xref: /openbmc/linux/drivers/net/xen-netfront.c (revision 9a29ad52)
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 			       void *accel_priv, select_queue_fallback_t fallback)
549 {
550 	unsigned int num_queues = dev->real_num_tx_queues;
551 	u32 hash;
552 	u16 queue_idx;
553 
554 	/* First, check if there is only one queue */
555 	if (num_queues == 1) {
556 		queue_idx = 0;
557 	} else {
558 		hash = skb_get_hash(skb);
559 		queue_idx = hash % num_queues;
560 	}
561 
562 	return queue_idx;
563 }
564 
565 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
566 
567 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
568 {
569 	struct netfront_info *np = netdev_priv(dev);
570 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
571 	struct xen_netif_tx_request *tx, *first_tx;
572 	unsigned int i;
573 	int notify;
574 	int slots;
575 	struct page *page;
576 	unsigned int offset;
577 	unsigned int len;
578 	unsigned long flags;
579 	struct netfront_queue *queue = NULL;
580 	unsigned int num_queues = dev->real_num_tx_queues;
581 	u16 queue_index;
582 	struct sk_buff *nskb;
583 
584 	/* Drop the packet if no queues are set up */
585 	if (num_queues < 1)
586 		goto drop;
587 	/* Determine which queue to transmit this SKB on */
588 	queue_index = skb_get_queue_mapping(skb);
589 	queue = &np->queues[queue_index];
590 
591 	/* If skb->len is too big for wire format, drop skb and alert
592 	 * user about misconfiguration.
593 	 */
594 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
595 		net_alert_ratelimited(
596 			"xennet: skb->len = %u, too big for wire format\n",
597 			skb->len);
598 		goto drop;
599 	}
600 
601 	slots = xennet_count_skb_slots(skb);
602 	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
603 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
604 				    slots, skb->len);
605 		if (skb_linearize(skb))
606 			goto drop;
607 	}
608 
609 	page = virt_to_page(skb->data);
610 	offset = offset_in_page(skb->data);
611 
612 	/* The first req should be at least ETH_HLEN size or the packet will be
613 	 * dropped by netback.
614 	 */
615 	if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
616 		nskb = skb_copy(skb, GFP_ATOMIC);
617 		if (!nskb)
618 			goto drop;
619 		dev_consume_skb_any(skb);
620 		skb = nskb;
621 		page = virt_to_page(skb->data);
622 		offset = offset_in_page(skb->data);
623 	}
624 
625 	len = skb_headlen(skb);
626 
627 	spin_lock_irqsave(&queue->tx_lock, flags);
628 
629 	if (unlikely(!netif_carrier_ok(dev) ||
630 		     (slots > 1 && !xennet_can_sg(dev)) ||
631 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
632 		spin_unlock_irqrestore(&queue->tx_lock, flags);
633 		goto drop;
634 	}
635 
636 	/* First request for the linear area. */
637 	first_tx = tx = xennet_make_first_txreq(queue, skb,
638 						page, offset, len);
639 	offset += tx->size;
640 	if (offset == PAGE_SIZE) {
641 		page++;
642 		offset = 0;
643 	}
644 	len -= tx->size;
645 
646 	if (skb->ip_summed == CHECKSUM_PARTIAL)
647 		/* local packet? */
648 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
649 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
650 		/* remote but checksummed. */
651 		tx->flags |= XEN_NETTXF_data_validated;
652 
653 	/* Optional extra info after the first request. */
654 	if (skb_shinfo(skb)->gso_size) {
655 		struct xen_netif_extra_info *gso;
656 
657 		gso = (struct xen_netif_extra_info *)
658 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
659 
660 		tx->flags |= XEN_NETTXF_extra_info;
661 
662 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
663 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
664 			XEN_NETIF_GSO_TYPE_TCPV6 :
665 			XEN_NETIF_GSO_TYPE_TCPV4;
666 		gso->u.gso.pad = 0;
667 		gso->u.gso.features = 0;
668 
669 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
670 		gso->flags = 0;
671 	}
672 
673 	/* Requests for the rest of the linear area. */
674 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
675 
676 	/* Requests for all the frags. */
677 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
678 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
679 		tx = xennet_make_txreqs(queue, tx, skb,
680 					skb_frag_page(frag), frag->page_offset,
681 					skb_frag_size(frag));
682 	}
683 
684 	/* First request has the packet length. */
685 	first_tx->size = skb->len;
686 
687 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
688 	if (notify)
689 		notify_remote_via_irq(queue->tx_irq);
690 
691 	u64_stats_update_begin(&tx_stats->syncp);
692 	tx_stats->bytes += skb->len;
693 	tx_stats->packets++;
694 	u64_stats_update_end(&tx_stats->syncp);
695 
696 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
697 	xennet_tx_buf_gc(queue);
698 
699 	if (!netfront_tx_slot_available(queue))
700 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
701 
702 	spin_unlock_irqrestore(&queue->tx_lock, flags);
703 
704 	return NETDEV_TX_OK;
705 
706  drop:
707 	dev->stats.tx_dropped++;
708 	dev_kfree_skb_any(skb);
709 	return NETDEV_TX_OK;
710 }
711 
712 static int xennet_close(struct net_device *dev)
713 {
714 	struct netfront_info *np = netdev_priv(dev);
715 	unsigned int num_queues = dev->real_num_tx_queues;
716 	unsigned int i;
717 	struct netfront_queue *queue;
718 	netif_tx_stop_all_queues(np->netdev);
719 	for (i = 0; i < num_queues; ++i) {
720 		queue = &np->queues[i];
721 		napi_disable(&queue->napi);
722 	}
723 	return 0;
724 }
725 
726 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
727 				grant_ref_t ref)
728 {
729 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
730 
731 	BUG_ON(queue->rx_skbs[new]);
732 	queue->rx_skbs[new] = skb;
733 	queue->grant_rx_ref[new] = ref;
734 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
735 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
736 	queue->rx.req_prod_pvt++;
737 }
738 
739 static int xennet_get_extras(struct netfront_queue *queue,
740 			     struct xen_netif_extra_info *extras,
741 			     RING_IDX rp)
742 
743 {
744 	struct xen_netif_extra_info *extra;
745 	struct device *dev = &queue->info->netdev->dev;
746 	RING_IDX cons = queue->rx.rsp_cons;
747 	int err = 0;
748 
749 	do {
750 		struct sk_buff *skb;
751 		grant_ref_t ref;
752 
753 		if (unlikely(cons + 1 == rp)) {
754 			if (net_ratelimit())
755 				dev_warn(dev, "Missing extra info\n");
756 			err = -EBADR;
757 			break;
758 		}
759 
760 		extra = (struct xen_netif_extra_info *)
761 			RING_GET_RESPONSE(&queue->rx, ++cons);
762 
763 		if (unlikely(!extra->type ||
764 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
765 			if (net_ratelimit())
766 				dev_warn(dev, "Invalid extra type: %d\n",
767 					extra->type);
768 			err = -EINVAL;
769 		} else {
770 			memcpy(&extras[extra->type - 1], extra,
771 			       sizeof(*extra));
772 		}
773 
774 		skb = xennet_get_rx_skb(queue, cons);
775 		ref = xennet_get_rx_ref(queue, cons);
776 		xennet_move_rx_slot(queue, skb, ref);
777 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
778 
779 	queue->rx.rsp_cons = cons;
780 	return err;
781 }
782 
783 static int xennet_get_responses(struct netfront_queue *queue,
784 				struct netfront_rx_info *rinfo, RING_IDX rp,
785 				struct sk_buff_head *list)
786 {
787 	struct xen_netif_rx_response *rx = &rinfo->rx;
788 	struct xen_netif_extra_info *extras = rinfo->extras;
789 	struct device *dev = &queue->info->netdev->dev;
790 	RING_IDX cons = queue->rx.rsp_cons;
791 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
792 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
793 	int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
794 	int slots = 1;
795 	int err = 0;
796 	unsigned long ret;
797 
798 	if (rx->flags & XEN_NETRXF_extra_info) {
799 		err = xennet_get_extras(queue, extras, rp);
800 		cons = queue->rx.rsp_cons;
801 	}
802 
803 	for (;;) {
804 		if (unlikely(rx->status < 0 ||
805 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
806 			if (net_ratelimit())
807 				dev_warn(dev, "rx->offset: %u, size: %d\n",
808 					 rx->offset, rx->status);
809 			xennet_move_rx_slot(queue, skb, ref);
810 			err = -EINVAL;
811 			goto next;
812 		}
813 
814 		/*
815 		 * This definitely indicates a bug, either in this driver or in
816 		 * the backend driver. In future this should flag the bad
817 		 * situation to the system controller to reboot the backend.
818 		 */
819 		if (ref == GRANT_INVALID_REF) {
820 			if (net_ratelimit())
821 				dev_warn(dev, "Bad rx response id %d.\n",
822 					 rx->id);
823 			err = -EINVAL;
824 			goto next;
825 		}
826 
827 		ret = gnttab_end_foreign_access_ref(ref, 0);
828 		BUG_ON(!ret);
829 
830 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
831 
832 		__skb_queue_tail(list, skb);
833 
834 next:
835 		if (!(rx->flags & XEN_NETRXF_more_data))
836 			break;
837 
838 		if (cons + slots == rp) {
839 			if (net_ratelimit())
840 				dev_warn(dev, "Need more slots\n");
841 			err = -ENOENT;
842 			break;
843 		}
844 
845 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
846 		skb = xennet_get_rx_skb(queue, cons + slots);
847 		ref = xennet_get_rx_ref(queue, cons + slots);
848 		slots++;
849 	}
850 
851 	if (unlikely(slots > max)) {
852 		if (net_ratelimit())
853 			dev_warn(dev, "Too many slots\n");
854 		err = -E2BIG;
855 	}
856 
857 	if (unlikely(err))
858 		queue->rx.rsp_cons = cons + slots;
859 
860 	return err;
861 }
862 
863 static int xennet_set_skb_gso(struct sk_buff *skb,
864 			      struct xen_netif_extra_info *gso)
865 {
866 	if (!gso->u.gso.size) {
867 		if (net_ratelimit())
868 			pr_warn("GSO size must not be zero\n");
869 		return -EINVAL;
870 	}
871 
872 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
873 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
874 		if (net_ratelimit())
875 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
876 		return -EINVAL;
877 	}
878 
879 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
880 	skb_shinfo(skb)->gso_type =
881 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
882 		SKB_GSO_TCPV4 :
883 		SKB_GSO_TCPV6;
884 
885 	/* Header must be checked, and gso_segs computed. */
886 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
887 	skb_shinfo(skb)->gso_segs = 0;
888 
889 	return 0;
890 }
891 
892 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
893 				  struct sk_buff *skb,
894 				  struct sk_buff_head *list)
895 {
896 	struct skb_shared_info *shinfo = skb_shinfo(skb);
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 (shinfo->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 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
912 
913 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
914 				rx->offset, rx->status, PAGE_SIZE);
915 
916 		skb_shinfo(nskb)->nr_frags = 0;
917 		kfree_skb(nskb);
918 	}
919 
920 	return cons;
921 }
922 
923 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
924 {
925 	bool recalculate_partial_csum = false;
926 
927 	/*
928 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
929 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
930 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
931 	 * recalculate the partial checksum.
932 	 */
933 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
934 		struct netfront_info *np = netdev_priv(dev);
935 		atomic_inc(&np->rx_gso_checksum_fixup);
936 		skb->ip_summed = CHECKSUM_PARTIAL;
937 		recalculate_partial_csum = true;
938 	}
939 
940 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
941 	if (skb->ip_summed != CHECKSUM_PARTIAL)
942 		return 0;
943 
944 	return skb_checksum_setup(skb, recalculate_partial_csum);
945 }
946 
947 static int handle_incoming_queue(struct netfront_queue *queue,
948 				 struct sk_buff_head *rxq)
949 {
950 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
951 	int packets_dropped = 0;
952 	struct sk_buff *skb;
953 
954 	while ((skb = __skb_dequeue(rxq)) != NULL) {
955 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
956 
957 		if (pull_to > skb_headlen(skb))
958 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
959 
960 		/* Ethernet work: Delayed to here as it peeks the header. */
961 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
962 		skb_reset_network_header(skb);
963 
964 		if (checksum_setup(queue->info->netdev, skb)) {
965 			kfree_skb(skb);
966 			packets_dropped++;
967 			queue->info->netdev->stats.rx_errors++;
968 			continue;
969 		}
970 
971 		u64_stats_update_begin(&rx_stats->syncp);
972 		rx_stats->packets++;
973 		rx_stats->bytes += skb->len;
974 		u64_stats_update_end(&rx_stats->syncp);
975 
976 		/* Pass it up. */
977 		napi_gro_receive(&queue->napi, skb);
978 	}
979 
980 	return packets_dropped;
981 }
982 
983 static int xennet_poll(struct napi_struct *napi, int budget)
984 {
985 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
986 	struct net_device *dev = queue->info->netdev;
987 	struct sk_buff *skb;
988 	struct netfront_rx_info rinfo;
989 	struct xen_netif_rx_response *rx = &rinfo.rx;
990 	struct xen_netif_extra_info *extras = rinfo.extras;
991 	RING_IDX i, rp;
992 	int work_done;
993 	struct sk_buff_head rxq;
994 	struct sk_buff_head errq;
995 	struct sk_buff_head tmpq;
996 	int err;
997 
998 	spin_lock(&queue->rx_lock);
999 
1000 	skb_queue_head_init(&rxq);
1001 	skb_queue_head_init(&errq);
1002 	skb_queue_head_init(&tmpq);
1003 
1004 	rp = queue->rx.sring->rsp_prod;
1005 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1006 
1007 	i = queue->rx.rsp_cons;
1008 	work_done = 0;
1009 	while ((i != rp) && (work_done < budget)) {
1010 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1011 		memset(extras, 0, sizeof(rinfo.extras));
1012 
1013 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1014 
1015 		if (unlikely(err)) {
1016 err:
1017 			while ((skb = __skb_dequeue(&tmpq)))
1018 				__skb_queue_tail(&errq, skb);
1019 			dev->stats.rx_errors++;
1020 			i = queue->rx.rsp_cons;
1021 			continue;
1022 		}
1023 
1024 		skb = __skb_dequeue(&tmpq);
1025 
1026 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1027 			struct xen_netif_extra_info *gso;
1028 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1029 
1030 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1031 				__skb_queue_head(&tmpq, skb);
1032 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1033 				goto err;
1034 			}
1035 		}
1036 
1037 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1038 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1039 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1040 
1041 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1042 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1043 		skb->data_len = rx->status;
1044 		skb->len += rx->status;
1045 
1046 		i = xennet_fill_frags(queue, skb, &tmpq);
1047 
1048 		if (rx->flags & XEN_NETRXF_csum_blank)
1049 			skb->ip_summed = CHECKSUM_PARTIAL;
1050 		else if (rx->flags & XEN_NETRXF_data_validated)
1051 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1052 
1053 		__skb_queue_tail(&rxq, skb);
1054 
1055 		queue->rx.rsp_cons = ++i;
1056 		work_done++;
1057 	}
1058 
1059 	__skb_queue_purge(&errq);
1060 
1061 	work_done -= handle_incoming_queue(queue, &rxq);
1062 
1063 	xennet_alloc_rx_buffers(queue);
1064 
1065 	if (work_done < budget) {
1066 		int more_to_do = 0;
1067 
1068 		napi_complete_done(napi, work_done);
1069 
1070 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1071 		if (more_to_do)
1072 			napi_schedule(napi);
1073 	}
1074 
1075 	spin_unlock(&queue->rx_lock);
1076 
1077 	return work_done;
1078 }
1079 
1080 static int xennet_change_mtu(struct net_device *dev, int mtu)
1081 {
1082 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1083 
1084 	if (mtu > max)
1085 		return -EINVAL;
1086 	dev->mtu = mtu;
1087 	return 0;
1088 }
1089 
1090 static void xennet_get_stats64(struct net_device *dev,
1091 			       struct rtnl_link_stats64 *tot)
1092 {
1093 	struct netfront_info *np = netdev_priv(dev);
1094 	int cpu;
1095 
1096 	for_each_possible_cpu(cpu) {
1097 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1098 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1099 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1100 		unsigned int start;
1101 
1102 		do {
1103 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1104 			tx_packets = tx_stats->packets;
1105 			tx_bytes = tx_stats->bytes;
1106 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1107 
1108 		do {
1109 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1110 			rx_packets = rx_stats->packets;
1111 			rx_bytes = rx_stats->bytes;
1112 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1113 
1114 		tot->rx_packets += rx_packets;
1115 		tot->tx_packets += tx_packets;
1116 		tot->rx_bytes   += rx_bytes;
1117 		tot->tx_bytes   += tx_bytes;
1118 	}
1119 
1120 	tot->rx_errors  = dev->stats.rx_errors;
1121 	tot->tx_dropped = dev->stats.tx_dropped;
1122 }
1123 
1124 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1125 {
1126 	struct sk_buff *skb;
1127 	int i;
1128 
1129 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1130 		/* Skip over entries which are actually freelist references */
1131 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1132 			continue;
1133 
1134 		skb = queue->tx_skbs[i].skb;
1135 		get_page(queue->grant_tx_page[i]);
1136 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1137 					  GNTMAP_readonly,
1138 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1139 		queue->grant_tx_page[i] = NULL;
1140 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1141 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1142 		dev_kfree_skb_irq(skb);
1143 	}
1144 }
1145 
1146 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1147 {
1148 	int id, ref;
1149 
1150 	spin_lock_bh(&queue->rx_lock);
1151 
1152 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1153 		struct sk_buff *skb;
1154 		struct page *page;
1155 
1156 		skb = queue->rx_skbs[id];
1157 		if (!skb)
1158 			continue;
1159 
1160 		ref = queue->grant_rx_ref[id];
1161 		if (ref == GRANT_INVALID_REF)
1162 			continue;
1163 
1164 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1165 
1166 		/* gnttab_end_foreign_access() needs a page ref until
1167 		 * foreign access is ended (which may be deferred).
1168 		 */
1169 		get_page(page);
1170 		gnttab_end_foreign_access(ref, 0,
1171 					  (unsigned long)page_address(page));
1172 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1173 
1174 		kfree_skb(skb);
1175 	}
1176 
1177 	spin_unlock_bh(&queue->rx_lock);
1178 }
1179 
1180 static netdev_features_t xennet_fix_features(struct net_device *dev,
1181 	netdev_features_t features)
1182 {
1183 	struct netfront_info *np = netdev_priv(dev);
1184 
1185 	if (features & NETIF_F_SG &&
1186 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1187 		features &= ~NETIF_F_SG;
1188 
1189 	if (features & NETIF_F_IPV6_CSUM &&
1190 	    !xenbus_read_unsigned(np->xbdev->otherend,
1191 				  "feature-ipv6-csum-offload", 0))
1192 		features &= ~NETIF_F_IPV6_CSUM;
1193 
1194 	if (features & NETIF_F_TSO &&
1195 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1196 		features &= ~NETIF_F_TSO;
1197 
1198 	if (features & NETIF_F_TSO6 &&
1199 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1200 		features &= ~NETIF_F_TSO6;
1201 
1202 	return features;
1203 }
1204 
1205 static int xennet_set_features(struct net_device *dev,
1206 	netdev_features_t features)
1207 {
1208 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1209 		netdev_info(dev, "Reducing MTU because no SG offload");
1210 		dev->mtu = ETH_DATA_LEN;
1211 	}
1212 
1213 	return 0;
1214 }
1215 
1216 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1217 {
1218 	struct netfront_queue *queue = dev_id;
1219 	unsigned long flags;
1220 
1221 	spin_lock_irqsave(&queue->tx_lock, flags);
1222 	xennet_tx_buf_gc(queue);
1223 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1224 
1225 	return IRQ_HANDLED;
1226 }
1227 
1228 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1229 {
1230 	struct netfront_queue *queue = dev_id;
1231 	struct net_device *dev = queue->info->netdev;
1232 
1233 	if (likely(netif_carrier_ok(dev) &&
1234 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1235 		napi_schedule(&queue->napi);
1236 
1237 	return IRQ_HANDLED;
1238 }
1239 
1240 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1241 {
1242 	xennet_tx_interrupt(irq, dev_id);
1243 	xennet_rx_interrupt(irq, dev_id);
1244 	return IRQ_HANDLED;
1245 }
1246 
1247 #ifdef CONFIG_NET_POLL_CONTROLLER
1248 static void xennet_poll_controller(struct net_device *dev)
1249 {
1250 	/* Poll each queue */
1251 	struct netfront_info *info = netdev_priv(dev);
1252 	unsigned int num_queues = dev->real_num_tx_queues;
1253 	unsigned int i;
1254 	for (i = 0; i < num_queues; ++i)
1255 		xennet_interrupt(0, &info->queues[i]);
1256 }
1257 #endif
1258 
1259 static const struct net_device_ops xennet_netdev_ops = {
1260 	.ndo_open            = xennet_open,
1261 	.ndo_stop            = xennet_close,
1262 	.ndo_start_xmit      = xennet_start_xmit,
1263 	.ndo_change_mtu	     = xennet_change_mtu,
1264 	.ndo_get_stats64     = xennet_get_stats64,
1265 	.ndo_set_mac_address = eth_mac_addr,
1266 	.ndo_validate_addr   = eth_validate_addr,
1267 	.ndo_fix_features    = xennet_fix_features,
1268 	.ndo_set_features    = xennet_set_features,
1269 	.ndo_select_queue    = xennet_select_queue,
1270 #ifdef CONFIG_NET_POLL_CONTROLLER
1271 	.ndo_poll_controller = xennet_poll_controller,
1272 #endif
1273 };
1274 
1275 static void xennet_free_netdev(struct net_device *netdev)
1276 {
1277 	struct netfront_info *np = netdev_priv(netdev);
1278 
1279 	free_percpu(np->rx_stats);
1280 	free_percpu(np->tx_stats);
1281 	free_netdev(netdev);
1282 }
1283 
1284 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1285 {
1286 	int err;
1287 	struct net_device *netdev;
1288 	struct netfront_info *np;
1289 
1290 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1291 	if (!netdev)
1292 		return ERR_PTR(-ENOMEM);
1293 
1294 	np                   = netdev_priv(netdev);
1295 	np->xbdev            = dev;
1296 
1297 	np->queues = NULL;
1298 
1299 	err = -ENOMEM;
1300 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1301 	if (np->rx_stats == NULL)
1302 		goto exit;
1303 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1304 	if (np->tx_stats == NULL)
1305 		goto exit;
1306 
1307 	netdev->netdev_ops	= &xennet_netdev_ops;
1308 
1309 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1310 				  NETIF_F_GSO_ROBUST;
1311 	netdev->hw_features	= NETIF_F_SG |
1312 				  NETIF_F_IPV6_CSUM |
1313 				  NETIF_F_TSO | NETIF_F_TSO6;
1314 
1315 	/*
1316          * Assume that all hw features are available for now. This set
1317          * will be adjusted by the call to netdev_update_features() in
1318          * xennet_connect() which is the earliest point where we can
1319          * negotiate with the backend regarding supported features.
1320          */
1321 	netdev->features |= netdev->hw_features;
1322 
1323 	netdev->ethtool_ops = &xennet_ethtool_ops;
1324 	netdev->min_mtu = ETH_MIN_MTU;
1325 	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1326 	SET_NETDEV_DEV(netdev, &dev->dev);
1327 
1328 	np->netdev = netdev;
1329 
1330 	netif_carrier_off(netdev);
1331 
1332 	xenbus_switch_state(dev, XenbusStateInitialising);
1333 	return netdev;
1334 
1335  exit:
1336 	xennet_free_netdev(netdev);
1337 	return ERR_PTR(err);
1338 }
1339 
1340 /**
1341  * Entry point to this code when a new device is created.  Allocate the basic
1342  * structures and the ring buffers for communication with the backend, and
1343  * inform the backend of the appropriate details for those.
1344  */
1345 static int netfront_probe(struct xenbus_device *dev,
1346 			  const struct xenbus_device_id *id)
1347 {
1348 	int err;
1349 	struct net_device *netdev;
1350 	struct netfront_info *info;
1351 
1352 	netdev = xennet_create_dev(dev);
1353 	if (IS_ERR(netdev)) {
1354 		err = PTR_ERR(netdev);
1355 		xenbus_dev_fatal(dev, err, "creating netdev");
1356 		return err;
1357 	}
1358 
1359 	info = netdev_priv(netdev);
1360 	dev_set_drvdata(&dev->dev, info);
1361 #ifdef CONFIG_SYSFS
1362 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1363 #endif
1364 
1365 	return 0;
1366 }
1367 
1368 static void xennet_end_access(int ref, void *page)
1369 {
1370 	/* This frees the page as a side-effect */
1371 	if (ref != GRANT_INVALID_REF)
1372 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1373 }
1374 
1375 static void xennet_disconnect_backend(struct netfront_info *info)
1376 {
1377 	unsigned int i = 0;
1378 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1379 
1380 	netif_carrier_off(info->netdev);
1381 
1382 	for (i = 0; i < num_queues && info->queues; ++i) {
1383 		struct netfront_queue *queue = &info->queues[i];
1384 
1385 		del_timer_sync(&queue->rx_refill_timer);
1386 
1387 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1388 			unbind_from_irqhandler(queue->tx_irq, queue);
1389 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1390 			unbind_from_irqhandler(queue->tx_irq, queue);
1391 			unbind_from_irqhandler(queue->rx_irq, queue);
1392 		}
1393 		queue->tx_evtchn = queue->rx_evtchn = 0;
1394 		queue->tx_irq = queue->rx_irq = 0;
1395 
1396 		if (netif_running(info->netdev))
1397 			napi_synchronize(&queue->napi);
1398 
1399 		xennet_release_tx_bufs(queue);
1400 		xennet_release_rx_bufs(queue);
1401 		gnttab_free_grant_references(queue->gref_tx_head);
1402 		gnttab_free_grant_references(queue->gref_rx_head);
1403 
1404 		/* End access and free the pages */
1405 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1406 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1407 
1408 		queue->tx_ring_ref = GRANT_INVALID_REF;
1409 		queue->rx_ring_ref = GRANT_INVALID_REF;
1410 		queue->tx.sring = NULL;
1411 		queue->rx.sring = NULL;
1412 	}
1413 }
1414 
1415 /**
1416  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1417  * driver restart.  We tear down our netif structure and recreate it, but
1418  * leave the device-layer structures intact so that this is transparent to the
1419  * rest of the kernel.
1420  */
1421 static int netfront_resume(struct xenbus_device *dev)
1422 {
1423 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1424 
1425 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1426 
1427 	xennet_disconnect_backend(info);
1428 	return 0;
1429 }
1430 
1431 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1432 {
1433 	char *s, *e, *macstr;
1434 	int i;
1435 
1436 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1437 	if (IS_ERR(macstr))
1438 		return PTR_ERR(macstr);
1439 
1440 	for (i = 0; i < ETH_ALEN; i++) {
1441 		mac[i] = simple_strtoul(s, &e, 16);
1442 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1443 			kfree(macstr);
1444 			return -ENOENT;
1445 		}
1446 		s = e+1;
1447 	}
1448 
1449 	kfree(macstr);
1450 	return 0;
1451 }
1452 
1453 static int setup_netfront_single(struct netfront_queue *queue)
1454 {
1455 	int err;
1456 
1457 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1458 	if (err < 0)
1459 		goto fail;
1460 
1461 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1462 					xennet_interrupt,
1463 					0, queue->info->netdev->name, queue);
1464 	if (err < 0)
1465 		goto bind_fail;
1466 	queue->rx_evtchn = queue->tx_evtchn;
1467 	queue->rx_irq = queue->tx_irq = err;
1468 
1469 	return 0;
1470 
1471 bind_fail:
1472 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1473 	queue->tx_evtchn = 0;
1474 fail:
1475 	return err;
1476 }
1477 
1478 static int setup_netfront_split(struct netfront_queue *queue)
1479 {
1480 	int err;
1481 
1482 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1483 	if (err < 0)
1484 		goto fail;
1485 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1486 	if (err < 0)
1487 		goto alloc_rx_evtchn_fail;
1488 
1489 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1490 		 "%s-tx", queue->name);
1491 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1492 					xennet_tx_interrupt,
1493 					0, queue->tx_irq_name, queue);
1494 	if (err < 0)
1495 		goto bind_tx_fail;
1496 	queue->tx_irq = err;
1497 
1498 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1499 		 "%s-rx", queue->name);
1500 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1501 					xennet_rx_interrupt,
1502 					0, queue->rx_irq_name, queue);
1503 	if (err < 0)
1504 		goto bind_rx_fail;
1505 	queue->rx_irq = err;
1506 
1507 	return 0;
1508 
1509 bind_rx_fail:
1510 	unbind_from_irqhandler(queue->tx_irq, queue);
1511 	queue->tx_irq = 0;
1512 bind_tx_fail:
1513 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1514 	queue->rx_evtchn = 0;
1515 alloc_rx_evtchn_fail:
1516 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1517 	queue->tx_evtchn = 0;
1518 fail:
1519 	return err;
1520 }
1521 
1522 static int setup_netfront(struct xenbus_device *dev,
1523 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1524 {
1525 	struct xen_netif_tx_sring *txs;
1526 	struct xen_netif_rx_sring *rxs;
1527 	grant_ref_t gref;
1528 	int err;
1529 
1530 	queue->tx_ring_ref = GRANT_INVALID_REF;
1531 	queue->rx_ring_ref = GRANT_INVALID_REF;
1532 	queue->rx.sring = NULL;
1533 	queue->tx.sring = NULL;
1534 
1535 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1536 	if (!txs) {
1537 		err = -ENOMEM;
1538 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1539 		goto fail;
1540 	}
1541 	SHARED_RING_INIT(txs);
1542 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1543 
1544 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1545 	if (err < 0)
1546 		goto grant_tx_ring_fail;
1547 	queue->tx_ring_ref = gref;
1548 
1549 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1550 	if (!rxs) {
1551 		err = -ENOMEM;
1552 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1553 		goto alloc_rx_ring_fail;
1554 	}
1555 	SHARED_RING_INIT(rxs);
1556 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1557 
1558 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1559 	if (err < 0)
1560 		goto grant_rx_ring_fail;
1561 	queue->rx_ring_ref = gref;
1562 
1563 	if (feature_split_evtchn)
1564 		err = setup_netfront_split(queue);
1565 	/* setup single event channel if
1566 	 *  a) feature-split-event-channels == 0
1567 	 *  b) feature-split-event-channels == 1 but failed to setup
1568 	 */
1569 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1570 		err = setup_netfront_single(queue);
1571 
1572 	if (err)
1573 		goto alloc_evtchn_fail;
1574 
1575 	return 0;
1576 
1577 	/* If we fail to setup netfront, it is safe to just revoke access to
1578 	 * granted pages because backend is not accessing it at this point.
1579 	 */
1580 alloc_evtchn_fail:
1581 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1582 grant_rx_ring_fail:
1583 	free_page((unsigned long)rxs);
1584 alloc_rx_ring_fail:
1585 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1586 grant_tx_ring_fail:
1587 	free_page((unsigned long)txs);
1588 fail:
1589 	return err;
1590 }
1591 
1592 /* Queue-specific initialisation
1593  * This used to be done in xennet_create_dev() but must now
1594  * be run per-queue.
1595  */
1596 static int xennet_init_queue(struct netfront_queue *queue)
1597 {
1598 	unsigned short i;
1599 	int err = 0;
1600 
1601 	spin_lock_init(&queue->tx_lock);
1602 	spin_lock_init(&queue->rx_lock);
1603 
1604 	timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1605 
1606 	snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1607 		 queue->info->netdev->name, queue->id);
1608 
1609 	/* Initialise tx_skbs as a free chain containing every entry. */
1610 	queue->tx_skb_freelist = 0;
1611 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1612 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1613 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1614 		queue->grant_tx_page[i] = NULL;
1615 	}
1616 
1617 	/* Clear out rx_skbs */
1618 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1619 		queue->rx_skbs[i] = NULL;
1620 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1621 	}
1622 
1623 	/* A grant for every tx ring slot */
1624 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1625 					  &queue->gref_tx_head) < 0) {
1626 		pr_alert("can't alloc tx grant refs\n");
1627 		err = -ENOMEM;
1628 		goto exit;
1629 	}
1630 
1631 	/* A grant for every rx ring slot */
1632 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1633 					  &queue->gref_rx_head) < 0) {
1634 		pr_alert("can't alloc rx grant refs\n");
1635 		err = -ENOMEM;
1636 		goto exit_free_tx;
1637 	}
1638 
1639 	return 0;
1640 
1641  exit_free_tx:
1642 	gnttab_free_grant_references(queue->gref_tx_head);
1643  exit:
1644 	return err;
1645 }
1646 
1647 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1648 			   struct xenbus_transaction *xbt, int write_hierarchical)
1649 {
1650 	/* Write the queue-specific keys into XenStore in the traditional
1651 	 * way for a single queue, or in a queue subkeys for multiple
1652 	 * queues.
1653 	 */
1654 	struct xenbus_device *dev = queue->info->xbdev;
1655 	int err;
1656 	const char *message;
1657 	char *path;
1658 	size_t pathsize;
1659 
1660 	/* Choose the correct place to write the keys */
1661 	if (write_hierarchical) {
1662 		pathsize = strlen(dev->nodename) + 10;
1663 		path = kzalloc(pathsize, GFP_KERNEL);
1664 		if (!path) {
1665 			err = -ENOMEM;
1666 			message = "out of memory while writing ring references";
1667 			goto error;
1668 		}
1669 		snprintf(path, pathsize, "%s/queue-%u",
1670 				dev->nodename, queue->id);
1671 	} else {
1672 		path = (char *)dev->nodename;
1673 	}
1674 
1675 	/* Write ring references */
1676 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1677 			queue->tx_ring_ref);
1678 	if (err) {
1679 		message = "writing tx-ring-ref";
1680 		goto error;
1681 	}
1682 
1683 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1684 			queue->rx_ring_ref);
1685 	if (err) {
1686 		message = "writing rx-ring-ref";
1687 		goto error;
1688 	}
1689 
1690 	/* Write event channels; taking into account both shared
1691 	 * and split event channel scenarios.
1692 	 */
1693 	if (queue->tx_evtchn == queue->rx_evtchn) {
1694 		/* Shared event channel */
1695 		err = xenbus_printf(*xbt, path,
1696 				"event-channel", "%u", queue->tx_evtchn);
1697 		if (err) {
1698 			message = "writing event-channel";
1699 			goto error;
1700 		}
1701 	} else {
1702 		/* Split event channels */
1703 		err = xenbus_printf(*xbt, path,
1704 				"event-channel-tx", "%u", queue->tx_evtchn);
1705 		if (err) {
1706 			message = "writing event-channel-tx";
1707 			goto error;
1708 		}
1709 
1710 		err = xenbus_printf(*xbt, path,
1711 				"event-channel-rx", "%u", queue->rx_evtchn);
1712 		if (err) {
1713 			message = "writing event-channel-rx";
1714 			goto error;
1715 		}
1716 	}
1717 
1718 	if (write_hierarchical)
1719 		kfree(path);
1720 	return 0;
1721 
1722 error:
1723 	if (write_hierarchical)
1724 		kfree(path);
1725 	xenbus_dev_fatal(dev, err, "%s", message);
1726 	return err;
1727 }
1728 
1729 static void xennet_destroy_queues(struct netfront_info *info)
1730 {
1731 	unsigned int i;
1732 
1733 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1734 		struct netfront_queue *queue = &info->queues[i];
1735 
1736 		if (netif_running(info->netdev))
1737 			napi_disable(&queue->napi);
1738 		netif_napi_del(&queue->napi);
1739 	}
1740 
1741 	kfree(info->queues);
1742 	info->queues = NULL;
1743 }
1744 
1745 static int xennet_create_queues(struct netfront_info *info,
1746 				unsigned int *num_queues)
1747 {
1748 	unsigned int i;
1749 	int ret;
1750 
1751 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1752 			       GFP_KERNEL);
1753 	if (!info->queues)
1754 		return -ENOMEM;
1755 
1756 	for (i = 0; i < *num_queues; i++) {
1757 		struct netfront_queue *queue = &info->queues[i];
1758 
1759 		queue->id = i;
1760 		queue->info = info;
1761 
1762 		ret = xennet_init_queue(queue);
1763 		if (ret < 0) {
1764 			dev_warn(&info->xbdev->dev,
1765 				 "only created %d queues\n", i);
1766 			*num_queues = i;
1767 			break;
1768 		}
1769 
1770 		netif_napi_add(queue->info->netdev, &queue->napi,
1771 			       xennet_poll, 64);
1772 		if (netif_running(info->netdev))
1773 			napi_enable(&queue->napi);
1774 	}
1775 
1776 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
1777 
1778 	if (*num_queues == 0) {
1779 		dev_err(&info->xbdev->dev, "no queues\n");
1780 		return -EINVAL;
1781 	}
1782 	return 0;
1783 }
1784 
1785 /* Common code used when first setting up, and when resuming. */
1786 static int talk_to_netback(struct xenbus_device *dev,
1787 			   struct netfront_info *info)
1788 {
1789 	const char *message;
1790 	struct xenbus_transaction xbt;
1791 	int err;
1792 	unsigned int feature_split_evtchn;
1793 	unsigned int i = 0;
1794 	unsigned int max_queues = 0;
1795 	struct netfront_queue *queue = NULL;
1796 	unsigned int num_queues = 1;
1797 
1798 	info->netdev->irq = 0;
1799 
1800 	/* Check if backend supports multiple queues */
1801 	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1802 					  "multi-queue-max-queues", 1);
1803 	num_queues = min(max_queues, xennet_max_queues);
1804 
1805 	/* Check feature-split-event-channels */
1806 	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1807 					"feature-split-event-channels", 0);
1808 
1809 	/* Read mac addr. */
1810 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
1811 	if (err) {
1812 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1813 		goto out_unlocked;
1814 	}
1815 
1816 	rtnl_lock();
1817 	if (info->queues)
1818 		xennet_destroy_queues(info);
1819 
1820 	err = xennet_create_queues(info, &num_queues);
1821 	if (err < 0) {
1822 		xenbus_dev_fatal(dev, err, "creating queues");
1823 		kfree(info->queues);
1824 		info->queues = NULL;
1825 		goto out;
1826 	}
1827 	rtnl_unlock();
1828 
1829 	/* Create shared ring, alloc event channel -- for each queue */
1830 	for (i = 0; i < num_queues; ++i) {
1831 		queue = &info->queues[i];
1832 		err = setup_netfront(dev, queue, feature_split_evtchn);
1833 		if (err)
1834 			goto destroy_ring;
1835 	}
1836 
1837 again:
1838 	err = xenbus_transaction_start(&xbt);
1839 	if (err) {
1840 		xenbus_dev_fatal(dev, err, "starting transaction");
1841 		goto destroy_ring;
1842 	}
1843 
1844 	if (xenbus_exists(XBT_NIL,
1845 			  info->xbdev->otherend, "multi-queue-max-queues")) {
1846 		/* Write the number of queues */
1847 		err = xenbus_printf(xbt, dev->nodename,
1848 				    "multi-queue-num-queues", "%u", num_queues);
1849 		if (err) {
1850 			message = "writing multi-queue-num-queues";
1851 			goto abort_transaction_no_dev_fatal;
1852 		}
1853 	}
1854 
1855 	if (num_queues == 1) {
1856 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1857 		if (err)
1858 			goto abort_transaction_no_dev_fatal;
1859 	} else {
1860 		/* Write the keys for each queue */
1861 		for (i = 0; i < num_queues; ++i) {
1862 			queue = &info->queues[i];
1863 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1864 			if (err)
1865 				goto abort_transaction_no_dev_fatal;
1866 		}
1867 	}
1868 
1869 	/* The remaining keys are not queue-specific */
1870 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1871 			    1);
1872 	if (err) {
1873 		message = "writing request-rx-copy";
1874 		goto abort_transaction;
1875 	}
1876 
1877 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1878 	if (err) {
1879 		message = "writing feature-rx-notify";
1880 		goto abort_transaction;
1881 	}
1882 
1883 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1884 	if (err) {
1885 		message = "writing feature-sg";
1886 		goto abort_transaction;
1887 	}
1888 
1889 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1890 	if (err) {
1891 		message = "writing feature-gso-tcpv4";
1892 		goto abort_transaction;
1893 	}
1894 
1895 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1896 	if (err) {
1897 		message = "writing feature-gso-tcpv6";
1898 		goto abort_transaction;
1899 	}
1900 
1901 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1902 			   "1");
1903 	if (err) {
1904 		message = "writing feature-ipv6-csum-offload";
1905 		goto abort_transaction;
1906 	}
1907 
1908 	err = xenbus_transaction_end(xbt, 0);
1909 	if (err) {
1910 		if (err == -EAGAIN)
1911 			goto again;
1912 		xenbus_dev_fatal(dev, err, "completing transaction");
1913 		goto destroy_ring;
1914 	}
1915 
1916 	return 0;
1917 
1918  abort_transaction:
1919 	xenbus_dev_fatal(dev, err, "%s", message);
1920 abort_transaction_no_dev_fatal:
1921 	xenbus_transaction_end(xbt, 1);
1922  destroy_ring:
1923 	xennet_disconnect_backend(info);
1924 	rtnl_lock();
1925 	xennet_destroy_queues(info);
1926  out:
1927 	rtnl_unlock();
1928 out_unlocked:
1929 	device_unregister(&dev->dev);
1930 	return err;
1931 }
1932 
1933 static int xennet_connect(struct net_device *dev)
1934 {
1935 	struct netfront_info *np = netdev_priv(dev);
1936 	unsigned int num_queues = 0;
1937 	int err;
1938 	unsigned int j = 0;
1939 	struct netfront_queue *queue = NULL;
1940 
1941 	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1942 		dev_info(&dev->dev,
1943 			 "backend does not support copying receive path\n");
1944 		return -ENODEV;
1945 	}
1946 
1947 	err = talk_to_netback(np->xbdev, np);
1948 	if (err)
1949 		return err;
1950 
1951 	/* talk_to_netback() sets the correct number of queues */
1952 	num_queues = dev->real_num_tx_queues;
1953 
1954 	if (dev->reg_state == NETREG_UNINITIALIZED) {
1955 		err = register_netdev(dev);
1956 		if (err) {
1957 			pr_warn("%s: register_netdev err=%d\n", __func__, err);
1958 			device_unregister(&np->xbdev->dev);
1959 			return err;
1960 		}
1961 	}
1962 
1963 	rtnl_lock();
1964 	netdev_update_features(dev);
1965 	rtnl_unlock();
1966 
1967 	/*
1968 	 * All public and private state should now be sane.  Get
1969 	 * ready to start sending and receiving packets and give the driver
1970 	 * domain a kick because we've probably just requeued some
1971 	 * packets.
1972 	 */
1973 	netif_carrier_on(np->netdev);
1974 	for (j = 0; j < num_queues; ++j) {
1975 		queue = &np->queues[j];
1976 
1977 		notify_remote_via_irq(queue->tx_irq);
1978 		if (queue->tx_irq != queue->rx_irq)
1979 			notify_remote_via_irq(queue->rx_irq);
1980 
1981 		spin_lock_irq(&queue->tx_lock);
1982 		xennet_tx_buf_gc(queue);
1983 		spin_unlock_irq(&queue->tx_lock);
1984 
1985 		spin_lock_bh(&queue->rx_lock);
1986 		xennet_alloc_rx_buffers(queue);
1987 		spin_unlock_bh(&queue->rx_lock);
1988 	}
1989 
1990 	return 0;
1991 }
1992 
1993 /**
1994  * Callback received when the backend's state changes.
1995  */
1996 static void netback_changed(struct xenbus_device *dev,
1997 			    enum xenbus_state backend_state)
1998 {
1999 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2000 	struct net_device *netdev = np->netdev;
2001 
2002 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2003 
2004 	switch (backend_state) {
2005 	case XenbusStateInitialising:
2006 	case XenbusStateInitialised:
2007 	case XenbusStateReconfiguring:
2008 	case XenbusStateReconfigured:
2009 		break;
2010 
2011 	case XenbusStateUnknown:
2012 		wake_up_all(&module_unload_q);
2013 		break;
2014 
2015 	case XenbusStateInitWait:
2016 		if (dev->state != XenbusStateInitialising)
2017 			break;
2018 		if (xennet_connect(netdev) != 0)
2019 			break;
2020 		xenbus_switch_state(dev, XenbusStateConnected);
2021 		break;
2022 
2023 	case XenbusStateConnected:
2024 		netdev_notify_peers(netdev);
2025 		break;
2026 
2027 	case XenbusStateClosed:
2028 		wake_up_all(&module_unload_q);
2029 		if (dev->state == XenbusStateClosed)
2030 			break;
2031 		/* Missed the backend's CLOSING state -- fallthrough */
2032 	case XenbusStateClosing:
2033 		wake_up_all(&module_unload_q);
2034 		xenbus_frontend_closed(dev);
2035 		break;
2036 	}
2037 }
2038 
2039 static const struct xennet_stat {
2040 	char name[ETH_GSTRING_LEN];
2041 	u16 offset;
2042 } xennet_stats[] = {
2043 	{
2044 		"rx_gso_checksum_fixup",
2045 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2046 	},
2047 };
2048 
2049 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2050 {
2051 	switch (string_set) {
2052 	case ETH_SS_STATS:
2053 		return ARRAY_SIZE(xennet_stats);
2054 	default:
2055 		return -EINVAL;
2056 	}
2057 }
2058 
2059 static void xennet_get_ethtool_stats(struct net_device *dev,
2060 				     struct ethtool_stats *stats, u64 * data)
2061 {
2062 	void *np = netdev_priv(dev);
2063 	int i;
2064 
2065 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2066 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2067 }
2068 
2069 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2070 {
2071 	int i;
2072 
2073 	switch (stringset) {
2074 	case ETH_SS_STATS:
2075 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2076 			memcpy(data + i * ETH_GSTRING_LEN,
2077 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2078 		break;
2079 	}
2080 }
2081 
2082 static const struct ethtool_ops xennet_ethtool_ops =
2083 {
2084 	.get_link = ethtool_op_get_link,
2085 
2086 	.get_sset_count = xennet_get_sset_count,
2087 	.get_ethtool_stats = xennet_get_ethtool_stats,
2088 	.get_strings = xennet_get_strings,
2089 };
2090 
2091 #ifdef CONFIG_SYSFS
2092 static ssize_t show_rxbuf(struct device *dev,
2093 			  struct device_attribute *attr, char *buf)
2094 {
2095 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2096 }
2097 
2098 static ssize_t store_rxbuf(struct device *dev,
2099 			   struct device_attribute *attr,
2100 			   const char *buf, size_t len)
2101 {
2102 	char *endp;
2103 	unsigned long target;
2104 
2105 	if (!capable(CAP_NET_ADMIN))
2106 		return -EPERM;
2107 
2108 	target = simple_strtoul(buf, &endp, 0);
2109 	if (endp == buf)
2110 		return -EBADMSG;
2111 
2112 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2113 
2114 	return len;
2115 }
2116 
2117 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2118 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2119 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2120 
2121 static struct attribute *xennet_dev_attrs[] = {
2122 	&dev_attr_rxbuf_min.attr,
2123 	&dev_attr_rxbuf_max.attr,
2124 	&dev_attr_rxbuf_cur.attr,
2125 	NULL
2126 };
2127 
2128 static const struct attribute_group xennet_dev_group = {
2129 	.attrs = xennet_dev_attrs
2130 };
2131 #endif /* CONFIG_SYSFS */
2132 
2133 static int xennet_remove(struct xenbus_device *dev)
2134 {
2135 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2136 
2137 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2138 
2139 	if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) {
2140 		xenbus_switch_state(dev, XenbusStateClosing);
2141 		wait_event(module_unload_q,
2142 			   xenbus_read_driver_state(dev->otherend) ==
2143 			   XenbusStateClosing ||
2144 			   xenbus_read_driver_state(dev->otherend) ==
2145 			   XenbusStateUnknown);
2146 
2147 		xenbus_switch_state(dev, XenbusStateClosed);
2148 		wait_event(module_unload_q,
2149 			   xenbus_read_driver_state(dev->otherend) ==
2150 			   XenbusStateClosed ||
2151 			   xenbus_read_driver_state(dev->otherend) ==
2152 			   XenbusStateUnknown);
2153 	}
2154 
2155 	xennet_disconnect_backend(info);
2156 
2157 	if (info->netdev->reg_state == NETREG_REGISTERED)
2158 		unregister_netdev(info->netdev);
2159 
2160 	if (info->queues) {
2161 		rtnl_lock();
2162 		xennet_destroy_queues(info);
2163 		rtnl_unlock();
2164 	}
2165 	xennet_free_netdev(info->netdev);
2166 
2167 	return 0;
2168 }
2169 
2170 static const struct xenbus_device_id netfront_ids[] = {
2171 	{ "vif" },
2172 	{ "" }
2173 };
2174 
2175 static struct xenbus_driver netfront_driver = {
2176 	.ids = netfront_ids,
2177 	.probe = netfront_probe,
2178 	.remove = xennet_remove,
2179 	.resume = netfront_resume,
2180 	.otherend_changed = netback_changed,
2181 };
2182 
2183 static int __init netif_init(void)
2184 {
2185 	if (!xen_domain())
2186 		return -ENODEV;
2187 
2188 	if (!xen_has_pv_nic_devices())
2189 		return -ENODEV;
2190 
2191 	pr_info("Initialising Xen virtual ethernet driver\n");
2192 
2193 	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2194 	 * specified a value.
2195 	 */
2196 	if (xennet_max_queues == 0)
2197 		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2198 					  num_online_cpus());
2199 
2200 	return xenbus_register_frontend(&netfront_driver);
2201 }
2202 module_init(netif_init);
2203 
2204 
2205 static void __exit netif_exit(void)
2206 {
2207 	xenbus_unregister_driver(&netfront_driver);
2208 }
2209 module_exit(netif_exit);
2210 
2211 MODULE_DESCRIPTION("Xen virtual network device frontend");
2212 MODULE_LICENSE("GPL");
2213 MODULE_ALIAS("xen:vif");
2214 MODULE_ALIAS("xennet");
2215