xref: /openbmc/linux/drivers/net/xen-netfront.c (revision e2f1cf25)
1 /*
2  * Virtual network driver for conversing with remote driver backends.
3  *
4  * Copyright (c) 2002-2005, K A Fraser
5  * Copyright (c) 2005, XenSource Ltd
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License version 2
9  * as published by the Free Software Foundation; or, when distributed
10  * separately from the Linux kernel or incorporated into other
11  * software packages, subject to the following license:
12  *
13  * Permission is hereby granted, free of charge, to any person obtaining a copy
14  * of this source file (the "Software"), to deal in the Software without
15  * restriction, including without limitation the rights to use, copy, modify,
16  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17  * and to permit persons to whom the Software is furnished to do so, subject to
18  * the following conditions:
19  *
20  * The above copyright notice and this permission notice shall be included in
21  * all copies or substantial portions of the Software.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29  * IN THE SOFTWARE.
30  */
31 
32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33 
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_ether.h>
41 #include <net/tcp.h>
42 #include <linux/udp.h>
43 #include <linux/moduleparam.h>
44 #include <linux/mm.h>
45 #include <linux/slab.h>
46 #include <net/ip.h>
47 
48 #include <xen/xen.h>
49 #include <xen/xenbus.h>
50 #include <xen/events.h>
51 #include <xen/page.h>
52 #include <xen/platform_pci.h>
53 #include <xen/grant_table.h>
54 
55 #include <xen/interface/io/netif.h>
56 #include <xen/interface/memory.h>
57 #include <xen/interface/grant_table.h>
58 
59 /* Module parameters */
60 static unsigned int xennet_max_queues;
61 module_param_named(max_queues, xennet_max_queues, uint, 0644);
62 MODULE_PARM_DESC(max_queues,
63 		 "Maximum number of queues per virtual interface");
64 
65 static const struct ethtool_ops xennet_ethtool_ops;
66 
67 struct netfront_cb {
68 	int pull_to;
69 };
70 
71 #define NETFRONT_SKB_CB(skb)	((struct netfront_cb *)((skb)->cb))
72 
73 #define RX_COPY_THRESHOLD 256
74 
75 #define GRANT_INVALID_REF	0
76 
77 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
78 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
79 
80 /* Minimum number of Rx slots (includes slot for GSO metadata). */
81 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
82 
83 /* Queue name is interface name with "-qNNN" appended */
84 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
85 
86 /* IRQ name is queue name with "-tx" or "-rx" appended */
87 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
88 
89 struct netfront_stats {
90 	u64			packets;
91 	u64			bytes;
92 	struct u64_stats_sync	syncp;
93 };
94 
95 struct netfront_info;
96 
97 struct netfront_queue {
98 	unsigned int id; /* Queue ID, 0-based */
99 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
100 	struct netfront_info *info;
101 
102 	struct napi_struct napi;
103 
104 	/* Split event channels support, tx_* == rx_* when using
105 	 * single event channel.
106 	 */
107 	unsigned int tx_evtchn, rx_evtchn;
108 	unsigned int tx_irq, rx_irq;
109 	/* Only used when split event channels support is enabled */
110 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
111 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
112 
113 	spinlock_t   tx_lock;
114 	struct xen_netif_tx_front_ring tx;
115 	int tx_ring_ref;
116 
117 	/*
118 	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
119 	 * are linked from tx_skb_freelist through skb_entry.link.
120 	 *
121 	 *  NB. Freelist index entries are always going to be less than
122 	 *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
123 	 *  greater than PAGE_OFFSET: we use this property to distinguish
124 	 *  them.
125 	 */
126 	union skb_entry {
127 		struct sk_buff *skb;
128 		unsigned long link;
129 	} tx_skbs[NET_TX_RING_SIZE];
130 	grant_ref_t gref_tx_head;
131 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
132 	struct page *grant_tx_page[NET_TX_RING_SIZE];
133 	unsigned tx_skb_freelist;
134 
135 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
136 	struct xen_netif_rx_front_ring rx;
137 	int rx_ring_ref;
138 
139 	struct timer_list rx_refill_timer;
140 
141 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
142 	grant_ref_t gref_rx_head;
143 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
144 };
145 
146 struct netfront_info {
147 	struct list_head list;
148 	struct net_device *netdev;
149 
150 	struct xenbus_device *xbdev;
151 
152 	/* Multi-queue support */
153 	struct netfront_queue *queues;
154 
155 	/* Statistics */
156 	struct netfront_stats __percpu *rx_stats;
157 	struct netfront_stats __percpu *tx_stats;
158 
159 	atomic_t rx_gso_checksum_fixup;
160 };
161 
162 struct netfront_rx_info {
163 	struct xen_netif_rx_response rx;
164 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
165 };
166 
167 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
168 {
169 	list->link = id;
170 }
171 
172 static int skb_entry_is_link(const union skb_entry *list)
173 {
174 	BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
175 	return (unsigned long)list->skb < PAGE_OFFSET;
176 }
177 
178 /*
179  * Access macros for acquiring freeing slots in tx_skbs[].
180  */
181 
182 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
183 			       unsigned short id)
184 {
185 	skb_entry_set_link(&list[id], *head);
186 	*head = id;
187 }
188 
189 static unsigned short get_id_from_freelist(unsigned *head,
190 					   union skb_entry *list)
191 {
192 	unsigned int id = *head;
193 	*head = list[id].link;
194 	return id;
195 }
196 
197 static int xennet_rxidx(RING_IDX idx)
198 {
199 	return idx & (NET_RX_RING_SIZE - 1);
200 }
201 
202 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
203 					 RING_IDX ri)
204 {
205 	int i = xennet_rxidx(ri);
206 	struct sk_buff *skb = queue->rx_skbs[i];
207 	queue->rx_skbs[i] = NULL;
208 	return skb;
209 }
210 
211 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
212 					    RING_IDX ri)
213 {
214 	int i = xennet_rxidx(ri);
215 	grant_ref_t ref = queue->grant_rx_ref[i];
216 	queue->grant_rx_ref[i] = GRANT_INVALID_REF;
217 	return ref;
218 }
219 
220 #ifdef CONFIG_SYSFS
221 static const struct attribute_group xennet_dev_group;
222 #endif
223 
224 static bool xennet_can_sg(struct net_device *dev)
225 {
226 	return dev->features & NETIF_F_SG;
227 }
228 
229 
230 static void rx_refill_timeout(unsigned long data)
231 {
232 	struct netfront_queue *queue = (struct netfront_queue *)data;
233 	napi_schedule(&queue->napi);
234 }
235 
236 static int netfront_tx_slot_available(struct netfront_queue *queue)
237 {
238 	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
239 		(NET_TX_RING_SIZE - MAX_SKB_FRAGS - 2);
240 }
241 
242 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
243 {
244 	struct net_device *dev = queue->info->netdev;
245 	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
246 
247 	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
248 	    netfront_tx_slot_available(queue) &&
249 	    likely(netif_running(dev)))
250 		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
251 }
252 
253 
254 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
255 {
256 	struct sk_buff *skb;
257 	struct page *page;
258 
259 	skb = __netdev_alloc_skb(queue->info->netdev,
260 				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
261 				 GFP_ATOMIC | __GFP_NOWARN);
262 	if (unlikely(!skb))
263 		return NULL;
264 
265 	page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
266 	if (!page) {
267 		kfree_skb(skb);
268 		return NULL;
269 	}
270 	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
271 
272 	/* Align ip header to a 16 bytes boundary */
273 	skb_reserve(skb, NET_IP_ALIGN);
274 	skb->dev = queue->info->netdev;
275 
276 	return skb;
277 }
278 
279 
280 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
281 {
282 	RING_IDX req_prod = queue->rx.req_prod_pvt;
283 	int notify;
284 
285 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
286 		return;
287 
288 	for (req_prod = queue->rx.req_prod_pvt;
289 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
290 	     req_prod++) {
291 		struct sk_buff *skb;
292 		unsigned short id;
293 		grant_ref_t ref;
294 		unsigned long pfn;
295 		struct xen_netif_rx_request *req;
296 
297 		skb = xennet_alloc_one_rx_buffer(queue);
298 		if (!skb)
299 			break;
300 
301 		id = xennet_rxidx(req_prod);
302 
303 		BUG_ON(queue->rx_skbs[id]);
304 		queue->rx_skbs[id] = skb;
305 
306 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
307 		BUG_ON((signed short)ref < 0);
308 		queue->grant_rx_ref[id] = ref;
309 
310 		pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
311 
312 		req = RING_GET_REQUEST(&queue->rx, req_prod);
313 		gnttab_grant_foreign_access_ref(ref,
314 						queue->info->xbdev->otherend_id,
315 						pfn_to_mfn(pfn),
316 						0);
317 
318 		req->id = id;
319 		req->gref = ref;
320 	}
321 
322 	queue->rx.req_prod_pvt = req_prod;
323 
324 	/* Not enough requests? Try again later. */
325 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN) {
326 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
327 		return;
328 	}
329 
330 	wmb();		/* barrier so backend seens requests */
331 
332 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
333 	if (notify)
334 		notify_remote_via_irq(queue->rx_irq);
335 }
336 
337 static int xennet_open(struct net_device *dev)
338 {
339 	struct netfront_info *np = netdev_priv(dev);
340 	unsigned int num_queues = dev->real_num_tx_queues;
341 	unsigned int i = 0;
342 	struct netfront_queue *queue = NULL;
343 
344 	for (i = 0; i < num_queues; ++i) {
345 		queue = &np->queues[i];
346 		napi_enable(&queue->napi);
347 
348 		spin_lock_bh(&queue->rx_lock);
349 		if (netif_carrier_ok(dev)) {
350 			xennet_alloc_rx_buffers(queue);
351 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
352 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
353 				napi_schedule(&queue->napi);
354 		}
355 		spin_unlock_bh(&queue->rx_lock);
356 	}
357 
358 	netif_tx_start_all_queues(dev);
359 
360 	return 0;
361 }
362 
363 static void xennet_tx_buf_gc(struct netfront_queue *queue)
364 {
365 	RING_IDX cons, prod;
366 	unsigned short id;
367 	struct sk_buff *skb;
368 
369 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
370 
371 	do {
372 		prod = queue->tx.sring->rsp_prod;
373 		rmb(); /* Ensure we see responses up to 'rp'. */
374 
375 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
376 			struct xen_netif_tx_response *txrsp;
377 
378 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
379 			if (txrsp->status == XEN_NETIF_RSP_NULL)
380 				continue;
381 
382 			id  = txrsp->id;
383 			skb = queue->tx_skbs[id].skb;
384 			if (unlikely(gnttab_query_foreign_access(
385 				queue->grant_tx_ref[id]) != 0)) {
386 				pr_alert("%s: warning -- grant still in use by backend domain\n",
387 					 __func__);
388 				BUG();
389 			}
390 			gnttab_end_foreign_access_ref(
391 				queue->grant_tx_ref[id], GNTMAP_readonly);
392 			gnttab_release_grant_reference(
393 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
394 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
395 			queue->grant_tx_page[id] = NULL;
396 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
397 			dev_kfree_skb_irq(skb);
398 		}
399 
400 		queue->tx.rsp_cons = prod;
401 
402 		/*
403 		 * Set a new event, then check for race with update of tx_cons.
404 		 * Note that it is essential to schedule a callback, no matter
405 		 * how few buffers are pending. Even if there is space in the
406 		 * transmit ring, higher layers may be blocked because too much
407 		 * data is outstanding: in such cases notification from Xen is
408 		 * likely to be the only kick that we'll get.
409 		 */
410 		queue->tx.sring->rsp_event =
411 			prod + ((queue->tx.sring->req_prod - prod) >> 1) + 1;
412 		mb();		/* update shared area */
413 	} while ((cons == prod) && (prod != queue->tx.sring->rsp_prod));
414 
415 	xennet_maybe_wake_tx(queue);
416 }
417 
418 static struct xen_netif_tx_request *xennet_make_one_txreq(
419 	struct netfront_queue *queue, struct sk_buff *skb,
420 	struct page *page, unsigned int offset, unsigned int len)
421 {
422 	unsigned int id;
423 	struct xen_netif_tx_request *tx;
424 	grant_ref_t ref;
425 
426 	len = min_t(unsigned int, PAGE_SIZE - offset, len);
427 
428 	id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
429 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
430 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
431 	BUG_ON((signed short)ref < 0);
432 
433 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
434 					page_to_mfn(page), GNTMAP_readonly);
435 
436 	queue->tx_skbs[id].skb = skb;
437 	queue->grant_tx_page[id] = page;
438 	queue->grant_tx_ref[id] = ref;
439 
440 	tx->id = id;
441 	tx->gref = ref;
442 	tx->offset = offset;
443 	tx->size = len;
444 	tx->flags = 0;
445 
446 	return tx;
447 }
448 
449 static struct xen_netif_tx_request *xennet_make_txreqs(
450 	struct netfront_queue *queue, struct xen_netif_tx_request *tx,
451 	struct sk_buff *skb, struct page *page,
452 	unsigned int offset, unsigned int len)
453 {
454 	/* Skip unused frames from start of page */
455 	page += offset >> PAGE_SHIFT;
456 	offset &= ~PAGE_MASK;
457 
458 	while (len) {
459 		tx->flags |= XEN_NETTXF_more_data;
460 		tx = xennet_make_one_txreq(queue, skb_get(skb),
461 					   page, offset, len);
462 		page++;
463 		offset = 0;
464 		len -= tx->size;
465 	}
466 
467 	return tx;
468 }
469 
470 /*
471  * Count how many ring slots are required to send this skb. Each frag
472  * might be a compound page.
473  */
474 static int xennet_count_skb_slots(struct sk_buff *skb)
475 {
476 	int i, frags = skb_shinfo(skb)->nr_frags;
477 	int pages;
478 
479 	pages = PFN_UP(offset_in_page(skb->data) + skb_headlen(skb));
480 
481 	for (i = 0; i < frags; i++) {
482 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
483 		unsigned long size = skb_frag_size(frag);
484 		unsigned long offset = frag->page_offset;
485 
486 		/* Skip unused frames from start of page */
487 		offset &= ~PAGE_MASK;
488 
489 		pages += PFN_UP(offset + size);
490 	}
491 
492 	return pages;
493 }
494 
495 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
496 			       void *accel_priv, select_queue_fallback_t fallback)
497 {
498 	unsigned int num_queues = dev->real_num_tx_queues;
499 	u32 hash;
500 	u16 queue_idx;
501 
502 	/* First, check if there is only one queue */
503 	if (num_queues == 1) {
504 		queue_idx = 0;
505 	} else {
506 		hash = skb_get_hash(skb);
507 		queue_idx = hash % num_queues;
508 	}
509 
510 	return queue_idx;
511 }
512 
513 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
514 {
515 	struct netfront_info *np = netdev_priv(dev);
516 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
517 	struct xen_netif_tx_request *tx, *first_tx;
518 	unsigned int i;
519 	int notify;
520 	int slots;
521 	struct page *page;
522 	unsigned int offset;
523 	unsigned int len;
524 	unsigned long flags;
525 	struct netfront_queue *queue = NULL;
526 	unsigned int num_queues = dev->real_num_tx_queues;
527 	u16 queue_index;
528 
529 	/* Drop the packet if no queues are set up */
530 	if (num_queues < 1)
531 		goto drop;
532 	/* Determine which queue to transmit this SKB on */
533 	queue_index = skb_get_queue_mapping(skb);
534 	queue = &np->queues[queue_index];
535 
536 	/* If skb->len is too big for wire format, drop skb and alert
537 	 * user about misconfiguration.
538 	 */
539 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
540 		net_alert_ratelimited(
541 			"xennet: skb->len = %u, too big for wire format\n",
542 			skb->len);
543 		goto drop;
544 	}
545 
546 	slots = xennet_count_skb_slots(skb);
547 	if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
548 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
549 				    slots, skb->len);
550 		if (skb_linearize(skb))
551 			goto drop;
552 	}
553 
554 	page = virt_to_page(skb->data);
555 	offset = offset_in_page(skb->data);
556 	len = skb_headlen(skb);
557 
558 	spin_lock_irqsave(&queue->tx_lock, flags);
559 
560 	if (unlikely(!netif_carrier_ok(dev) ||
561 		     (slots > 1 && !xennet_can_sg(dev)) ||
562 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
563 		spin_unlock_irqrestore(&queue->tx_lock, flags);
564 		goto drop;
565 	}
566 
567 	/* First request for the linear area. */
568 	first_tx = tx = xennet_make_one_txreq(queue, skb,
569 					      page, offset, len);
570 	page++;
571 	offset = 0;
572 	len -= tx->size;
573 
574 	if (skb->ip_summed == CHECKSUM_PARTIAL)
575 		/* local packet? */
576 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
577 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
578 		/* remote but checksummed. */
579 		tx->flags |= XEN_NETTXF_data_validated;
580 
581 	/* Optional extra info after the first request. */
582 	if (skb_shinfo(skb)->gso_size) {
583 		struct xen_netif_extra_info *gso;
584 
585 		gso = (struct xen_netif_extra_info *)
586 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
587 
588 		tx->flags |= XEN_NETTXF_extra_info;
589 
590 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
591 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
592 			XEN_NETIF_GSO_TYPE_TCPV6 :
593 			XEN_NETIF_GSO_TYPE_TCPV4;
594 		gso->u.gso.pad = 0;
595 		gso->u.gso.features = 0;
596 
597 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
598 		gso->flags = 0;
599 	}
600 
601 	/* Requests for the rest of the linear area. */
602 	tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
603 
604 	/* Requests for all the frags. */
605 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
606 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
607 		tx = xennet_make_txreqs(queue, tx, skb,
608 					skb_frag_page(frag), frag->page_offset,
609 					skb_frag_size(frag));
610 	}
611 
612 	/* First request has the packet length. */
613 	first_tx->size = skb->len;
614 
615 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
616 	if (notify)
617 		notify_remote_via_irq(queue->tx_irq);
618 
619 	u64_stats_update_begin(&tx_stats->syncp);
620 	tx_stats->bytes += skb->len;
621 	tx_stats->packets++;
622 	u64_stats_update_end(&tx_stats->syncp);
623 
624 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
625 	xennet_tx_buf_gc(queue);
626 
627 	if (!netfront_tx_slot_available(queue))
628 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
629 
630 	spin_unlock_irqrestore(&queue->tx_lock, flags);
631 
632 	return NETDEV_TX_OK;
633 
634  drop:
635 	dev->stats.tx_dropped++;
636 	dev_kfree_skb_any(skb);
637 	return NETDEV_TX_OK;
638 }
639 
640 static int xennet_close(struct net_device *dev)
641 {
642 	struct netfront_info *np = netdev_priv(dev);
643 	unsigned int num_queues = dev->real_num_tx_queues;
644 	unsigned int i;
645 	struct netfront_queue *queue;
646 	netif_tx_stop_all_queues(np->netdev);
647 	for (i = 0; i < num_queues; ++i) {
648 		queue = &np->queues[i];
649 		napi_disable(&queue->napi);
650 	}
651 	return 0;
652 }
653 
654 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
655 				grant_ref_t ref)
656 {
657 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
658 
659 	BUG_ON(queue->rx_skbs[new]);
660 	queue->rx_skbs[new] = skb;
661 	queue->grant_rx_ref[new] = ref;
662 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
663 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
664 	queue->rx.req_prod_pvt++;
665 }
666 
667 static int xennet_get_extras(struct netfront_queue *queue,
668 			     struct xen_netif_extra_info *extras,
669 			     RING_IDX rp)
670 
671 {
672 	struct xen_netif_extra_info *extra;
673 	struct device *dev = &queue->info->netdev->dev;
674 	RING_IDX cons = queue->rx.rsp_cons;
675 	int err = 0;
676 
677 	do {
678 		struct sk_buff *skb;
679 		grant_ref_t ref;
680 
681 		if (unlikely(cons + 1 == rp)) {
682 			if (net_ratelimit())
683 				dev_warn(dev, "Missing extra info\n");
684 			err = -EBADR;
685 			break;
686 		}
687 
688 		extra = (struct xen_netif_extra_info *)
689 			RING_GET_RESPONSE(&queue->rx, ++cons);
690 
691 		if (unlikely(!extra->type ||
692 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
693 			if (net_ratelimit())
694 				dev_warn(dev, "Invalid extra type: %d\n",
695 					extra->type);
696 			err = -EINVAL;
697 		} else {
698 			memcpy(&extras[extra->type - 1], extra,
699 			       sizeof(*extra));
700 		}
701 
702 		skb = xennet_get_rx_skb(queue, cons);
703 		ref = xennet_get_rx_ref(queue, cons);
704 		xennet_move_rx_slot(queue, skb, ref);
705 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
706 
707 	queue->rx.rsp_cons = cons;
708 	return err;
709 }
710 
711 static int xennet_get_responses(struct netfront_queue *queue,
712 				struct netfront_rx_info *rinfo, RING_IDX rp,
713 				struct sk_buff_head *list)
714 {
715 	struct xen_netif_rx_response *rx = &rinfo->rx;
716 	struct xen_netif_extra_info *extras = rinfo->extras;
717 	struct device *dev = &queue->info->netdev->dev;
718 	RING_IDX cons = queue->rx.rsp_cons;
719 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
720 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
721 	int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
722 	int slots = 1;
723 	int err = 0;
724 	unsigned long ret;
725 
726 	if (rx->flags & XEN_NETRXF_extra_info) {
727 		err = xennet_get_extras(queue, extras, rp);
728 		cons = queue->rx.rsp_cons;
729 	}
730 
731 	for (;;) {
732 		if (unlikely(rx->status < 0 ||
733 			     rx->offset + rx->status > PAGE_SIZE)) {
734 			if (net_ratelimit())
735 				dev_warn(dev, "rx->offset: %u, size: %d\n",
736 					 rx->offset, rx->status);
737 			xennet_move_rx_slot(queue, skb, ref);
738 			err = -EINVAL;
739 			goto next;
740 		}
741 
742 		/*
743 		 * This definitely indicates a bug, either in this driver or in
744 		 * the backend driver. In future this should flag the bad
745 		 * situation to the system controller to reboot the backend.
746 		 */
747 		if (ref == GRANT_INVALID_REF) {
748 			if (net_ratelimit())
749 				dev_warn(dev, "Bad rx response id %d.\n",
750 					 rx->id);
751 			err = -EINVAL;
752 			goto next;
753 		}
754 
755 		ret = gnttab_end_foreign_access_ref(ref, 0);
756 		BUG_ON(!ret);
757 
758 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
759 
760 		__skb_queue_tail(list, skb);
761 
762 next:
763 		if (!(rx->flags & XEN_NETRXF_more_data))
764 			break;
765 
766 		if (cons + slots == rp) {
767 			if (net_ratelimit())
768 				dev_warn(dev, "Need more slots\n");
769 			err = -ENOENT;
770 			break;
771 		}
772 
773 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
774 		skb = xennet_get_rx_skb(queue, cons + slots);
775 		ref = xennet_get_rx_ref(queue, cons + slots);
776 		slots++;
777 	}
778 
779 	if (unlikely(slots > max)) {
780 		if (net_ratelimit())
781 			dev_warn(dev, "Too many slots\n");
782 		err = -E2BIG;
783 	}
784 
785 	if (unlikely(err))
786 		queue->rx.rsp_cons = cons + slots;
787 
788 	return err;
789 }
790 
791 static int xennet_set_skb_gso(struct sk_buff *skb,
792 			      struct xen_netif_extra_info *gso)
793 {
794 	if (!gso->u.gso.size) {
795 		if (net_ratelimit())
796 			pr_warn("GSO size must not be zero\n");
797 		return -EINVAL;
798 	}
799 
800 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
801 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
802 		if (net_ratelimit())
803 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
804 		return -EINVAL;
805 	}
806 
807 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
808 	skb_shinfo(skb)->gso_type =
809 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
810 		SKB_GSO_TCPV4 :
811 		SKB_GSO_TCPV6;
812 
813 	/* Header must be checked, and gso_segs computed. */
814 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
815 	skb_shinfo(skb)->gso_segs = 0;
816 
817 	return 0;
818 }
819 
820 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
821 				  struct sk_buff *skb,
822 				  struct sk_buff_head *list)
823 {
824 	struct skb_shared_info *shinfo = skb_shinfo(skb);
825 	RING_IDX cons = queue->rx.rsp_cons;
826 	struct sk_buff *nskb;
827 
828 	while ((nskb = __skb_dequeue(list))) {
829 		struct xen_netif_rx_response *rx =
830 			RING_GET_RESPONSE(&queue->rx, ++cons);
831 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
832 
833 		if (shinfo->nr_frags == MAX_SKB_FRAGS) {
834 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
835 
836 			BUG_ON(pull_to <= skb_headlen(skb));
837 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
838 		}
839 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
840 
841 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
842 				rx->offset, rx->status, PAGE_SIZE);
843 
844 		skb_shinfo(nskb)->nr_frags = 0;
845 		kfree_skb(nskb);
846 	}
847 
848 	return cons;
849 }
850 
851 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
852 {
853 	bool recalculate_partial_csum = false;
854 
855 	/*
856 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
857 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
858 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
859 	 * recalculate the partial checksum.
860 	 */
861 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
862 		struct netfront_info *np = netdev_priv(dev);
863 		atomic_inc(&np->rx_gso_checksum_fixup);
864 		skb->ip_summed = CHECKSUM_PARTIAL;
865 		recalculate_partial_csum = true;
866 	}
867 
868 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
869 	if (skb->ip_summed != CHECKSUM_PARTIAL)
870 		return 0;
871 
872 	return skb_checksum_setup(skb, recalculate_partial_csum);
873 }
874 
875 static int handle_incoming_queue(struct netfront_queue *queue,
876 				 struct sk_buff_head *rxq)
877 {
878 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
879 	int packets_dropped = 0;
880 	struct sk_buff *skb;
881 
882 	while ((skb = __skb_dequeue(rxq)) != NULL) {
883 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
884 
885 		if (pull_to > skb_headlen(skb))
886 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
887 
888 		/* Ethernet work: Delayed to here as it peeks the header. */
889 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
890 		skb_reset_network_header(skb);
891 
892 		if (checksum_setup(queue->info->netdev, skb)) {
893 			kfree_skb(skb);
894 			packets_dropped++;
895 			queue->info->netdev->stats.rx_errors++;
896 			continue;
897 		}
898 
899 		u64_stats_update_begin(&rx_stats->syncp);
900 		rx_stats->packets++;
901 		rx_stats->bytes += skb->len;
902 		u64_stats_update_end(&rx_stats->syncp);
903 
904 		/* Pass it up. */
905 		napi_gro_receive(&queue->napi, skb);
906 	}
907 
908 	return packets_dropped;
909 }
910 
911 static int xennet_poll(struct napi_struct *napi, int budget)
912 {
913 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
914 	struct net_device *dev = queue->info->netdev;
915 	struct sk_buff *skb;
916 	struct netfront_rx_info rinfo;
917 	struct xen_netif_rx_response *rx = &rinfo.rx;
918 	struct xen_netif_extra_info *extras = rinfo.extras;
919 	RING_IDX i, rp;
920 	int work_done;
921 	struct sk_buff_head rxq;
922 	struct sk_buff_head errq;
923 	struct sk_buff_head tmpq;
924 	int err;
925 
926 	spin_lock(&queue->rx_lock);
927 
928 	skb_queue_head_init(&rxq);
929 	skb_queue_head_init(&errq);
930 	skb_queue_head_init(&tmpq);
931 
932 	rp = queue->rx.sring->rsp_prod;
933 	rmb(); /* Ensure we see queued responses up to 'rp'. */
934 
935 	i = queue->rx.rsp_cons;
936 	work_done = 0;
937 	while ((i != rp) && (work_done < budget)) {
938 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
939 		memset(extras, 0, sizeof(rinfo.extras));
940 
941 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
942 
943 		if (unlikely(err)) {
944 err:
945 			while ((skb = __skb_dequeue(&tmpq)))
946 				__skb_queue_tail(&errq, skb);
947 			dev->stats.rx_errors++;
948 			i = queue->rx.rsp_cons;
949 			continue;
950 		}
951 
952 		skb = __skb_dequeue(&tmpq);
953 
954 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
955 			struct xen_netif_extra_info *gso;
956 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
957 
958 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
959 				__skb_queue_head(&tmpq, skb);
960 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
961 				goto err;
962 			}
963 		}
964 
965 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
966 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
967 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
968 
969 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
970 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
971 		skb->data_len = rx->status;
972 		skb->len += rx->status;
973 
974 		i = xennet_fill_frags(queue, skb, &tmpq);
975 
976 		if (rx->flags & XEN_NETRXF_csum_blank)
977 			skb->ip_summed = CHECKSUM_PARTIAL;
978 		else if (rx->flags & XEN_NETRXF_data_validated)
979 			skb->ip_summed = CHECKSUM_UNNECESSARY;
980 
981 		__skb_queue_tail(&rxq, skb);
982 
983 		queue->rx.rsp_cons = ++i;
984 		work_done++;
985 	}
986 
987 	__skb_queue_purge(&errq);
988 
989 	work_done -= handle_incoming_queue(queue, &rxq);
990 
991 	xennet_alloc_rx_buffers(queue);
992 
993 	if (work_done < budget) {
994 		int more_to_do = 0;
995 
996 		napi_complete(napi);
997 
998 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
999 		if (more_to_do)
1000 			napi_schedule(napi);
1001 	}
1002 
1003 	spin_unlock(&queue->rx_lock);
1004 
1005 	return work_done;
1006 }
1007 
1008 static int xennet_change_mtu(struct net_device *dev, int mtu)
1009 {
1010 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1011 
1012 	if (mtu > max)
1013 		return -EINVAL;
1014 	dev->mtu = mtu;
1015 	return 0;
1016 }
1017 
1018 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1019 						    struct rtnl_link_stats64 *tot)
1020 {
1021 	struct netfront_info *np = netdev_priv(dev);
1022 	int cpu;
1023 
1024 	for_each_possible_cpu(cpu) {
1025 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1026 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1027 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1028 		unsigned int start;
1029 
1030 		do {
1031 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1032 			tx_packets = tx_stats->packets;
1033 			tx_bytes = tx_stats->bytes;
1034 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1035 
1036 		do {
1037 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1038 			rx_packets = rx_stats->packets;
1039 			rx_bytes = rx_stats->bytes;
1040 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1041 
1042 		tot->rx_packets += rx_packets;
1043 		tot->tx_packets += tx_packets;
1044 		tot->rx_bytes   += rx_bytes;
1045 		tot->tx_bytes   += tx_bytes;
1046 	}
1047 
1048 	tot->rx_errors  = dev->stats.rx_errors;
1049 	tot->tx_dropped = dev->stats.tx_dropped;
1050 
1051 	return tot;
1052 }
1053 
1054 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1055 {
1056 	struct sk_buff *skb;
1057 	int i;
1058 
1059 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1060 		/* Skip over entries which are actually freelist references */
1061 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1062 			continue;
1063 
1064 		skb = queue->tx_skbs[i].skb;
1065 		get_page(queue->grant_tx_page[i]);
1066 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1067 					  GNTMAP_readonly,
1068 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1069 		queue->grant_tx_page[i] = NULL;
1070 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1071 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1072 		dev_kfree_skb_irq(skb);
1073 	}
1074 }
1075 
1076 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1077 {
1078 	int id, ref;
1079 
1080 	spin_lock_bh(&queue->rx_lock);
1081 
1082 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1083 		struct sk_buff *skb;
1084 		struct page *page;
1085 
1086 		skb = queue->rx_skbs[id];
1087 		if (!skb)
1088 			continue;
1089 
1090 		ref = queue->grant_rx_ref[id];
1091 		if (ref == GRANT_INVALID_REF)
1092 			continue;
1093 
1094 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1095 
1096 		/* gnttab_end_foreign_access() needs a page ref until
1097 		 * foreign access is ended (which may be deferred).
1098 		 */
1099 		get_page(page);
1100 		gnttab_end_foreign_access(ref, 0,
1101 					  (unsigned long)page_address(page));
1102 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1103 
1104 		kfree_skb(skb);
1105 	}
1106 
1107 	spin_unlock_bh(&queue->rx_lock);
1108 }
1109 
1110 static netdev_features_t xennet_fix_features(struct net_device *dev,
1111 	netdev_features_t features)
1112 {
1113 	struct netfront_info *np = netdev_priv(dev);
1114 	int val;
1115 
1116 	if (features & NETIF_F_SG) {
1117 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1118 				 "%d", &val) < 0)
1119 			val = 0;
1120 
1121 		if (!val)
1122 			features &= ~NETIF_F_SG;
1123 	}
1124 
1125 	if (features & NETIF_F_IPV6_CSUM) {
1126 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1127 				 "feature-ipv6-csum-offload", "%d", &val) < 0)
1128 			val = 0;
1129 
1130 		if (!val)
1131 			features &= ~NETIF_F_IPV6_CSUM;
1132 	}
1133 
1134 	if (features & NETIF_F_TSO) {
1135 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1136 				 "feature-gso-tcpv4", "%d", &val) < 0)
1137 			val = 0;
1138 
1139 		if (!val)
1140 			features &= ~NETIF_F_TSO;
1141 	}
1142 
1143 	if (features & NETIF_F_TSO6) {
1144 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1145 				 "feature-gso-tcpv6", "%d", &val) < 0)
1146 			val = 0;
1147 
1148 		if (!val)
1149 			features &= ~NETIF_F_TSO6;
1150 	}
1151 
1152 	return features;
1153 }
1154 
1155 static int xennet_set_features(struct net_device *dev,
1156 	netdev_features_t features)
1157 {
1158 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1159 		netdev_info(dev, "Reducing MTU because no SG offload");
1160 		dev->mtu = ETH_DATA_LEN;
1161 	}
1162 
1163 	return 0;
1164 }
1165 
1166 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1167 {
1168 	struct netfront_queue *queue = dev_id;
1169 	unsigned long flags;
1170 
1171 	spin_lock_irqsave(&queue->tx_lock, flags);
1172 	xennet_tx_buf_gc(queue);
1173 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1174 
1175 	return IRQ_HANDLED;
1176 }
1177 
1178 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1179 {
1180 	struct netfront_queue *queue = dev_id;
1181 	struct net_device *dev = queue->info->netdev;
1182 
1183 	if (likely(netif_carrier_ok(dev) &&
1184 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1185 		napi_schedule(&queue->napi);
1186 
1187 	return IRQ_HANDLED;
1188 }
1189 
1190 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1191 {
1192 	xennet_tx_interrupt(irq, dev_id);
1193 	xennet_rx_interrupt(irq, dev_id);
1194 	return IRQ_HANDLED;
1195 }
1196 
1197 #ifdef CONFIG_NET_POLL_CONTROLLER
1198 static void xennet_poll_controller(struct net_device *dev)
1199 {
1200 	/* Poll each queue */
1201 	struct netfront_info *info = netdev_priv(dev);
1202 	unsigned int num_queues = dev->real_num_tx_queues;
1203 	unsigned int i;
1204 	for (i = 0; i < num_queues; ++i)
1205 		xennet_interrupt(0, &info->queues[i]);
1206 }
1207 #endif
1208 
1209 static const struct net_device_ops xennet_netdev_ops = {
1210 	.ndo_open            = xennet_open,
1211 	.ndo_stop            = xennet_close,
1212 	.ndo_start_xmit      = xennet_start_xmit,
1213 	.ndo_change_mtu	     = xennet_change_mtu,
1214 	.ndo_get_stats64     = xennet_get_stats64,
1215 	.ndo_set_mac_address = eth_mac_addr,
1216 	.ndo_validate_addr   = eth_validate_addr,
1217 	.ndo_fix_features    = xennet_fix_features,
1218 	.ndo_set_features    = xennet_set_features,
1219 	.ndo_select_queue    = xennet_select_queue,
1220 #ifdef CONFIG_NET_POLL_CONTROLLER
1221 	.ndo_poll_controller = xennet_poll_controller,
1222 #endif
1223 };
1224 
1225 static void xennet_free_netdev(struct net_device *netdev)
1226 {
1227 	struct netfront_info *np = netdev_priv(netdev);
1228 
1229 	free_percpu(np->rx_stats);
1230 	free_percpu(np->tx_stats);
1231 	free_netdev(netdev);
1232 }
1233 
1234 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1235 {
1236 	int err;
1237 	struct net_device *netdev;
1238 	struct netfront_info *np;
1239 
1240 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1241 	if (!netdev)
1242 		return ERR_PTR(-ENOMEM);
1243 
1244 	np                   = netdev_priv(netdev);
1245 	np->xbdev            = dev;
1246 
1247 	np->queues = NULL;
1248 
1249 	err = -ENOMEM;
1250 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1251 	if (np->rx_stats == NULL)
1252 		goto exit;
1253 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1254 	if (np->tx_stats == NULL)
1255 		goto exit;
1256 
1257 	netdev->netdev_ops	= &xennet_netdev_ops;
1258 
1259 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1260 				  NETIF_F_GSO_ROBUST;
1261 	netdev->hw_features	= NETIF_F_SG |
1262 				  NETIF_F_IPV6_CSUM |
1263 				  NETIF_F_TSO | NETIF_F_TSO6;
1264 
1265 	/*
1266          * Assume that all hw features are available for now. This set
1267          * will be adjusted by the call to netdev_update_features() in
1268          * xennet_connect() which is the earliest point where we can
1269          * negotiate with the backend regarding supported features.
1270          */
1271 	netdev->features |= netdev->hw_features;
1272 
1273 	netdev->ethtool_ops = &xennet_ethtool_ops;
1274 	SET_NETDEV_DEV(netdev, &dev->dev);
1275 
1276 	np->netdev = netdev;
1277 
1278 	netif_carrier_off(netdev);
1279 
1280 	return netdev;
1281 
1282  exit:
1283 	xennet_free_netdev(netdev);
1284 	return ERR_PTR(err);
1285 }
1286 
1287 /**
1288  * Entry point to this code when a new device is created.  Allocate the basic
1289  * structures and the ring buffers for communication with the backend, and
1290  * inform the backend of the appropriate details for those.
1291  */
1292 static int netfront_probe(struct xenbus_device *dev,
1293 			  const struct xenbus_device_id *id)
1294 {
1295 	int err;
1296 	struct net_device *netdev;
1297 	struct netfront_info *info;
1298 
1299 	netdev = xennet_create_dev(dev);
1300 	if (IS_ERR(netdev)) {
1301 		err = PTR_ERR(netdev);
1302 		xenbus_dev_fatal(dev, err, "creating netdev");
1303 		return err;
1304 	}
1305 
1306 	info = netdev_priv(netdev);
1307 	dev_set_drvdata(&dev->dev, info);
1308 #ifdef CONFIG_SYSFS
1309 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1310 #endif
1311 	err = register_netdev(info->netdev);
1312 	if (err) {
1313 		pr_warn("%s: register_netdev err=%d\n", __func__, err);
1314 		goto fail;
1315 	}
1316 
1317 	return 0;
1318 
1319  fail:
1320 	xennet_free_netdev(netdev);
1321 	dev_set_drvdata(&dev->dev, NULL);
1322 	return err;
1323 }
1324 
1325 static void xennet_end_access(int ref, void *page)
1326 {
1327 	/* This frees the page as a side-effect */
1328 	if (ref != GRANT_INVALID_REF)
1329 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1330 }
1331 
1332 static void xennet_disconnect_backend(struct netfront_info *info)
1333 {
1334 	unsigned int i = 0;
1335 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1336 
1337 	netif_carrier_off(info->netdev);
1338 
1339 	for (i = 0; i < num_queues; ++i) {
1340 		struct netfront_queue *queue = &info->queues[i];
1341 
1342 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1343 			unbind_from_irqhandler(queue->tx_irq, queue);
1344 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1345 			unbind_from_irqhandler(queue->tx_irq, queue);
1346 			unbind_from_irqhandler(queue->rx_irq, queue);
1347 		}
1348 		queue->tx_evtchn = queue->rx_evtchn = 0;
1349 		queue->tx_irq = queue->rx_irq = 0;
1350 
1351 		napi_synchronize(&queue->napi);
1352 
1353 		xennet_release_tx_bufs(queue);
1354 		xennet_release_rx_bufs(queue);
1355 		gnttab_free_grant_references(queue->gref_tx_head);
1356 		gnttab_free_grant_references(queue->gref_rx_head);
1357 
1358 		/* End access and free the pages */
1359 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1360 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1361 
1362 		queue->tx_ring_ref = GRANT_INVALID_REF;
1363 		queue->rx_ring_ref = GRANT_INVALID_REF;
1364 		queue->tx.sring = NULL;
1365 		queue->rx.sring = NULL;
1366 	}
1367 }
1368 
1369 /**
1370  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1371  * driver restart.  We tear down our netif structure and recreate it, but
1372  * leave the device-layer structures intact so that this is transparent to the
1373  * rest of the kernel.
1374  */
1375 static int netfront_resume(struct xenbus_device *dev)
1376 {
1377 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1378 
1379 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1380 
1381 	xennet_disconnect_backend(info);
1382 	return 0;
1383 }
1384 
1385 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1386 {
1387 	char *s, *e, *macstr;
1388 	int i;
1389 
1390 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1391 	if (IS_ERR(macstr))
1392 		return PTR_ERR(macstr);
1393 
1394 	for (i = 0; i < ETH_ALEN; i++) {
1395 		mac[i] = simple_strtoul(s, &e, 16);
1396 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1397 			kfree(macstr);
1398 			return -ENOENT;
1399 		}
1400 		s = e+1;
1401 	}
1402 
1403 	kfree(macstr);
1404 	return 0;
1405 }
1406 
1407 static int setup_netfront_single(struct netfront_queue *queue)
1408 {
1409 	int err;
1410 
1411 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1412 	if (err < 0)
1413 		goto fail;
1414 
1415 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1416 					xennet_interrupt,
1417 					0, queue->info->netdev->name, queue);
1418 	if (err < 0)
1419 		goto bind_fail;
1420 	queue->rx_evtchn = queue->tx_evtchn;
1421 	queue->rx_irq = queue->tx_irq = err;
1422 
1423 	return 0;
1424 
1425 bind_fail:
1426 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1427 	queue->tx_evtchn = 0;
1428 fail:
1429 	return err;
1430 }
1431 
1432 static int setup_netfront_split(struct netfront_queue *queue)
1433 {
1434 	int err;
1435 
1436 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1437 	if (err < 0)
1438 		goto fail;
1439 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1440 	if (err < 0)
1441 		goto alloc_rx_evtchn_fail;
1442 
1443 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1444 		 "%s-tx", queue->name);
1445 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1446 					xennet_tx_interrupt,
1447 					0, queue->tx_irq_name, queue);
1448 	if (err < 0)
1449 		goto bind_tx_fail;
1450 	queue->tx_irq = err;
1451 
1452 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1453 		 "%s-rx", queue->name);
1454 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1455 					xennet_rx_interrupt,
1456 					0, queue->rx_irq_name, queue);
1457 	if (err < 0)
1458 		goto bind_rx_fail;
1459 	queue->rx_irq = err;
1460 
1461 	return 0;
1462 
1463 bind_rx_fail:
1464 	unbind_from_irqhandler(queue->tx_irq, queue);
1465 	queue->tx_irq = 0;
1466 bind_tx_fail:
1467 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1468 	queue->rx_evtchn = 0;
1469 alloc_rx_evtchn_fail:
1470 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1471 	queue->tx_evtchn = 0;
1472 fail:
1473 	return err;
1474 }
1475 
1476 static int setup_netfront(struct xenbus_device *dev,
1477 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1478 {
1479 	struct xen_netif_tx_sring *txs;
1480 	struct xen_netif_rx_sring *rxs;
1481 	grant_ref_t gref;
1482 	int err;
1483 
1484 	queue->tx_ring_ref = GRANT_INVALID_REF;
1485 	queue->rx_ring_ref = GRANT_INVALID_REF;
1486 	queue->rx.sring = NULL;
1487 	queue->tx.sring = NULL;
1488 
1489 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1490 	if (!txs) {
1491 		err = -ENOMEM;
1492 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1493 		goto fail;
1494 	}
1495 	SHARED_RING_INIT(txs);
1496 	FRONT_RING_INIT(&queue->tx, txs, PAGE_SIZE);
1497 
1498 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1499 	if (err < 0)
1500 		goto grant_tx_ring_fail;
1501 	queue->tx_ring_ref = gref;
1502 
1503 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1504 	if (!rxs) {
1505 		err = -ENOMEM;
1506 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1507 		goto alloc_rx_ring_fail;
1508 	}
1509 	SHARED_RING_INIT(rxs);
1510 	FRONT_RING_INIT(&queue->rx, rxs, PAGE_SIZE);
1511 
1512 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
1513 	if (err < 0)
1514 		goto grant_rx_ring_fail;
1515 	queue->rx_ring_ref = gref;
1516 
1517 	if (feature_split_evtchn)
1518 		err = setup_netfront_split(queue);
1519 	/* setup single event channel if
1520 	 *  a) feature-split-event-channels == 0
1521 	 *  b) feature-split-event-channels == 1 but failed to setup
1522 	 */
1523 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1524 		err = setup_netfront_single(queue);
1525 
1526 	if (err)
1527 		goto alloc_evtchn_fail;
1528 
1529 	return 0;
1530 
1531 	/* If we fail to setup netfront, it is safe to just revoke access to
1532 	 * granted pages because backend is not accessing it at this point.
1533 	 */
1534 alloc_evtchn_fail:
1535 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1536 grant_rx_ring_fail:
1537 	free_page((unsigned long)rxs);
1538 alloc_rx_ring_fail:
1539 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1540 grant_tx_ring_fail:
1541 	free_page((unsigned long)txs);
1542 fail:
1543 	return err;
1544 }
1545 
1546 /* Queue-specific initialisation
1547  * This used to be done in xennet_create_dev() but must now
1548  * be run per-queue.
1549  */
1550 static int xennet_init_queue(struct netfront_queue *queue)
1551 {
1552 	unsigned short i;
1553 	int err = 0;
1554 
1555 	spin_lock_init(&queue->tx_lock);
1556 	spin_lock_init(&queue->rx_lock);
1557 
1558 	setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1559 		    (unsigned long)queue);
1560 
1561 	snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1562 		 queue->info->netdev->name, queue->id);
1563 
1564 	/* Initialise tx_skbs as a free chain containing every entry. */
1565 	queue->tx_skb_freelist = 0;
1566 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1567 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1568 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1569 		queue->grant_tx_page[i] = NULL;
1570 	}
1571 
1572 	/* Clear out rx_skbs */
1573 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1574 		queue->rx_skbs[i] = NULL;
1575 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1576 	}
1577 
1578 	/* A grant for every tx ring slot */
1579 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1580 					  &queue->gref_tx_head) < 0) {
1581 		pr_alert("can't alloc tx grant refs\n");
1582 		err = -ENOMEM;
1583 		goto exit;
1584 	}
1585 
1586 	/* A grant for every rx ring slot */
1587 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1588 					  &queue->gref_rx_head) < 0) {
1589 		pr_alert("can't alloc rx grant refs\n");
1590 		err = -ENOMEM;
1591 		goto exit_free_tx;
1592 	}
1593 
1594 	return 0;
1595 
1596  exit_free_tx:
1597 	gnttab_free_grant_references(queue->gref_tx_head);
1598  exit:
1599 	return err;
1600 }
1601 
1602 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1603 			   struct xenbus_transaction *xbt, int write_hierarchical)
1604 {
1605 	/* Write the queue-specific keys into XenStore in the traditional
1606 	 * way for a single queue, or in a queue subkeys for multiple
1607 	 * queues.
1608 	 */
1609 	struct xenbus_device *dev = queue->info->xbdev;
1610 	int err;
1611 	const char *message;
1612 	char *path;
1613 	size_t pathsize;
1614 
1615 	/* Choose the correct place to write the keys */
1616 	if (write_hierarchical) {
1617 		pathsize = strlen(dev->nodename) + 10;
1618 		path = kzalloc(pathsize, GFP_KERNEL);
1619 		if (!path) {
1620 			err = -ENOMEM;
1621 			message = "out of memory while writing ring references";
1622 			goto error;
1623 		}
1624 		snprintf(path, pathsize, "%s/queue-%u",
1625 				dev->nodename, queue->id);
1626 	} else {
1627 		path = (char *)dev->nodename;
1628 	}
1629 
1630 	/* Write ring references */
1631 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1632 			queue->tx_ring_ref);
1633 	if (err) {
1634 		message = "writing tx-ring-ref";
1635 		goto error;
1636 	}
1637 
1638 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1639 			queue->rx_ring_ref);
1640 	if (err) {
1641 		message = "writing rx-ring-ref";
1642 		goto error;
1643 	}
1644 
1645 	/* Write event channels; taking into account both shared
1646 	 * and split event channel scenarios.
1647 	 */
1648 	if (queue->tx_evtchn == queue->rx_evtchn) {
1649 		/* Shared event channel */
1650 		err = xenbus_printf(*xbt, path,
1651 				"event-channel", "%u", queue->tx_evtchn);
1652 		if (err) {
1653 			message = "writing event-channel";
1654 			goto error;
1655 		}
1656 	} else {
1657 		/* Split event channels */
1658 		err = xenbus_printf(*xbt, path,
1659 				"event-channel-tx", "%u", queue->tx_evtchn);
1660 		if (err) {
1661 			message = "writing event-channel-tx";
1662 			goto error;
1663 		}
1664 
1665 		err = xenbus_printf(*xbt, path,
1666 				"event-channel-rx", "%u", queue->rx_evtchn);
1667 		if (err) {
1668 			message = "writing event-channel-rx";
1669 			goto error;
1670 		}
1671 	}
1672 
1673 	if (write_hierarchical)
1674 		kfree(path);
1675 	return 0;
1676 
1677 error:
1678 	if (write_hierarchical)
1679 		kfree(path);
1680 	xenbus_dev_fatal(dev, err, "%s", message);
1681 	return err;
1682 }
1683 
1684 static void xennet_destroy_queues(struct netfront_info *info)
1685 {
1686 	unsigned int i;
1687 
1688 	rtnl_lock();
1689 
1690 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1691 		struct netfront_queue *queue = &info->queues[i];
1692 
1693 		if (netif_running(info->netdev))
1694 			napi_disable(&queue->napi);
1695 		del_timer_sync(&queue->rx_refill_timer);
1696 		netif_napi_del(&queue->napi);
1697 	}
1698 
1699 	rtnl_unlock();
1700 
1701 	kfree(info->queues);
1702 	info->queues = NULL;
1703 }
1704 
1705 static int xennet_create_queues(struct netfront_info *info,
1706 				unsigned int num_queues)
1707 {
1708 	unsigned int i;
1709 	int ret;
1710 
1711 	info->queues = kcalloc(num_queues, sizeof(struct netfront_queue),
1712 			       GFP_KERNEL);
1713 	if (!info->queues)
1714 		return -ENOMEM;
1715 
1716 	rtnl_lock();
1717 
1718 	for (i = 0; i < num_queues; i++) {
1719 		struct netfront_queue *queue = &info->queues[i];
1720 
1721 		queue->id = i;
1722 		queue->info = info;
1723 
1724 		ret = xennet_init_queue(queue);
1725 		if (ret < 0) {
1726 			dev_warn(&info->netdev->dev,
1727 				 "only created %d queues\n", i);
1728 			num_queues = i;
1729 			break;
1730 		}
1731 
1732 		netif_napi_add(queue->info->netdev, &queue->napi,
1733 			       xennet_poll, 64);
1734 		if (netif_running(info->netdev))
1735 			napi_enable(&queue->napi);
1736 	}
1737 
1738 	netif_set_real_num_tx_queues(info->netdev, num_queues);
1739 
1740 	rtnl_unlock();
1741 
1742 	if (num_queues == 0) {
1743 		dev_err(&info->netdev->dev, "no queues\n");
1744 		return -EINVAL;
1745 	}
1746 	return 0;
1747 }
1748 
1749 /* Common code used when first setting up, and when resuming. */
1750 static int talk_to_netback(struct xenbus_device *dev,
1751 			   struct netfront_info *info)
1752 {
1753 	const char *message;
1754 	struct xenbus_transaction xbt;
1755 	int err;
1756 	unsigned int feature_split_evtchn;
1757 	unsigned int i = 0;
1758 	unsigned int max_queues = 0;
1759 	struct netfront_queue *queue = NULL;
1760 	unsigned int num_queues = 1;
1761 
1762 	info->netdev->irq = 0;
1763 
1764 	/* Check if backend supports multiple queues */
1765 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1766 			   "multi-queue-max-queues", "%u", &max_queues);
1767 	if (err < 0)
1768 		max_queues = 1;
1769 	num_queues = min(max_queues, xennet_max_queues);
1770 
1771 	/* Check feature-split-event-channels */
1772 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1773 			   "feature-split-event-channels", "%u",
1774 			   &feature_split_evtchn);
1775 	if (err < 0)
1776 		feature_split_evtchn = 0;
1777 
1778 	/* Read mac addr. */
1779 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
1780 	if (err) {
1781 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1782 		goto out;
1783 	}
1784 
1785 	if (info->queues)
1786 		xennet_destroy_queues(info);
1787 
1788 	err = xennet_create_queues(info, num_queues);
1789 	if (err < 0)
1790 		goto destroy_ring;
1791 
1792 	/* Create shared ring, alloc event channel -- for each queue */
1793 	for (i = 0; i < num_queues; ++i) {
1794 		queue = &info->queues[i];
1795 		err = setup_netfront(dev, queue, feature_split_evtchn);
1796 		if (err) {
1797 			/* setup_netfront() will tidy up the current
1798 			 * queue on error, but we need to clean up
1799 			 * those already allocated.
1800 			 */
1801 			if (i > 0) {
1802 				rtnl_lock();
1803 				netif_set_real_num_tx_queues(info->netdev, i);
1804 				rtnl_unlock();
1805 				goto destroy_ring;
1806 			} else {
1807 				goto out;
1808 			}
1809 		}
1810 	}
1811 
1812 again:
1813 	err = xenbus_transaction_start(&xbt);
1814 	if (err) {
1815 		xenbus_dev_fatal(dev, err, "starting transaction");
1816 		goto destroy_ring;
1817 	}
1818 
1819 	if (num_queues == 1) {
1820 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1821 		if (err)
1822 			goto abort_transaction_no_dev_fatal;
1823 	} else {
1824 		/* Write the number of queues */
1825 		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues",
1826 				    "%u", num_queues);
1827 		if (err) {
1828 			message = "writing multi-queue-num-queues";
1829 			goto abort_transaction_no_dev_fatal;
1830 		}
1831 
1832 		/* Write the keys for each queue */
1833 		for (i = 0; i < num_queues; ++i) {
1834 			queue = &info->queues[i];
1835 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1836 			if (err)
1837 				goto abort_transaction_no_dev_fatal;
1838 		}
1839 	}
1840 
1841 	/* The remaining keys are not queue-specific */
1842 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1843 			    1);
1844 	if (err) {
1845 		message = "writing request-rx-copy";
1846 		goto abort_transaction;
1847 	}
1848 
1849 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1850 	if (err) {
1851 		message = "writing feature-rx-notify";
1852 		goto abort_transaction;
1853 	}
1854 
1855 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1856 	if (err) {
1857 		message = "writing feature-sg";
1858 		goto abort_transaction;
1859 	}
1860 
1861 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1862 	if (err) {
1863 		message = "writing feature-gso-tcpv4";
1864 		goto abort_transaction;
1865 	}
1866 
1867 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1868 	if (err) {
1869 		message = "writing feature-gso-tcpv6";
1870 		goto abort_transaction;
1871 	}
1872 
1873 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1874 			   "1");
1875 	if (err) {
1876 		message = "writing feature-ipv6-csum-offload";
1877 		goto abort_transaction;
1878 	}
1879 
1880 	err = xenbus_transaction_end(xbt, 0);
1881 	if (err) {
1882 		if (err == -EAGAIN)
1883 			goto again;
1884 		xenbus_dev_fatal(dev, err, "completing transaction");
1885 		goto destroy_ring;
1886 	}
1887 
1888 	return 0;
1889 
1890  abort_transaction:
1891 	xenbus_dev_fatal(dev, err, "%s", message);
1892 abort_transaction_no_dev_fatal:
1893 	xenbus_transaction_end(xbt, 1);
1894  destroy_ring:
1895 	xennet_disconnect_backend(info);
1896 	kfree(info->queues);
1897 	info->queues = NULL;
1898  out:
1899 	return err;
1900 }
1901 
1902 static int xennet_connect(struct net_device *dev)
1903 {
1904 	struct netfront_info *np = netdev_priv(dev);
1905 	unsigned int num_queues = 0;
1906 	int err;
1907 	unsigned int feature_rx_copy;
1908 	unsigned int j = 0;
1909 	struct netfront_queue *queue = NULL;
1910 
1911 	err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1912 			   "feature-rx-copy", "%u", &feature_rx_copy);
1913 	if (err != 1)
1914 		feature_rx_copy = 0;
1915 
1916 	if (!feature_rx_copy) {
1917 		dev_info(&dev->dev,
1918 			 "backend does not support copying receive path\n");
1919 		return -ENODEV;
1920 	}
1921 
1922 	err = talk_to_netback(np->xbdev, np);
1923 	if (err)
1924 		return err;
1925 
1926 	/* talk_to_netback() sets the correct number of queues */
1927 	num_queues = dev->real_num_tx_queues;
1928 
1929 	rtnl_lock();
1930 	netdev_update_features(dev);
1931 	rtnl_unlock();
1932 
1933 	/*
1934 	 * All public and private state should now be sane.  Get
1935 	 * ready to start sending and receiving packets and give the driver
1936 	 * domain a kick because we've probably just requeued some
1937 	 * packets.
1938 	 */
1939 	netif_carrier_on(np->netdev);
1940 	for (j = 0; j < num_queues; ++j) {
1941 		queue = &np->queues[j];
1942 
1943 		notify_remote_via_irq(queue->tx_irq);
1944 		if (queue->tx_irq != queue->rx_irq)
1945 			notify_remote_via_irq(queue->rx_irq);
1946 
1947 		spin_lock_irq(&queue->tx_lock);
1948 		xennet_tx_buf_gc(queue);
1949 		spin_unlock_irq(&queue->tx_lock);
1950 
1951 		spin_lock_bh(&queue->rx_lock);
1952 		xennet_alloc_rx_buffers(queue);
1953 		spin_unlock_bh(&queue->rx_lock);
1954 	}
1955 
1956 	return 0;
1957 }
1958 
1959 /**
1960  * Callback received when the backend's state changes.
1961  */
1962 static void netback_changed(struct xenbus_device *dev,
1963 			    enum xenbus_state backend_state)
1964 {
1965 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
1966 	struct net_device *netdev = np->netdev;
1967 
1968 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1969 
1970 	switch (backend_state) {
1971 	case XenbusStateInitialising:
1972 	case XenbusStateInitialised:
1973 	case XenbusStateReconfiguring:
1974 	case XenbusStateReconfigured:
1975 	case XenbusStateUnknown:
1976 		break;
1977 
1978 	case XenbusStateInitWait:
1979 		if (dev->state != XenbusStateInitialising)
1980 			break;
1981 		if (xennet_connect(netdev) != 0)
1982 			break;
1983 		xenbus_switch_state(dev, XenbusStateConnected);
1984 		break;
1985 
1986 	case XenbusStateConnected:
1987 		netdev_notify_peers(netdev);
1988 		break;
1989 
1990 	case XenbusStateClosed:
1991 		if (dev->state == XenbusStateClosed)
1992 			break;
1993 		/* Missed the backend's CLOSING state -- fallthrough */
1994 	case XenbusStateClosing:
1995 		xenbus_frontend_closed(dev);
1996 		break;
1997 	}
1998 }
1999 
2000 static const struct xennet_stat {
2001 	char name[ETH_GSTRING_LEN];
2002 	u16 offset;
2003 } xennet_stats[] = {
2004 	{
2005 		"rx_gso_checksum_fixup",
2006 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2007 	},
2008 };
2009 
2010 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2011 {
2012 	switch (string_set) {
2013 	case ETH_SS_STATS:
2014 		return ARRAY_SIZE(xennet_stats);
2015 	default:
2016 		return -EINVAL;
2017 	}
2018 }
2019 
2020 static void xennet_get_ethtool_stats(struct net_device *dev,
2021 				     struct ethtool_stats *stats, u64 * data)
2022 {
2023 	void *np = netdev_priv(dev);
2024 	int i;
2025 
2026 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2027 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2028 }
2029 
2030 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2031 {
2032 	int i;
2033 
2034 	switch (stringset) {
2035 	case ETH_SS_STATS:
2036 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2037 			memcpy(data + i * ETH_GSTRING_LEN,
2038 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2039 		break;
2040 	}
2041 }
2042 
2043 static const struct ethtool_ops xennet_ethtool_ops =
2044 {
2045 	.get_link = ethtool_op_get_link,
2046 
2047 	.get_sset_count = xennet_get_sset_count,
2048 	.get_ethtool_stats = xennet_get_ethtool_stats,
2049 	.get_strings = xennet_get_strings,
2050 };
2051 
2052 #ifdef CONFIG_SYSFS
2053 static ssize_t show_rxbuf(struct device *dev,
2054 			  struct device_attribute *attr, char *buf)
2055 {
2056 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2057 }
2058 
2059 static ssize_t store_rxbuf(struct device *dev,
2060 			   struct device_attribute *attr,
2061 			   const char *buf, size_t len)
2062 {
2063 	char *endp;
2064 	unsigned long target;
2065 
2066 	if (!capable(CAP_NET_ADMIN))
2067 		return -EPERM;
2068 
2069 	target = simple_strtoul(buf, &endp, 0);
2070 	if (endp == buf)
2071 		return -EBADMSG;
2072 
2073 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2074 
2075 	return len;
2076 }
2077 
2078 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2079 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2080 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2081 
2082 static struct attribute *xennet_dev_attrs[] = {
2083 	&dev_attr_rxbuf_min.attr,
2084 	&dev_attr_rxbuf_max.attr,
2085 	&dev_attr_rxbuf_cur.attr,
2086 	NULL
2087 };
2088 
2089 static const struct attribute_group xennet_dev_group = {
2090 	.attrs = xennet_dev_attrs
2091 };
2092 #endif /* CONFIG_SYSFS */
2093 
2094 static int xennet_remove(struct xenbus_device *dev)
2095 {
2096 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2097 
2098 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2099 
2100 	xennet_disconnect_backend(info);
2101 
2102 	unregister_netdev(info->netdev);
2103 
2104 	xennet_destroy_queues(info);
2105 	xennet_free_netdev(info->netdev);
2106 
2107 	return 0;
2108 }
2109 
2110 static const struct xenbus_device_id netfront_ids[] = {
2111 	{ "vif" },
2112 	{ "" }
2113 };
2114 
2115 static struct xenbus_driver netfront_driver = {
2116 	.ids = netfront_ids,
2117 	.probe = netfront_probe,
2118 	.remove = xennet_remove,
2119 	.resume = netfront_resume,
2120 	.otherend_changed = netback_changed,
2121 };
2122 
2123 static int __init netif_init(void)
2124 {
2125 	if (!xen_domain())
2126 		return -ENODEV;
2127 
2128 	if (!xen_has_pv_nic_devices())
2129 		return -ENODEV;
2130 
2131 	pr_info("Initialising Xen virtual ethernet driver\n");
2132 
2133 	/* Allow as many queues as there are CPUs, by default */
2134 	xennet_max_queues = num_online_cpus();
2135 
2136 	return xenbus_register_frontend(&netfront_driver);
2137 }
2138 module_init(netif_init);
2139 
2140 
2141 static void __exit netif_exit(void)
2142 {
2143 	xenbus_unregister_driver(&netfront_driver);
2144 }
2145 module_exit(netif_exit);
2146 
2147 MODULE_DESCRIPTION("Xen virtual network device frontend");
2148 MODULE_LICENSE("GPL");
2149 MODULE_ALIAS("xen:vif");
2150 MODULE_ALIAS("xennet");
2151