xref: /openbmc/linux/drivers/net/xen-netfront.c (revision 0c874100)
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_wq);
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 	RING_IDX cons = queue->rx.rsp_cons;
898 	struct sk_buff *nskb;
899 
900 	while ((nskb = __skb_dequeue(list))) {
901 		struct xen_netif_rx_response *rx =
902 			RING_GET_RESPONSE(&queue->rx, ++cons);
903 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
904 
905 		if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
906 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
907 
908 			BUG_ON(pull_to <= skb_headlen(skb));
909 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
910 		}
911 		if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
912 			queue->rx.rsp_cons = ++cons;
913 			kfree_skb(nskb);
914 			return ~0U;
915 		}
916 
917 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
918 				skb_frag_page(nfrag),
919 				rx->offset, rx->status, PAGE_SIZE);
920 
921 		skb_shinfo(nskb)->nr_frags = 0;
922 		kfree_skb(nskb);
923 	}
924 
925 	return cons;
926 }
927 
928 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
929 {
930 	bool recalculate_partial_csum = false;
931 
932 	/*
933 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
934 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
935 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
936 	 * recalculate the partial checksum.
937 	 */
938 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
939 		struct netfront_info *np = netdev_priv(dev);
940 		atomic_inc(&np->rx_gso_checksum_fixup);
941 		skb->ip_summed = CHECKSUM_PARTIAL;
942 		recalculate_partial_csum = true;
943 	}
944 
945 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
946 	if (skb->ip_summed != CHECKSUM_PARTIAL)
947 		return 0;
948 
949 	return skb_checksum_setup(skb, recalculate_partial_csum);
950 }
951 
952 static int handle_incoming_queue(struct netfront_queue *queue,
953 				 struct sk_buff_head *rxq)
954 {
955 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
956 	int packets_dropped = 0;
957 	struct sk_buff *skb;
958 
959 	while ((skb = __skb_dequeue(rxq)) != NULL) {
960 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
961 
962 		if (pull_to > skb_headlen(skb))
963 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
964 
965 		/* Ethernet work: Delayed to here as it peeks the header. */
966 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
967 		skb_reset_network_header(skb);
968 
969 		if (checksum_setup(queue->info->netdev, skb)) {
970 			kfree_skb(skb);
971 			packets_dropped++;
972 			queue->info->netdev->stats.rx_errors++;
973 			continue;
974 		}
975 
976 		u64_stats_update_begin(&rx_stats->syncp);
977 		rx_stats->packets++;
978 		rx_stats->bytes += skb->len;
979 		u64_stats_update_end(&rx_stats->syncp);
980 
981 		/* Pass it up. */
982 		napi_gro_receive(&queue->napi, skb);
983 	}
984 
985 	return packets_dropped;
986 }
987 
988 static int xennet_poll(struct napi_struct *napi, int budget)
989 {
990 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
991 	struct net_device *dev = queue->info->netdev;
992 	struct sk_buff *skb;
993 	struct netfront_rx_info rinfo;
994 	struct xen_netif_rx_response *rx = &rinfo.rx;
995 	struct xen_netif_extra_info *extras = rinfo.extras;
996 	RING_IDX i, rp;
997 	int work_done;
998 	struct sk_buff_head rxq;
999 	struct sk_buff_head errq;
1000 	struct sk_buff_head tmpq;
1001 	int err;
1002 
1003 	spin_lock(&queue->rx_lock);
1004 
1005 	skb_queue_head_init(&rxq);
1006 	skb_queue_head_init(&errq);
1007 	skb_queue_head_init(&tmpq);
1008 
1009 	rp = queue->rx.sring->rsp_prod;
1010 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1011 
1012 	i = queue->rx.rsp_cons;
1013 	work_done = 0;
1014 	while ((i != rp) && (work_done < budget)) {
1015 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1016 		memset(extras, 0, sizeof(rinfo.extras));
1017 
1018 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1019 
1020 		if (unlikely(err)) {
1021 err:
1022 			while ((skb = __skb_dequeue(&tmpq)))
1023 				__skb_queue_tail(&errq, skb);
1024 			dev->stats.rx_errors++;
1025 			i = queue->rx.rsp_cons;
1026 			continue;
1027 		}
1028 
1029 		skb = __skb_dequeue(&tmpq);
1030 
1031 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1032 			struct xen_netif_extra_info *gso;
1033 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1034 
1035 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1036 				__skb_queue_head(&tmpq, skb);
1037 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1038 				goto err;
1039 			}
1040 		}
1041 
1042 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1043 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1044 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1045 
1046 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1047 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1048 		skb->data_len = rx->status;
1049 		skb->len += rx->status;
1050 
1051 		i = xennet_fill_frags(queue, skb, &tmpq);
1052 		if (unlikely(i == ~0U))
1053 			goto err;
1054 
1055 		if (rx->flags & XEN_NETRXF_csum_blank)
1056 			skb->ip_summed = CHECKSUM_PARTIAL;
1057 		else if (rx->flags & XEN_NETRXF_data_validated)
1058 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1059 
1060 		__skb_queue_tail(&rxq, skb);
1061 
1062 		queue->rx.rsp_cons = ++i;
1063 		work_done++;
1064 	}
1065 
1066 	__skb_queue_purge(&errq);
1067 
1068 	work_done -= handle_incoming_queue(queue, &rxq);
1069 
1070 	xennet_alloc_rx_buffers(queue);
1071 
1072 	if (work_done < budget) {
1073 		int more_to_do = 0;
1074 
1075 		napi_complete_done(napi, work_done);
1076 
1077 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1078 		if (more_to_do)
1079 			napi_schedule(napi);
1080 	}
1081 
1082 	spin_unlock(&queue->rx_lock);
1083 
1084 	return work_done;
1085 }
1086 
1087 static int xennet_change_mtu(struct net_device *dev, int mtu)
1088 {
1089 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1090 
1091 	if (mtu > max)
1092 		return -EINVAL;
1093 	dev->mtu = mtu;
1094 	return 0;
1095 }
1096 
1097 static void xennet_get_stats64(struct net_device *dev,
1098 			       struct rtnl_link_stats64 *tot)
1099 {
1100 	struct netfront_info *np = netdev_priv(dev);
1101 	int cpu;
1102 
1103 	for_each_possible_cpu(cpu) {
1104 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1105 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1106 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1107 		unsigned int start;
1108 
1109 		do {
1110 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1111 			tx_packets = tx_stats->packets;
1112 			tx_bytes = tx_stats->bytes;
1113 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1114 
1115 		do {
1116 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1117 			rx_packets = rx_stats->packets;
1118 			rx_bytes = rx_stats->bytes;
1119 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1120 
1121 		tot->rx_packets += rx_packets;
1122 		tot->tx_packets += tx_packets;
1123 		tot->rx_bytes   += rx_bytes;
1124 		tot->tx_bytes   += tx_bytes;
1125 	}
1126 
1127 	tot->rx_errors  = dev->stats.rx_errors;
1128 	tot->tx_dropped = dev->stats.tx_dropped;
1129 }
1130 
1131 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1132 {
1133 	struct sk_buff *skb;
1134 	int i;
1135 
1136 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1137 		/* Skip over entries which are actually freelist references */
1138 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1139 			continue;
1140 
1141 		skb = queue->tx_skbs[i].skb;
1142 		get_page(queue->grant_tx_page[i]);
1143 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1144 					  GNTMAP_readonly,
1145 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1146 		queue->grant_tx_page[i] = NULL;
1147 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1148 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1149 		dev_kfree_skb_irq(skb);
1150 	}
1151 }
1152 
1153 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1154 {
1155 	int id, ref;
1156 
1157 	spin_lock_bh(&queue->rx_lock);
1158 
1159 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1160 		struct sk_buff *skb;
1161 		struct page *page;
1162 
1163 		skb = queue->rx_skbs[id];
1164 		if (!skb)
1165 			continue;
1166 
1167 		ref = queue->grant_rx_ref[id];
1168 		if (ref == GRANT_INVALID_REF)
1169 			continue;
1170 
1171 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1172 
1173 		/* gnttab_end_foreign_access() needs a page ref until
1174 		 * foreign access is ended (which may be deferred).
1175 		 */
1176 		get_page(page);
1177 		gnttab_end_foreign_access(ref, 0,
1178 					  (unsigned long)page_address(page));
1179 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1180 
1181 		kfree_skb(skb);
1182 	}
1183 
1184 	spin_unlock_bh(&queue->rx_lock);
1185 }
1186 
1187 static netdev_features_t xennet_fix_features(struct net_device *dev,
1188 	netdev_features_t features)
1189 {
1190 	struct netfront_info *np = netdev_priv(dev);
1191 
1192 	if (features & NETIF_F_SG &&
1193 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1194 		features &= ~NETIF_F_SG;
1195 
1196 	if (features & NETIF_F_IPV6_CSUM &&
1197 	    !xenbus_read_unsigned(np->xbdev->otherend,
1198 				  "feature-ipv6-csum-offload", 0))
1199 		features &= ~NETIF_F_IPV6_CSUM;
1200 
1201 	if (features & NETIF_F_TSO &&
1202 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1203 		features &= ~NETIF_F_TSO;
1204 
1205 	if (features & NETIF_F_TSO6 &&
1206 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1207 		features &= ~NETIF_F_TSO6;
1208 
1209 	return features;
1210 }
1211 
1212 static int xennet_set_features(struct net_device *dev,
1213 	netdev_features_t features)
1214 {
1215 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1216 		netdev_info(dev, "Reducing MTU because no SG offload");
1217 		dev->mtu = ETH_DATA_LEN;
1218 	}
1219 
1220 	return 0;
1221 }
1222 
1223 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1224 {
1225 	struct netfront_queue *queue = dev_id;
1226 	unsigned long flags;
1227 
1228 	spin_lock_irqsave(&queue->tx_lock, flags);
1229 	xennet_tx_buf_gc(queue);
1230 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1231 
1232 	return IRQ_HANDLED;
1233 }
1234 
1235 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1236 {
1237 	struct netfront_queue *queue = dev_id;
1238 	struct net_device *dev = queue->info->netdev;
1239 
1240 	if (likely(netif_carrier_ok(dev) &&
1241 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1242 		napi_schedule(&queue->napi);
1243 
1244 	return IRQ_HANDLED;
1245 }
1246 
1247 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1248 {
1249 	xennet_tx_interrupt(irq, dev_id);
1250 	xennet_rx_interrupt(irq, dev_id);
1251 	return IRQ_HANDLED;
1252 }
1253 
1254 #ifdef CONFIG_NET_POLL_CONTROLLER
1255 static void xennet_poll_controller(struct net_device *dev)
1256 {
1257 	/* Poll each queue */
1258 	struct netfront_info *info = netdev_priv(dev);
1259 	unsigned int num_queues = dev->real_num_tx_queues;
1260 	unsigned int i;
1261 	for (i = 0; i < num_queues; ++i)
1262 		xennet_interrupt(0, &info->queues[i]);
1263 }
1264 #endif
1265 
1266 static const struct net_device_ops xennet_netdev_ops = {
1267 	.ndo_open            = xennet_open,
1268 	.ndo_stop            = xennet_close,
1269 	.ndo_start_xmit      = xennet_start_xmit,
1270 	.ndo_change_mtu	     = xennet_change_mtu,
1271 	.ndo_get_stats64     = xennet_get_stats64,
1272 	.ndo_set_mac_address = eth_mac_addr,
1273 	.ndo_validate_addr   = eth_validate_addr,
1274 	.ndo_fix_features    = xennet_fix_features,
1275 	.ndo_set_features    = xennet_set_features,
1276 	.ndo_select_queue    = xennet_select_queue,
1277 #ifdef CONFIG_NET_POLL_CONTROLLER
1278 	.ndo_poll_controller = xennet_poll_controller,
1279 #endif
1280 };
1281 
1282 static void xennet_free_netdev(struct net_device *netdev)
1283 {
1284 	struct netfront_info *np = netdev_priv(netdev);
1285 
1286 	free_percpu(np->rx_stats);
1287 	free_percpu(np->tx_stats);
1288 	free_netdev(netdev);
1289 }
1290 
1291 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1292 {
1293 	int err;
1294 	struct net_device *netdev;
1295 	struct netfront_info *np;
1296 
1297 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1298 	if (!netdev)
1299 		return ERR_PTR(-ENOMEM);
1300 
1301 	np                   = netdev_priv(netdev);
1302 	np->xbdev            = dev;
1303 
1304 	np->queues = NULL;
1305 
1306 	err = -ENOMEM;
1307 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1308 	if (np->rx_stats == NULL)
1309 		goto exit;
1310 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1311 	if (np->tx_stats == NULL)
1312 		goto exit;
1313 
1314 	netdev->netdev_ops	= &xennet_netdev_ops;
1315 
1316 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1317 				  NETIF_F_GSO_ROBUST;
1318 	netdev->hw_features	= NETIF_F_SG |
1319 				  NETIF_F_IPV6_CSUM |
1320 				  NETIF_F_TSO | NETIF_F_TSO6;
1321 
1322 	/*
1323          * Assume that all hw features are available for now. This set
1324          * will be adjusted by the call to netdev_update_features() in
1325          * xennet_connect() which is the earliest point where we can
1326          * negotiate with the backend regarding supported features.
1327          */
1328 	netdev->features |= netdev->hw_features;
1329 
1330 	netdev->ethtool_ops = &xennet_ethtool_ops;
1331 	netdev->min_mtu = ETH_MIN_MTU;
1332 	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1333 	SET_NETDEV_DEV(netdev, &dev->dev);
1334 
1335 	np->netdev = netdev;
1336 
1337 	netif_carrier_off(netdev);
1338 
1339 	xenbus_switch_state(dev, XenbusStateInitialising);
1340 	wait_event(module_wq,
1341 		   xenbus_read_driver_state(dev->otherend) !=
1342 		   XenbusStateClosed &&
1343 		   xenbus_read_driver_state(dev->otherend) !=
1344 		   XenbusStateUnknown);
1345 	return netdev;
1346 
1347  exit:
1348 	xennet_free_netdev(netdev);
1349 	return ERR_PTR(err);
1350 }
1351 
1352 /**
1353  * Entry point to this code when a new device is created.  Allocate the basic
1354  * structures and the ring buffers for communication with the backend, and
1355  * inform the backend of the appropriate details for those.
1356  */
1357 static int netfront_probe(struct xenbus_device *dev,
1358 			  const struct xenbus_device_id *id)
1359 {
1360 	int err;
1361 	struct net_device *netdev;
1362 	struct netfront_info *info;
1363 
1364 	netdev = xennet_create_dev(dev);
1365 	if (IS_ERR(netdev)) {
1366 		err = PTR_ERR(netdev);
1367 		xenbus_dev_fatal(dev, err, "creating netdev");
1368 		return err;
1369 	}
1370 
1371 	info = netdev_priv(netdev);
1372 	dev_set_drvdata(&dev->dev, info);
1373 #ifdef CONFIG_SYSFS
1374 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1375 #endif
1376 
1377 	return 0;
1378 }
1379 
1380 static void xennet_end_access(int ref, void *page)
1381 {
1382 	/* This frees the page as a side-effect */
1383 	if (ref != GRANT_INVALID_REF)
1384 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1385 }
1386 
1387 static void xennet_disconnect_backend(struct netfront_info *info)
1388 {
1389 	unsigned int i = 0;
1390 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1391 
1392 	netif_carrier_off(info->netdev);
1393 
1394 	for (i = 0; i < num_queues && info->queues; ++i) {
1395 		struct netfront_queue *queue = &info->queues[i];
1396 
1397 		del_timer_sync(&queue->rx_refill_timer);
1398 
1399 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1400 			unbind_from_irqhandler(queue->tx_irq, queue);
1401 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1402 			unbind_from_irqhandler(queue->tx_irq, queue);
1403 			unbind_from_irqhandler(queue->rx_irq, queue);
1404 		}
1405 		queue->tx_evtchn = queue->rx_evtchn = 0;
1406 		queue->tx_irq = queue->rx_irq = 0;
1407 
1408 		if (netif_running(info->netdev))
1409 			napi_synchronize(&queue->napi);
1410 
1411 		xennet_release_tx_bufs(queue);
1412 		xennet_release_rx_bufs(queue);
1413 		gnttab_free_grant_references(queue->gref_tx_head);
1414 		gnttab_free_grant_references(queue->gref_rx_head);
1415 
1416 		/* End access and free the pages */
1417 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1418 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1419 
1420 		queue->tx_ring_ref = GRANT_INVALID_REF;
1421 		queue->rx_ring_ref = GRANT_INVALID_REF;
1422 		queue->tx.sring = NULL;
1423 		queue->rx.sring = NULL;
1424 	}
1425 }
1426 
1427 /**
1428  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1429  * driver restart.  We tear down our netif structure and recreate it, but
1430  * leave the device-layer structures intact so that this is transparent to the
1431  * rest of the kernel.
1432  */
1433 static int netfront_resume(struct xenbus_device *dev)
1434 {
1435 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1436 
1437 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1438 
1439 	xennet_disconnect_backend(info);
1440 	return 0;
1441 }
1442 
1443 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1444 {
1445 	char *s, *e, *macstr;
1446 	int i;
1447 
1448 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1449 	if (IS_ERR(macstr))
1450 		return PTR_ERR(macstr);
1451 
1452 	for (i = 0; i < ETH_ALEN; i++) {
1453 		mac[i] = simple_strtoul(s, &e, 16);
1454 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1455 			kfree(macstr);
1456 			return -ENOENT;
1457 		}
1458 		s = e+1;
1459 	}
1460 
1461 	kfree(macstr);
1462 	return 0;
1463 }
1464 
1465 static int setup_netfront_single(struct netfront_queue *queue)
1466 {
1467 	int err;
1468 
1469 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1470 	if (err < 0)
1471 		goto fail;
1472 
1473 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1474 					xennet_interrupt,
1475 					0, queue->info->netdev->name, queue);
1476 	if (err < 0)
1477 		goto bind_fail;
1478 	queue->rx_evtchn = queue->tx_evtchn;
1479 	queue->rx_irq = queue->tx_irq = err;
1480 
1481 	return 0;
1482 
1483 bind_fail:
1484 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1485 	queue->tx_evtchn = 0;
1486 fail:
1487 	return err;
1488 }
1489 
1490 static int setup_netfront_split(struct netfront_queue *queue)
1491 {
1492 	int err;
1493 
1494 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1495 	if (err < 0)
1496 		goto fail;
1497 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1498 	if (err < 0)
1499 		goto alloc_rx_evtchn_fail;
1500 
1501 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1502 		 "%s-tx", queue->name);
1503 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1504 					xennet_tx_interrupt,
1505 					0, queue->tx_irq_name, queue);
1506 	if (err < 0)
1507 		goto bind_tx_fail;
1508 	queue->tx_irq = err;
1509 
1510 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1511 		 "%s-rx", queue->name);
1512 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1513 					xennet_rx_interrupt,
1514 					0, queue->rx_irq_name, queue);
1515 	if (err < 0)
1516 		goto bind_rx_fail;
1517 	queue->rx_irq = err;
1518 
1519 	return 0;
1520 
1521 bind_rx_fail:
1522 	unbind_from_irqhandler(queue->tx_irq, queue);
1523 	queue->tx_irq = 0;
1524 bind_tx_fail:
1525 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1526 	queue->rx_evtchn = 0;
1527 alloc_rx_evtchn_fail:
1528 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1529 	queue->tx_evtchn = 0;
1530 fail:
1531 	return err;
1532 }
1533 
1534 static int setup_netfront(struct xenbus_device *dev,
1535 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1536 {
1537 	struct xen_netif_tx_sring *txs;
1538 	struct xen_netif_rx_sring *rxs;
1539 	grant_ref_t gref;
1540 	int err;
1541 
1542 	queue->tx_ring_ref = GRANT_INVALID_REF;
1543 	queue->rx_ring_ref = GRANT_INVALID_REF;
1544 	queue->rx.sring = NULL;
1545 	queue->tx.sring = NULL;
1546 
1547 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1548 	if (!txs) {
1549 		err = -ENOMEM;
1550 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1551 		goto fail;
1552 	}
1553 	SHARED_RING_INIT(txs);
1554 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1555 
1556 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1557 	if (err < 0)
1558 		goto grant_tx_ring_fail;
1559 	queue->tx_ring_ref = gref;
1560 
1561 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1562 	if (!rxs) {
1563 		err = -ENOMEM;
1564 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1565 		goto alloc_rx_ring_fail;
1566 	}
1567 	SHARED_RING_INIT(rxs);
1568 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1569 
1570 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1571 	if (err < 0)
1572 		goto grant_rx_ring_fail;
1573 	queue->rx_ring_ref = gref;
1574 
1575 	if (feature_split_evtchn)
1576 		err = setup_netfront_split(queue);
1577 	/* setup single event channel if
1578 	 *  a) feature-split-event-channels == 0
1579 	 *  b) feature-split-event-channels == 1 but failed to setup
1580 	 */
1581 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1582 		err = setup_netfront_single(queue);
1583 
1584 	if (err)
1585 		goto alloc_evtchn_fail;
1586 
1587 	return 0;
1588 
1589 	/* If we fail to setup netfront, it is safe to just revoke access to
1590 	 * granted pages because backend is not accessing it at this point.
1591 	 */
1592 alloc_evtchn_fail:
1593 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1594 grant_rx_ring_fail:
1595 	free_page((unsigned long)rxs);
1596 alloc_rx_ring_fail:
1597 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1598 grant_tx_ring_fail:
1599 	free_page((unsigned long)txs);
1600 fail:
1601 	return err;
1602 }
1603 
1604 /* Queue-specific initialisation
1605  * This used to be done in xennet_create_dev() but must now
1606  * be run per-queue.
1607  */
1608 static int xennet_init_queue(struct netfront_queue *queue)
1609 {
1610 	unsigned short i;
1611 	int err = 0;
1612 	char *devid;
1613 
1614 	spin_lock_init(&queue->tx_lock);
1615 	spin_lock_init(&queue->rx_lock);
1616 
1617 	timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1618 
1619 	devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1620 	snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1621 		 devid, queue->id);
1622 
1623 	/* Initialise tx_skbs as a free chain containing every entry. */
1624 	queue->tx_skb_freelist = 0;
1625 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1626 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1627 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1628 		queue->grant_tx_page[i] = NULL;
1629 	}
1630 
1631 	/* Clear out rx_skbs */
1632 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1633 		queue->rx_skbs[i] = NULL;
1634 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1635 	}
1636 
1637 	/* A grant for every tx ring slot */
1638 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1639 					  &queue->gref_tx_head) < 0) {
1640 		pr_alert("can't alloc tx grant refs\n");
1641 		err = -ENOMEM;
1642 		goto exit;
1643 	}
1644 
1645 	/* A grant for every rx ring slot */
1646 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1647 					  &queue->gref_rx_head) < 0) {
1648 		pr_alert("can't alloc rx grant refs\n");
1649 		err = -ENOMEM;
1650 		goto exit_free_tx;
1651 	}
1652 
1653 	return 0;
1654 
1655  exit_free_tx:
1656 	gnttab_free_grant_references(queue->gref_tx_head);
1657  exit:
1658 	return err;
1659 }
1660 
1661 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1662 			   struct xenbus_transaction *xbt, int write_hierarchical)
1663 {
1664 	/* Write the queue-specific keys into XenStore in the traditional
1665 	 * way for a single queue, or in a queue subkeys for multiple
1666 	 * queues.
1667 	 */
1668 	struct xenbus_device *dev = queue->info->xbdev;
1669 	int err;
1670 	const char *message;
1671 	char *path;
1672 	size_t pathsize;
1673 
1674 	/* Choose the correct place to write the keys */
1675 	if (write_hierarchical) {
1676 		pathsize = strlen(dev->nodename) + 10;
1677 		path = kzalloc(pathsize, GFP_KERNEL);
1678 		if (!path) {
1679 			err = -ENOMEM;
1680 			message = "out of memory while writing ring references";
1681 			goto error;
1682 		}
1683 		snprintf(path, pathsize, "%s/queue-%u",
1684 				dev->nodename, queue->id);
1685 	} else {
1686 		path = (char *)dev->nodename;
1687 	}
1688 
1689 	/* Write ring references */
1690 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1691 			queue->tx_ring_ref);
1692 	if (err) {
1693 		message = "writing tx-ring-ref";
1694 		goto error;
1695 	}
1696 
1697 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1698 			queue->rx_ring_ref);
1699 	if (err) {
1700 		message = "writing rx-ring-ref";
1701 		goto error;
1702 	}
1703 
1704 	/* Write event channels; taking into account both shared
1705 	 * and split event channel scenarios.
1706 	 */
1707 	if (queue->tx_evtchn == queue->rx_evtchn) {
1708 		/* Shared event channel */
1709 		err = xenbus_printf(*xbt, path,
1710 				"event-channel", "%u", queue->tx_evtchn);
1711 		if (err) {
1712 			message = "writing event-channel";
1713 			goto error;
1714 		}
1715 	} else {
1716 		/* Split event channels */
1717 		err = xenbus_printf(*xbt, path,
1718 				"event-channel-tx", "%u", queue->tx_evtchn);
1719 		if (err) {
1720 			message = "writing event-channel-tx";
1721 			goto error;
1722 		}
1723 
1724 		err = xenbus_printf(*xbt, path,
1725 				"event-channel-rx", "%u", queue->rx_evtchn);
1726 		if (err) {
1727 			message = "writing event-channel-rx";
1728 			goto error;
1729 		}
1730 	}
1731 
1732 	if (write_hierarchical)
1733 		kfree(path);
1734 	return 0;
1735 
1736 error:
1737 	if (write_hierarchical)
1738 		kfree(path);
1739 	xenbus_dev_fatal(dev, err, "%s", message);
1740 	return err;
1741 }
1742 
1743 static void xennet_destroy_queues(struct netfront_info *info)
1744 {
1745 	unsigned int i;
1746 
1747 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1748 		struct netfront_queue *queue = &info->queues[i];
1749 
1750 		if (netif_running(info->netdev))
1751 			napi_disable(&queue->napi);
1752 		netif_napi_del(&queue->napi);
1753 	}
1754 
1755 	kfree(info->queues);
1756 	info->queues = NULL;
1757 }
1758 
1759 static int xennet_create_queues(struct netfront_info *info,
1760 				unsigned int *num_queues)
1761 {
1762 	unsigned int i;
1763 	int ret;
1764 
1765 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1766 			       GFP_KERNEL);
1767 	if (!info->queues)
1768 		return -ENOMEM;
1769 
1770 	for (i = 0; i < *num_queues; i++) {
1771 		struct netfront_queue *queue = &info->queues[i];
1772 
1773 		queue->id = i;
1774 		queue->info = info;
1775 
1776 		ret = xennet_init_queue(queue);
1777 		if (ret < 0) {
1778 			dev_warn(&info->xbdev->dev,
1779 				 "only created %d queues\n", i);
1780 			*num_queues = i;
1781 			break;
1782 		}
1783 
1784 		netif_napi_add(queue->info->netdev, &queue->napi,
1785 			       xennet_poll, 64);
1786 		if (netif_running(info->netdev))
1787 			napi_enable(&queue->napi);
1788 	}
1789 
1790 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
1791 
1792 	if (*num_queues == 0) {
1793 		dev_err(&info->xbdev->dev, "no queues\n");
1794 		return -EINVAL;
1795 	}
1796 	return 0;
1797 }
1798 
1799 /* Common code used when first setting up, and when resuming. */
1800 static int talk_to_netback(struct xenbus_device *dev,
1801 			   struct netfront_info *info)
1802 {
1803 	const char *message;
1804 	struct xenbus_transaction xbt;
1805 	int err;
1806 	unsigned int feature_split_evtchn;
1807 	unsigned int i = 0;
1808 	unsigned int max_queues = 0;
1809 	struct netfront_queue *queue = NULL;
1810 	unsigned int num_queues = 1;
1811 
1812 	info->netdev->irq = 0;
1813 
1814 	/* Check if backend supports multiple queues */
1815 	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1816 					  "multi-queue-max-queues", 1);
1817 	num_queues = min(max_queues, xennet_max_queues);
1818 
1819 	/* Check feature-split-event-channels */
1820 	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1821 					"feature-split-event-channels", 0);
1822 
1823 	/* Read mac addr. */
1824 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
1825 	if (err) {
1826 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1827 		goto out_unlocked;
1828 	}
1829 
1830 	rtnl_lock();
1831 	if (info->queues)
1832 		xennet_destroy_queues(info);
1833 
1834 	err = xennet_create_queues(info, &num_queues);
1835 	if (err < 0) {
1836 		xenbus_dev_fatal(dev, err, "creating queues");
1837 		kfree(info->queues);
1838 		info->queues = NULL;
1839 		goto out;
1840 	}
1841 	rtnl_unlock();
1842 
1843 	/* Create shared ring, alloc event channel -- for each queue */
1844 	for (i = 0; i < num_queues; ++i) {
1845 		queue = &info->queues[i];
1846 		err = setup_netfront(dev, queue, feature_split_evtchn);
1847 		if (err)
1848 			goto destroy_ring;
1849 	}
1850 
1851 again:
1852 	err = xenbus_transaction_start(&xbt);
1853 	if (err) {
1854 		xenbus_dev_fatal(dev, err, "starting transaction");
1855 		goto destroy_ring;
1856 	}
1857 
1858 	if (xenbus_exists(XBT_NIL,
1859 			  info->xbdev->otherend, "multi-queue-max-queues")) {
1860 		/* Write the number of queues */
1861 		err = xenbus_printf(xbt, dev->nodename,
1862 				    "multi-queue-num-queues", "%u", num_queues);
1863 		if (err) {
1864 			message = "writing multi-queue-num-queues";
1865 			goto abort_transaction_no_dev_fatal;
1866 		}
1867 	}
1868 
1869 	if (num_queues == 1) {
1870 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1871 		if (err)
1872 			goto abort_transaction_no_dev_fatal;
1873 	} else {
1874 		/* Write the keys for each queue */
1875 		for (i = 0; i < num_queues; ++i) {
1876 			queue = &info->queues[i];
1877 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1878 			if (err)
1879 				goto abort_transaction_no_dev_fatal;
1880 		}
1881 	}
1882 
1883 	/* The remaining keys are not queue-specific */
1884 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1885 			    1);
1886 	if (err) {
1887 		message = "writing request-rx-copy";
1888 		goto abort_transaction;
1889 	}
1890 
1891 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1892 	if (err) {
1893 		message = "writing feature-rx-notify";
1894 		goto abort_transaction;
1895 	}
1896 
1897 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1898 	if (err) {
1899 		message = "writing feature-sg";
1900 		goto abort_transaction;
1901 	}
1902 
1903 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1904 	if (err) {
1905 		message = "writing feature-gso-tcpv4";
1906 		goto abort_transaction;
1907 	}
1908 
1909 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1910 	if (err) {
1911 		message = "writing feature-gso-tcpv6";
1912 		goto abort_transaction;
1913 	}
1914 
1915 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1916 			   "1");
1917 	if (err) {
1918 		message = "writing feature-ipv6-csum-offload";
1919 		goto abort_transaction;
1920 	}
1921 
1922 	err = xenbus_transaction_end(xbt, 0);
1923 	if (err) {
1924 		if (err == -EAGAIN)
1925 			goto again;
1926 		xenbus_dev_fatal(dev, err, "completing transaction");
1927 		goto destroy_ring;
1928 	}
1929 
1930 	return 0;
1931 
1932  abort_transaction:
1933 	xenbus_dev_fatal(dev, err, "%s", message);
1934 abort_transaction_no_dev_fatal:
1935 	xenbus_transaction_end(xbt, 1);
1936  destroy_ring:
1937 	xennet_disconnect_backend(info);
1938 	rtnl_lock();
1939 	xennet_destroy_queues(info);
1940  out:
1941 	rtnl_unlock();
1942 out_unlocked:
1943 	device_unregister(&dev->dev);
1944 	return err;
1945 }
1946 
1947 static int xennet_connect(struct net_device *dev)
1948 {
1949 	struct netfront_info *np = netdev_priv(dev);
1950 	unsigned int num_queues = 0;
1951 	int err;
1952 	unsigned int j = 0;
1953 	struct netfront_queue *queue = NULL;
1954 
1955 	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1956 		dev_info(&dev->dev,
1957 			 "backend does not support copying receive path\n");
1958 		return -ENODEV;
1959 	}
1960 
1961 	err = talk_to_netback(np->xbdev, np);
1962 	if (err)
1963 		return err;
1964 
1965 	/* talk_to_netback() sets the correct number of queues */
1966 	num_queues = dev->real_num_tx_queues;
1967 
1968 	if (dev->reg_state == NETREG_UNINITIALIZED) {
1969 		err = register_netdev(dev);
1970 		if (err) {
1971 			pr_warn("%s: register_netdev err=%d\n", __func__, err);
1972 			device_unregister(&np->xbdev->dev);
1973 			return err;
1974 		}
1975 	}
1976 
1977 	rtnl_lock();
1978 	netdev_update_features(dev);
1979 	rtnl_unlock();
1980 
1981 	/*
1982 	 * All public and private state should now be sane.  Get
1983 	 * ready to start sending and receiving packets and give the driver
1984 	 * domain a kick because we've probably just requeued some
1985 	 * packets.
1986 	 */
1987 	netif_carrier_on(np->netdev);
1988 	for (j = 0; j < num_queues; ++j) {
1989 		queue = &np->queues[j];
1990 
1991 		notify_remote_via_irq(queue->tx_irq);
1992 		if (queue->tx_irq != queue->rx_irq)
1993 			notify_remote_via_irq(queue->rx_irq);
1994 
1995 		spin_lock_irq(&queue->tx_lock);
1996 		xennet_tx_buf_gc(queue);
1997 		spin_unlock_irq(&queue->tx_lock);
1998 
1999 		spin_lock_bh(&queue->rx_lock);
2000 		xennet_alloc_rx_buffers(queue);
2001 		spin_unlock_bh(&queue->rx_lock);
2002 	}
2003 
2004 	return 0;
2005 }
2006 
2007 /**
2008  * Callback received when the backend's state changes.
2009  */
2010 static void netback_changed(struct xenbus_device *dev,
2011 			    enum xenbus_state backend_state)
2012 {
2013 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2014 	struct net_device *netdev = np->netdev;
2015 
2016 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2017 
2018 	wake_up_all(&module_wq);
2019 
2020 	switch (backend_state) {
2021 	case XenbusStateInitialising:
2022 	case XenbusStateInitialised:
2023 	case XenbusStateReconfiguring:
2024 	case XenbusStateReconfigured:
2025 	case XenbusStateUnknown:
2026 		break;
2027 
2028 	case XenbusStateInitWait:
2029 		if (dev->state != XenbusStateInitialising)
2030 			break;
2031 		if (xennet_connect(netdev) != 0)
2032 			break;
2033 		xenbus_switch_state(dev, XenbusStateConnected);
2034 		break;
2035 
2036 	case XenbusStateConnected:
2037 		netdev_notify_peers(netdev);
2038 		break;
2039 
2040 	case XenbusStateClosed:
2041 		if (dev->state == XenbusStateClosed)
2042 			break;
2043 		/* Missed the backend's CLOSING state -- fallthrough */
2044 	case XenbusStateClosing:
2045 		xenbus_frontend_closed(dev);
2046 		break;
2047 	}
2048 }
2049 
2050 static const struct xennet_stat {
2051 	char name[ETH_GSTRING_LEN];
2052 	u16 offset;
2053 } xennet_stats[] = {
2054 	{
2055 		"rx_gso_checksum_fixup",
2056 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2057 	},
2058 };
2059 
2060 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2061 {
2062 	switch (string_set) {
2063 	case ETH_SS_STATS:
2064 		return ARRAY_SIZE(xennet_stats);
2065 	default:
2066 		return -EINVAL;
2067 	}
2068 }
2069 
2070 static void xennet_get_ethtool_stats(struct net_device *dev,
2071 				     struct ethtool_stats *stats, u64 * data)
2072 {
2073 	void *np = netdev_priv(dev);
2074 	int i;
2075 
2076 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2077 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2078 }
2079 
2080 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2081 {
2082 	int i;
2083 
2084 	switch (stringset) {
2085 	case ETH_SS_STATS:
2086 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2087 			memcpy(data + i * ETH_GSTRING_LEN,
2088 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2089 		break;
2090 	}
2091 }
2092 
2093 static const struct ethtool_ops xennet_ethtool_ops =
2094 {
2095 	.get_link = ethtool_op_get_link,
2096 
2097 	.get_sset_count = xennet_get_sset_count,
2098 	.get_ethtool_stats = xennet_get_ethtool_stats,
2099 	.get_strings = xennet_get_strings,
2100 };
2101 
2102 #ifdef CONFIG_SYSFS
2103 static ssize_t show_rxbuf(struct device *dev,
2104 			  struct device_attribute *attr, char *buf)
2105 {
2106 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2107 }
2108 
2109 static ssize_t store_rxbuf(struct device *dev,
2110 			   struct device_attribute *attr,
2111 			   const char *buf, size_t len)
2112 {
2113 	char *endp;
2114 	unsigned long target;
2115 
2116 	if (!capable(CAP_NET_ADMIN))
2117 		return -EPERM;
2118 
2119 	target = simple_strtoul(buf, &endp, 0);
2120 	if (endp == buf)
2121 		return -EBADMSG;
2122 
2123 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2124 
2125 	return len;
2126 }
2127 
2128 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2129 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2130 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2131 
2132 static struct attribute *xennet_dev_attrs[] = {
2133 	&dev_attr_rxbuf_min.attr,
2134 	&dev_attr_rxbuf_max.attr,
2135 	&dev_attr_rxbuf_cur.attr,
2136 	NULL
2137 };
2138 
2139 static const struct attribute_group xennet_dev_group = {
2140 	.attrs = xennet_dev_attrs
2141 };
2142 #endif /* CONFIG_SYSFS */
2143 
2144 static int xennet_remove(struct xenbus_device *dev)
2145 {
2146 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2147 
2148 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2149 
2150 	if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) {
2151 		xenbus_switch_state(dev, XenbusStateClosing);
2152 		wait_event(module_wq,
2153 			   xenbus_read_driver_state(dev->otherend) ==
2154 			   XenbusStateClosing ||
2155 			   xenbus_read_driver_state(dev->otherend) ==
2156 			   XenbusStateUnknown);
2157 
2158 		xenbus_switch_state(dev, XenbusStateClosed);
2159 		wait_event(module_wq,
2160 			   xenbus_read_driver_state(dev->otherend) ==
2161 			   XenbusStateClosed ||
2162 			   xenbus_read_driver_state(dev->otherend) ==
2163 			   XenbusStateUnknown);
2164 	}
2165 
2166 	xennet_disconnect_backend(info);
2167 
2168 	if (info->netdev->reg_state == NETREG_REGISTERED)
2169 		unregister_netdev(info->netdev);
2170 
2171 	if (info->queues) {
2172 		rtnl_lock();
2173 		xennet_destroy_queues(info);
2174 		rtnl_unlock();
2175 	}
2176 	xennet_free_netdev(info->netdev);
2177 
2178 	return 0;
2179 }
2180 
2181 static const struct xenbus_device_id netfront_ids[] = {
2182 	{ "vif" },
2183 	{ "" }
2184 };
2185 
2186 static struct xenbus_driver netfront_driver = {
2187 	.ids = netfront_ids,
2188 	.probe = netfront_probe,
2189 	.remove = xennet_remove,
2190 	.resume = netfront_resume,
2191 	.otherend_changed = netback_changed,
2192 };
2193 
2194 static int __init netif_init(void)
2195 {
2196 	if (!xen_domain())
2197 		return -ENODEV;
2198 
2199 	if (!xen_has_pv_nic_devices())
2200 		return -ENODEV;
2201 
2202 	pr_info("Initialising Xen virtual ethernet driver\n");
2203 
2204 	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2205 	 * specified a value.
2206 	 */
2207 	if (xennet_max_queues == 0)
2208 		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2209 					  num_online_cpus());
2210 
2211 	return xenbus_register_frontend(&netfront_driver);
2212 }
2213 module_init(netif_init);
2214 
2215 
2216 static void __exit netif_exit(void)
2217 {
2218 	xenbus_unregister_driver(&netfront_driver);
2219 }
2220 module_exit(netif_exit);
2221 
2222 MODULE_DESCRIPTION("Xen virtual network device frontend");
2223 MODULE_LICENSE("GPL");
2224 MODULE_ALIAS("xen:vif");
2225 MODULE_ALIAS("xennet");
2226