xref: /openbmc/linux/drivers/net/xen-netfront.c (revision e8f6f3b4)
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 <asm/xen/page.h>
49 #include <xen/xen.h>
50 #include <xen/xenbus.h>
51 #include <xen/events.h>
52 #include <xen/page.h>
53 #include <xen/platform_pci.h>
54 #include <xen/grant_table.h>
55 
56 #include <xen/interface/io/netif.h>
57 #include <xen/interface/memory.h>
58 #include <xen/interface/grant_table.h>
59 
60 /* Module parameters */
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, PAGE_SIZE)
79 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, 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 struct netfront_stats {
91 	u64			packets;
92 	u64			bytes;
93 	struct u64_stats_sync	syncp;
94 };
95 
96 struct netfront_info;
97 
98 struct netfront_queue {
99 	unsigned int id; /* Queue ID, 0-based */
100 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
101 	struct netfront_info *info;
102 
103 	struct napi_struct napi;
104 
105 	/* Split event channels support, tx_* == rx_* when using
106 	 * single event channel.
107 	 */
108 	unsigned int tx_evtchn, rx_evtchn;
109 	unsigned int tx_irq, rx_irq;
110 	/* Only used when split event channels support is enabled */
111 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
112 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
113 
114 	spinlock_t   tx_lock;
115 	struct xen_netif_tx_front_ring tx;
116 	int tx_ring_ref;
117 
118 	/*
119 	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
120 	 * are linked from tx_skb_freelist through skb_entry.link.
121 	 *
122 	 *  NB. Freelist index entries are always going to be less than
123 	 *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
124 	 *  greater than PAGE_OFFSET: we use this property to distinguish
125 	 *  them.
126 	 */
127 	union skb_entry {
128 		struct sk_buff *skb;
129 		unsigned long link;
130 	} tx_skbs[NET_TX_RING_SIZE];
131 	grant_ref_t gref_tx_head;
132 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
133 	struct page *grant_tx_page[NET_TX_RING_SIZE];
134 	unsigned tx_skb_freelist;
135 
136 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
137 	struct xen_netif_rx_front_ring rx;
138 	int rx_ring_ref;
139 
140 	struct timer_list rx_refill_timer;
141 
142 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
143 	grant_ref_t gref_rx_head;
144 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
145 
146 	unsigned long rx_pfn_array[NET_RX_RING_SIZE];
147 	struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
148 	struct mmu_update rx_mmu[NET_RX_RING_SIZE];
149 };
150 
151 struct netfront_info {
152 	struct list_head list;
153 	struct net_device *netdev;
154 
155 	struct xenbus_device *xbdev;
156 
157 	/* Multi-queue support */
158 	struct netfront_queue *queues;
159 
160 	/* Statistics */
161 	struct netfront_stats __percpu *rx_stats;
162 	struct netfront_stats __percpu *tx_stats;
163 
164 	atomic_t rx_gso_checksum_fixup;
165 };
166 
167 struct netfront_rx_info {
168 	struct xen_netif_rx_response rx;
169 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
170 };
171 
172 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
173 {
174 	list->link = id;
175 }
176 
177 static int skb_entry_is_link(const union skb_entry *list)
178 {
179 	BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
180 	return (unsigned long)list->skb < PAGE_OFFSET;
181 }
182 
183 /*
184  * Access macros for acquiring freeing slots in tx_skbs[].
185  */
186 
187 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
188 			       unsigned short id)
189 {
190 	skb_entry_set_link(&list[id], *head);
191 	*head = id;
192 }
193 
194 static unsigned short get_id_from_freelist(unsigned *head,
195 					   union skb_entry *list)
196 {
197 	unsigned int id = *head;
198 	*head = list[id].link;
199 	return id;
200 }
201 
202 static int xennet_rxidx(RING_IDX idx)
203 {
204 	return idx & (NET_RX_RING_SIZE - 1);
205 }
206 
207 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
208 					 RING_IDX ri)
209 {
210 	int i = xennet_rxidx(ri);
211 	struct sk_buff *skb = queue->rx_skbs[i];
212 	queue->rx_skbs[i] = NULL;
213 	return skb;
214 }
215 
216 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
217 					    RING_IDX ri)
218 {
219 	int i = xennet_rxidx(ri);
220 	grant_ref_t ref = queue->grant_rx_ref[i];
221 	queue->grant_rx_ref[i] = GRANT_INVALID_REF;
222 	return ref;
223 }
224 
225 #ifdef CONFIG_SYSFS
226 static int xennet_sysfs_addif(struct net_device *netdev);
227 static void xennet_sysfs_delif(struct net_device *netdev);
228 #else /* !CONFIG_SYSFS */
229 #define xennet_sysfs_addif(dev) (0)
230 #define xennet_sysfs_delif(dev) do { } while (0)
231 #endif
232 
233 static bool xennet_can_sg(struct net_device *dev)
234 {
235 	return dev->features & NETIF_F_SG;
236 }
237 
238 
239 static void rx_refill_timeout(unsigned long data)
240 {
241 	struct netfront_queue *queue = (struct netfront_queue *)data;
242 	napi_schedule(&queue->napi);
243 }
244 
245 static int netfront_tx_slot_available(struct netfront_queue *queue)
246 {
247 	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
248 		(NET_TX_RING_SIZE - MAX_SKB_FRAGS - 2);
249 }
250 
251 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
252 {
253 	struct net_device *dev = queue->info->netdev;
254 	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
255 
256 	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
257 	    netfront_tx_slot_available(queue) &&
258 	    likely(netif_running(dev)))
259 		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
260 }
261 
262 
263 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
264 {
265 	struct sk_buff *skb;
266 	struct page *page;
267 
268 	skb = __netdev_alloc_skb(queue->info->netdev,
269 				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
270 				 GFP_ATOMIC | __GFP_NOWARN);
271 	if (unlikely(!skb))
272 		return NULL;
273 
274 	page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
275 	if (!page) {
276 		kfree_skb(skb);
277 		return NULL;
278 	}
279 	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
280 
281 	/* Align ip header to a 16 bytes boundary */
282 	skb_reserve(skb, NET_IP_ALIGN);
283 	skb->dev = queue->info->netdev;
284 
285 	return skb;
286 }
287 
288 
289 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
290 {
291 	RING_IDX req_prod = queue->rx.req_prod_pvt;
292 	int notify;
293 
294 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
295 		return;
296 
297 	for (req_prod = queue->rx.req_prod_pvt;
298 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
299 	     req_prod++) {
300 		struct sk_buff *skb;
301 		unsigned short id;
302 		grant_ref_t ref;
303 		unsigned long pfn;
304 		struct xen_netif_rx_request *req;
305 
306 		skb = xennet_alloc_one_rx_buffer(queue);
307 		if (!skb)
308 			break;
309 
310 		id = xennet_rxidx(req_prod);
311 
312 		BUG_ON(queue->rx_skbs[id]);
313 		queue->rx_skbs[id] = skb;
314 
315 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
316 		BUG_ON((signed short)ref < 0);
317 		queue->grant_rx_ref[id] = ref;
318 
319 		pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
320 
321 		req = RING_GET_REQUEST(&queue->rx, req_prod);
322 		gnttab_grant_foreign_access_ref(ref,
323 						queue->info->xbdev->otherend_id,
324 						pfn_to_mfn(pfn),
325 						0);
326 
327 		req->id = id;
328 		req->gref = ref;
329 	}
330 
331 	queue->rx.req_prod_pvt = req_prod;
332 
333 	/* Not enough requests? Try again later. */
334 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN) {
335 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
336 		return;
337 	}
338 
339 	wmb();		/* barrier so backend seens requests */
340 
341 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
342 	if (notify)
343 		notify_remote_via_irq(queue->rx_irq);
344 }
345 
346 static int xennet_open(struct net_device *dev)
347 {
348 	struct netfront_info *np = netdev_priv(dev);
349 	unsigned int num_queues = dev->real_num_tx_queues;
350 	unsigned int i = 0;
351 	struct netfront_queue *queue = NULL;
352 
353 	for (i = 0; i < num_queues; ++i) {
354 		queue = &np->queues[i];
355 		napi_enable(&queue->napi);
356 
357 		spin_lock_bh(&queue->rx_lock);
358 		if (netif_carrier_ok(dev)) {
359 			xennet_alloc_rx_buffers(queue);
360 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
361 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
362 				napi_schedule(&queue->napi);
363 		}
364 		spin_unlock_bh(&queue->rx_lock);
365 	}
366 
367 	netif_tx_start_all_queues(dev);
368 
369 	return 0;
370 }
371 
372 static void xennet_tx_buf_gc(struct netfront_queue *queue)
373 {
374 	RING_IDX cons, prod;
375 	unsigned short id;
376 	struct sk_buff *skb;
377 
378 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
379 
380 	do {
381 		prod = queue->tx.sring->rsp_prod;
382 		rmb(); /* Ensure we see responses up to 'rp'. */
383 
384 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
385 			struct xen_netif_tx_response *txrsp;
386 
387 			txrsp = RING_GET_RESPONSE(&queue->tx, cons);
388 			if (txrsp->status == XEN_NETIF_RSP_NULL)
389 				continue;
390 
391 			id  = txrsp->id;
392 			skb = queue->tx_skbs[id].skb;
393 			if (unlikely(gnttab_query_foreign_access(
394 				queue->grant_tx_ref[id]) != 0)) {
395 				pr_alert("%s: warning -- grant still in use by backend domain\n",
396 					 __func__);
397 				BUG();
398 			}
399 			gnttab_end_foreign_access_ref(
400 				queue->grant_tx_ref[id], GNTMAP_readonly);
401 			gnttab_release_grant_reference(
402 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
403 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
404 			queue->grant_tx_page[id] = NULL;
405 			add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
406 			dev_kfree_skb_irq(skb);
407 		}
408 
409 		queue->tx.rsp_cons = prod;
410 
411 		/*
412 		 * Set a new event, then check for race with update of tx_cons.
413 		 * Note that it is essential to schedule a callback, no matter
414 		 * how few buffers are pending. Even if there is space in the
415 		 * transmit ring, higher layers may be blocked because too much
416 		 * data is outstanding: in such cases notification from Xen is
417 		 * likely to be the only kick that we'll get.
418 		 */
419 		queue->tx.sring->rsp_event =
420 			prod + ((queue->tx.sring->req_prod - prod) >> 1) + 1;
421 		mb();		/* update shared area */
422 	} while ((cons == prod) && (prod != queue->tx.sring->rsp_prod));
423 
424 	xennet_maybe_wake_tx(queue);
425 }
426 
427 static void xennet_make_frags(struct sk_buff *skb, struct netfront_queue *queue,
428 			      struct xen_netif_tx_request *tx)
429 {
430 	char *data = skb->data;
431 	unsigned long mfn;
432 	RING_IDX prod = queue->tx.req_prod_pvt;
433 	int frags = skb_shinfo(skb)->nr_frags;
434 	unsigned int offset = offset_in_page(data);
435 	unsigned int len = skb_headlen(skb);
436 	unsigned int id;
437 	grant_ref_t ref;
438 	int i;
439 
440 	/* While the header overlaps a page boundary (including being
441 	   larger than a page), split it it into page-sized chunks. */
442 	while (len > PAGE_SIZE - offset) {
443 		tx->size = PAGE_SIZE - offset;
444 		tx->flags |= XEN_NETTXF_more_data;
445 		len -= tx->size;
446 		data += tx->size;
447 		offset = 0;
448 
449 		id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
450 		queue->tx_skbs[id].skb = skb_get(skb);
451 		tx = RING_GET_REQUEST(&queue->tx, prod++);
452 		tx->id = id;
453 		ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
454 		BUG_ON((signed short)ref < 0);
455 
456 		mfn = virt_to_mfn(data);
457 		gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
458 						mfn, GNTMAP_readonly);
459 
460 		queue->grant_tx_page[id] = virt_to_page(data);
461 		tx->gref = queue->grant_tx_ref[id] = ref;
462 		tx->offset = offset;
463 		tx->size = len;
464 		tx->flags = 0;
465 	}
466 
467 	/* Grant backend access to each skb fragment page. */
468 	for (i = 0; i < frags; i++) {
469 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
470 		struct page *page = skb_frag_page(frag);
471 
472 		len = skb_frag_size(frag);
473 		offset = frag->page_offset;
474 
475 		/* Skip unused frames from start of page */
476 		page += offset >> PAGE_SHIFT;
477 		offset &= ~PAGE_MASK;
478 
479 		while (len > 0) {
480 			unsigned long bytes;
481 
482 			bytes = PAGE_SIZE - offset;
483 			if (bytes > len)
484 				bytes = len;
485 
486 			tx->flags |= XEN_NETTXF_more_data;
487 
488 			id = get_id_from_freelist(&queue->tx_skb_freelist,
489 						  queue->tx_skbs);
490 			queue->tx_skbs[id].skb = skb_get(skb);
491 			tx = RING_GET_REQUEST(&queue->tx, prod++);
492 			tx->id = id;
493 			ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
494 			BUG_ON((signed short)ref < 0);
495 
496 			mfn = pfn_to_mfn(page_to_pfn(page));
497 			gnttab_grant_foreign_access_ref(ref,
498 							queue->info->xbdev->otherend_id,
499 							mfn, GNTMAP_readonly);
500 
501 			queue->grant_tx_page[id] = page;
502 			tx->gref = queue->grant_tx_ref[id] = ref;
503 			tx->offset = offset;
504 			tx->size = bytes;
505 			tx->flags = 0;
506 
507 			offset += bytes;
508 			len -= bytes;
509 
510 			/* Next frame */
511 			if (offset == PAGE_SIZE && len) {
512 				BUG_ON(!PageCompound(page));
513 				page++;
514 				offset = 0;
515 			}
516 		}
517 	}
518 
519 	queue->tx.req_prod_pvt = prod;
520 }
521 
522 /*
523  * Count how many ring slots are required to send the frags of this
524  * skb. Each frag might be a compound page.
525  */
526 static int xennet_count_skb_frag_slots(struct sk_buff *skb)
527 {
528 	int i, frags = skb_shinfo(skb)->nr_frags;
529 	int pages = 0;
530 
531 	for (i = 0; i < frags; i++) {
532 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
533 		unsigned long size = skb_frag_size(frag);
534 		unsigned long offset = frag->page_offset;
535 
536 		/* Skip unused frames from start of page */
537 		offset &= ~PAGE_MASK;
538 
539 		pages += PFN_UP(offset + size);
540 	}
541 
542 	return pages;
543 }
544 
545 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
546 			       void *accel_priv, select_queue_fallback_t fallback)
547 {
548 	unsigned int num_queues = dev->real_num_tx_queues;
549 	u32 hash;
550 	u16 queue_idx;
551 
552 	/* First, check if there is only one queue */
553 	if (num_queues == 1) {
554 		queue_idx = 0;
555 	} else {
556 		hash = skb_get_hash(skb);
557 		queue_idx = hash % num_queues;
558 	}
559 
560 	return queue_idx;
561 }
562 
563 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
564 {
565 	unsigned short id;
566 	struct netfront_info *np = netdev_priv(dev);
567 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
568 	struct xen_netif_tx_request *tx;
569 	char *data = skb->data;
570 	RING_IDX i;
571 	grant_ref_t ref;
572 	unsigned long mfn;
573 	int notify;
574 	int slots;
575 	unsigned int offset = offset_in_page(data);
576 	unsigned int len = skb_headlen(skb);
577 	unsigned long flags;
578 	struct netfront_queue *queue = NULL;
579 	unsigned int num_queues = dev->real_num_tx_queues;
580 	u16 queue_index;
581 
582 	/* Drop the packet if no queues are set up */
583 	if (num_queues < 1)
584 		goto drop;
585 	/* Determine which queue to transmit this SKB on */
586 	queue_index = skb_get_queue_mapping(skb);
587 	queue = &np->queues[queue_index];
588 
589 	/* If skb->len is too big for wire format, drop skb and alert
590 	 * user about misconfiguration.
591 	 */
592 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
593 		net_alert_ratelimited(
594 			"xennet: skb->len = %u, too big for wire format\n",
595 			skb->len);
596 		goto drop;
597 	}
598 
599 	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) +
600 		xennet_count_skb_frag_slots(skb);
601 	if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
602 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
603 				    slots, skb->len);
604 		if (skb_linearize(skb))
605 			goto drop;
606 		data = skb->data;
607 		offset = offset_in_page(data);
608 		len = skb_headlen(skb);
609 	}
610 
611 	spin_lock_irqsave(&queue->tx_lock, flags);
612 
613 	if (unlikely(!netif_carrier_ok(dev) ||
614 		     (slots > 1 && !xennet_can_sg(dev)) ||
615 		     netif_needs_gso(dev, skb, netif_skb_features(skb)))) {
616 		spin_unlock_irqrestore(&queue->tx_lock, flags);
617 		goto drop;
618 	}
619 
620 	i = queue->tx.req_prod_pvt;
621 
622 	id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
623 	queue->tx_skbs[id].skb = skb;
624 
625 	tx = RING_GET_REQUEST(&queue->tx, i);
626 
627 	tx->id   = id;
628 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
629 	BUG_ON((signed short)ref < 0);
630 	mfn = virt_to_mfn(data);
631 	gnttab_grant_foreign_access_ref(
632 		ref, queue->info->xbdev->otherend_id, mfn, GNTMAP_readonly);
633 	queue->grant_tx_page[id] = virt_to_page(data);
634 	tx->gref = queue->grant_tx_ref[id] = ref;
635 	tx->offset = offset;
636 	tx->size = len;
637 
638 	tx->flags = 0;
639 	if (skb->ip_summed == CHECKSUM_PARTIAL)
640 		/* local packet? */
641 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
642 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
643 		/* remote but checksummed. */
644 		tx->flags |= XEN_NETTXF_data_validated;
645 
646 	if (skb_shinfo(skb)->gso_size) {
647 		struct xen_netif_extra_info *gso;
648 
649 		gso = (struct xen_netif_extra_info *)
650 			RING_GET_REQUEST(&queue->tx, ++i);
651 
652 		tx->flags |= XEN_NETTXF_extra_info;
653 
654 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
655 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
656 			XEN_NETIF_GSO_TYPE_TCPV6 :
657 			XEN_NETIF_GSO_TYPE_TCPV4;
658 		gso->u.gso.pad = 0;
659 		gso->u.gso.features = 0;
660 
661 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
662 		gso->flags = 0;
663 	}
664 
665 	queue->tx.req_prod_pvt = i + 1;
666 
667 	xennet_make_frags(skb, queue, tx);
668 	tx->size = skb->len;
669 
670 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
671 	if (notify)
672 		notify_remote_via_irq(queue->tx_irq);
673 
674 	u64_stats_update_begin(&tx_stats->syncp);
675 	tx_stats->bytes += skb->len;
676 	tx_stats->packets++;
677 	u64_stats_update_end(&tx_stats->syncp);
678 
679 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
680 	xennet_tx_buf_gc(queue);
681 
682 	if (!netfront_tx_slot_available(queue))
683 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
684 
685 	spin_unlock_irqrestore(&queue->tx_lock, flags);
686 
687 	return NETDEV_TX_OK;
688 
689  drop:
690 	dev->stats.tx_dropped++;
691 	dev_kfree_skb_any(skb);
692 	return NETDEV_TX_OK;
693 }
694 
695 static int xennet_close(struct net_device *dev)
696 {
697 	struct netfront_info *np = netdev_priv(dev);
698 	unsigned int num_queues = dev->real_num_tx_queues;
699 	unsigned int i;
700 	struct netfront_queue *queue;
701 	netif_tx_stop_all_queues(np->netdev);
702 	for (i = 0; i < num_queues; ++i) {
703 		queue = &np->queues[i];
704 		napi_disable(&queue->napi);
705 	}
706 	return 0;
707 }
708 
709 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
710 				grant_ref_t ref)
711 {
712 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
713 
714 	BUG_ON(queue->rx_skbs[new]);
715 	queue->rx_skbs[new] = skb;
716 	queue->grant_rx_ref[new] = ref;
717 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
718 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
719 	queue->rx.req_prod_pvt++;
720 }
721 
722 static int xennet_get_extras(struct netfront_queue *queue,
723 			     struct xen_netif_extra_info *extras,
724 			     RING_IDX rp)
725 
726 {
727 	struct xen_netif_extra_info *extra;
728 	struct device *dev = &queue->info->netdev->dev;
729 	RING_IDX cons = queue->rx.rsp_cons;
730 	int err = 0;
731 
732 	do {
733 		struct sk_buff *skb;
734 		grant_ref_t ref;
735 
736 		if (unlikely(cons + 1 == rp)) {
737 			if (net_ratelimit())
738 				dev_warn(dev, "Missing extra info\n");
739 			err = -EBADR;
740 			break;
741 		}
742 
743 		extra = (struct xen_netif_extra_info *)
744 			RING_GET_RESPONSE(&queue->rx, ++cons);
745 
746 		if (unlikely(!extra->type ||
747 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
748 			if (net_ratelimit())
749 				dev_warn(dev, "Invalid extra type: %d\n",
750 					extra->type);
751 			err = -EINVAL;
752 		} else {
753 			memcpy(&extras[extra->type - 1], extra,
754 			       sizeof(*extra));
755 		}
756 
757 		skb = xennet_get_rx_skb(queue, cons);
758 		ref = xennet_get_rx_ref(queue, cons);
759 		xennet_move_rx_slot(queue, skb, ref);
760 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
761 
762 	queue->rx.rsp_cons = cons;
763 	return err;
764 }
765 
766 static int xennet_get_responses(struct netfront_queue *queue,
767 				struct netfront_rx_info *rinfo, RING_IDX rp,
768 				struct sk_buff_head *list)
769 {
770 	struct xen_netif_rx_response *rx = &rinfo->rx;
771 	struct xen_netif_extra_info *extras = rinfo->extras;
772 	struct device *dev = &queue->info->netdev->dev;
773 	RING_IDX cons = queue->rx.rsp_cons;
774 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
775 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
776 	int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
777 	int slots = 1;
778 	int err = 0;
779 	unsigned long ret;
780 
781 	if (rx->flags & XEN_NETRXF_extra_info) {
782 		err = xennet_get_extras(queue, extras, rp);
783 		cons = queue->rx.rsp_cons;
784 	}
785 
786 	for (;;) {
787 		if (unlikely(rx->status < 0 ||
788 			     rx->offset + rx->status > PAGE_SIZE)) {
789 			if (net_ratelimit())
790 				dev_warn(dev, "rx->offset: %x, size: %u\n",
791 					 rx->offset, rx->status);
792 			xennet_move_rx_slot(queue, skb, ref);
793 			err = -EINVAL;
794 			goto next;
795 		}
796 
797 		/*
798 		 * This definitely indicates a bug, either in this driver or in
799 		 * the backend driver. In future this should flag the bad
800 		 * situation to the system controller to reboot the backend.
801 		 */
802 		if (ref == GRANT_INVALID_REF) {
803 			if (net_ratelimit())
804 				dev_warn(dev, "Bad rx response id %d.\n",
805 					 rx->id);
806 			err = -EINVAL;
807 			goto next;
808 		}
809 
810 		ret = gnttab_end_foreign_access_ref(ref, 0);
811 		BUG_ON(!ret);
812 
813 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
814 
815 		__skb_queue_tail(list, skb);
816 
817 next:
818 		if (!(rx->flags & XEN_NETRXF_more_data))
819 			break;
820 
821 		if (cons + slots == rp) {
822 			if (net_ratelimit())
823 				dev_warn(dev, "Need more slots\n");
824 			err = -ENOENT;
825 			break;
826 		}
827 
828 		rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
829 		skb = xennet_get_rx_skb(queue, cons + slots);
830 		ref = xennet_get_rx_ref(queue, cons + slots);
831 		slots++;
832 	}
833 
834 	if (unlikely(slots > max)) {
835 		if (net_ratelimit())
836 			dev_warn(dev, "Too many slots\n");
837 		err = -E2BIG;
838 	}
839 
840 	if (unlikely(err))
841 		queue->rx.rsp_cons = cons + slots;
842 
843 	return err;
844 }
845 
846 static int xennet_set_skb_gso(struct sk_buff *skb,
847 			      struct xen_netif_extra_info *gso)
848 {
849 	if (!gso->u.gso.size) {
850 		if (net_ratelimit())
851 			pr_warn("GSO size must not be zero\n");
852 		return -EINVAL;
853 	}
854 
855 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
856 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
857 		if (net_ratelimit())
858 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
859 		return -EINVAL;
860 	}
861 
862 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
863 	skb_shinfo(skb)->gso_type =
864 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
865 		SKB_GSO_TCPV4 :
866 		SKB_GSO_TCPV6;
867 
868 	/* Header must be checked, and gso_segs computed. */
869 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
870 	skb_shinfo(skb)->gso_segs = 0;
871 
872 	return 0;
873 }
874 
875 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
876 				  struct sk_buff *skb,
877 				  struct sk_buff_head *list)
878 {
879 	struct skb_shared_info *shinfo = skb_shinfo(skb);
880 	RING_IDX cons = queue->rx.rsp_cons;
881 	struct sk_buff *nskb;
882 
883 	while ((nskb = __skb_dequeue(list))) {
884 		struct xen_netif_rx_response *rx =
885 			RING_GET_RESPONSE(&queue->rx, ++cons);
886 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
887 
888 		if (shinfo->nr_frags == MAX_SKB_FRAGS) {
889 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
890 
891 			BUG_ON(pull_to <= skb_headlen(skb));
892 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
893 		}
894 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
895 
896 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
897 				rx->offset, rx->status, PAGE_SIZE);
898 
899 		skb_shinfo(nskb)->nr_frags = 0;
900 		kfree_skb(nskb);
901 	}
902 
903 	return cons;
904 }
905 
906 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
907 {
908 	bool recalculate_partial_csum = false;
909 
910 	/*
911 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
912 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
913 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
914 	 * recalculate the partial checksum.
915 	 */
916 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
917 		struct netfront_info *np = netdev_priv(dev);
918 		atomic_inc(&np->rx_gso_checksum_fixup);
919 		skb->ip_summed = CHECKSUM_PARTIAL;
920 		recalculate_partial_csum = true;
921 	}
922 
923 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
924 	if (skb->ip_summed != CHECKSUM_PARTIAL)
925 		return 0;
926 
927 	return skb_checksum_setup(skb, recalculate_partial_csum);
928 }
929 
930 static int handle_incoming_queue(struct netfront_queue *queue,
931 				 struct sk_buff_head *rxq)
932 {
933 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
934 	int packets_dropped = 0;
935 	struct sk_buff *skb;
936 
937 	while ((skb = __skb_dequeue(rxq)) != NULL) {
938 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
939 
940 		if (pull_to > skb_headlen(skb))
941 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
942 
943 		/* Ethernet work: Delayed to here as it peeks the header. */
944 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
945 		skb_reset_network_header(skb);
946 
947 		if (checksum_setup(queue->info->netdev, skb)) {
948 			kfree_skb(skb);
949 			packets_dropped++;
950 			queue->info->netdev->stats.rx_errors++;
951 			continue;
952 		}
953 
954 		u64_stats_update_begin(&rx_stats->syncp);
955 		rx_stats->packets++;
956 		rx_stats->bytes += skb->len;
957 		u64_stats_update_end(&rx_stats->syncp);
958 
959 		/* Pass it up. */
960 		napi_gro_receive(&queue->napi, skb);
961 	}
962 
963 	return packets_dropped;
964 }
965 
966 static int xennet_poll(struct napi_struct *napi, int budget)
967 {
968 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
969 	struct net_device *dev = queue->info->netdev;
970 	struct sk_buff *skb;
971 	struct netfront_rx_info rinfo;
972 	struct xen_netif_rx_response *rx = &rinfo.rx;
973 	struct xen_netif_extra_info *extras = rinfo.extras;
974 	RING_IDX i, rp;
975 	int work_done;
976 	struct sk_buff_head rxq;
977 	struct sk_buff_head errq;
978 	struct sk_buff_head tmpq;
979 	int err;
980 
981 	spin_lock(&queue->rx_lock);
982 
983 	skb_queue_head_init(&rxq);
984 	skb_queue_head_init(&errq);
985 	skb_queue_head_init(&tmpq);
986 
987 	rp = queue->rx.sring->rsp_prod;
988 	rmb(); /* Ensure we see queued responses up to 'rp'. */
989 
990 	i = queue->rx.rsp_cons;
991 	work_done = 0;
992 	while ((i != rp) && (work_done < budget)) {
993 		memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
994 		memset(extras, 0, sizeof(rinfo.extras));
995 
996 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
997 
998 		if (unlikely(err)) {
999 err:
1000 			while ((skb = __skb_dequeue(&tmpq)))
1001 				__skb_queue_tail(&errq, skb);
1002 			dev->stats.rx_errors++;
1003 			i = queue->rx.rsp_cons;
1004 			continue;
1005 		}
1006 
1007 		skb = __skb_dequeue(&tmpq);
1008 
1009 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1010 			struct xen_netif_extra_info *gso;
1011 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1012 
1013 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1014 				__skb_queue_head(&tmpq, skb);
1015 				queue->rx.rsp_cons += skb_queue_len(&tmpq);
1016 				goto err;
1017 			}
1018 		}
1019 
1020 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1021 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1022 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1023 
1024 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1025 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1026 		skb->data_len = rx->status;
1027 		skb->len += rx->status;
1028 
1029 		i = xennet_fill_frags(queue, skb, &tmpq);
1030 
1031 		if (rx->flags & XEN_NETRXF_csum_blank)
1032 			skb->ip_summed = CHECKSUM_PARTIAL;
1033 		else if (rx->flags & XEN_NETRXF_data_validated)
1034 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1035 
1036 		__skb_queue_tail(&rxq, skb);
1037 
1038 		queue->rx.rsp_cons = ++i;
1039 		work_done++;
1040 	}
1041 
1042 	__skb_queue_purge(&errq);
1043 
1044 	work_done -= handle_incoming_queue(queue, &rxq);
1045 
1046 	xennet_alloc_rx_buffers(queue);
1047 
1048 	if (work_done < budget) {
1049 		int more_to_do = 0;
1050 
1051 		napi_complete(napi);
1052 
1053 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1054 		if (more_to_do)
1055 			napi_schedule(napi);
1056 	}
1057 
1058 	spin_unlock(&queue->rx_lock);
1059 
1060 	return work_done;
1061 }
1062 
1063 static int xennet_change_mtu(struct net_device *dev, int mtu)
1064 {
1065 	int max = xennet_can_sg(dev) ?
1066 		XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1067 
1068 	if (mtu > max)
1069 		return -EINVAL;
1070 	dev->mtu = mtu;
1071 	return 0;
1072 }
1073 
1074 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1075 						    struct rtnl_link_stats64 *tot)
1076 {
1077 	struct netfront_info *np = netdev_priv(dev);
1078 	int cpu;
1079 
1080 	for_each_possible_cpu(cpu) {
1081 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1082 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1083 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1084 		unsigned int start;
1085 
1086 		do {
1087 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1088 			tx_packets = tx_stats->packets;
1089 			tx_bytes = tx_stats->bytes;
1090 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1091 
1092 		do {
1093 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1094 			rx_packets = rx_stats->packets;
1095 			rx_bytes = rx_stats->bytes;
1096 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1097 
1098 		tot->rx_packets += rx_packets;
1099 		tot->tx_packets += tx_packets;
1100 		tot->rx_bytes   += rx_bytes;
1101 		tot->tx_bytes   += tx_bytes;
1102 	}
1103 
1104 	tot->rx_errors  = dev->stats.rx_errors;
1105 	tot->tx_dropped = dev->stats.tx_dropped;
1106 
1107 	return tot;
1108 }
1109 
1110 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1111 {
1112 	struct sk_buff *skb;
1113 	int i;
1114 
1115 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1116 		/* Skip over entries which are actually freelist references */
1117 		if (skb_entry_is_link(&queue->tx_skbs[i]))
1118 			continue;
1119 
1120 		skb = queue->tx_skbs[i].skb;
1121 		get_page(queue->grant_tx_page[i]);
1122 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1123 					  GNTMAP_readonly,
1124 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1125 		queue->grant_tx_page[i] = NULL;
1126 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1127 		add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1128 		dev_kfree_skb_irq(skb);
1129 	}
1130 }
1131 
1132 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1133 {
1134 	int id, ref;
1135 
1136 	spin_lock_bh(&queue->rx_lock);
1137 
1138 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1139 		struct sk_buff *skb;
1140 		struct page *page;
1141 
1142 		skb = queue->rx_skbs[id];
1143 		if (!skb)
1144 			continue;
1145 
1146 		ref = queue->grant_rx_ref[id];
1147 		if (ref == GRANT_INVALID_REF)
1148 			continue;
1149 
1150 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1151 
1152 		/* gnttab_end_foreign_access() needs a page ref until
1153 		 * foreign access is ended (which may be deferred).
1154 		 */
1155 		get_page(page);
1156 		gnttab_end_foreign_access(ref, 0,
1157 					  (unsigned long)page_address(page));
1158 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1159 
1160 		kfree_skb(skb);
1161 	}
1162 
1163 	spin_unlock_bh(&queue->rx_lock);
1164 }
1165 
1166 static netdev_features_t xennet_fix_features(struct net_device *dev,
1167 	netdev_features_t features)
1168 {
1169 	struct netfront_info *np = netdev_priv(dev);
1170 	int val;
1171 
1172 	if (features & NETIF_F_SG) {
1173 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1174 				 "%d", &val) < 0)
1175 			val = 0;
1176 
1177 		if (!val)
1178 			features &= ~NETIF_F_SG;
1179 	}
1180 
1181 	if (features & NETIF_F_IPV6_CSUM) {
1182 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1183 				 "feature-ipv6-csum-offload", "%d", &val) < 0)
1184 			val = 0;
1185 
1186 		if (!val)
1187 			features &= ~NETIF_F_IPV6_CSUM;
1188 	}
1189 
1190 	if (features & NETIF_F_TSO) {
1191 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1192 				 "feature-gso-tcpv4", "%d", &val) < 0)
1193 			val = 0;
1194 
1195 		if (!val)
1196 			features &= ~NETIF_F_TSO;
1197 	}
1198 
1199 	if (features & NETIF_F_TSO6) {
1200 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1201 				 "feature-gso-tcpv6", "%d", &val) < 0)
1202 			val = 0;
1203 
1204 		if (!val)
1205 			features &= ~NETIF_F_TSO6;
1206 	}
1207 
1208 	return features;
1209 }
1210 
1211 static int xennet_set_features(struct net_device *dev,
1212 	netdev_features_t features)
1213 {
1214 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1215 		netdev_info(dev, "Reducing MTU because no SG offload");
1216 		dev->mtu = ETH_DATA_LEN;
1217 	}
1218 
1219 	return 0;
1220 }
1221 
1222 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1223 {
1224 	struct netfront_queue *queue = dev_id;
1225 	unsigned long flags;
1226 
1227 	spin_lock_irqsave(&queue->tx_lock, flags);
1228 	xennet_tx_buf_gc(queue);
1229 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1230 
1231 	return IRQ_HANDLED;
1232 }
1233 
1234 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1235 {
1236 	struct netfront_queue *queue = dev_id;
1237 	struct net_device *dev = queue->info->netdev;
1238 
1239 	if (likely(netif_carrier_ok(dev) &&
1240 		   RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1241 		napi_schedule(&queue->napi);
1242 
1243 	return IRQ_HANDLED;
1244 }
1245 
1246 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1247 {
1248 	xennet_tx_interrupt(irq, dev_id);
1249 	xennet_rx_interrupt(irq, dev_id);
1250 	return IRQ_HANDLED;
1251 }
1252 
1253 #ifdef CONFIG_NET_POLL_CONTROLLER
1254 static void xennet_poll_controller(struct net_device *dev)
1255 {
1256 	/* Poll each queue */
1257 	struct netfront_info *info = netdev_priv(dev);
1258 	unsigned int num_queues = dev->real_num_tx_queues;
1259 	unsigned int i;
1260 	for (i = 0; i < num_queues; ++i)
1261 		xennet_interrupt(0, &info->queues[i]);
1262 }
1263 #endif
1264 
1265 static const struct net_device_ops xennet_netdev_ops = {
1266 	.ndo_open            = xennet_open,
1267 	.ndo_stop            = xennet_close,
1268 	.ndo_start_xmit      = xennet_start_xmit,
1269 	.ndo_change_mtu	     = xennet_change_mtu,
1270 	.ndo_get_stats64     = xennet_get_stats64,
1271 	.ndo_set_mac_address = eth_mac_addr,
1272 	.ndo_validate_addr   = eth_validate_addr,
1273 	.ndo_fix_features    = xennet_fix_features,
1274 	.ndo_set_features    = xennet_set_features,
1275 	.ndo_select_queue    = xennet_select_queue,
1276 #ifdef CONFIG_NET_POLL_CONTROLLER
1277 	.ndo_poll_controller = xennet_poll_controller,
1278 #endif
1279 };
1280 
1281 static void xennet_free_netdev(struct net_device *netdev)
1282 {
1283 	struct netfront_info *np = netdev_priv(netdev);
1284 
1285 	free_percpu(np->rx_stats);
1286 	free_percpu(np->tx_stats);
1287 	free_netdev(netdev);
1288 }
1289 
1290 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1291 {
1292 	int err;
1293 	struct net_device *netdev;
1294 	struct netfront_info *np;
1295 
1296 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1297 	if (!netdev)
1298 		return ERR_PTR(-ENOMEM);
1299 
1300 	np                   = netdev_priv(netdev);
1301 	np->xbdev            = dev;
1302 
1303 	/* No need to use rtnl_lock() before the call below as it
1304 	 * happens before register_netdev().
1305 	 */
1306 	netif_set_real_num_tx_queues(netdev, 0);
1307 	np->queues = NULL;
1308 
1309 	err = -ENOMEM;
1310 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1311 	if (np->rx_stats == NULL)
1312 		goto exit;
1313 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1314 	if (np->tx_stats == NULL)
1315 		goto exit;
1316 
1317 	netdev->netdev_ops	= &xennet_netdev_ops;
1318 
1319 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1320 				  NETIF_F_GSO_ROBUST;
1321 	netdev->hw_features	= NETIF_F_SG |
1322 				  NETIF_F_IPV6_CSUM |
1323 				  NETIF_F_TSO | NETIF_F_TSO6;
1324 
1325 	/*
1326          * Assume that all hw features are available for now. This set
1327          * will be adjusted by the call to netdev_update_features() in
1328          * xennet_connect() which is the earliest point where we can
1329          * negotiate with the backend regarding supported features.
1330          */
1331 	netdev->features |= netdev->hw_features;
1332 
1333 	netdev->ethtool_ops = &xennet_ethtool_ops;
1334 	SET_NETDEV_DEV(netdev, &dev->dev);
1335 
1336 	netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1337 
1338 	np->netdev = netdev;
1339 
1340 	netif_carrier_off(netdev);
1341 
1342 	return netdev;
1343 
1344  exit:
1345 	xennet_free_netdev(netdev);
1346 	return ERR_PTR(err);
1347 }
1348 
1349 /**
1350  * Entry point to this code when a new device is created.  Allocate the basic
1351  * structures and the ring buffers for communication with the backend, and
1352  * inform the backend of the appropriate details for those.
1353  */
1354 static int netfront_probe(struct xenbus_device *dev,
1355 			  const struct xenbus_device_id *id)
1356 {
1357 	int err;
1358 	struct net_device *netdev;
1359 	struct netfront_info *info;
1360 
1361 	netdev = xennet_create_dev(dev);
1362 	if (IS_ERR(netdev)) {
1363 		err = PTR_ERR(netdev);
1364 		xenbus_dev_fatal(dev, err, "creating netdev");
1365 		return err;
1366 	}
1367 
1368 	info = netdev_priv(netdev);
1369 	dev_set_drvdata(&dev->dev, info);
1370 
1371 	err = register_netdev(info->netdev);
1372 	if (err) {
1373 		pr_warn("%s: register_netdev err=%d\n", __func__, err);
1374 		goto fail;
1375 	}
1376 
1377 	err = xennet_sysfs_addif(info->netdev);
1378 	if (err) {
1379 		unregister_netdev(info->netdev);
1380 		pr_warn("%s: add sysfs failed err=%d\n", __func__, err);
1381 		goto fail;
1382 	}
1383 
1384 	return 0;
1385 
1386  fail:
1387 	xennet_free_netdev(netdev);
1388 	dev_set_drvdata(&dev->dev, NULL);
1389 	return err;
1390 }
1391 
1392 static void xennet_end_access(int ref, void *page)
1393 {
1394 	/* This frees the page as a side-effect */
1395 	if (ref != GRANT_INVALID_REF)
1396 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1397 }
1398 
1399 static void xennet_disconnect_backend(struct netfront_info *info)
1400 {
1401 	unsigned int i = 0;
1402 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1403 
1404 	netif_carrier_off(info->netdev);
1405 
1406 	for (i = 0; i < num_queues; ++i) {
1407 		struct netfront_queue *queue = &info->queues[i];
1408 
1409 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1410 			unbind_from_irqhandler(queue->tx_irq, queue);
1411 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1412 			unbind_from_irqhandler(queue->tx_irq, queue);
1413 			unbind_from_irqhandler(queue->rx_irq, queue);
1414 		}
1415 		queue->tx_evtchn = queue->rx_evtchn = 0;
1416 		queue->tx_irq = queue->rx_irq = 0;
1417 
1418 		napi_synchronize(&queue->napi);
1419 
1420 		xennet_release_tx_bufs(queue);
1421 		xennet_release_rx_bufs(queue);
1422 		gnttab_free_grant_references(queue->gref_tx_head);
1423 		gnttab_free_grant_references(queue->gref_rx_head);
1424 
1425 		/* End access and free the pages */
1426 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1427 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1428 
1429 		queue->tx_ring_ref = GRANT_INVALID_REF;
1430 		queue->rx_ring_ref = GRANT_INVALID_REF;
1431 		queue->tx.sring = NULL;
1432 		queue->rx.sring = NULL;
1433 	}
1434 }
1435 
1436 /**
1437  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1438  * driver restart.  We tear down our netif structure and recreate it, but
1439  * leave the device-layer structures intact so that this is transparent to the
1440  * rest of the kernel.
1441  */
1442 static int netfront_resume(struct xenbus_device *dev)
1443 {
1444 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1445 
1446 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1447 
1448 	xennet_disconnect_backend(info);
1449 	return 0;
1450 }
1451 
1452 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1453 {
1454 	char *s, *e, *macstr;
1455 	int i;
1456 
1457 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1458 	if (IS_ERR(macstr))
1459 		return PTR_ERR(macstr);
1460 
1461 	for (i = 0; i < ETH_ALEN; i++) {
1462 		mac[i] = simple_strtoul(s, &e, 16);
1463 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1464 			kfree(macstr);
1465 			return -ENOENT;
1466 		}
1467 		s = e+1;
1468 	}
1469 
1470 	kfree(macstr);
1471 	return 0;
1472 }
1473 
1474 static int setup_netfront_single(struct netfront_queue *queue)
1475 {
1476 	int err;
1477 
1478 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1479 	if (err < 0)
1480 		goto fail;
1481 
1482 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1483 					xennet_interrupt,
1484 					0, queue->info->netdev->name, queue);
1485 	if (err < 0)
1486 		goto bind_fail;
1487 	queue->rx_evtchn = queue->tx_evtchn;
1488 	queue->rx_irq = queue->tx_irq = err;
1489 
1490 	return 0;
1491 
1492 bind_fail:
1493 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1494 	queue->tx_evtchn = 0;
1495 fail:
1496 	return err;
1497 }
1498 
1499 static int setup_netfront_split(struct netfront_queue *queue)
1500 {
1501 	int err;
1502 
1503 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1504 	if (err < 0)
1505 		goto fail;
1506 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1507 	if (err < 0)
1508 		goto alloc_rx_evtchn_fail;
1509 
1510 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1511 		 "%s-tx", queue->name);
1512 	err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1513 					xennet_tx_interrupt,
1514 					0, queue->tx_irq_name, queue);
1515 	if (err < 0)
1516 		goto bind_tx_fail;
1517 	queue->tx_irq = err;
1518 
1519 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1520 		 "%s-rx", queue->name);
1521 	err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1522 					xennet_rx_interrupt,
1523 					0, queue->rx_irq_name, queue);
1524 	if (err < 0)
1525 		goto bind_rx_fail;
1526 	queue->rx_irq = err;
1527 
1528 	return 0;
1529 
1530 bind_rx_fail:
1531 	unbind_from_irqhandler(queue->tx_irq, queue);
1532 	queue->tx_irq = 0;
1533 bind_tx_fail:
1534 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1535 	queue->rx_evtchn = 0;
1536 alloc_rx_evtchn_fail:
1537 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1538 	queue->tx_evtchn = 0;
1539 fail:
1540 	return err;
1541 }
1542 
1543 static int setup_netfront(struct xenbus_device *dev,
1544 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1545 {
1546 	struct xen_netif_tx_sring *txs;
1547 	struct xen_netif_rx_sring *rxs;
1548 	int err;
1549 
1550 	queue->tx_ring_ref = GRANT_INVALID_REF;
1551 	queue->rx_ring_ref = GRANT_INVALID_REF;
1552 	queue->rx.sring = NULL;
1553 	queue->tx.sring = NULL;
1554 
1555 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1556 	if (!txs) {
1557 		err = -ENOMEM;
1558 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1559 		goto fail;
1560 	}
1561 	SHARED_RING_INIT(txs);
1562 	FRONT_RING_INIT(&queue->tx, txs, PAGE_SIZE);
1563 
1564 	err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1565 	if (err < 0)
1566 		goto grant_tx_ring_fail;
1567 	queue->tx_ring_ref = err;
1568 
1569 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1570 	if (!rxs) {
1571 		err = -ENOMEM;
1572 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1573 		goto alloc_rx_ring_fail;
1574 	}
1575 	SHARED_RING_INIT(rxs);
1576 	FRONT_RING_INIT(&queue->rx, rxs, PAGE_SIZE);
1577 
1578 	err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1579 	if (err < 0)
1580 		goto grant_rx_ring_fail;
1581 	queue->rx_ring_ref = err;
1582 
1583 	if (feature_split_evtchn)
1584 		err = setup_netfront_split(queue);
1585 	/* setup single event channel if
1586 	 *  a) feature-split-event-channels == 0
1587 	 *  b) feature-split-event-channels == 1 but failed to setup
1588 	 */
1589 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1590 		err = setup_netfront_single(queue);
1591 
1592 	if (err)
1593 		goto alloc_evtchn_fail;
1594 
1595 	return 0;
1596 
1597 	/* If we fail to setup netfront, it is safe to just revoke access to
1598 	 * granted pages because backend is not accessing it at this point.
1599 	 */
1600 alloc_evtchn_fail:
1601 	gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1602 grant_rx_ring_fail:
1603 	free_page((unsigned long)rxs);
1604 alloc_rx_ring_fail:
1605 	gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1606 grant_tx_ring_fail:
1607 	free_page((unsigned long)txs);
1608 fail:
1609 	return err;
1610 }
1611 
1612 /* Queue-specific initialisation
1613  * This used to be done in xennet_create_dev() but must now
1614  * be run per-queue.
1615  */
1616 static int xennet_init_queue(struct netfront_queue *queue)
1617 {
1618 	unsigned short i;
1619 	int err = 0;
1620 
1621 	spin_lock_init(&queue->tx_lock);
1622 	spin_lock_init(&queue->rx_lock);
1623 
1624 	init_timer(&queue->rx_refill_timer);
1625 	queue->rx_refill_timer.data = (unsigned long)queue;
1626 	queue->rx_refill_timer.function = rx_refill_timeout;
1627 
1628 	snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1629 		 queue->info->netdev->name, queue->id);
1630 
1631 	/* Initialise tx_skbs as a free chain containing every entry. */
1632 	queue->tx_skb_freelist = 0;
1633 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1634 		skb_entry_set_link(&queue->tx_skbs[i], i+1);
1635 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1636 		queue->grant_tx_page[i] = NULL;
1637 	}
1638 
1639 	/* Clear out rx_skbs */
1640 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1641 		queue->rx_skbs[i] = NULL;
1642 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1643 	}
1644 
1645 	/* A grant for every tx ring slot */
1646 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1647 					  &queue->gref_tx_head) < 0) {
1648 		pr_alert("can't alloc tx grant refs\n");
1649 		err = -ENOMEM;
1650 		goto exit;
1651 	}
1652 
1653 	/* A grant for every rx ring slot */
1654 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1655 					  &queue->gref_rx_head) < 0) {
1656 		pr_alert("can't alloc rx grant refs\n");
1657 		err = -ENOMEM;
1658 		goto exit_free_tx;
1659 	}
1660 
1661 	return 0;
1662 
1663  exit_free_tx:
1664 	gnttab_free_grant_references(queue->gref_tx_head);
1665  exit:
1666 	return err;
1667 }
1668 
1669 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1670 			   struct xenbus_transaction *xbt, int write_hierarchical)
1671 {
1672 	/* Write the queue-specific keys into XenStore in the traditional
1673 	 * way for a single queue, or in a queue subkeys for multiple
1674 	 * queues.
1675 	 */
1676 	struct xenbus_device *dev = queue->info->xbdev;
1677 	int err;
1678 	const char *message;
1679 	char *path;
1680 	size_t pathsize;
1681 
1682 	/* Choose the correct place to write the keys */
1683 	if (write_hierarchical) {
1684 		pathsize = strlen(dev->nodename) + 10;
1685 		path = kzalloc(pathsize, GFP_KERNEL);
1686 		if (!path) {
1687 			err = -ENOMEM;
1688 			message = "out of memory while writing ring references";
1689 			goto error;
1690 		}
1691 		snprintf(path, pathsize, "%s/queue-%u",
1692 				dev->nodename, queue->id);
1693 	} else {
1694 		path = (char *)dev->nodename;
1695 	}
1696 
1697 	/* Write ring references */
1698 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1699 			queue->tx_ring_ref);
1700 	if (err) {
1701 		message = "writing tx-ring-ref";
1702 		goto error;
1703 	}
1704 
1705 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1706 			queue->rx_ring_ref);
1707 	if (err) {
1708 		message = "writing rx-ring-ref";
1709 		goto error;
1710 	}
1711 
1712 	/* Write event channels; taking into account both shared
1713 	 * and split event channel scenarios.
1714 	 */
1715 	if (queue->tx_evtchn == queue->rx_evtchn) {
1716 		/* Shared event channel */
1717 		err = xenbus_printf(*xbt, path,
1718 				"event-channel", "%u", queue->tx_evtchn);
1719 		if (err) {
1720 			message = "writing event-channel";
1721 			goto error;
1722 		}
1723 	} else {
1724 		/* Split event channels */
1725 		err = xenbus_printf(*xbt, path,
1726 				"event-channel-tx", "%u", queue->tx_evtchn);
1727 		if (err) {
1728 			message = "writing event-channel-tx";
1729 			goto error;
1730 		}
1731 
1732 		err = xenbus_printf(*xbt, path,
1733 				"event-channel-rx", "%u", queue->rx_evtchn);
1734 		if (err) {
1735 			message = "writing event-channel-rx";
1736 			goto error;
1737 		}
1738 	}
1739 
1740 	if (write_hierarchical)
1741 		kfree(path);
1742 	return 0;
1743 
1744 error:
1745 	if (write_hierarchical)
1746 		kfree(path);
1747 	xenbus_dev_fatal(dev, err, "%s", message);
1748 	return err;
1749 }
1750 
1751 static void xennet_destroy_queues(struct netfront_info *info)
1752 {
1753 	unsigned int i;
1754 
1755 	rtnl_lock();
1756 
1757 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1758 		struct netfront_queue *queue = &info->queues[i];
1759 
1760 		if (netif_running(info->netdev))
1761 			napi_disable(&queue->napi);
1762 		netif_napi_del(&queue->napi);
1763 	}
1764 
1765 	rtnl_unlock();
1766 
1767 	kfree(info->queues);
1768 	info->queues = NULL;
1769 }
1770 
1771 static int xennet_create_queues(struct netfront_info *info,
1772 				unsigned int num_queues)
1773 {
1774 	unsigned int i;
1775 	int ret;
1776 
1777 	info->queues = kcalloc(num_queues, sizeof(struct netfront_queue),
1778 			       GFP_KERNEL);
1779 	if (!info->queues)
1780 		return -ENOMEM;
1781 
1782 	rtnl_lock();
1783 
1784 	for (i = 0; i < num_queues; i++) {
1785 		struct netfront_queue *queue = &info->queues[i];
1786 
1787 		queue->id = i;
1788 		queue->info = info;
1789 
1790 		ret = xennet_init_queue(queue);
1791 		if (ret < 0) {
1792 			dev_warn(&info->netdev->dev,
1793 				 "only created %d queues\n", i);
1794 			num_queues = i;
1795 			break;
1796 		}
1797 
1798 		netif_napi_add(queue->info->netdev, &queue->napi,
1799 			       xennet_poll, 64);
1800 		if (netif_running(info->netdev))
1801 			napi_enable(&queue->napi);
1802 	}
1803 
1804 	netif_set_real_num_tx_queues(info->netdev, num_queues);
1805 
1806 	rtnl_unlock();
1807 
1808 	if (num_queues == 0) {
1809 		dev_err(&info->netdev->dev, "no queues\n");
1810 		return -EINVAL;
1811 	}
1812 	return 0;
1813 }
1814 
1815 /* Common code used when first setting up, and when resuming. */
1816 static int talk_to_netback(struct xenbus_device *dev,
1817 			   struct netfront_info *info)
1818 {
1819 	const char *message;
1820 	struct xenbus_transaction xbt;
1821 	int err;
1822 	unsigned int feature_split_evtchn;
1823 	unsigned int i = 0;
1824 	unsigned int max_queues = 0;
1825 	struct netfront_queue *queue = NULL;
1826 	unsigned int num_queues = 1;
1827 
1828 	info->netdev->irq = 0;
1829 
1830 	/* Check if backend supports multiple queues */
1831 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1832 			   "multi-queue-max-queues", "%u", &max_queues);
1833 	if (err < 0)
1834 		max_queues = 1;
1835 	num_queues = min(max_queues, xennet_max_queues);
1836 
1837 	/* Check feature-split-event-channels */
1838 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1839 			   "feature-split-event-channels", "%u",
1840 			   &feature_split_evtchn);
1841 	if (err < 0)
1842 		feature_split_evtchn = 0;
1843 
1844 	/* Read mac addr. */
1845 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
1846 	if (err) {
1847 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1848 		goto out;
1849 	}
1850 
1851 	if (info->queues)
1852 		xennet_destroy_queues(info);
1853 
1854 	err = xennet_create_queues(info, num_queues);
1855 	if (err < 0)
1856 		goto destroy_ring;
1857 
1858 	/* Create shared ring, alloc event channel -- for each queue */
1859 	for (i = 0; i < num_queues; ++i) {
1860 		queue = &info->queues[i];
1861 		err = setup_netfront(dev, queue, feature_split_evtchn);
1862 		if (err) {
1863 			/* setup_netfront() will tidy up the current
1864 			 * queue on error, but we need to clean up
1865 			 * those already allocated.
1866 			 */
1867 			if (i > 0) {
1868 				rtnl_lock();
1869 				netif_set_real_num_tx_queues(info->netdev, i);
1870 				rtnl_unlock();
1871 				goto destroy_ring;
1872 			} else {
1873 				goto out;
1874 			}
1875 		}
1876 	}
1877 
1878 again:
1879 	err = xenbus_transaction_start(&xbt);
1880 	if (err) {
1881 		xenbus_dev_fatal(dev, err, "starting transaction");
1882 		goto destroy_ring;
1883 	}
1884 
1885 	if (num_queues == 1) {
1886 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1887 		if (err)
1888 			goto abort_transaction_no_dev_fatal;
1889 	} else {
1890 		/* Write the number of queues */
1891 		err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues",
1892 				    "%u", num_queues);
1893 		if (err) {
1894 			message = "writing multi-queue-num-queues";
1895 			goto abort_transaction_no_dev_fatal;
1896 		}
1897 
1898 		/* Write the keys for each queue */
1899 		for (i = 0; i < num_queues; ++i) {
1900 			queue = &info->queues[i];
1901 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1902 			if (err)
1903 				goto abort_transaction_no_dev_fatal;
1904 		}
1905 	}
1906 
1907 	/* The remaining keys are not queue-specific */
1908 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1909 			    1);
1910 	if (err) {
1911 		message = "writing request-rx-copy";
1912 		goto abort_transaction;
1913 	}
1914 
1915 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1916 	if (err) {
1917 		message = "writing feature-rx-notify";
1918 		goto abort_transaction;
1919 	}
1920 
1921 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1922 	if (err) {
1923 		message = "writing feature-sg";
1924 		goto abort_transaction;
1925 	}
1926 
1927 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1928 	if (err) {
1929 		message = "writing feature-gso-tcpv4";
1930 		goto abort_transaction;
1931 	}
1932 
1933 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1934 	if (err) {
1935 		message = "writing feature-gso-tcpv6";
1936 		goto abort_transaction;
1937 	}
1938 
1939 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1940 			   "1");
1941 	if (err) {
1942 		message = "writing feature-ipv6-csum-offload";
1943 		goto abort_transaction;
1944 	}
1945 
1946 	err = xenbus_transaction_end(xbt, 0);
1947 	if (err) {
1948 		if (err == -EAGAIN)
1949 			goto again;
1950 		xenbus_dev_fatal(dev, err, "completing transaction");
1951 		goto destroy_ring;
1952 	}
1953 
1954 	return 0;
1955 
1956  abort_transaction:
1957 	xenbus_dev_fatal(dev, err, "%s", message);
1958 abort_transaction_no_dev_fatal:
1959 	xenbus_transaction_end(xbt, 1);
1960  destroy_ring:
1961 	xennet_disconnect_backend(info);
1962 	kfree(info->queues);
1963 	info->queues = NULL;
1964 	rtnl_lock();
1965 	netif_set_real_num_tx_queues(info->netdev, 0);
1966 	rtnl_unlock();
1967  out:
1968 	return err;
1969 }
1970 
1971 static int xennet_connect(struct net_device *dev)
1972 {
1973 	struct netfront_info *np = netdev_priv(dev);
1974 	unsigned int num_queues = 0;
1975 	int err;
1976 	unsigned int feature_rx_copy;
1977 	unsigned int j = 0;
1978 	struct netfront_queue *queue = NULL;
1979 
1980 	err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1981 			   "feature-rx-copy", "%u", &feature_rx_copy);
1982 	if (err != 1)
1983 		feature_rx_copy = 0;
1984 
1985 	if (!feature_rx_copy) {
1986 		dev_info(&dev->dev,
1987 			 "backend does not support copying receive path\n");
1988 		return -ENODEV;
1989 	}
1990 
1991 	err = talk_to_netback(np->xbdev, np);
1992 	if (err)
1993 		return err;
1994 
1995 	/* talk_to_netback() sets the correct number of queues */
1996 	num_queues = dev->real_num_tx_queues;
1997 
1998 	rtnl_lock();
1999 	netdev_update_features(dev);
2000 	rtnl_unlock();
2001 
2002 	/*
2003 	 * All public and private state should now be sane.  Get
2004 	 * ready to start sending and receiving packets and give the driver
2005 	 * domain a kick because we've probably just requeued some
2006 	 * packets.
2007 	 */
2008 	netif_carrier_on(np->netdev);
2009 	for (j = 0; j < num_queues; ++j) {
2010 		queue = &np->queues[j];
2011 
2012 		notify_remote_via_irq(queue->tx_irq);
2013 		if (queue->tx_irq != queue->rx_irq)
2014 			notify_remote_via_irq(queue->rx_irq);
2015 
2016 		spin_lock_irq(&queue->tx_lock);
2017 		xennet_tx_buf_gc(queue);
2018 		spin_unlock_irq(&queue->tx_lock);
2019 
2020 		spin_lock_bh(&queue->rx_lock);
2021 		xennet_alloc_rx_buffers(queue);
2022 		spin_unlock_bh(&queue->rx_lock);
2023 	}
2024 
2025 	return 0;
2026 }
2027 
2028 /**
2029  * Callback received when the backend's state changes.
2030  */
2031 static void netback_changed(struct xenbus_device *dev,
2032 			    enum xenbus_state backend_state)
2033 {
2034 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2035 	struct net_device *netdev = np->netdev;
2036 
2037 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2038 
2039 	switch (backend_state) {
2040 	case XenbusStateInitialising:
2041 	case XenbusStateInitialised:
2042 	case XenbusStateReconfiguring:
2043 	case XenbusStateReconfigured:
2044 	case XenbusStateUnknown:
2045 		break;
2046 
2047 	case XenbusStateInitWait:
2048 		if (dev->state != XenbusStateInitialising)
2049 			break;
2050 		if (xennet_connect(netdev) != 0)
2051 			break;
2052 		xenbus_switch_state(dev, XenbusStateConnected);
2053 		break;
2054 
2055 	case XenbusStateConnected:
2056 		netdev_notify_peers(netdev);
2057 		break;
2058 
2059 	case XenbusStateClosed:
2060 		if (dev->state == XenbusStateClosed)
2061 			break;
2062 		/* Missed the backend's CLOSING state -- fallthrough */
2063 	case XenbusStateClosing:
2064 		xenbus_frontend_closed(dev);
2065 		break;
2066 	}
2067 }
2068 
2069 static const struct xennet_stat {
2070 	char name[ETH_GSTRING_LEN];
2071 	u16 offset;
2072 } xennet_stats[] = {
2073 	{
2074 		"rx_gso_checksum_fixup",
2075 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2076 	},
2077 };
2078 
2079 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2080 {
2081 	switch (string_set) {
2082 	case ETH_SS_STATS:
2083 		return ARRAY_SIZE(xennet_stats);
2084 	default:
2085 		return -EINVAL;
2086 	}
2087 }
2088 
2089 static void xennet_get_ethtool_stats(struct net_device *dev,
2090 				     struct ethtool_stats *stats, u64 * data)
2091 {
2092 	void *np = netdev_priv(dev);
2093 	int i;
2094 
2095 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2096 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2097 }
2098 
2099 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2100 {
2101 	int i;
2102 
2103 	switch (stringset) {
2104 	case ETH_SS_STATS:
2105 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2106 			memcpy(data + i * ETH_GSTRING_LEN,
2107 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2108 		break;
2109 	}
2110 }
2111 
2112 static const struct ethtool_ops xennet_ethtool_ops =
2113 {
2114 	.get_link = ethtool_op_get_link,
2115 
2116 	.get_sset_count = xennet_get_sset_count,
2117 	.get_ethtool_stats = xennet_get_ethtool_stats,
2118 	.get_strings = xennet_get_strings,
2119 };
2120 
2121 #ifdef CONFIG_SYSFS
2122 static ssize_t show_rxbuf(struct device *dev,
2123 			  struct device_attribute *attr, char *buf)
2124 {
2125 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2126 }
2127 
2128 static ssize_t store_rxbuf(struct device *dev,
2129 			   struct device_attribute *attr,
2130 			   const char *buf, size_t len)
2131 {
2132 	char *endp;
2133 	unsigned long target;
2134 
2135 	if (!capable(CAP_NET_ADMIN))
2136 		return -EPERM;
2137 
2138 	target = simple_strtoul(buf, &endp, 0);
2139 	if (endp == buf)
2140 		return -EBADMSG;
2141 
2142 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2143 
2144 	return len;
2145 }
2146 
2147 static struct device_attribute xennet_attrs[] = {
2148 	__ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf),
2149 	__ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf),
2150 	__ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL),
2151 };
2152 
2153 static int xennet_sysfs_addif(struct net_device *netdev)
2154 {
2155 	int i;
2156 	int err;
2157 
2158 	for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
2159 		err = device_create_file(&netdev->dev,
2160 					   &xennet_attrs[i]);
2161 		if (err)
2162 			goto fail;
2163 	}
2164 	return 0;
2165 
2166  fail:
2167 	while (--i >= 0)
2168 		device_remove_file(&netdev->dev, &xennet_attrs[i]);
2169 	return err;
2170 }
2171 
2172 static void xennet_sysfs_delif(struct net_device *netdev)
2173 {
2174 	int i;
2175 
2176 	for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
2177 		device_remove_file(&netdev->dev, &xennet_attrs[i]);
2178 }
2179 
2180 #endif /* CONFIG_SYSFS */
2181 
2182 static int xennet_remove(struct xenbus_device *dev)
2183 {
2184 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2185 	unsigned int num_queues = info->netdev->real_num_tx_queues;
2186 	struct netfront_queue *queue = NULL;
2187 	unsigned int i = 0;
2188 
2189 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2190 
2191 	xennet_disconnect_backend(info);
2192 
2193 	xennet_sysfs_delif(info->netdev);
2194 
2195 	unregister_netdev(info->netdev);
2196 
2197 	for (i = 0; i < num_queues; ++i) {
2198 		queue = &info->queues[i];
2199 		del_timer_sync(&queue->rx_refill_timer);
2200 	}
2201 
2202 	if (num_queues) {
2203 		kfree(info->queues);
2204 		info->queues = NULL;
2205 	}
2206 
2207 	xennet_free_netdev(info->netdev);
2208 
2209 	return 0;
2210 }
2211 
2212 static const struct xenbus_device_id netfront_ids[] = {
2213 	{ "vif" },
2214 	{ "" }
2215 };
2216 
2217 static struct xenbus_driver netfront_driver = {
2218 	.ids = netfront_ids,
2219 	.probe = netfront_probe,
2220 	.remove = xennet_remove,
2221 	.resume = netfront_resume,
2222 	.otherend_changed = netback_changed,
2223 };
2224 
2225 static int __init netif_init(void)
2226 {
2227 	if (!xen_domain())
2228 		return -ENODEV;
2229 
2230 	if (!xen_has_pv_nic_devices())
2231 		return -ENODEV;
2232 
2233 	pr_info("Initialising Xen virtual ethernet driver\n");
2234 
2235 	/* Allow as many queues as there are CPUs, by default */
2236 	xennet_max_queues = num_online_cpus();
2237 
2238 	return xenbus_register_frontend(&netfront_driver);
2239 }
2240 module_init(netif_init);
2241 
2242 
2243 static void __exit netif_exit(void)
2244 {
2245 	xenbus_unregister_driver(&netfront_driver);
2246 }
2247 module_exit(netif_exit);
2248 
2249 MODULE_DESCRIPTION("Xen virtual network device frontend");
2250 MODULE_LICENSE("GPL");
2251 MODULE_ALIAS("xen:vif");
2252 MODULE_ALIAS("xennet");
2253