 	/* A grant for every rx ring slot */
1631 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1632 					  &queue->gref_rx_head) < 0) {
1633 		pr_alert("can't alloc rx grant refs\n");
1634 		err = -ENOMEM;
1635 		goto exit_free_tx;
1636 	}
1637 
1638 	return 0;
1639 
1640  exit_free_tx:
1641 	gnttab_free_grant_references(queue->gref_tx_head);
1642  exit:
1643 	return err;
1644 }
1645 
1646 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1647 			   struct xenbus_transaction *xbt, int write_hierarchical)
1648 {
1649 	/* Write the queue-specific keys into XenStore in the traditional
1650 	 * way for a single queue, or in a queue subkeys for multiple
1651 	 * queues.
1652 	 */
1653 	struct xenbus_device *dev = queue->info->xbdev;
1654 	int err;
1655 	const char *message;
1656 	char *path;
1657 	size_t pathsize;
1658 
1659 	/* Choose the correct place to write the keys */
1660 	if (write_hierarchical) {
1661 		pathsize = strlen(dev->nodename) + 10;
1662 		path = kzalloc(pathsize, GFP_KERNEL);
1663 		if (!path) {
1664 			err = -ENOMEM;
1665 			message = "out of memory while writing ring references";
1666 			goto error;
1667 		}
1668 		snprintf(path, pathsize, "%s/queue-%u",
1669 				dev->nodename, queue->id);
1670 	} else {
1671 		path = (char *)dev->nodename;
1672 	}
1673 
1674 	/* Write ring references */
1675 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1676 			queue->tx_ring_ref);
1677 	if (err) {
1678 		message = "writing tx-ring-ref";
1679 		goto error;
1680 	}
1681 
1682 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1683 			queue->rx_ring_ref);
1684 	if (err) {
1685 		message = "writing rx-ring-ref";
1686 		goto error;
1687 	}
1688 
1689 	/* Write event channels; taking into account both shared
1690 	 * and split event channel scenarios.
1691 	 */
1692 	if (queue->tx_evtchn == queue->rx_evtchn) {
1693 		/* Shared event channel */
1694 		err = xenbus_printf(*xbt, path,
1695 				"event-channel", "%u", queue->tx_evtchn);
1696 		if (err) {
1697 			message = "writing event-channel";
1698 			goto error;
1699 		}
1700 	} else {
1701 		/* Split event channels */
1702 		err = xenbus_printf(*xbt, path,
1703 				"event-channel-tx", "%u", queue->tx_evtchn);
1704 		if (err) {
1705 			message = "writing event-channel-tx";
1706 			goto error;
1707 		}
1708 
1709 		err = xenbus_printf(*xbt, path,
1710 				"event-channel-rx", "%u", queue->rx_evtchn);
1711 		if (err) {
1712 			message = "writing event-channel-rx";
1713 			goto error;
1714 		}
1715 	}
1716 
1717 	if (write_hierarchical)
1718 		kfree(path);
1719 	return 0;
1720 
1721 error:
1722 	if (write_hierarchical)
1723 		kfree(path);
1724 	xenbus_dev_fatal(dev, err, "%s", message);
1725 	return err;
1726 }
1727 
1728 static void xennet_destroy_queues(struct netfront_info *info)
1729 {
1730 	unsigned int i;
1731 
1732 	rtnl_lock();
1733 
1734 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1735 		struct netfront_queue *queue = &info->queues[i];
1736 
1737 		if (netif_running(info->netdev))
1738 			napi_disable(&queue->napi);
1739 		del_timer_sync(&queue->rx_refill_timer);
1740 		netif_napi_del(&queue->napi);
1741 	}
1742 
1743 	rtnl_unlock();
1744 
1745 	kfree(info->queues);
1746 	info->queues = NULL;
1747 }
1748 
1749 static int xennet_create_queues(struct netfront_info *info,
1750 				unsigned int *num_queues)
1751 {
1752 	unsigned int i;
1753 	int ret;
1754 
1755 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1756 			       GFP_KERNEL);
1757 	if (!info->queues)
1758 		return -ENOMEM;
1759 
1760 	rtnl_lock();
1761 
1762 	for (i = 0; i < *num_queues; i++) {
1763 		struct netfront_queue *queue = &info->queues[i];
1764 
1765 		queue->id = i;
1766 		queue->info = info;
1767 
1768 		ret = xennet_init_queue(queue);
1769 		if (ret < 0) {
1770 			dev_warn(&info->netdev->dev,
1771 				 "only created %d queues\n", i);
1772 			*num_queues = i;
1773 			break;
1774 		}
1775 
1776 		netif_napi_add(queue->info->netdev, &queue->napi,
1777 			       xennet_poll, 64);
1778 		if (netif_running(info->netdev))
1779 			napi_enable(&queue->napi);
1780 	}
1781 
1782 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
1783 
1784 	rtnl_unlock();
1785 
1786 	if (*num_queues == 0) {
1787 		dev_err(&info->netdev->dev, "no queues\n");
1788 		return -EINVAL;
1789 	}
1790 	return 0;
1791 }
1792 
1793 /* Common code used when first setting up, and when resuming. */
1794 static int talk_to_netback(struct xenbus_device *dev,
1795 			   struct netfront_info *info)
1796 {
1797 	const char *message;
1798 	struct xenbus_transaction xbt;
1799 	int err;
1800 	unsigned int feature_split_evtchn;
1801 	unsigned int i = 0;
1802 	unsigned int max_queues = 0;
1803 	struct netfront_queue *queue = NULL;
1804 	unsigned int num_queues = 1;
1805 
1806 	info->netdev->irq = 0;
1807 
1808 	/* Check if backend supports multiple queues */
1809 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1810 			   "multi-queue-max-queues", "%u", &max_queues);
1811 	if (err < 0)
1812 		max_queues = 1;
1813 	num_queues = min(max_queues, xennet_max_queues);
1814 
1815 	/* Check feature-split-event-channels */
1816 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1817 			   "feature-split-event-channels", "%u",
1818 			   &feature_split_evtchn);
1819 	if (err < 0)
1820 		feature_split_evtchn = 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 		goto destroy_ring;
1835 
1836 	/* Create shared ring, alloc event channel -- for each queue */
1837 	for (i = 0; i < num_queues; ++i) {
1838 		queue = &info->queues[i];
1839 		err = setup_netfront(dev, queue, feature_split_evtchn);
1840 		if (err) {
1841 			/* setup_netfront() will tidy up the current
1842 			 * queue on error, but we need to clean up
1843 			 * those already allocated.
1844 			 */
1845 			if (i > 0) {
1846 				rtnl_lock();
1847 				netif_set_real_num_tx_queues(info->netdev, i);
1848 				rtnl_unlock();
1849 				goto destroy_ring;
1850 			} else {
1851 				goto out;
1852 			}
1853 		}
1854 	}
1855 
1856 again:
1857 	err = xenbus_transaction_start(&xbt);
1858 	if (err) {
1859 		xenbus_dev_fatal(dev, err, "starting transaction");
1860 		goto destroy_ring;
1861 	}
1862 
1863 	if (xenbus_exists(XBT_NIL,
1864 			  info->xbdev->otherend, "multi-queue-max-queues")) {
1865 		/* Write the number of queues */
1866 		err = xenbus_printf(xbt, dev->nodename,
1867 				    "multi-queue-num-queues", "%u", num_queues);
1868 		if (err) {
1869 			message = "writing multi-queue-num-queues";
1870 			goto abort_transaction_no_dev_fatal;
1871 		}
1872 	}
1873 
1874 	if (num_queues == 1) {
1875 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1876 		if (err)
1877 			goto abort_transaction_no_dev_fatal;
1878 	} else {
1879 		/* Write the keys for each queue */
1880 		for (i = 0; i < num_queues; ++i) {
1881 			queue = &info->queues[i];
1882 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1883 			if (err)
1884 				goto abort_transaction_no_dev_fatal;
1885 		}
1886 	}
1887 
1888 	/* The remaining keys are not queue-specific */
1889 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1890 			    1);
1891 	if (err) {
1892 		message = "writing request-rx-copy";
1893 		goto abort_transaction;
1894 	}
1895 
1896 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1897 	if (err) {
1898 		message = "writing feature-rx-notify";
1899 		goto abort_transaction;
1900 	}
1901 
1902 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1903 	if (err) {
1904 		message = "writing feature-sg";
1905 		goto abort_transaction;
1906 	}
1907 
1908 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1909 	if (err) {
1910 		message = "writing feature-gso-tcpv4";
1911 		goto abort_transaction;
1912 	}
1913 
1914 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1915 	if (err) {
1916 		message = "writing feature-gso-tcpv6";
1917 		goto abort_transaction;
1918 	}
1919 
1920 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1921 			   "1");
1922 	if (err) {
1923 		message = "writing feature-ipv6-csum-offload";
1924 		goto abort_transaction;
1925 	}
1926 
1927 	err = xenbus_transaction_end(xbt, 0);
1928 	if (err) {
1929 		if (err == -EAGAIN)
1930 			goto again;
1931 		xenbus_dev_fatal(dev, err, "completing transaction");
1932 		goto destroy_ring;
1933 	}
1934 
1935 	return 0;
1936 
1937  abort_transaction:
1938 	xenbus_dev_fatal(dev, err, "%s", message);
1939 abort_transaction_no_dev_fatal:
1940 	xenbus_transaction_end(xbt, 1);
1941  destroy_ring:
1942 	xennet_disconnect_backend(info);
1943 	kfree(info->queues);
1944 	info->queues = NULL;
1945  out:
1946 	return err;
1947 }
1948 
1949 static int xennet_connect(struct net_device *dev)
1950 {
1951 	struct netfront_info *np = netdev_priv(dev);
1952 	unsigned int num_queues = 0;
1953 	int err;
1954 	unsigned int feature_rx_copy;
1955 	unsigned int j = 0;
1956 	struct netfront_queue *queue = NULL;
1957 
1958 	err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1959 			   "feature-rx-copy", "%u", &feature_rx_copy);
1960 	if (err != 1)
1961 		feature_rx_copy = 0;
1962 
1963 	if (!feature_rx_copy) {
1964 		dev_info(&dev->dev,
1965 			 "backend does not support copying receive path\n");
1966 		return -ENODEV;
1967 	}
1968 
1969 	err = talk_to_netback(np->xbdev, np);
1970 	if (err)
1971 		return err;
1972 
1973 	/* talk_to_netback() sets the correct number of queues */
1974 	num_queues = dev->real_num_tx_queues;
1975 
1976 	rtnl_lock();
1977 	netdev_update_features(dev);
1978 	rtnl_unlock();
1979 
1980 	/*
1981 	 * All public and private state should now be sane.  Get
1982 	 * ready to start sending and receiving packets and give the driver
1983 	 * domain a kick because we've probably just requeued some
1984 	 * packets.
1985 	 */
1986 	netif_carrier_on(np->netdev);
1987 	for (j = 0; j < num_queues; ++j) {
1988 		queue = &np->queues[j];
1989 
1990 		notify_remote_via_irq(queue->tx_irq);
1991 		if (queue->tx_irq != queue->rx_irq)
1992 			notify_remote_via_irq(queue->rx_irq);
1993 
1994 		spin_lock_irq(&queue->tx_lock);
1995 		xennet_tx_buf_gc(queue);
1996 		spin_unlock_irq(&queue->tx_lock);
1997 
1998 		spin_lock_bh(&queue->rx_lock);
1999 		xennet_alloc_rx_buffers(queue);
2000 		spin_unlock_bh(&queue->rx_lock);
2001 	}
2002 
2003 	return 0;
2004 }
2005 
2006 /**
2007  * Callback received when the backend's state changes.
2008  */
2009 static void netback_changed(struct xenbus_device *dev,
2010 			    enum xenbus_state backend_state)
2011 {
2012 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2013 	struct net_device *netdev = np->netdev;
2014 
2015 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2016 
2017 	switch (backend_state) {
2018 	case XenbusStateInitialising:
2019 	case XenbusStateInitialised:
2020 	case XenbusStateReconfiguring:
2021 	case XenbusStateReconfigured:
2022 	case XenbusStateUnknown:
2023 		break;
2024 
2025 	case XenbusStateInitWait:
2026 		if (dev->state != XenbusStateInitialising)
2027 			break;
2028 		if (xennet_connect(netdev) != 0)
2029 			break;
2030 		xenbus_switch_state(dev, XenbusStateConnected);
2031 		break;
2032 
2033 	case XenbusStateConnected:
2034 		netdev_notify_peers(netdev);
2035 		break;
2036 
2037 	case XenbusStateClosed:
2038 		if (dev->state == XenbusStateClosed)
2039 			break;
2040 		/* Missed the backend's CLOSING state -- fallthrough */
2041 	case XenbusStateClosing:
2042 		xenbus_frontend_closed(dev);
2043 		break;
2044 	}
2045 }
2046 
2047 static const struct xennet_stat {
2048 	char name[ETH_GSTRING_LEN];
2049 	u16 offset;
2050 } xennet_stats[] = {
2051 	{
2052 		"rx_gso_checksum_fixup",
2053 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2054 	},
2055 };
2056 
2057 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2058 {
2059 	switch (string_set) {
2060 	case ETH_SS_STATS:
2061 		return ARRAY_SIZE(xennet_stats);
2062 	default:
2063 		return -EINVAL;
2064 	}
2065 }
2066 
2067 static void xennet_get_ethtool_stats(struct net_device *dev,
2068 				     struct ethtool_stats *stats, u64 * data)
2069 {
2070 	void *np = netdev_priv(dev);
2071 	int i;
2072 
2073 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2074 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2075 }
2076 
2077 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2078 {
2079 	int i;
2080 
2081 	switch (stringset) {
2082 	case ETH_SS_STATS:
2083 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2084 			memcpy(data + i * ETH_GSTRING_LEN,
2085 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2086 		break;
2087 	}
2088 }
2089 
2090 static const struct ethtool_ops xennet_ethtool_ops =
2091 {
2092 	.get_link = ethtool_op_get_link,
2093 
2094 	.get_sset_count = xennet_get_sset_count,
2095 	.get_ethtool_stats = xennet_get_ethtool_stats,
2096 	.get_strings = xennet_get_strings,
2097 };
2098 
2099 #ifdef CONFIG_SYSFS
2100 static ssize_t show_rxbuf(struct device *dev,
2101 			  struct device_attribute *attr, char *buf)
2102 {
2103 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2104 }
2105 
2106 static ssize_t store_rxbuf(struct device *dev,
2107 			   struct device_attribute *attr,
2108 			   const char *buf, size_t len)
2109 {
2110 	char *endp;
2111 	unsigned long target;
2112 
2113 	if (!capable(CAP_NET_ADMIN))
2114 		return -EPERM;
2115 
2116 	target = simple_strtoul(buf, &endp, 0);
2117 	if (endp == buf)
2118 		return -EBADMSG;
2119 
2120 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2121 
2122 	return len;
2123 }
2124 
2125 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2126 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2127 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2128 
2129 static struct attribute *xennet_dev_attrs[] = {
2130 	&dev_attr_rxbuf_min.attr,
2131 	&dev_attr_rxbuf_max.attr,
2132 	&dev_attr_rxbuf_cur.attr,
2133 	NULL
2134 };
2135 
2136 static const struct attribute_group xennet_dev_group = {
2137 	.attrs = xennet_dev_attrs
2138 };
2139 #endif /* CONFIG_SYSFS */
2140 
2141 static int xennet_remove(struct xenbus_device *dev)
2142 {
2143 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2144 
2145 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2146 
2147 	xennet_disconnect_backend(info);
2148 
2149 	unregister_netdev(info->netdev);
2150 
2151 	if (info->queues)
2152 		xennet_destroy_queues(info);
2153 	xennet_free_netdev(info->netdev);
2154 
2155 	return 0;
2156 }
2157 
2158 static const struct xenbus_device_id netfront_ids[] = {
2159 	{ "vif" },
2160 	{ "" }
2161 };
2162 
2163 static struct xenbus_driver netfront_driver = {
2164 	.ids = netfront_ids,
2165 	.probe = netfront_probe,
2166 	.remove = xennet_remove,
2167 	.resume = netfront_resume,
2168 	.otherend_changed = netback_changed,
2169 };
2170 
2171 static int __init netif_init(void)
2172 {
2173 	if (!xen_domain())
2174 		return -ENODEV;
2175 
2176 	if (!xen_has_pv_nic_devices())
2177 		return -ENODEV;
2178 
2179 	pr_info("Initialising Xen virtual ethernet driver\n");
2180 
2181 	/* Allow as many queues as there are CPUs if user has not
2182 	 * specified a value.
2183 	 */
2184 	if (xennet_max_queues == 0)
2185 		xennet_max_queues = num_online_cpus();
2186 
2187 	return xenbus_register_frontend(&netfront_driver);
2188 }
2189 module_init(netif_init);
2190 
2191 
2192 static void __exit netif_exit(void)
2193 {
2194 	xenbus_unregister_driver(&netfront_driver);
2195 }
2196 module_exit(netif_exit);
2197 
2198 MODULE_DESCRIPTION("Xen virtual network device frontend");
2199 MODULE_LICENSE("GPL");
2200 MODULE_ALIAS("xen:vif");
2201 MODULE_ALIAS("xennet");
2202