xref: /openbmc/linux/drivers/net/xen-netfront.c (revision a0ae2562c6c4b2721d9fddba63b7286c13517d9f)
1 /*
2  * Virtual network driver for conversing with remote driver backends.
3  *
4  * Copyright (c) 2002-2005, K A Fraser
5  * Copyright (c) 2005, XenSource Ltd
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License version 2
9  * as published by the Free Software Foundation; or, when distributed
10  * separately from the Linux kernel or incorporated into other
11  * software packages, subject to the following license:
12  *
13  * Permission is hereby granted, free of charge, to any person obtaining a copy
14  * of this source file (the "Software"), to deal in the Software without
15  * restriction, including without limitation the rights to use, copy, modify,
16  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17  * and to permit persons to whom the Software is furnished to do so, subject to
18  * the following conditions:
19  *
20  * The above copyright notice and this permission notice shall be included in
21  * all copies or substantial portions of the Software.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29  * IN THE SOFTWARE.
30  */
31 
32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33 
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_ether.h>
41 #include <net/tcp.h>
42 #include <linux/udp.h>
43 #include <linux/moduleparam.h>
44 #include <linux/mm.h>
45 #include <linux/slab.h>
46 #include <net/ip.h>
47 
48 #include <xen/xen.h>
49 #include <xen/xenbus.h>
50 #include <xen/events.h>
51 #include <xen/page.h>
52 #include <xen/platform_pci.h>
53 #include <xen/grant_table.h>
54 
55 #include <xen/interface/io/netif.h>
56 #include <xen/interface/memory.h>
57 #include <xen/interface/grant_table.h>
58 
59 /* Module parameters */
60 #define MAX_QUEUES_DEFAULT 8
61 static unsigned int xennet_max_queues;
62 module_param_named(max_queues, xennet_max_queues, uint, 0644);
63 MODULE_PARM_DESC(max_queues,
64 		 "Maximum number of queues per virtual interface");
65 
66 static const struct ethtool_ops xennet_ethtool_ops;
67 
68 struct netfront_cb {
69 	int pull_to;
70 };
71 
72 #define NETFRONT_SKB_CB(skb)	((struct netfront_cb *)((skb)->cb))
73 
74 #define RX_COPY_THRESHOLD 256
75 
76 #define GRANT_INVALID_REF	0
77 
78 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
79 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
80 
81 /* Minimum number of Rx slots (includes slot for GSO metadata). */
82 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
83 
84 /* Queue name is interface name with "-qNNN" appended */
85 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
86 
87 /* IRQ name is queue name with "-tx" or "-rx" appended */
88 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
89 
90 static DECLARE_WAIT_QUEUE_HEAD(module_unload_q);
91 
92 struct netfront_stats {
93 	u64			packets;
94 	u64			bytes;
95 	struct u64_stats_sync	syncp;
96 };
97 
98 struct netfront_info;
99 
100 struct netfront_queue {
101 	unsigned int id; /* Queue ID, 0-based */
102 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
103 	struct netfront_info *info;
104 
105 	struct napi_struct napi;
106 
107 	/* Split event channels support, tx_* == rx_* when using
108 	 * single event channel.
109 	 */
110 	unsigned int tx_evtchn, rx_evtchn;
111 	unsigned int tx_irq, rx_irq;
112 	/* Only used when split event channels support is enabled */
113 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
114 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
115 
116 	spinlock_t   tx_lock;
117 	struct xen_netif_tx_front_ring tx;
118 	int tx_ring_ref;
119 
120 	/*
121 	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
122 	 * are linked from tx_skb_freelist through skb_entry.link.
123 	 *
124 	 *  NB. Freelist index entries are always going to be less than
125 	 *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
126 	 *  greater than PAGE_OFFSET: we use this property to distinguish
127 	 *  them.
128 	 */
129 	union skb_entry {
130 		struct sk_buff *skb;
131 		unsigned long link;
132 	} tx_skbs[NET_TX_RING_SIZE];
133 	grant_ref_t gref_tx_head;
134 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
135 	struct page *grant_tx_page[NET_TX_RING_SIZE];
136 	unsigned tx_skb_freelist;
137 
138 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
139 	struct xen_netif_rx_front_ring rx;
140 	int rx_ring_ref;
141 
142 	struct timer_list rx_refill_timer;
143 
144 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
145 	grant_ref_t gref_rx_head;
146 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
147 };
148 
149 struct netfront_info {
150 	struct list_head list;
151 	struct net_device *netdev;
152 
153 	struct xenbus_device *xbdev;
154 
155 	/* Multi-queue support */
156 	struct netfront_queue *queues;
157 
158 	/* Statistics */
159 	struct netfront_stats __percpu *rx_stats;
160 	struct netfront_stats __percpu *tx_stats;
161 
162 	atomic_t rx_gso_checksum_fixup;
163 };
164 
165 struct netfront_rx_info {
166 	struct xen_netif_rx_response rx;
167 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
168 };
169 
170 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
171 {
172 	list->link = id;
173 }
174 
175 static int skb_entry_is_link(const union skb_entry *list)
176 {
177 	BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
178 	return (unsigned long)list->skb < PAGE_OFFSET;
179 }
180 
181 /*
182  * Access macros for acquiring freeing slots in tx_skbs[].
183  */
184 
185 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
186 			       unsigned short id)
187 {
188 	skb_entry_set_link(&list[id], *head);
189 	*head = id;
190 }
191 
192 static unsigned short get_id_from_freelist(unsigned *head,
193 					   union skb_entry *list)
194 {
195 	unsigned int id = *head;
196 	*head = list[id].link;
197 	return id;
198 }
199 
200 static int xennet_rxidx(RING_IDX idx)
201 {
202 	return idx & (NET_RX_RING_SIZE - 1);
203 }
204 
205 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
206 					 RING_IDX ri)
207 {
208 	int i = xennet_rxidx(ri);
209 	struct sk_buff *skb = queue->rx_skbs[i];
210 	queue->rx_skbs[i] = NULL;
211 	return skb;
212 }
213 
214 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
215 					    RING_IDX ri)
216 {
217 	int i = xennet_rxidx(ri);
218 	grant_ref_t ref = queue->grant_rx_ref[i];
219 	queue->grant_rx_ref[i] = GRANT_INVALID_REF;
220 	return ref;
221 }
222 
223 #ifdef CONFIG_SYSFS
224 static const struct attribute_group xennet_dev_group;
225 #endif
226 
227 static bool xennet_can_sg(struct net_device *dev)
228 {
229 	return dev->features & NETIF_F_SG;
230 }
231 
232 
233 static void rx_refill_timeout(struct timer_list *t)
234 {
235 	struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
236 	napi_schedule(&queue->napi);
237 }
238 
239 static int netfront_tx_slot_available(struct netfront_queue *queue)
240 {
241 	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
242 		(NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
243 }
244 
245 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
246 {
247 	struct net_device *dev = queue->info->netdev;
248 	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
249 
250 	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
251 	    netfront_tx_slot_available(queue) &&
252 	    likely(netif_running(dev)))
253 		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
254 }
255 
256 
257 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
258 {
259 	struct sk_buff *skb;
260 	struct page *page;
261 
262 	skb = __netdev_alloc_skb(queue->info->netdev,
263 				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
264 				 GFP_ATOMIC | __GFP_NOWARN);
265 	if (unlikely(!skb))
266 		return NULL;
267 
268 	page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
269 	if (!page) {
270 		kfree_skb(skb);
271 		return NULL;
272 	}
273 	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
274 
275 	/* Align ip header to a 16 bytes boundary */
276 	skb_reserve(skb, NET_IP_ALIGN);
277 	skb->dev = queue->info->netdev;
278 
279 	return skb;
280 }
281 
282 
283 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
284 {
285 	RING_IDX req_prod = queue->rx.req_prod_pvt;
286 	int notify;
287 	int err = 0;
288 
289 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
290 		return;
291 
292 	for (req_prod = queue->rx.req_prod_pvt;
293 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
294 	     req_prod++) {
295 		struct sk_buff *skb;
296 		unsigned short id;
297 		grant_ref_t ref;
298 		struct page *page;
299 		struct xen_netif_rx_request *req;
300 
301 		skb = xennet_alloc_one_rx_buffer(queue);
302 		if (!skb) {
303 			err = -ENOMEM;
304 			break;
305 		}
306 
307 		id = xennet_rxidx(req_prod);
308 
309 		BUG_ON(queue->rx_skbs[id]);
310 		queue->rx_skbs[id] = skb;
311 
312 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
313 		WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
314 		queue->grant_rx_ref[id] = ref;
315 
316 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
317 
318 		req = RING_GET_REQUEST(&queue->rx, req_prod);
319 		gnttab_page_grant_foreign_access_ref_one(ref,
320 							 queue->info->xbdev->otherend_id,
321 							 page,
322 							 0);
323 		req->id = id;
324 		req->gref = ref;
325 	}
326 
327 	queue->rx.req_prod_pvt = req_prod;
328 
329 	/* Try again later if there are not enough requests or skb allocation
330 	 * failed.
331 	 * Enough requests is quantified as the sum of newly created slots and
332 	 * the unconsumed slots at the backend.
333 	 */
334 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
335 	    unlikely(err)) {
336 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
337 		return;
338 	}
339 
340 	wmb();		/* barrier so backend seens requests */
341 
342 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
343 	if (notify)
344 		notify_remote_via_irq(queue->rx_irq);
345 }
346 
347 static int xennet_open(struct net_device *dev)
348 {
349 	struct netfront_info *np = netdev_priv(dev);
350 	unsigned int num_queues = dev->real_num_tx_queues;
351 	unsigned int i = 0;
352 	struct netfront_queue *queue = NULL;
353 
354 	if (!np->queues)
355 		return -ENODEV;
356 
357 	for (i = 0; i < num_queues; ++i) {
358 		queue = &np->queues[i];
359 		napi_enable(&queue->napi);
360 
361 		spin_lock_bh(&queue->rx_lock);
362 		if (netif_carrier_ok(dev)) {
363 			xennet_alloc_rx_buffers(queue);
364 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
365 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
366 				napi_schedule(&queue->napi);
367 		}
368 		spin_unlock_bh(&queue->rx_lock);
369 	}
370 
371 	netif_tx_start_all_queues(dev);
372 
373 	return 0;
374 }
375 
376 static void xennet_tx_buf_gc(struct netfront_queue *queue)
377 {
378 	RING_IDX cons, prod;
379 	unsigned short id;
380 	struct sk_buff *skb;
381 	bool more_to_do;
382 
383 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
384 
385 	do {
386 		prod = queue->tx.sring->rsp_prod;
387 		rmb(); /* Ensure we see responses up to 'rp'. */
388 
389 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
390 			struct xen_netif_tx_response *txrsp;
391 
392 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
393 			if (txrsp->status == XEN_NETIF_RSP_NULL)
394 				continue;
395 
396 			id  = txrsp->id;
397 			skb = queue->tx_skbs[id].skb;
398 			if (unlikely(gnttab_query_foreign_access(
399 				queue->grant_tx_ref[id]) != 0)) {
400 				pr_alert("%s: warning -- grant still in use by backend domain\n",
401 					 __func__);
402 				BUG();
403 			}
404 			gnttab_end_foreign_access_ref(
405 				queue->grant_tx_ref[id], GNTMAP_readonly);
406 			gnttab_release_grant_reference(
407 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
408 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
409 			queue->grant_tx_page[id] = NULL;
410 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
411 			dev_kfree_skb_irq(skb);
412 		}
413 
414 		queue->tx.rsp_cons = prod;
415 
416 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
417 	} while (more_to_do);
418 
419 	xennet_maybe_wake_tx(queue);
420 }
421 
422 struct xennet_gnttab_make_txreq {
423 	struct netfront_queue *queue;
424 	struct sk_buff *skb;
425 	struct page *page;
426 	struct xen_netif_tx_request *tx; /* Last request */
427 	unsigned int size;
428 };
429 
430 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
431 				  unsigned int len, void *data)
432 {
433 	struct xennet_gnttab_make_txreq *info = data;
434 	unsigned int id;
435 	struct xen_netif_tx_request *tx;
436 	grant_ref_t ref;
437 	/* convenient aliases */
438 	struct page *page = info->page;
439 	struct netfront_queue *queue = info->queue;
440 	struct sk_buff *skb = info->skb;
441 
442 	id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
443 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
444 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
445 	WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
446 
447 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
448 					gfn, GNTMAP_readonly);
449 
450 	queue->tx_skbs[id].skb = skb;
451 	queue->grant_tx_page[id] = page;
452 	queue->grant_tx_ref[id] = ref;
453 
454 	tx->id = id;
455 	tx->gref = ref;
456 	tx->offset = offset;
457 	tx->size = len;
458 	tx->flags = 0;
459 
460 	info->tx = tx;
461 	info->size += tx->size;
462 }
463 
464 static struct xen_netif_tx_request *xennet_make_first_txreq(
465 	struct netfront_queue *queue, struct sk_buff *skb,
466 	struct page *page, unsigned int offset, unsigned int len)
467 {
468 	struct xennet_gnttab_make_txreq info = {
469 		.queue = queue,
470 		.skb = skb,
471 		.page = page,
472 		.size = 0,
473 	};
474 
475 	gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
476 
477 	return info.tx;
478 }
479 
480 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
481 				  unsigned int len, void *data)
482 {
483 	struct xennet_gnttab_make_txreq *info = data;
484 
485 	info->tx->flags |= XEN_NETTXF_more_data;
486 	skb_get(info->skb);
487 	xennet_tx_setup_grant(gfn, offset, len, data);
488 }
489 
490 static struct xen_netif_tx_request *xennet_make_txreqs(
491 	struct netfront_queue *queue, struct xen_netif_tx_request *tx,
492 	struct sk_buff *skb, struct page *page,
493 	unsigned int offset, unsigned int len)
494 {
495 	struct xennet_gnttab_make_txreq info = {
496 		.queue = queue,
497 		.skb = skb,
498 		.tx = tx,
499 	};
500 
501 	/* Skip unused frames from start of page */
502 	page += offset >> PAGE_SHIFT;
503 	offset &= ~PAGE_MASK;
504 
505 	while (len) {
506 		info.page = page;
507 		info.size = 0;
508 
509 		gnttab_foreach_grant_in_range(page, offset, len,
510 					      xennet_make_one_txreq,
511 					      &info);
512 
513 		page++;
514 		offset = 0;
515 		len -= info.size;
516 	}
517 
518 	return info.tx;
519 }
520 
521 /*
522  * Count how many ring slots are required to send this skb. Each frag
523  * might be a compound page.
524  */
525 static int xennet_count_skb_slots(struct sk_buff *skb)
526 {
527 	int i, frags = skb_shinfo(skb)->nr_frags;
528 	int slots;
529 
530 	slots = gnttab_count_grant(offset_in_page(skb->data),
531 				   skb_headlen(skb));
532 
533 	for (i = 0; i < frags; i++) {
534 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
535 		unsigned long size = skb_frag_size(frag);
536 		unsigned long offset = frag->page_offset;
537 
538 		/* Skip unused frames from start of page */
539 		offset &= ~PAGE_MASK;
540 
541 		slots += gnttab_count_grant(offset, size);
542 	}
543 
544 	return slots;
545 }
546 
547 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
548 			       struct net_device *sb_dev,
549 			       select_queue_fallback_t fallback)
550 {
551 	unsigned int num_queues = dev->real_num_tx_queues;
552 	u32 hash;
553 	u16 queue_idx;
554 
555 	/* First, check if there is only one queue */
556 	if (num_queues == 1) {
557 		queue_idx = 0;
558 	} else {
559 		hash = skb_get_hash(skb);
560 		queue_idx = hash % num_queues;
561 	}
562 
563 	return queue_idx;
564 }
565 
566 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
567 
568 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
569 {
570 	struct netfront_info *np = netdev_priv(dev);
571 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
572 	struct xen_netif_tx_request *tx, *first_tx;
573 	unsigned int i;
574 	int notify;
575 	int slots;
576 	struct page *page;
577 	unsigned int offset;
578 	unsigned int len;
579 	unsigned long flags;
580 	struct netfront_queue *queue = NULL;
581 	unsigned int num_queues = dev->real_num_tx_queues;
582 	u16 queue_index;
583 	struct sk_buff *nskb;
584 
585 	/* Drop the packet if no queues are set up */
586 	if (num_queues < 1)
587 		goto drop;
588 	/* Determine which queue to transmit this SKB on */
589 	queue_index = skb_get_queue_mapping(skb);
590 	queue = &np->queues[queue_index];
591 
592 	/* If skb->len is too big for wire format, drop skb and alert
593 	 * user about misconfiguration.
594 	 */
595 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
596 		net_alert_ratelimited(
597 			"xennet: skb->len = %u, too big for wire format\n",
598 			skb->len);
599 		goto drop;
600 	}
601 
602 	slots = xennet_count_skb_slots(skb);
603 	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
604 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
605 				    slots, skb->len);
606 		if (skb_linearize(skb))
607 			goto drop;
608 	}
609 
610 	page = virt_to_page(skb->data);
611 	offset = offset_in_page(skb->data);
612 
613 	/* The first req should be at least ETH_HLEN size or the packet will be
614 	 * dropped by netback.
615 	 */
616 	if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
617 		nskb = skb_copy(skb, GFP_ATOMIC);
618 		if (!nskb)
619 			goto drop;
620 		dev_consume_skb_any(skb);
621 		skb = nskb;
622 		page = virt_to_page(skb->data);
623 		offset = offset_in_page(skb->data);
624 	}
625 
626 	len = skb_headlen(skb);
627 
628 	spin_lock_irqsave(&queue->tx_lock, flags);
629 
630 	if (unlikely(!netif_carrier_ok(dev) ||
631 		     (slots > 1 && !xennet_can_sg(dev)) ||
632 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
633 		spin_unlock_irqrestore(&queue->tx_lock, flags);
634 		goto drop;
635 	}
636 
637 	/* First request for the linear area. */
638 	first_tx = tx = xennet_make_first_txreq(queue, skb,
639 						page, offset, len);
640 	offset += tx->size;
641 	if (offset == PAGE_SIZE) {
642 		page++;
643 		offset = 0;
644 	}
645 	len -= tx->size;
646 
647 	if (skb->ip_summed == CHECKSUM_PARTIAL)
648 		/* local packet? */
649 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
650 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
651 		/* remote but checksummed. */
652 		tx->flags |= XEN_NETTXF_data_validated;
653 
654 	/* Optional extra info after the first request. */
655 	if (skb_shinfo(skb)->gso_size) {
656 		struct xen_netif_extra_info *gso;
657 
658 		gso = (struct xen_netif_extra_info *)
659 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
660 
661 		tx->flags |= XEN_NETTXF_extra_info;
662 
663 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
664 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
665 			XEN_NETIF_GSO_TYPE_TCPV6 :
666 			XEN_NETIF_GSO_TYPE_TCPV4;
667 		gso->u.gso.pad = 0;
668 		gso->u.gso.features = 0;
669 
670 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
671 		gso->flags = 0;
672 	}
673 
674 	/* Requests for the rest of the linear area. */
675 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
676 
677 	/* Requests for all the frags. */
678 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
679 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
680 		tx = xennet_make_txreqs(queue, tx, skb,
681 					skb_frag_page(frag), frag->page_offset,
682 					skb_frag_size(frag));
683 	}
684 
685 	/* First request has the packet length. */
686 	first_tx->size = skb->len;
687 
688 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
689 	if (notify)
690 		notify_remote_via_irq(queue->tx_irq);
691 
692 	u64_stats_update_begin(&tx_stats->syncp);
693 	tx_stats->bytes += skb->len;
694 	tx_stats->packets++;
695 	u64_stats_update_end(&tx_stats->syncp);
696 
697 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
698 	xennet_tx_buf_gc(queue);
699 
700 	if (!netfront_tx_slot_available(queue))
701 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
702 
703 	spin_unlock_irqrestore(&queue->tx_lock, flags);
704 
705 	return NETDEV_TX_OK;
706 
707  drop:
708 	dev->stats.tx_dropped++;
709 	dev_kfree_skb_any(skb);
710 	return NETDEV_TX_OK;
711 }
712 
713 static int xennet_close(struct net_device *dev)
714 {
715 	struct netfront_info *np = netdev_priv(dev);
716 	unsigned int num_queues = dev->real_num_tx_queues;
717 	unsigned int i;
718 	struct netfront_queue *queue;
719 	netif_tx_stop_all_queues(np->netdev);
720 	for (i = 0; i < num_queues; ++i) {
721 		queue = &np->queues[i];
722 		napi_disable(&queue->napi);
723 	}
724 	return 0;
725 }
726 
727 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
728 				grant_ref_t ref)
729 {
730 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
731 
732 	BUG_ON(queue->rx_skbs[new]);
733 	queue->rx_skbs[new] = skb;
734 	queue->grant_rx_ref[new] = ref;
735 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
736 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
737 	queue->rx.req_prod_pvt++;
738 }
739 
740 static int xennet_get_extras(struct netfront_queue *queue,
741 			     struct xen_netif_extra_info *extras,
742 			     RING_IDX rp)
743 
744 {
745 	struct xen_netif_extra_info *extra;
746 	struct device *dev = &queue->info->netdev->dev;
747 	RING_IDX cons = queue->rx.rsp_cons;
748 	int err = 0;
749 
750 	do {
751 		struct sk_buff *skb;
752 		grant_ref_t ref;
753 
754 		if (unlikely(cons + 1 == rp)) {
755 			if (net_ratelimit())
756 				dev_warn(dev, "Missing extra info\n");
757 			err = -EBADR;
758 			break;
759 		}
760 
761 		extra = (struct xen_netif_extra_info *)
762 			RING_GET_RESPONSE(&queue->rx, ++cons);
763 
764 		if (unlikely(!extra->type ||
765 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
766 			if (net_ratelimit())
767 				dev_warn(dev, "Invalid extra type: %d\n",
768 					extra->type);
769 			err = -EINVAL;
770 		} else {
771 			memcpy(&extras[extra->type - 1], extra,
772 			       sizeof(*extra));
773 		}
774 
775 		skb = xennet_get_rx_skb(queue, cons);
776 		ref = xennet_get_rx_ref(queue, cons);
777 		xennet_move_rx_slot(queue, skb, ref);
778 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
779 
780 	queue->rx.rsp_cons = cons;
781 	return err;
782 }
783 
784 static int xennet_get_responses(struct netfront_queue *queue,
785 				struct netfront_rx_info *rinfo, RING_IDX rp,
786 				struct sk_buff_head *list)
787 {
788 	struct xen_netif_rx_response *rx = &rinfo->rx;
789 	struct xen_netif_extra_info *extras = rinfo->extras;
790 	struct device *dev = &queue->info->netdev->dev;
791 	RING_IDX cons = queue->rx.rsp_cons;
792 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
793 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
794 	int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
795 	int slots = 1;
796 	int err = 0;
797 	unsigned long ret;
798 
799 	if (rx->flags & XEN_NETRXF_extra_info) {
800 		err = xennet_get_extras(queue, extras, rp);
801 		cons = queue->rx.rsp_cons;
802 	}
803 
804 	for (;;) {
805 		if (unlikely(rx->status < 0 ||
806 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
807 			if (net_ratelimit())
808 				dev_warn(dev, "rx->offset: %u, size: %d\n",
809 					 rx->offset, rx->status);
810 			xennet_move_rx_slot(queue, skb, ref);
811 			err = -EINVAL;
812 			goto next;
813 		}
814 
815 		/*
816 		 * This definitely indicates a bug, either in this driver or in
817 		 * the backend driver. In future this should flag the bad
818 		 * situation to the system controller to reboot the backend.
819 		 */
820 		if (ref == GRANT_INVALID_REF) {
821 			if (net_ratelimit())
822 				dev_warn(dev, "Bad rx response id %d.\n",
823 					 rx->id);
824 			err = -EINVAL;
825 			goto next;
826 		}
827 
828 		ret = gnttab_end_foreign_access_ref(ref, 0);
829 		BUG_ON(!ret);
830 
831 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
832 
833 		__skb_queue_tail(list, skb);
834 
835 next:
836 		if (!(rx->flags & XEN_NETRXF_more_data))
837 			break;
838 
839 		if (cons + slots == rp) {
840 			if (net_ratelimit())
841 				dev_warn(dev, "Need more slots\n");
842 			err = -ENOENT;
843 			break;
844 		}
845 
846 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
847 		skb = xennet_get_rx_skb(queue, cons + slots);
848 		ref = xennet_get_rx_ref(queue, cons + slots);
849 		slots++;
850 	}
851 
852 	if (unlikely(slots > max)) {
853 		if (net_ratelimit())
854 			dev_warn(dev, "Too many slots\n");
855 		err = -E2BIG;
856 	}
857 
858 	if (unlikely(err))
859 		queue->rx.rsp_cons = cons + slots;
860 
861 	return err;
862 }
863 
864 static int xennet_set_skb_gso(struct sk_buff *skb,
865 			      struct xen_netif_extra_info *gso)
866 {
867 	if (!gso->u.gso.size) {
868 		if (net_ratelimit())
869 			pr_warn("GSO size must not be zero\n");
870 		return -EINVAL;
871 	}
872 
873 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
874 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
875 		if (net_ratelimit())
876 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
877 		return -EINVAL;
878 	}
879 
880 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
881 	skb_shinfo(skb)->gso_type =
882 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
883 		SKB_GSO_TCPV4 :
884 		SKB_GSO_TCPV6;
885 
886 	/* Header must be checked, and gso_segs computed. */
887 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
888 	skb_shinfo(skb)->gso_segs = 0;
889 
890 	return 0;
891 }
892 
893 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
894 				  struct sk_buff *skb,
895 				  struct sk_buff_head *list)
896 {
897 	struct skb_shared_info *shinfo = skb_shinfo(skb);
898 	RING_IDX cons = queue->rx.rsp_cons;
899 	struct sk_buff *nskb;
900 
901 	while ((nskb = __skb_dequeue(list))) {
902 		struct xen_netif_rx_response *rx =
903 			RING_GET_RESPONSE(&queue->rx, ++cons);
904 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
905 
906 		if (shinfo->nr_frags == MAX_SKB_FRAGS) {
907 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
908 
909 			BUG_ON(pull_to <= skb_headlen(skb));
910 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
911 		}
912 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
913 
914 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
915 				rx->offset, rx->status, PAGE_SIZE);
916 
917 		skb_shinfo(nskb)->nr_frags = 0;
918 		kfree_skb(nskb);
919 	}
920 
921 	return cons;
922 }
923 
924 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
925 {
926 	bool recalculate_partial_csum = false;
927 
928 	/*
929 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
930 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
931 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
932 	 * recalculate the partial checksum.
933 	 */
934 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
935 		struct netfront_info *np = netdev_priv(dev);
936 		atomic_inc(&np->rx_gso_checksum_fixup);
937 		skb->ip_summed = CHECKSUM_PARTIAL;
938 		recalculate_partial_csum = true;
939 	}
940 
941 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
942 	if (skb->ip_summed != CHECKSUM_PARTIAL)
943 		return 0;
944 
945 	return skb_checksum_setup(skb, recalculate_partial_csum);
946 }
947 
948 static int handle_incoming_queue(struct netfront_queue *queue,
949 				 struct sk_buff_head *rxq)
950 {
951 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
952 	int packets_dropped = 0;
953 	struct sk_buff *skb;
954 
955 	while ((skb = __skb_dequeue(rxq)) != NULL) {
956 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
957 
958 		if (pull_to > skb_headlen(skb))
959 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
960 
961 		/* Ethernet work: Delayed to here as it peeks the header. */
962 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
963 		skb_reset_network_header(skb);
964 
965 		if (checksum_setup(queue->info->netdev, skb)) {
966 			kfree_skb(skb);
967 			packets_dropped++;
968 			queue->info->netdev->stats.rx_errors++;
969 			continue;
970 		}
971 
972 		u64_stats_update_begin(&rx_stats->syncp);
973 		rx_stats->packets++;
974 		rx_stats->bytes += skb->len;
975 		u64_stats_update_end(&rx_stats->syncp);
976 
977 		/* Pass it up. */
978 		napi_gro_receive(&queue->napi, skb);
979 	}
980 
981 	return packets_dropped;
982 }
983 
984 static int xennet_poll(struct napi_struct *napi, int budget)
985 {
986 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
987 	struct net_device *dev = queue->info->netdev;
988 	struct sk_buff *skb;
989 	struct netfront_rx_info rinfo;
990 	struct xen_netif_rx_response *rx = &rinfo.rx;
991 	struct xen_netif_extra_info *extras = rinfo.extras;
992 	RING_IDX i, rp;
993 	int work_done;
994 	struct sk_buff_head rxq;
995 	struct sk_buff_head errq;
996 	struct sk_buff_head tmpq;
997 	int err;
998 
999 	spin_lock(&queue->rx_lock);
1000 
1001 	skb_queue_head_init(&rxq);
1002 	skb_queue_head_init(&errq);
1003 	skb_queue_head_init(&tmpq);
1004 
1005 	rp = queue->rx.sring->rsp_prod;
1006 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1007 
1008 	i = queue->rx.rsp_cons;
1009 	work_done = 0;
1010 	while ((i != rp) && (work_done < budget)) {
1011 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1012 		memset(extras, 0, sizeof(rinfo.extras));
1013 
1014 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1015 
1016 		if (unlikely(err)) {
1017 err:
1018 			while ((skb = __skb_dequeue(&tmpq)))
1019 				__skb_queue_tail(&errq, skb);
1020 			dev->stats.rx_errors++;
1021 			i = queue->rx.rsp_cons;
1022 			continue;
1023 		}
1024 
1025 		skb = __skb_dequeue(&tmpq);
1026 
1027 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1028 			struct xen_netif_extra_info *gso;
1029 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1030 
1031 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1032 				__skb_queue_head(&tmpq, skb);
1033 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1034 				goto err;
1035 			}
1036 		}
1037 
1038 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1039 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1040 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1041 
1042 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1043 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1044 		skb->data_len = rx->status;
1045 		skb->len += rx->status;
1046 
1047 		i = xennet_fill_frags(queue, skb, &tmpq);
1048 
1049 		if (rx->flags & XEN_NETRXF_csum_blank)
1050 			skb->ip_summed = CHECKSUM_PARTIAL;
1051 		else if (rx->flags & XEN_NETRXF_data_validated)
1052 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1053 
1054 		__skb_queue_tail(&rxq, skb);
1055 
1056 		queue->rx.rsp_cons = ++i;
1057 		work_done++;
1058 	}
1059 
1060 	__skb_queue_purge(&errq);
1061 
1062 	work_done -= handle_incoming_queue(queue, &rxq);
1063 
1064 	xennet_alloc_rx_buffers(queue);
1065 
1066 	if (work_done < budget) {
1067 		int more_to_do = 0;
1068 
1069 		napi_complete_done(napi, work_done);
1070 
1071 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1072 		if (more_to_do)
1073 			napi_schedule(napi);
1074 	}
1075 
1076 	spin_unlock(&queue->rx_lock);
1077 
1078 	return work_done;
1079 }
1080 
1081 static int xennet_change_mtu(struct net_device *dev, int mtu)
1082 {
1083 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1084 
1085 	if (mtu > max)
1086 		return -EINVAL;
1087 	dev->mtu = mtu;
1088 	return 0;
1089 }
1090 
1091 static void xennet_get_stats64(struct net_device *dev,
1092 			       struct rtnl_link_stats64 *tot)
1093 {
1094 	struct netfront_info *np = netdev_priv(dev);
1095 	int cpu;
1096 
1097 	for_each_possible_cpu(cpu) {
1098 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1099 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1100 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1101 		unsigned int start;
1102 
1103 		do {
1104 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1105 			tx_packets = tx_stats->packets;
1106 			tx_bytes = tx_stats->bytes;
1107 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1108 
1109 		do {
1110 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1111 			rx_packets = rx_stats->packets;
1112 			rx_bytes = rx_stats->bytes;
1113 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1114 
1115 		tot->rx_packets += rx_packets;
1116 		tot->tx_packets += tx_packets;
1117 		tot->rx_bytes   += rx_bytes;
1118 		tot->tx_bytes   += tx_bytes;
1119 	}
1120 
1121 	tot->rx_errors  = dev->stats.rx_errors;
1122 	tot->tx_dropped = dev->stats.tx_dropped;
1123 }
1124 
1125 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1126 {
1127 	struct sk_buff *skb;
1128 	int i;
1129 
1130 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1131 		/* Skip over entries which are actually freelist references */
1132 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1133 			continue;
1134 
1135 		skb = queue->tx_skbs[i].skb;
1136 		get_page(queue->grant_tx_page[i]);
1137 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1138 					  GNTMAP_readonly,
1139 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1140 		queue->grant_tx_page[i] = NULL;
1141 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1142 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1143 		dev_kfree_skb_irq(skb);
1144 	}
1145 }
1146 
1147 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1148 {
1149 	int id, ref;
1150 
1151 	spin_lock_bh(&queue->rx_lock);
1152 
1153 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1154 		struct sk_buff *skb;
1155 		struct page *page;
1156 
1157 		skb = queue->rx_skbs[id];
1158 		if (!skb)
1159 			continue;
1160 
1161 		ref = queue->grant_rx_ref[id];
1162 		if (ref == GRANT_INVALID_REF)
1163 			continue;
1164 
1165 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1166 
1167 		/* gnttab_end_foreign_access() needs a page ref until
1168 		 * foreign access is ended (which may be deferred).
1169 		 */
1170 		get_page(page);
1171 		gnttab_end_foreign_access(ref, 0,
1172 					  (unsigned long)page_address(page));
1173 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1174 
1175 		kfree_skb(skb);
1176 	}
1177 
1178 	spin_unlock_bh(&queue->rx_lock);
1179 }
1180 
1181 static netdev_features_t xennet_fix_features(struct net_device *dev,
1182 	netdev_features_t features)
1183 {
1184 	struct netfront_info *np = netdev_priv(dev);
1185 
1186 	if (features & NETIF_F_SG &&
1187 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1188 		features &= ~NETIF_F_SG;
1189 
1190 	if (features & NETIF_F_IPV6_CSUM &&
1191 	    !xenbus_read_unsigned(np->xbdev->otherend,
1192 				  "feature-ipv6-csum-offload", 0))
1193 		features &= ~NETIF_F_IPV6_CSUM;
1194 
1195 	if (features & NETIF_F_TSO &&
1196 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1197 		features &= ~NETIF_F_TSO;
1198 
1199 	if (features & NETIF_F_TSO6 &&
1200 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1201 		features &= ~NETIF_F_TSO6;
1202 
1203 	return features;
1204 }
1205 
1206 static int xennet_set_features(struct net_device *dev,
1207 	netdev_features_t features)
1208 {
1209 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1210 		netdev_info(dev, "Reducing MTU because no SG offload");
1211 		dev->mtu = ETH_DATA_LEN;
1212 	}
1213 
1214 	return 0;
1215 }
1216 
1217 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1218 {
1219 	struct netfront_queue *queue = dev_id;
1220 	unsigned long flags;
1221 
1222 	spin_lock_irqsave(&queue->tx_lock, flags);
1223 	xennet_tx_buf_gc(queue);
1224 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1225 
1226 	return IRQ_HANDLED;
1227 }
1228 
1229 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1230 {
1231 	struct netfront_queue *queue = dev_id;
1232 	struct net_device *dev = queue->info->netdev;
1233 
1234 	if (likely(netif_carrier_ok(dev) &&
1235 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1236 		napi_schedule(&queue->napi);
1237 
1238 	return IRQ_HANDLED;
1239 }
1240 
1241 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1242 {
1243 	xennet_tx_interrupt(irq, dev_id);
1244 	xennet_rx_interrupt(irq, dev_id);
1245 	return IRQ_HANDLED;
1246 }
1247 
1248 #ifdef CONFIG_NET_POLL_CONTROLLER
1249 static void xennet_poll_controller(struct net_device *dev)
1250 {
1251 	/* Poll each queue */
1252 	struct netfront_info *info = netdev_priv(dev);
1253 	unsigned int num_queues = dev->real_num_tx_queues;
1254 	unsigned int i;
1255 	for (i = 0; i < num_queues; ++i)
1256 		xennet_interrupt(0, &info->queues[i]);
1257 }
1258 #endif
1259 
1260 static const struct net_device_ops xennet_netdev_ops = {
1261 	.ndo_open            = xennet_open,
1262 	.ndo_stop            = xennet_close,
1263 	.ndo_start_xmit      = xennet_start_xmit,
1264 	.ndo_change_mtu	     = xennet_change_mtu,
1265 	.ndo_get_stats64     = xennet_get_stats64,
1266 	.ndo_set_mac_address = eth_mac_addr,
1267 	.ndo_validate_addr   = eth_validate_addr,
1268 	.ndo_fix_features    = xennet_fix_features,
1269 	.ndo_set_features    = xennet_set_features,
1270 	.ndo_select_queue    = xennet_select_queue,
1271 #ifdef CONFIG_NET_POLL_CONTROLLER
1272 	.ndo_poll_controller = xennet_poll_controller,
1273 #endif
1274 };
1275 
1276 static void xennet_free_netdev(struct net_device *netdev)
1277 {
1278 	struct netfront_info *np = netdev_priv(netdev);
1279 
1280 	free_percpu(np->rx_stats);
1281 	free_percpu(np->tx_stats);
1282 	free_netdev(netdev);
1283 }
1284 
1285 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1286 {
1287 	int err;
1288 	struct net_device *netdev;
1289 	struct netfront_info *np;
1290 
1291 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1292 	if (!netdev)
1293 		return ERR_PTR(-ENOMEM);
1294 
1295 	np                   = netdev_priv(netdev);
1296 	np->xbdev            = dev;
1297 
1298 	np->queues = NULL;
1299 
1300 	err = -ENOMEM;
1301 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1302 	if (np->rx_stats == NULL)
1303 		goto exit;
1304 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1305 	if (np->tx_stats == NULL)
1306 		goto exit;
1307 
1308 	netdev->netdev_ops	= &xennet_netdev_ops;
1309 
1310 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1311 				  NETIF_F_GSO_ROBUST;
1312 	netdev->hw_features	= NETIF_F_SG |
1313 				  NETIF_F_IPV6_CSUM |
1314 				  NETIF_F_TSO | NETIF_F_TSO6;
1315 
1316 	/*
1317          * Assume that all hw features are available for now. This set
1318          * will be adjusted by the call to netdev_update_features() in
1319          * xennet_connect() which is the earliest point where we can
1320          * negotiate with the backend regarding supported features.
1321          */
1322 	netdev->features |= netdev->hw_features;
1323 
1324 	netdev->ethtool_ops = &xennet_ethtool_ops;
1325 	netdev->min_mtu = ETH_MIN_MTU;
1326 	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1327 	SET_NETDEV_DEV(netdev, &dev->dev);
1328 
1329 	np->netdev = netdev;
1330 
1331 	netif_carrier_off(netdev);
1332 
1333 	xenbus_switch_state(dev, XenbusStateInitialising);
1334 	return netdev;
1335 
1336  exit:
1337 	xennet_free_netdev(netdev);
1338 	return ERR_PTR(err);
1339 }
1340 
1341 /**
1342  * Entry point to this code when a new device is created.  Allocate the basic
1343  * structures and the ring buffers for communication with the backend, and
1344  * inform the backend of the appropriate details for those.
1345  */
1346 static int netfront_probe(struct xenbus_device *dev,
1347 			  const struct xenbus_device_id *id)
1348 {
1349 	int err;
1350 	struct net_device *netdev;
1351 	struct netfront_info *info;
1352 
1353 	netdev = xennet_create_dev(dev);
1354 	if (IS_ERR(netdev)) {
1355 		err = PTR_ERR(netdev);
1356 		xenbus_dev_fatal(dev, err, "creating netdev");
1357 		return err;
1358 	}
1359 
1360 	info = netdev_priv(netdev);
1361 	dev_set_drvdata(&dev->dev, info);
1362 #ifdef CONFIG_SYSFS
1363 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1364 #endif
1365 
1366 	return 0;
1367 }
1368 
1369 static void xennet_end_access(int ref, void *page)
1370 {
1371 	/* This frees the page as a side-effect */
1372 	if (ref != GRANT_INVALID_REF)
1373 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1374 }
1375 
1376 static void xennet_disconnect_backend(struct netfront_info *info)
1377 {
1378 	unsigned int i = 0;
1379 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1380 
1381 	netif_carrier_off(info->netdev);
1382 
1383 	for (i = 0; i < num_queues && info->queues; ++i) {
1384 		struct netfront_queue *queue = &info->queues[i];
1385 
1386 		del_timer_sync(&queue->rx_refill_timer);
1387 
1388 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1389 			unbind_from_irqhandler(queue->tx_irq, queue);
1390 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1391 			unbind_from_irqhandler(queue->tx_irq, queue);
1392 			unbind_from_irqhandler(queue->rx_irq, queue);
1393 		}
1394 		queue->tx_evtchn = queue->rx_evtchn = 0;
1395 		queue->tx_irq = queue->rx_irq = 0;
1396 
1397 		if (netif_running(info->netdev))
1398 			napi_synchronize(&queue->napi);
1399 
1400 		xennet_release_tx_bufs(queue);
1401 		xennet_release_rx_bufs(queue);
1402 		gnttab_free_grant_references(queue->gref_tx_head);
1403 		gnttab_free_grant_references(queue->gref_rx_head);
1404 
1405 		/* End access and free the pages */
1406 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1407 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1408 
1409 		queue->tx_ring_ref = GRANT_INVALID_REF;
1410 		queue->rx_ring_ref = GRANT_INVALID_REF;
1411 		queue->tx.sring = NULL;
1412 		queue->rx.sring = NULL;
1413 	}
1414 }
1415 
1416 /**
1417  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1418  * driver restart.  We tear down our netif structure and recreate it, but
1419  * leave the device-layer structures intact so that this is transparent to the
1420  * rest of the kernel.
1421  */
1422 static int netfront_resume(struct xenbus_device *dev)
1423 {
1424 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1425 
1426 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1427 
1428 	xennet_disconnect_backend(info);
1429 	return 0;
1430 }
1431 
1432 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1433 {
1434 	char *s, *e, *macstr;
1435 	int i;
1436 
1437 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1438 	if (IS_ERR(macstr))
1439 		return PTR_ERR(macstr);
1440 
1441 	for (i = 0; i < ETH_ALEN; i++) {
1442 		mac[i] = simple_strtoul(s, &e, 16);
1443 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1444 			kfree(macstr);
1445 			return -ENOENT;
1446 		}
1447 		s = e+1;
1448 	}
1449 
1450 	kfree(macstr);
1451 	return 0;
1452 }
1453 
1454 static int setup_netfront_single(struct netfront_queue *queue)
1455 {
1456 	int err;
1457 
1458 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1459 	if (err < 0)
1460 		goto fail;
1461 
1462 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1463 					xennet_interrupt,
1464 					0, queue->info->netdev->name, queue);
1465 	if (err < 0)
1466 		goto bind_fail;
1467 	queue->rx_evtchn = queue->tx_evtchn;
1468 	queue->rx_irq = queue->tx_irq = err;
1469 
1470 	return 0;
1471 
1472 bind_fail:
1473 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1474 	queue->tx_evtchn = 0;
1475 fail:
1476 	return err;
1477 }
1478 
1479 static int setup_netfront_split(struct netfront_queue *queue)
1480 {
1481 	int err;
1482 
1483 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1484 	if (err < 0)
1485 		goto fail;
1486 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1487 	if (err < 0)
1488 		goto alloc_rx_evtchn_fail;
1489 
1490 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1491 		 "%s-tx", queue->name);
1492 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1493 					xennet_tx_interrupt,
1494 					0, queue->tx_irq_name, queue);
1495 	if (err < 0)
1496 		goto bind_tx_fail;
1497 	queue->tx_irq = err;
1498 
1499 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1500 		 "%s-rx", queue->name);
1501 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1502 					xennet_rx_interrupt,
1503 					0, queue->rx_irq_name, queue);
1504 	if (err < 0)
1505 		goto bind_rx_fail;
1506 	queue->rx_irq = err;
1507 
1508 	return 0;
1509 
1510 bind_rx_fail:
1511 	unbind_from_irqhandler(queue->tx_irq, queue);
1512 	queue->tx_irq = 0;
1513 bind_tx_fail:
1514 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1515 	queue->rx_evtchn = 0;
1516 alloc_rx_evtchn_fail:
1517 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1518 	queue->tx_evtchn = 0;
1519 fail:
1520 	return err;
1521 }
1522 
1523 static int setup_netfront(struct xenbus_device *dev,
1524 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1525 {
1526 	struct xen_netif_tx_sring *txs;
1527 	struct xen_netif_rx_sring *rxs;
1528 	grant_ref_t gref;
1529 	int err;
1530 
1531 	queue->tx_ring_ref = GRANT_INVALID_REF;
1532 	queue->rx_ring_ref = GRANT_INVALID_REF;
1533 	queue->rx.sring = NULL;
1534 	queue->tx.sring = NULL;
1535 
1536 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1537 	if (!txs) {
1538 		err = -ENOMEM;
1539 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1540 		goto fail;
1541 	}
1542 	SHARED_RING_INIT(txs);
1543 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1544 
1545 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1546 	if (err < 0)
1547 		goto grant_tx_ring_fail;
1548 	queue->tx_ring_ref = gref;
1549 
1550 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1551 	if (!rxs) {
1552 		err = -ENOMEM;
1553 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1554 		goto alloc_rx_ring_fail;
1555 	}
1556 	SHARED_RING_INIT(rxs);
1557 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1558 
1559 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1560 	if (err < 0)
1561 		goto grant_rx_ring_fail;
1562 	queue->rx_ring_ref = gref;
1563 
1564 	if (feature_split_evtchn)
1565 		err = setup_netfront_split(queue);
1566 	/* setup single event channel if
1567 	 *  a) feature-split-event-channels == 0
1568 	 *  b) feature-split-event-channels == 1 but failed to setup
1569 	 */
1570 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1571 		err = setup_netfront_single(queue);
1572 
1573 	if (err)
1574 		goto alloc_evtchn_fail;
1575 
1576 	return 0;
1577 
1578 	/* If we fail to setup netfront, it is safe to just revoke access to
1579 	 * granted pages because backend is not accessing it at this point.
1580 	 */
1581 alloc_evtchn_fail:
1582 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1583 grant_rx_ring_fail:
1584 	free_page((unsigned long)rxs);
1585 alloc_rx_ring_fail:
1586 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1587 grant_tx_ring_fail:
1588 	free_page((unsigned long)txs);
1589 fail:
1590 	return err;
1591 }
1592 
1593 /* Queue-specific initialisation
1594  * This used to be done in xennet_create_dev() but must now
1595  * be run per-queue.
1596  */
1597 static int xennet_init_queue(struct netfront_queue *queue)
1598 {
1599 	unsigned short i;
1600 	int err = 0;
1601 
1602 	spin_lock_init(&queue->tx_lock);
1603 	spin_lock_init(&queue->rx_lock);
1604 
1605 	timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1606 
1607 	snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1608 		 queue->info->netdev->name, queue->id);
1609 
1610 	/* Initialise tx_skbs as a free chain containing every entry. */
1611 	queue->tx_skb_freelist = 0;
1612 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1613 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1614 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1615 		queue->grant_tx_page[i] = NULL;
1616 	}
1617 
1618 	/* Clear out rx_skbs */
1619 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1620 		queue->rx_skbs[i] = NULL;
1621 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1622 	}
1623 
1624 	/* A grant for every tx ring slot */
1625 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1626 					  &queue->gref_tx_head) < 0) {
1627 		pr_alert("can't alloc tx grant refs\n");
1628 		err = -ENOMEM;
1629 		goto exit;
1630 	}
1631 
1632 	/* A grant for every rx ring slot */
1633 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1634 					  &queue->gref_rx_head) < 0) {
1635 		pr_alert("can't alloc rx grant refs\n");
1636 		err = -ENOMEM;
1637 		goto exit_free_tx;
1638 	}
1639 
1640 	return 0;
1641 
1642  exit_free_tx:
1643 	gnttab_free_grant_references(queue->gref_tx_head);
1644  exit:
1645 	return err;
1646 }
1647 
1648 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1649 			   struct xenbus_transaction *xbt, int write_hierarchical)
1650 {
1651 	/* Write the queue-specific keys into XenStore in the traditional
1652 	 * way for a single queue, or in a queue subkeys for multiple
1653 	 * queues.
1654 	 */
1655 	struct xenbus_device *dev = queue->info->xbdev;
1656 	int err;
1657 	const char *message;
1658 	char *path;
1659 	size_t pathsize;
1660 
1661 	/* Choose the correct place to write the keys */
1662 	if (write_hierarchical) {
1663 		pathsize = strlen(dev->nodename) + 10;
1664 		path = kzalloc(pathsize, GFP_KERNEL);
1665 		if (!path) {
1666 			err = -ENOMEM;
1667 			message = "out of memory while writing ring references";
1668 			goto error;
1669 		}
1670 		snprintf(path, pathsize, "%s/queue-%u",
1671 				dev->nodename, queue->id);
1672 	} else {
1673 		path = (char *)dev->nodename;
1674 	}
1675 
1676 	/* Write ring references */
1677 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1678 			queue->tx_ring_ref);
1679 	if (err) {
1680 		message = "writing tx-ring-ref";
1681 		goto error;
1682 	}
1683 
1684 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1685 			queue->rx_ring_ref);
1686 	if (err) {
1687 		message = "writing rx-ring-ref";
1688 		goto error;
1689 	}
1690 
1691 	/* Write event channels; taking into account both shared
1692 	 * and split event channel scenarios.
1693 	 */
1694 	if (queue->tx_evtchn == queue->rx_evtchn) {
1695 		/* Shared event channel */
1696 		err = xenbus_printf(*xbt, path,
1697 				"event-channel", "%u", queue->tx_evtchn);
1698 		if (err) {
1699 			message = "writing event-channel";
1700 			goto error;
1701 		}
1702 	} else {
1703 		/* Split event channels */
1704 		err = xenbus_printf(*xbt, path,
1705 				"event-channel-tx", "%u", queue->tx_evtchn);
1706 		if (err) {
1707 			message = "writing event-channel-tx";
1708 			goto error;
1709 		}
1710 
1711 		err = xenbus_printf(*xbt, path,
1712 				"event-channel-rx", "%u", queue->rx_evtchn);
1713 		if (err) {
1714 			message = "writing event-channel-rx";
1715 			goto error;
1716 		}
1717 	}
1718 
1719 	if (write_hierarchical)
1720 		kfree(path);
1721 	return 0;
1722 
1723 error:
1724 	if (write_hierarchical)
1725 		kfree(path);
1726 	xenbus_dev_fatal(dev, err, "%s", message);
1727 	return err;
1728 }
1729 
1730 static void xennet_destroy_queues(struct netfront_info *info)
1731 {
1732 	unsigned int i;
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 		netif_napi_del(&queue->napi);
1740 	}
1741 
1742 	kfree(info->queues);
1743 	info->queues = NULL;
1744 }
1745 
1746 static int xennet_create_queues(struct netfront_info *info,
1747 				unsigned int *num_queues)
1748 {
1749 	unsigned int i;
1750 	int ret;
1751 
1752 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1753 			       GFP_KERNEL);
1754 	if (!info->queues)
1755 		return -ENOMEM;
1756 
1757 	for (i = 0; i < *num_queues; i++) {
1758 		struct netfront_queue *queue = &info->queues[i];
1759 
1760 		queue->id = i;
1761 		queue->info = info;
1762 
1763 		ret = xennet_init_queue(queue);
1764 		if (ret < 0) {
1765 			dev_warn(&info->xbdev->dev,
1766 				 "only created %d queues\n", i);
1767 			*num_queues = i;
1768 			break;
1769 		}
1770 
1771 		netif_napi_add(queue->info->netdev, &queue->napi,
1772 			       xennet_poll, 64);
1773 		if (netif_running(info->netdev))
1774 			napi_enable(&queue->napi);
1775 	}
1776 
1777 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
1778 
1779 	if (*num_queues == 0) {
1780 		dev_err(&info->xbdev->dev, "no queues\n");
1781 		return -EINVAL;
1782 	}
1783 	return 0;
1784 }
1785 
1786 /* Common code used when first setting up, and when resuming. */
1787 static int talk_to_netback(struct xenbus_device *dev,
1788 			   struct netfront_info *info)
1789 {
1790 	const char *message;
1791 	struct xenbus_transaction xbt;
1792 	int err;
1793 	unsigned int feature_split_evtchn;
1794 	unsigned int i = 0;
1795 	unsigned int max_queues = 0;
1796 	struct netfront_queue *queue = NULL;
1797 	unsigned int num_queues = 1;
1798 
1799 	info->netdev->irq = 0;
1800 
1801 	/* Check if backend supports multiple queues */
1802 	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1803 					  "multi-queue-max-queues", 1);
1804 	num_queues = min(max_queues, xennet_max_queues);
1805 
1806 	/* Check feature-split-event-channels */
1807 	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1808 					"feature-split-event-channels", 0);
1809 
1810 	/* Read mac addr. */
1811 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
1812 	if (err) {
1813 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1814 		goto out_unlocked;
1815 	}
1816 
1817 	rtnl_lock();
1818 	if (info->queues)
1819 		xennet_destroy_queues(info);
1820 
1821 	err = xennet_create_queues(info, &num_queues);
1822 	if (err < 0) {
1823 		xenbus_dev_fatal(dev, err, "creating queues");
1824 		kfree(info->queues);
1825 		info->queues = NULL;
1826 		goto out;
1827 	}
1828 	rtnl_unlock();
1829 
1830 	/* Create shared ring, alloc event channel -- for each queue */
1831 	for (i = 0; i < num_queues; ++i) {
1832 		queue = &info->queues[i];
1833 		err = setup_netfront(dev, queue, feature_split_evtchn);
1834 		if (err)
1835 			goto destroy_ring;
1836 	}
1837 
1838 again:
1839 	err = xenbus_transaction_start(&xbt);
1840 	if (err) {
1841 		xenbus_dev_fatal(dev, err, "starting transaction");
1842 		goto destroy_ring;
1843 	}
1844 
1845 	if (xenbus_exists(XBT_NIL,
1846 			  info->xbdev->otherend, "multi-queue-max-queues")) {
1847 		/* Write the number of queues */
1848 		err = xenbus_printf(xbt, dev->nodename,
1849 				    "multi-queue-num-queues", "%u", num_queues);
1850 		if (err) {
1851 			message = "writing multi-queue-num-queues";
1852 			goto abort_transaction_no_dev_fatal;
1853 		}
1854 	}
1855 
1856 	if (num_queues == 1) {
1857 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1858 		if (err)
1859 			goto abort_transaction_no_dev_fatal;
1860 	} else {
1861 		/* Write the keys for each queue */
1862 		for (i = 0; i < num_queues; ++i) {
1863 			queue = &info->queues[i];
1864 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1865 			if (err)
1866 				goto abort_transaction_no_dev_fatal;
1867 		}
1868 	}
1869 
1870 	/* The remaining keys are not queue-specific */
1871 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1872 			    1);
1873 	if (err) {
1874 		message = "writing request-rx-copy";
1875 		goto abort_transaction;
1876 	}
1877 
1878 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1879 	if (err) {
1880 		message = "writing feature-rx-notify";
1881 		goto abort_transaction;
1882 	}
1883 
1884 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1885 	if (err) {
1886 		message = "writing feature-sg";
1887 		goto abort_transaction;
1888 	}
1889 
1890 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1891 	if (err) {
1892 		message = "writing feature-gso-tcpv4";
1893 		goto abort_transaction;
1894 	}
1895 
1896 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1897 	if (err) {
1898 		message = "writing feature-gso-tcpv6";
1899 		goto abort_transaction;
1900 	}
1901 
1902 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1903 			   "1");
1904 	if (err) {
1905 		message = "writing feature-ipv6-csum-offload";
1906 		goto abort_transaction;
1907 	}
1908 
1909 	err = xenbus_transaction_end(xbt, 0);
1910 	if (err) {
1911 		if (err == -EAGAIN)
1912 			goto again;
1913 		xenbus_dev_fatal(dev, err, "completing transaction");
1914 		goto destroy_ring;
1915 	}
1916 
1917 	return 0;
1918 
1919  abort_transaction:
1920 	xenbus_dev_fatal(dev, err, "%s", message);
1921 abort_transaction_no_dev_fatal:
1922 	xenbus_transaction_end(xbt, 1);
1923  destroy_ring:
1924 	xennet_disconnect_backend(info);
1925 	rtnl_lock();
1926 	xennet_destroy_queues(info);
1927  out:
1928 	rtnl_unlock();
1929 out_unlocked:
1930 	device_unregister(&dev->dev);
1931 	return err;
1932 }
1933 
1934 static int xennet_connect(struct net_device *dev)
1935 {
1936 	struct netfront_info *np = netdev_priv(dev);
1937 	unsigned int num_queues = 0;
1938 	int err;
1939 	unsigned int j = 0;
1940 	struct netfront_queue *queue = NULL;
1941 
1942 	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1943 		dev_info(&dev->dev,
1944 			 "backend does not support copying receive path\n");
1945 		return -ENODEV;
1946 	}
1947 
1948 	err = talk_to_netback(np->xbdev, np);
1949 	if (err)
1950 		return err;
1951 
1952 	/* talk_to_netback() sets the correct number of queues */
1953 	num_queues = dev->real_num_tx_queues;
1954 
1955 	if (dev->reg_state == NETREG_UNINITIALIZED) {
1956 		err = register_netdev(dev);
1957 		if (err) {
1958 			pr_warn("%s: register_netdev err=%d\n", __func__, err);
1959 			device_unregister(&np->xbdev->dev);
1960 			return err;
1961 		}
1962 	}
1963 
1964 	rtnl_lock();
1965 	netdev_update_features(dev);
1966 	rtnl_unlock();
1967 
1968 	/*
1969 	 * All public and private state should now be sane.  Get
1970 	 * ready to start sending and receiving packets and give the driver
1971 	 * domain a kick because we've probably just requeued some
1972 	 * packets.
1973 	 */
1974 	netif_carrier_on(np->netdev);
1975 	for (j = 0; j < num_queues; ++j) {
1976 		queue = &np->queues[j];
1977 
1978 		notify_remote_via_irq(queue->tx_irq);
1979 		if (queue->tx_irq != queue->rx_irq)
1980 			notify_remote_via_irq(queue->rx_irq);
1981 
1982 		spin_lock_irq(&queue->tx_lock);
1983 		xennet_tx_buf_gc(queue);
1984 		spin_unlock_irq(&queue->tx_lock);
1985 
1986 		spin_lock_bh(&queue->rx_lock);
1987 		xennet_alloc_rx_buffers(queue);
1988 		spin_unlock_bh(&queue->rx_lock);
1989 	}
1990 
1991 	return 0;
1992 }
1993 
1994 /**
1995  * Callback received when the backend's state changes.
1996  */
1997 static void netback_changed(struct xenbus_device *dev,
1998 			    enum xenbus_state backend_state)
1999 {
2000 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2001 	struct net_device *netdev = np->netdev;
2002 
2003 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2004 
2005 	switch (backend_state) {
2006 	case XenbusStateInitialising:
2007 	case XenbusStateInitialised:
2008 	case XenbusStateReconfiguring:
2009 	case XenbusStateReconfigured:
2010 		break;
2011 
2012 	case XenbusStateUnknown:
2013 		wake_up_all(&module_unload_q);
2014 		break;
2015 
2016 	case XenbusStateInitWait:
2017 		if (dev->state != XenbusStateInitialising)
2018 			break;
2019 		if (xennet_connect(netdev) != 0)
2020 			break;
2021 		xenbus_switch_state(dev, XenbusStateConnected);
2022 		break;
2023 
2024 	case XenbusStateConnected:
2025 		netdev_notify_peers(netdev);
2026 		break;
2027 
2028 	case XenbusStateClosed:
2029 		wake_up_all(&module_unload_q);
2030 		if (dev->state == XenbusStateClosed)
2031 			break;
2032 		/* Missed the backend's CLOSING state -- fallthrough */
2033 	case XenbusStateClosing:
2034 		wake_up_all(&module_unload_q);
2035 		xenbus_frontend_closed(dev);
2036 		break;
2037 	}
2038 }
2039 
2040 static const struct xennet_stat {
2041 	char name[ETH_GSTRING_LEN];
2042 	u16 offset;
2043 } xennet_stats[] = {
2044 	{
2045 		"rx_gso_checksum_fixup",
2046 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2047 	},
2048 };
2049 
2050 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2051 {
2052 	switch (string_set) {
2053 	case ETH_SS_STATS:
2054 		return ARRAY_SIZE(xennet_stats);
2055 	default:
2056 		return -EINVAL;
2057 	}
2058 }
2059 
2060 static void xennet_get_ethtool_stats(struct net_device *dev,
2061 				     struct ethtool_stats *stats, u64 * data)
2062 {
2063 	void *np = netdev_priv(dev);
2064 	int i;
2065 
2066 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2067 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2068 }
2069 
2070 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2071 {
2072 	int i;
2073 
2074 	switch (stringset) {
2075 	case ETH_SS_STATS:
2076 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2077 			memcpy(data + i * ETH_GSTRING_LEN,
2078 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2079 		break;
2080 	}
2081 }
2082 
2083 static const struct ethtool_ops xennet_ethtool_ops =
2084 {
2085 	.get_link = ethtool_op_get_link,
2086 
2087 	.get_sset_count = xennet_get_sset_count,
2088 	.get_ethtool_stats = xennet_get_ethtool_stats,
2089 	.get_strings = xennet_get_strings,
2090 };
2091 
2092 #ifdef CONFIG_SYSFS
2093 static ssize_t show_rxbuf(struct device *dev,
2094 			  struct device_attribute *attr, char *buf)
2095 {
2096 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2097 }
2098 
2099 static ssize_t store_rxbuf(struct device *dev,
2100 			   struct device_attribute *attr,
2101 			   const char *buf, size_t len)
2102 {
2103 	char *endp;
2104 	unsigned long target;
2105 
2106 	if (!capable(CAP_NET_ADMIN))
2107 		return -EPERM;
2108 
2109 	target = simple_strtoul(buf, &endp, 0);
2110 	if (endp == buf)
2111 		return -EBADMSG;
2112 
2113 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2114 
2115 	return len;
2116 }
2117 
2118 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2119 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2120 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2121 
2122 static struct attribute *xennet_dev_attrs[] = {
2123 	&dev_attr_rxbuf_min.attr,
2124 	&dev_attr_rxbuf_max.attr,
2125 	&dev_attr_rxbuf_cur.attr,
2126 	NULL
2127 };
2128 
2129 static const struct attribute_group xennet_dev_group = {
2130 	.attrs = xennet_dev_attrs
2131 };
2132 #endif /* CONFIG_SYSFS */
2133 
2134 static int xennet_remove(struct xenbus_device *dev)
2135 {
2136 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2137 
2138 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2139 
2140 	if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) {
2141 		xenbus_switch_state(dev, XenbusStateClosing);
2142 		wait_event(module_unload_q,
2143 			   xenbus_read_driver_state(dev->otherend) ==
2144 			   XenbusStateClosing ||
2145 			   xenbus_read_driver_state(dev->otherend) ==
2146 			   XenbusStateUnknown);
2147 
2148 		xenbus_switch_state(dev, XenbusStateClosed);
2149 		wait_event(module_unload_q,
2150 			   xenbus_read_driver_state(dev->otherend) ==
2151 			   XenbusStateClosed ||
2152 			   xenbus_read_driver_state(dev->otherend) ==
2153 			   XenbusStateUnknown);
2154 	}
2155 
2156 	xennet_disconnect_backend(info);
2157 
2158 	if (info->netdev->reg_state == NETREG_REGISTERED)
2159 		unregister_netdev(info->netdev);
2160 
2161 	if (info->queues) {
2162 		rtnl_lock();
2163 		xennet_destroy_queues(info);
2164 		rtnl_unlock();
2165 	}
2166 	xennet_free_netdev(info->netdev);
2167 
2168 	return 0;
2169 }
2170 
2171 static const struct xenbus_device_id netfront_ids[] = {
2172 	{ "vif" },
2173 	{ "" }
2174 };
2175 
2176 static struct xenbus_driver netfront_driver = {
2177 	.ids = netfront_ids,
2178 	.probe = netfront_probe,
2179 	.remove = xennet_remove,
2180 	.resume = netfront_resume,
2181 	.otherend_changed = netback_changed,
2182 };
2183 
2184 static int __init netif_init(void)
2185 {
2186 	if (!xen_domain())
2187 		return -ENODEV;
2188 
2189 	if (!xen_has_pv_nic_devices())
2190 		return -ENODEV;
2191 
2192 	pr_info("Initialising Xen virtual ethernet driver\n");
2193 
2194 	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2195 	 * specified a value.
2196 	 */
2197 	if (xennet_max_queues == 0)
2198 		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2199 					  num_online_cpus());
2200 
2201 	return xenbus_register_frontend(&netfront_driver);
2202 }
2203 module_init(netif_init);
2204 
2205 
2206 static void __exit netif_exit(void)
2207 {
2208 	xenbus_unregister_driver(&netfront_driver);
2209 }
2210 module_exit(netif_exit);
2211 
2212 MODULE_DESCRIPTION("Xen virtual network device frontend");
2213 MODULE_LICENSE("GPL");
2214 MODULE_ALIAS("xen:vif");
2215 MODULE_ALIAS("xennet");
2216