xref: /openbmc/linux/drivers/net/xen-netfront.c (revision 5104d265)
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 static const struct ethtool_ops xennet_ethtool_ops;
61 
62 struct netfront_cb {
63 	int pull_to;
64 };
65 
66 #define NETFRONT_SKB_CB(skb)	((struct netfront_cb *)((skb)->cb))
67 
68 #define RX_COPY_THRESHOLD 256
69 
70 #define GRANT_INVALID_REF	0
71 
72 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
73 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
74 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
75 
76 struct netfront_stats {
77 	u64			rx_packets;
78 	u64			tx_packets;
79 	u64			rx_bytes;
80 	u64			tx_bytes;
81 	struct u64_stats_sync	syncp;
82 };
83 
84 struct netfront_info {
85 	struct list_head list;
86 	struct net_device *netdev;
87 
88 	struct napi_struct napi;
89 
90 	/* Split event channels support, tx_* == rx_* when using
91 	 * single event channel.
92 	 */
93 	unsigned int tx_evtchn, rx_evtchn;
94 	unsigned int tx_irq, rx_irq;
95 	/* Only used when split event channels support is enabled */
96 	char tx_irq_name[IFNAMSIZ+4]; /* DEVNAME-tx */
97 	char rx_irq_name[IFNAMSIZ+4]; /* DEVNAME-rx */
98 
99 	struct xenbus_device *xbdev;
100 
101 	spinlock_t   tx_lock;
102 	struct xen_netif_tx_front_ring tx;
103 	int tx_ring_ref;
104 
105 	/*
106 	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
107 	 * are linked from tx_skb_freelist through skb_entry.link.
108 	 *
109 	 *  NB. Freelist index entries are always going to be less than
110 	 *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
111 	 *  greater than PAGE_OFFSET: we use this property to distinguish
112 	 *  them.
113 	 */
114 	union skb_entry {
115 		struct sk_buff *skb;
116 		unsigned long link;
117 	} tx_skbs[NET_TX_RING_SIZE];
118 	grant_ref_t gref_tx_head;
119 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
120 	unsigned tx_skb_freelist;
121 
122 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
123 	struct xen_netif_rx_front_ring rx;
124 	int rx_ring_ref;
125 
126 	/* Receive-ring batched refills. */
127 #define RX_MIN_TARGET 8
128 #define RX_DFL_MIN_TARGET 64
129 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
130 	unsigned rx_min_target, rx_max_target, rx_target;
131 	struct sk_buff_head rx_batch;
132 
133 	struct timer_list rx_refill_timer;
134 
135 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
136 	grant_ref_t gref_rx_head;
137 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
138 
139 	unsigned long rx_pfn_array[NET_RX_RING_SIZE];
140 	struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
141 	struct mmu_update rx_mmu[NET_RX_RING_SIZE];
142 
143 	/* Statistics */
144 	struct netfront_stats __percpu *stats;
145 
146 	unsigned long rx_gso_checksum_fixup;
147 };
148 
149 struct netfront_rx_info {
150 	struct xen_netif_rx_response rx;
151 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
152 };
153 
154 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
155 {
156 	list->link = id;
157 }
158 
159 static int skb_entry_is_link(const union skb_entry *list)
160 {
161 	BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
162 	return (unsigned long)list->skb < PAGE_OFFSET;
163 }
164 
165 /*
166  * Access macros for acquiring freeing slots in tx_skbs[].
167  */
168 
169 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
170 			       unsigned short id)
171 {
172 	skb_entry_set_link(&list[id], *head);
173 	*head = id;
174 }
175 
176 static unsigned short get_id_from_freelist(unsigned *head,
177 					   union skb_entry *list)
178 {
179 	unsigned int id = *head;
180 	*head = list[id].link;
181 	return id;
182 }
183 
184 static int xennet_rxidx(RING_IDX idx)
185 {
186 	return idx & (NET_RX_RING_SIZE - 1);
187 }
188 
189 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
190 					 RING_IDX ri)
191 {
192 	int i = xennet_rxidx(ri);
193 	struct sk_buff *skb = np->rx_skbs[i];
194 	np->rx_skbs[i] = NULL;
195 	return skb;
196 }
197 
198 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
199 					    RING_IDX ri)
200 {
201 	int i = xennet_rxidx(ri);
202 	grant_ref_t ref = np->grant_rx_ref[i];
203 	np->grant_rx_ref[i] = GRANT_INVALID_REF;
204 	return ref;
205 }
206 
207 #ifdef CONFIG_SYSFS
208 static int xennet_sysfs_addif(struct net_device *netdev);
209 static void xennet_sysfs_delif(struct net_device *netdev);
210 #else /* !CONFIG_SYSFS */
211 #define xennet_sysfs_addif(dev) (0)
212 #define xennet_sysfs_delif(dev) do { } while (0)
213 #endif
214 
215 static bool xennet_can_sg(struct net_device *dev)
216 {
217 	return dev->features & NETIF_F_SG;
218 }
219 
220 
221 static void rx_refill_timeout(unsigned long data)
222 {
223 	struct net_device *dev = (struct net_device *)data;
224 	struct netfront_info *np = netdev_priv(dev);
225 	napi_schedule(&np->napi);
226 }
227 
228 static int netfront_tx_slot_available(struct netfront_info *np)
229 {
230 	return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
231 		(TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
232 }
233 
234 static void xennet_maybe_wake_tx(struct net_device *dev)
235 {
236 	struct netfront_info *np = netdev_priv(dev);
237 
238 	if (unlikely(netif_queue_stopped(dev)) &&
239 	    netfront_tx_slot_available(np) &&
240 	    likely(netif_running(dev)))
241 		netif_wake_queue(dev);
242 }
243 
244 static void xennet_alloc_rx_buffers(struct net_device *dev)
245 {
246 	unsigned short id;
247 	struct netfront_info *np = netdev_priv(dev);
248 	struct sk_buff *skb;
249 	struct page *page;
250 	int i, batch_target, notify;
251 	RING_IDX req_prod = np->rx.req_prod_pvt;
252 	grant_ref_t ref;
253 	unsigned long pfn;
254 	void *vaddr;
255 	struct xen_netif_rx_request *req;
256 
257 	if (unlikely(!netif_carrier_ok(dev)))
258 		return;
259 
260 	/*
261 	 * Allocate skbuffs greedily, even though we batch updates to the
262 	 * receive ring. This creates a less bursty demand on the memory
263 	 * allocator, so should reduce the chance of failed allocation requests
264 	 * both for ourself and for other kernel subsystems.
265 	 */
266 	batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
267 	for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
268 		skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
269 					 GFP_ATOMIC | __GFP_NOWARN);
270 		if (unlikely(!skb))
271 			goto no_skb;
272 
273 		/* Align ip header to a 16 bytes boundary */
274 		skb_reserve(skb, NET_IP_ALIGN);
275 
276 		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
277 		if (!page) {
278 			kfree_skb(skb);
279 no_skb:
280 			/* Any skbuffs queued for refill? Force them out. */
281 			if (i != 0)
282 				goto refill;
283 			/* Could not allocate any skbuffs. Try again later. */
284 			mod_timer(&np->rx_refill_timer,
285 				  jiffies + (HZ/10));
286 			break;
287 		}
288 
289 		skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
290 		__skb_queue_tail(&np->rx_batch, skb);
291 	}
292 
293 	/* Is the batch large enough to be worthwhile? */
294 	if (i < (np->rx_target/2)) {
295 		if (req_prod > np->rx.sring->req_prod)
296 			goto push;
297 		return;
298 	}
299 
300 	/* Adjust our fill target if we risked running out of buffers. */
301 	if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
302 	    ((np->rx_target *= 2) > np->rx_max_target))
303 		np->rx_target = np->rx_max_target;
304 
305  refill:
306 	for (i = 0; ; i++) {
307 		skb = __skb_dequeue(&np->rx_batch);
308 		if (skb == NULL)
309 			break;
310 
311 		skb->dev = dev;
312 
313 		id = xennet_rxidx(req_prod + i);
314 
315 		BUG_ON(np->rx_skbs[id]);
316 		np->rx_skbs[id] = skb;
317 
318 		ref = gnttab_claim_grant_reference(&np->gref_rx_head);
319 		BUG_ON((signed short)ref < 0);
320 		np->grant_rx_ref[id] = ref;
321 
322 		pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
323 		vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
324 
325 		req = RING_GET_REQUEST(&np->rx, req_prod + i);
326 		gnttab_grant_foreign_access_ref(ref,
327 						np->xbdev->otherend_id,
328 						pfn_to_mfn(pfn),
329 						0);
330 
331 		req->id = id;
332 		req->gref = ref;
333 	}
334 
335 	wmb();		/* barrier so backend seens requests */
336 
337 	/* Above is a suitable barrier to ensure backend will see requests. */
338 	np->rx.req_prod_pvt = req_prod + i;
339  push:
340 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
341 	if (notify)
342 		notify_remote_via_irq(np->rx_irq);
343 }
344 
345 static int xennet_open(struct net_device *dev)
346 {
347 	struct netfront_info *np = netdev_priv(dev);
348 
349 	napi_enable(&np->napi);
350 
351 	spin_lock_bh(&np->rx_lock);
352 	if (netif_carrier_ok(dev)) {
353 		xennet_alloc_rx_buffers(dev);
354 		np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
355 		if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
356 			napi_schedule(&np->napi);
357 	}
358 	spin_unlock_bh(&np->rx_lock);
359 
360 	netif_start_queue(dev);
361 
362 	return 0;
363 }
364 
365 static void xennet_tx_buf_gc(struct net_device *dev)
366 {
367 	RING_IDX cons, prod;
368 	unsigned short id;
369 	struct netfront_info *np = netdev_priv(dev);
370 	struct sk_buff *skb;
371 
372 	BUG_ON(!netif_carrier_ok(dev));
373 
374 	do {
375 		prod = np->tx.sring->rsp_prod;
376 		rmb(); /* Ensure we see responses up to 'rp'. */
377 
378 		for (cons = np->tx.rsp_cons; cons != prod; cons++) {
379 			struct xen_netif_tx_response *txrsp;
380 
381 			txrsp = RING_GET_RESPONSE(&np->tx, cons);
382 			if (txrsp->status == XEN_NETIF_RSP_NULL)
383 				continue;
384 
385 			id  = txrsp->id;
386 			skb = np->tx_skbs[id].skb;
387 			if (unlikely(gnttab_query_foreign_access(
388 				np->grant_tx_ref[id]) != 0)) {
389 				pr_alert("%s: warning -- grant still in use by backend domain\n",
390 					 __func__);
391 				BUG();
392 			}
393 			gnttab_end_foreign_access_ref(
394 				np->grant_tx_ref[id], GNTMAP_readonly);
395 			gnttab_release_grant_reference(
396 				&np->gref_tx_head, np->grant_tx_ref[id]);
397 			np->grant_tx_ref[id] = GRANT_INVALID_REF;
398 			add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
399 			dev_kfree_skb_irq(skb);
400 		}
401 
402 		np->tx.rsp_cons = prod;
403 
404 		/*
405 		 * Set a new event, then check for race with update of tx_cons.
406 		 * Note that it is essential to schedule a callback, no matter
407 		 * how few buffers are pending. Even if there is space in the
408 		 * transmit ring, higher layers may be blocked because too much
409 		 * data is outstanding: in such cases notification from Xen is
410 		 * likely to be the only kick that we'll get.
411 		 */
412 		np->tx.sring->rsp_event =
413 			prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
414 		mb();		/* update shared area */
415 	} while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
416 
417 	xennet_maybe_wake_tx(dev);
418 }
419 
420 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
421 			      struct xen_netif_tx_request *tx)
422 {
423 	struct netfront_info *np = netdev_priv(dev);
424 	char *data = skb->data;
425 	unsigned long mfn;
426 	RING_IDX prod = np->tx.req_prod_pvt;
427 	int frags = skb_shinfo(skb)->nr_frags;
428 	unsigned int offset = offset_in_page(data);
429 	unsigned int len = skb_headlen(skb);
430 	unsigned int id;
431 	grant_ref_t ref;
432 	int i;
433 
434 	/* While the header overlaps a page boundary (including being
435 	   larger than a page), split it it into page-sized chunks. */
436 	while (len > PAGE_SIZE - offset) {
437 		tx->size = PAGE_SIZE - offset;
438 		tx->flags |= XEN_NETTXF_more_data;
439 		len -= tx->size;
440 		data += tx->size;
441 		offset = 0;
442 
443 		id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
444 		np->tx_skbs[id].skb = skb_get(skb);
445 		tx = RING_GET_REQUEST(&np->tx, prod++);
446 		tx->id = id;
447 		ref = gnttab_claim_grant_reference(&np->gref_tx_head);
448 		BUG_ON((signed short)ref < 0);
449 
450 		mfn = virt_to_mfn(data);
451 		gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
452 						mfn, GNTMAP_readonly);
453 
454 		tx->gref = np->grant_tx_ref[id] = ref;
455 		tx->offset = offset;
456 		tx->size = len;
457 		tx->flags = 0;
458 	}
459 
460 	/* Grant backend access to each skb fragment page. */
461 	for (i = 0; i < frags; i++) {
462 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
463 		struct page *page = skb_frag_page(frag);
464 
465 		len = skb_frag_size(frag);
466 		offset = frag->page_offset;
467 
468 		/* Data must not cross a page boundary. */
469 		BUG_ON(len + offset > PAGE_SIZE<<compound_order(page));
470 
471 		/* Skip unused frames from start of page */
472 		page += offset >> PAGE_SHIFT;
473 		offset &= ~PAGE_MASK;
474 
475 		while (len > 0) {
476 			unsigned long bytes;
477 
478 			BUG_ON(offset >= PAGE_SIZE);
479 
480 			bytes = PAGE_SIZE - offset;
481 			if (bytes > len)
482 				bytes = len;
483 
484 			tx->flags |= XEN_NETTXF_more_data;
485 
486 			id = get_id_from_freelist(&np->tx_skb_freelist,
487 						  np->tx_skbs);
488 			np->tx_skbs[id].skb = skb_get(skb);
489 			tx = RING_GET_REQUEST(&np->tx, prod++);
490 			tx->id = id;
491 			ref = gnttab_claim_grant_reference(&np->gref_tx_head);
492 			BUG_ON((signed short)ref < 0);
493 
494 			mfn = pfn_to_mfn(page_to_pfn(page));
495 			gnttab_grant_foreign_access_ref(ref,
496 							np->xbdev->otherend_id,
497 							mfn, GNTMAP_readonly);
498 
499 			tx->gref = np->grant_tx_ref[id] = ref;
500 			tx->offset = offset;
501 			tx->size = bytes;
502 			tx->flags = 0;
503 
504 			offset += bytes;
505 			len -= bytes;
506 
507 			/* Next frame */
508 			if (offset == PAGE_SIZE && len) {
509 				BUG_ON(!PageCompound(page));
510 				page++;
511 				offset = 0;
512 			}
513 		}
514 	}
515 
516 	np->tx.req_prod_pvt = prod;
517 }
518 
519 /*
520  * Count how many ring slots are required to send the frags of this
521  * skb. Each frag might be a compound page.
522  */
523 static int xennet_count_skb_frag_slots(struct sk_buff *skb)
524 {
525 	int i, frags = skb_shinfo(skb)->nr_frags;
526 	int pages = 0;
527 
528 	for (i = 0; i < frags; i++) {
529 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
530 		unsigned long size = skb_frag_size(frag);
531 		unsigned long offset = frag->page_offset;
532 
533 		/* Skip unused frames from start of page */
534 		offset &= ~PAGE_MASK;
535 
536 		pages += PFN_UP(offset + size);
537 	}
538 
539 	return pages;
540 }
541 
542 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
543 {
544 	unsigned short id;
545 	struct netfront_info *np = netdev_priv(dev);
546 	struct netfront_stats *stats = this_cpu_ptr(np->stats);
547 	struct xen_netif_tx_request *tx;
548 	char *data = skb->data;
549 	RING_IDX i;
550 	grant_ref_t ref;
551 	unsigned long mfn;
552 	int notify;
553 	int slots;
554 	unsigned int offset = offset_in_page(data);
555 	unsigned int len = skb_headlen(skb);
556 	unsigned long flags;
557 
558 	/* If skb->len is too big for wire format, drop skb and alert
559 	 * user about misconfiguration.
560 	 */
561 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
562 		net_alert_ratelimited(
563 			"xennet: skb->len = %u, too big for wire format\n",
564 			skb->len);
565 		goto drop;
566 	}
567 
568 	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) +
569 		xennet_count_skb_frag_slots(skb);
570 	if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
571 		net_alert_ratelimited(
572 			"xennet: skb rides the rocket: %d slots\n", slots);
573 		goto drop;
574 	}
575 
576 	spin_lock_irqsave(&np->tx_lock, flags);
577 
578 	if (unlikely(!netif_carrier_ok(dev) ||
579 		     (slots > 1 && !xennet_can_sg(dev)) ||
580 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
581 		spin_unlock_irqrestore(&np->tx_lock, flags);
582 		goto drop;
583 	}
584 
585 	i = np->tx.req_prod_pvt;
586 
587 	id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
588 	np->tx_skbs[id].skb = skb;
589 
590 	tx = RING_GET_REQUEST(&np->tx, i);
591 
592 	tx->id   = id;
593 	ref = gnttab_claim_grant_reference(&np->gref_tx_head);
594 	BUG_ON((signed short)ref < 0);
595 	mfn = virt_to_mfn(data);
596 	gnttab_grant_foreign_access_ref(
597 		ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
598 	tx->gref = np->grant_tx_ref[id] = ref;
599 	tx->offset = offset;
600 	tx->size = len;
601 
602 	tx->flags = 0;
603 	if (skb->ip_summed == CHECKSUM_PARTIAL)
604 		/* local packet? */
605 		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
606 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
607 		/* remote but checksummed. */
608 		tx->flags |= XEN_NETTXF_data_validated;
609 
610 	if (skb_shinfo(skb)->gso_size) {
611 		struct xen_netif_extra_info *gso;
612 
613 		gso = (struct xen_netif_extra_info *)
614 			RING_GET_REQUEST(&np->tx, ++i);
615 
616 		tx->flags |= XEN_NETTXF_extra_info;
617 
618 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
619 		gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
620 		gso->u.gso.pad = 0;
621 		gso->u.gso.features = 0;
622 
623 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
624 		gso->flags = 0;
625 	}
626 
627 	np->tx.req_prod_pvt = i + 1;
628 
629 	xennet_make_frags(skb, dev, tx);
630 	tx->size = skb->len;
631 
632 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
633 	if (notify)
634 		notify_remote_via_irq(np->tx_irq);
635 
636 	u64_stats_update_begin(&stats->syncp);
637 	stats->tx_bytes += skb->len;
638 	stats->tx_packets++;
639 	u64_stats_update_end(&stats->syncp);
640 
641 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
642 	xennet_tx_buf_gc(dev);
643 
644 	if (!netfront_tx_slot_available(np))
645 		netif_stop_queue(dev);
646 
647 	spin_unlock_irqrestore(&np->tx_lock, flags);
648 
649 	return NETDEV_TX_OK;
650 
651  drop:
652 	dev->stats.tx_dropped++;
653 	dev_kfree_skb(skb);
654 	return NETDEV_TX_OK;
655 }
656 
657 static int xennet_close(struct net_device *dev)
658 {
659 	struct netfront_info *np = netdev_priv(dev);
660 	netif_stop_queue(np->netdev);
661 	napi_disable(&np->napi);
662 	return 0;
663 }
664 
665 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
666 				grant_ref_t ref)
667 {
668 	int new = xennet_rxidx(np->rx.req_prod_pvt);
669 
670 	BUG_ON(np->rx_skbs[new]);
671 	np->rx_skbs[new] = skb;
672 	np->grant_rx_ref[new] = ref;
673 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
674 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
675 	np->rx.req_prod_pvt++;
676 }
677 
678 static int xennet_get_extras(struct netfront_info *np,
679 			     struct xen_netif_extra_info *extras,
680 			     RING_IDX rp)
681 
682 {
683 	struct xen_netif_extra_info *extra;
684 	struct device *dev = &np->netdev->dev;
685 	RING_IDX cons = np->rx.rsp_cons;
686 	int err = 0;
687 
688 	do {
689 		struct sk_buff *skb;
690 		grant_ref_t ref;
691 
692 		if (unlikely(cons + 1 == rp)) {
693 			if (net_ratelimit())
694 				dev_warn(dev, "Missing extra info\n");
695 			err = -EBADR;
696 			break;
697 		}
698 
699 		extra = (struct xen_netif_extra_info *)
700 			RING_GET_RESPONSE(&np->rx, ++cons);
701 
702 		if (unlikely(!extra->type ||
703 			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
704 			if (net_ratelimit())
705 				dev_warn(dev, "Invalid extra type: %d\n",
706 					extra->type);
707 			err = -EINVAL;
708 		} else {
709 			memcpy(&extras[extra->type - 1], extra,
710 			       sizeof(*extra));
711 		}
712 
713 		skb = xennet_get_rx_skb(np, cons);
714 		ref = xennet_get_rx_ref(np, cons);
715 		xennet_move_rx_slot(np, skb, ref);
716 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
717 
718 	np->rx.rsp_cons = cons;
719 	return err;
720 }
721 
722 static int xennet_get_responses(struct netfront_info *np,
723 				struct netfront_rx_info *rinfo, RING_IDX rp,
724 				struct sk_buff_head *list)
725 {
726 	struct xen_netif_rx_response *rx = &rinfo->rx;
727 	struct xen_netif_extra_info *extras = rinfo->extras;
728 	struct device *dev = &np->netdev->dev;
729 	RING_IDX cons = np->rx.rsp_cons;
730 	struct sk_buff *skb = xennet_get_rx_skb(np, cons);
731 	grant_ref_t ref = xennet_get_rx_ref(np, cons);
732 	int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
733 	int slots = 1;
734 	int err = 0;
735 	unsigned long ret;
736 
737 	if (rx->flags & XEN_NETRXF_extra_info) {
738 		err = xennet_get_extras(np, extras, rp);
739 		cons = np->rx.rsp_cons;
740 	}
741 
742 	for (;;) {
743 		if (unlikely(rx->status < 0 ||
744 			     rx->offset + rx->status > PAGE_SIZE)) {
745 			if (net_ratelimit())
746 				dev_warn(dev, "rx->offset: %x, size: %u\n",
747 					 rx->offset, rx->status);
748 			xennet_move_rx_slot(np, skb, ref);
749 			err = -EINVAL;
750 			goto next;
751 		}
752 
753 		/*
754 		 * This definitely indicates a bug, either in this driver or in
755 		 * the backend driver. In future this should flag the bad
756 		 * situation to the system controller to reboot the backend.
757 		 */
758 		if (ref == GRANT_INVALID_REF) {
759 			if (net_ratelimit())
760 				dev_warn(dev, "Bad rx response id %d.\n",
761 					 rx->id);
762 			err = -EINVAL;
763 			goto next;
764 		}
765 
766 		ret = gnttab_end_foreign_access_ref(ref, 0);
767 		BUG_ON(!ret);
768 
769 		gnttab_release_grant_reference(&np->gref_rx_head, ref);
770 
771 		__skb_queue_tail(list, skb);
772 
773 next:
774 		if (!(rx->flags & XEN_NETRXF_more_data))
775 			break;
776 
777 		if (cons + slots == rp) {
778 			if (net_ratelimit())
779 				dev_warn(dev, "Need more slots\n");
780 			err = -ENOENT;
781 			break;
782 		}
783 
784 		rx = RING_GET_RESPONSE(&np->rx, cons + slots);
785 		skb = xennet_get_rx_skb(np, cons + slots);
786 		ref = xennet_get_rx_ref(np, cons + slots);
787 		slots++;
788 	}
789 
790 	if (unlikely(slots > max)) {
791 		if (net_ratelimit())
792 			dev_warn(dev, "Too many slots\n");
793 		err = -E2BIG;
794 	}
795 
796 	if (unlikely(err))
797 		np->rx.rsp_cons = cons + slots;
798 
799 	return err;
800 }
801 
802 static int xennet_set_skb_gso(struct sk_buff *skb,
803 			      struct xen_netif_extra_info *gso)
804 {
805 	if (!gso->u.gso.size) {
806 		if (net_ratelimit())
807 			pr_warn("GSO size must not be zero\n");
808 		return -EINVAL;
809 	}
810 
811 	/* Currently only TCPv4 S.O. is supported. */
812 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
813 		if (net_ratelimit())
814 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
815 		return -EINVAL;
816 	}
817 
818 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
819 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
820 
821 	/* Header must be checked, and gso_segs computed. */
822 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
823 	skb_shinfo(skb)->gso_segs = 0;
824 
825 	return 0;
826 }
827 
828 static RING_IDX xennet_fill_frags(struct netfront_info *np,
829 				  struct sk_buff *skb,
830 				  struct sk_buff_head *list)
831 {
832 	struct skb_shared_info *shinfo = skb_shinfo(skb);
833 	RING_IDX cons = np->rx.rsp_cons;
834 	struct sk_buff *nskb;
835 
836 	while ((nskb = __skb_dequeue(list))) {
837 		struct xen_netif_rx_response *rx =
838 			RING_GET_RESPONSE(&np->rx, ++cons);
839 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
840 
841 		if (shinfo->nr_frags == MAX_SKB_FRAGS) {
842 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
843 
844 			BUG_ON(pull_to <= skb_headlen(skb));
845 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
846 		}
847 		BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
848 
849 		skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
850 				rx->offset, rx->status, PAGE_SIZE);
851 
852 		skb_shinfo(nskb)->nr_frags = 0;
853 		kfree_skb(nskb);
854 	}
855 
856 	return cons;
857 }
858 
859 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
860 {
861 	struct iphdr *iph;
862 	int err = -EPROTO;
863 	int recalculate_partial_csum = 0;
864 
865 	/*
866 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
867 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
868 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
869 	 * recalculate the partial checksum.
870 	 */
871 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
872 		struct netfront_info *np = netdev_priv(dev);
873 		np->rx_gso_checksum_fixup++;
874 		skb->ip_summed = CHECKSUM_PARTIAL;
875 		recalculate_partial_csum = 1;
876 	}
877 
878 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
879 	if (skb->ip_summed != CHECKSUM_PARTIAL)
880 		return 0;
881 
882 	if (skb->protocol != htons(ETH_P_IP))
883 		goto out;
884 
885 	iph = (void *)skb->data;
886 
887 	switch (iph->protocol) {
888 	case IPPROTO_TCP:
889 		if (!skb_partial_csum_set(skb, 4 * iph->ihl,
890 					  offsetof(struct tcphdr, check)))
891 			goto out;
892 
893 		if (recalculate_partial_csum) {
894 			struct tcphdr *tcph = tcp_hdr(skb);
895 			tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
896 							 skb->len - iph->ihl*4,
897 							 IPPROTO_TCP, 0);
898 		}
899 		break;
900 	case IPPROTO_UDP:
901 		if (!skb_partial_csum_set(skb, 4 * iph->ihl,
902 					  offsetof(struct udphdr, check)))
903 			goto out;
904 
905 		if (recalculate_partial_csum) {
906 			struct udphdr *udph = udp_hdr(skb);
907 			udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
908 							 skb->len - iph->ihl*4,
909 							 IPPROTO_UDP, 0);
910 		}
911 		break;
912 	default:
913 		if (net_ratelimit())
914 			pr_err("Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
915 			       iph->protocol);
916 		goto out;
917 	}
918 
919 	err = 0;
920 
921 out:
922 	return err;
923 }
924 
925 static int handle_incoming_queue(struct net_device *dev,
926 				 struct sk_buff_head *rxq)
927 {
928 	struct netfront_info *np = netdev_priv(dev);
929 	struct netfront_stats *stats = this_cpu_ptr(np->stats);
930 	int packets_dropped = 0;
931 	struct sk_buff *skb;
932 
933 	while ((skb = __skb_dequeue(rxq)) != NULL) {
934 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
935 
936 		if (pull_to > skb_headlen(skb))
937 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
938 
939 		/* Ethernet work: Delayed to here as it peeks the header. */
940 		skb->protocol = eth_type_trans(skb, dev);
941 
942 		if (checksum_setup(dev, skb)) {
943 			kfree_skb(skb);
944 			packets_dropped++;
945 			dev->stats.rx_errors++;
946 			continue;
947 		}
948 
949 		u64_stats_update_begin(&stats->syncp);
950 		stats->rx_packets++;
951 		stats->rx_bytes += skb->len;
952 		u64_stats_update_end(&stats->syncp);
953 
954 		/* Pass it up. */
955 		netif_receive_skb(skb);
956 	}
957 
958 	return packets_dropped;
959 }
960 
961 static int xennet_poll(struct napi_struct *napi, int budget)
962 {
963 	struct netfront_info *np = container_of(napi, struct netfront_info, napi);
964 	struct net_device *dev = np->netdev;
965 	struct sk_buff *skb;
966 	struct netfront_rx_info rinfo;
967 	struct xen_netif_rx_response *rx = &rinfo.rx;
968 	struct xen_netif_extra_info *extras = rinfo.extras;
969 	RING_IDX i, rp;
970 	int work_done;
971 	struct sk_buff_head rxq;
972 	struct sk_buff_head errq;
973 	struct sk_buff_head tmpq;
974 	unsigned long flags;
975 	int err;
976 
977 	spin_lock(&np->rx_lock);
978 
979 	skb_queue_head_init(&rxq);
980 	skb_queue_head_init(&errq);
981 	skb_queue_head_init(&tmpq);
982 
983 	rp = np->rx.sring->rsp_prod;
984 	rmb(); /* Ensure we see queued responses up to 'rp'. */
985 
986 	i = np->rx.rsp_cons;
987 	work_done = 0;
988 	while ((i != rp) && (work_done < budget)) {
989 		memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
990 		memset(extras, 0, sizeof(rinfo.extras));
991 
992 		err = xennet_get_responses(np, &rinfo, rp, &tmpq);
993 
994 		if (unlikely(err)) {
995 err:
996 			while ((skb = __skb_dequeue(&tmpq)))
997 				__skb_queue_tail(&errq, skb);
998 			dev->stats.rx_errors++;
999 			i = np->rx.rsp_cons;
1000 			continue;
1001 		}
1002 
1003 		skb = __skb_dequeue(&tmpq);
1004 
1005 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1006 			struct xen_netif_extra_info *gso;
1007 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1008 
1009 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1010 				__skb_queue_head(&tmpq, skb);
1011 				np->rx.rsp_cons += skb_queue_len(&tmpq);
1012 				goto err;
1013 			}
1014 		}
1015 
1016 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1017 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1018 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1019 
1020 		skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1021 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1022 		skb->data_len = rx->status;
1023 		skb->len += rx->status;
1024 
1025 		i = xennet_fill_frags(np, skb, &tmpq);
1026 
1027 		if (rx->flags & XEN_NETRXF_csum_blank)
1028 			skb->ip_summed = CHECKSUM_PARTIAL;
1029 		else if (rx->flags & XEN_NETRXF_data_validated)
1030 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1031 
1032 		__skb_queue_tail(&rxq, skb);
1033 
1034 		np->rx.rsp_cons = ++i;
1035 		work_done++;
1036 	}
1037 
1038 	__skb_queue_purge(&errq);
1039 
1040 	work_done -= handle_incoming_queue(dev, &rxq);
1041 
1042 	/* If we get a callback with very few responses, reduce fill target. */
1043 	/* NB. Note exponential increase, linear decrease. */
1044 	if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1045 	     ((3*np->rx_target) / 4)) &&
1046 	    (--np->rx_target < np->rx_min_target))
1047 		np->rx_target = np->rx_min_target;
1048 
1049 	xennet_alloc_rx_buffers(dev);
1050 
1051 	if (work_done < budget) {
1052 		int more_to_do = 0;
1053 
1054 		local_irq_save(flags);
1055 
1056 		RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1057 		if (!more_to_do)
1058 			__napi_complete(napi);
1059 
1060 		local_irq_restore(flags);
1061 	}
1062 
1063 	spin_unlock(&np->rx_lock);
1064 
1065 	return work_done;
1066 }
1067 
1068 static int xennet_change_mtu(struct net_device *dev, int mtu)
1069 {
1070 	int max = xennet_can_sg(dev) ?
1071 		XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1072 
1073 	if (mtu > max)
1074 		return -EINVAL;
1075 	dev->mtu = mtu;
1076 	return 0;
1077 }
1078 
1079 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1080 						    struct rtnl_link_stats64 *tot)
1081 {
1082 	struct netfront_info *np = netdev_priv(dev);
1083 	int cpu;
1084 
1085 	for_each_possible_cpu(cpu) {
1086 		struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1087 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1088 		unsigned int start;
1089 
1090 		do {
1091 			start = u64_stats_fetch_begin_bh(&stats->syncp);
1092 
1093 			rx_packets = stats->rx_packets;
1094 			tx_packets = stats->tx_packets;
1095 			rx_bytes = stats->rx_bytes;
1096 			tx_bytes = stats->tx_bytes;
1097 		} while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1098 
1099 		tot->rx_packets += rx_packets;
1100 		tot->tx_packets += tx_packets;
1101 		tot->rx_bytes   += rx_bytes;
1102 		tot->tx_bytes   += tx_bytes;
1103 	}
1104 
1105 	tot->rx_errors  = dev->stats.rx_errors;
1106 	tot->tx_dropped = dev->stats.tx_dropped;
1107 
1108 	return tot;
1109 }
1110 
1111 static void xennet_release_tx_bufs(struct netfront_info *np)
1112 {
1113 	struct sk_buff *skb;
1114 	int i;
1115 
1116 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1117 		/* Skip over entries which are actually freelist references */
1118 		if (skb_entry_is_link(&np->tx_skbs[i]))
1119 			continue;
1120 
1121 		skb = np->tx_skbs[i].skb;
1122 		gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1123 					      GNTMAP_readonly);
1124 		gnttab_release_grant_reference(&np->gref_tx_head,
1125 					       np->grant_tx_ref[i]);
1126 		np->grant_tx_ref[i] = GRANT_INVALID_REF;
1127 		add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1128 		dev_kfree_skb_irq(skb);
1129 	}
1130 }
1131 
1132 static void xennet_release_rx_bufs(struct netfront_info *np)
1133 {
1134 	struct mmu_update      *mmu = np->rx_mmu;
1135 	struct multicall_entry *mcl = np->rx_mcl;
1136 	struct sk_buff_head free_list;
1137 	struct sk_buff *skb;
1138 	unsigned long mfn;
1139 	int xfer = 0, noxfer = 0, unused = 0;
1140 	int id, ref;
1141 
1142 	dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1143 			 __func__);
1144 	return;
1145 
1146 	skb_queue_head_init(&free_list);
1147 
1148 	spin_lock_bh(&np->rx_lock);
1149 
1150 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1151 		ref = np->grant_rx_ref[id];
1152 		if (ref == GRANT_INVALID_REF) {
1153 			unused++;
1154 			continue;
1155 		}
1156 
1157 		skb = np->rx_skbs[id];
1158 		mfn = gnttab_end_foreign_transfer_ref(ref);
1159 		gnttab_release_grant_reference(&np->gref_rx_head, ref);
1160 		np->grant_rx_ref[id] = GRANT_INVALID_REF;
1161 
1162 		if (0 == mfn) {
1163 			skb_shinfo(skb)->nr_frags = 0;
1164 			dev_kfree_skb(skb);
1165 			noxfer++;
1166 			continue;
1167 		}
1168 
1169 		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1170 			/* Remap the page. */
1171 			const struct page *page =
1172 				skb_frag_page(&skb_shinfo(skb)->frags[0]);
1173 			unsigned long pfn = page_to_pfn(page);
1174 			void *vaddr = page_address(page);
1175 
1176 			MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1177 						mfn_pte(mfn, PAGE_KERNEL),
1178 						0);
1179 			mcl++;
1180 			mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1181 				| MMU_MACHPHYS_UPDATE;
1182 			mmu->val = pfn;
1183 			mmu++;
1184 
1185 			set_phys_to_machine(pfn, mfn);
1186 		}
1187 		__skb_queue_tail(&free_list, skb);
1188 		xfer++;
1189 	}
1190 
1191 	dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1192 		 __func__, xfer, noxfer, unused);
1193 
1194 	if (xfer) {
1195 		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1196 			/* Do all the remapping work and M2P updates. */
1197 			MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1198 					 NULL, DOMID_SELF);
1199 			mcl++;
1200 			HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1201 		}
1202 	}
1203 
1204 	__skb_queue_purge(&free_list);
1205 
1206 	spin_unlock_bh(&np->rx_lock);
1207 }
1208 
1209 static void xennet_uninit(struct net_device *dev)
1210 {
1211 	struct netfront_info *np = netdev_priv(dev);
1212 	xennet_release_tx_bufs(np);
1213 	xennet_release_rx_bufs(np);
1214 	gnttab_free_grant_references(np->gref_tx_head);
1215 	gnttab_free_grant_references(np->gref_rx_head);
1216 }
1217 
1218 static netdev_features_t xennet_fix_features(struct net_device *dev,
1219 	netdev_features_t features)
1220 {
1221 	struct netfront_info *np = netdev_priv(dev);
1222 	int val;
1223 
1224 	if (features & NETIF_F_SG) {
1225 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1226 				 "%d", &val) < 0)
1227 			val = 0;
1228 
1229 		if (!val)
1230 			features &= ~NETIF_F_SG;
1231 	}
1232 
1233 	if (features & NETIF_F_TSO) {
1234 		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1235 				 "feature-gso-tcpv4", "%d", &val) < 0)
1236 			val = 0;
1237 
1238 		if (!val)
1239 			features &= ~NETIF_F_TSO;
1240 	}
1241 
1242 	return features;
1243 }
1244 
1245 static int xennet_set_features(struct net_device *dev,
1246 	netdev_features_t features)
1247 {
1248 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1249 		netdev_info(dev, "Reducing MTU because no SG offload");
1250 		dev->mtu = ETH_DATA_LEN;
1251 	}
1252 
1253 	return 0;
1254 }
1255 
1256 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1257 {
1258 	struct netfront_info *np = dev_id;
1259 	struct net_device *dev = np->netdev;
1260 	unsigned long flags;
1261 
1262 	spin_lock_irqsave(&np->tx_lock, flags);
1263 	xennet_tx_buf_gc(dev);
1264 	spin_unlock_irqrestore(&np->tx_lock, flags);
1265 
1266 	return IRQ_HANDLED;
1267 }
1268 
1269 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1270 {
1271 	struct netfront_info *np = dev_id;
1272 	struct net_device *dev = np->netdev;
1273 
1274 	if (likely(netif_carrier_ok(dev) &&
1275 		   RING_HAS_UNCONSUMED_RESPONSES(&np->rx)))
1276 			napi_schedule(&np->napi);
1277 
1278 	return IRQ_HANDLED;
1279 }
1280 
1281 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1282 {
1283 	xennet_tx_interrupt(irq, dev_id);
1284 	xennet_rx_interrupt(irq, dev_id);
1285 	return IRQ_HANDLED;
1286 }
1287 
1288 #ifdef CONFIG_NET_POLL_CONTROLLER
1289 static void xennet_poll_controller(struct net_device *dev)
1290 {
1291 	xennet_interrupt(0, dev);
1292 }
1293 #endif
1294 
1295 static const struct net_device_ops xennet_netdev_ops = {
1296 	.ndo_open            = xennet_open,
1297 	.ndo_uninit          = xennet_uninit,
1298 	.ndo_stop            = xennet_close,
1299 	.ndo_start_xmit      = xennet_start_xmit,
1300 	.ndo_change_mtu	     = xennet_change_mtu,
1301 	.ndo_get_stats64     = xennet_get_stats64,
1302 	.ndo_set_mac_address = eth_mac_addr,
1303 	.ndo_validate_addr   = eth_validate_addr,
1304 	.ndo_fix_features    = xennet_fix_features,
1305 	.ndo_set_features    = xennet_set_features,
1306 #ifdef CONFIG_NET_POLL_CONTROLLER
1307 	.ndo_poll_controller = xennet_poll_controller,
1308 #endif
1309 };
1310 
1311 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1312 {
1313 	int i, err;
1314 	struct net_device *netdev;
1315 	struct netfront_info *np;
1316 
1317 	netdev = alloc_etherdev(sizeof(struct netfront_info));
1318 	if (!netdev)
1319 		return ERR_PTR(-ENOMEM);
1320 
1321 	np                   = netdev_priv(netdev);
1322 	np->xbdev            = dev;
1323 
1324 	spin_lock_init(&np->tx_lock);
1325 	spin_lock_init(&np->rx_lock);
1326 
1327 	skb_queue_head_init(&np->rx_batch);
1328 	np->rx_target     = RX_DFL_MIN_TARGET;
1329 	np->rx_min_target = RX_DFL_MIN_TARGET;
1330 	np->rx_max_target = RX_MAX_TARGET;
1331 
1332 	init_timer(&np->rx_refill_timer);
1333 	np->rx_refill_timer.data = (unsigned long)netdev;
1334 	np->rx_refill_timer.function = rx_refill_timeout;
1335 
1336 	err = -ENOMEM;
1337 	np->stats = alloc_percpu(struct netfront_stats);
1338 	if (np->stats == NULL)
1339 		goto exit;
1340 
1341 	/* Initialise tx_skbs as a free chain containing every entry. */
1342 	np->tx_skb_freelist = 0;
1343 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1344 		skb_entry_set_link(&np->tx_skbs[i], i+1);
1345 		np->grant_tx_ref[i] = GRANT_INVALID_REF;
1346 	}
1347 
1348 	/* Clear out rx_skbs */
1349 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1350 		np->rx_skbs[i] = NULL;
1351 		np->grant_rx_ref[i] = GRANT_INVALID_REF;
1352 	}
1353 
1354 	/* A grant for every tx ring slot */
1355 	if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1356 					  &np->gref_tx_head) < 0) {
1357 		pr_alert("can't alloc tx grant refs\n");
1358 		err = -ENOMEM;
1359 		goto exit_free_stats;
1360 	}
1361 	/* A grant for every rx ring slot */
1362 	if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1363 					  &np->gref_rx_head) < 0) {
1364 		pr_alert("can't alloc rx grant refs\n");
1365 		err = -ENOMEM;
1366 		goto exit_free_tx;
1367 	}
1368 
1369 	netdev->netdev_ops	= &xennet_netdev_ops;
1370 
1371 	netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1372 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1373 				  NETIF_F_GSO_ROBUST;
1374 	netdev->hw_features	= NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1375 
1376 	/*
1377          * Assume that all hw features are available for now. This set
1378          * will be adjusted by the call to netdev_update_features() in
1379          * xennet_connect() which is the earliest point where we can
1380          * negotiate with the backend regarding supported features.
1381          */
1382 	netdev->features |= netdev->hw_features;
1383 
1384 	SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1385 	SET_NETDEV_DEV(netdev, &dev->dev);
1386 
1387 	netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1388 
1389 	np->netdev = netdev;
1390 
1391 	netif_carrier_off(netdev);
1392 
1393 	return netdev;
1394 
1395  exit_free_tx:
1396 	gnttab_free_grant_references(np->gref_tx_head);
1397  exit_free_stats:
1398 	free_percpu(np->stats);
1399  exit:
1400 	free_netdev(netdev);
1401 	return ERR_PTR(err);
1402 }
1403 
1404 /**
1405  * Entry point to this code when a new device is created.  Allocate the basic
1406  * structures and the ring buffers for communication with the backend, and
1407  * inform the backend of the appropriate details for those.
1408  */
1409 static int netfront_probe(struct xenbus_device *dev,
1410 			  const struct xenbus_device_id *id)
1411 {
1412 	int err;
1413 	struct net_device *netdev;
1414 	struct netfront_info *info;
1415 
1416 	netdev = xennet_create_dev(dev);
1417 	if (IS_ERR(netdev)) {
1418 		err = PTR_ERR(netdev);
1419 		xenbus_dev_fatal(dev, err, "creating netdev");
1420 		return err;
1421 	}
1422 
1423 	info = netdev_priv(netdev);
1424 	dev_set_drvdata(&dev->dev, info);
1425 
1426 	err = register_netdev(info->netdev);
1427 	if (err) {
1428 		pr_warn("%s: register_netdev err=%d\n", __func__, err);
1429 		goto fail;
1430 	}
1431 
1432 	err = xennet_sysfs_addif(info->netdev);
1433 	if (err) {
1434 		unregister_netdev(info->netdev);
1435 		pr_warn("%s: add sysfs failed err=%d\n", __func__, err);
1436 		goto fail;
1437 	}
1438 
1439 	return 0;
1440 
1441  fail:
1442 	free_netdev(netdev);
1443 	dev_set_drvdata(&dev->dev, NULL);
1444 	return err;
1445 }
1446 
1447 static void xennet_end_access(int ref, void *page)
1448 {
1449 	/* This frees the page as a side-effect */
1450 	if (ref != GRANT_INVALID_REF)
1451 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1452 }
1453 
1454 static void xennet_disconnect_backend(struct netfront_info *info)
1455 {
1456 	/* Stop old i/f to prevent errors whilst we rebuild the state. */
1457 	spin_lock_bh(&info->rx_lock);
1458 	spin_lock_irq(&info->tx_lock);
1459 	netif_carrier_off(info->netdev);
1460 	spin_unlock_irq(&info->tx_lock);
1461 	spin_unlock_bh(&info->rx_lock);
1462 
1463 	if (info->tx_irq && (info->tx_irq == info->rx_irq))
1464 		unbind_from_irqhandler(info->tx_irq, info);
1465 	if (info->tx_irq && (info->tx_irq != info->rx_irq)) {
1466 		unbind_from_irqhandler(info->tx_irq, info);
1467 		unbind_from_irqhandler(info->rx_irq, info);
1468 	}
1469 	info->tx_evtchn = info->rx_evtchn = 0;
1470 	info->tx_irq = info->rx_irq = 0;
1471 
1472 	/* End access and free the pages */
1473 	xennet_end_access(info->tx_ring_ref, info->tx.sring);
1474 	xennet_end_access(info->rx_ring_ref, info->rx.sring);
1475 
1476 	info->tx_ring_ref = GRANT_INVALID_REF;
1477 	info->rx_ring_ref = GRANT_INVALID_REF;
1478 	info->tx.sring = NULL;
1479 	info->rx.sring = NULL;
1480 }
1481 
1482 /**
1483  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1484  * driver restart.  We tear down our netif structure and recreate it, but
1485  * leave the device-layer structures intact so that this is transparent to the
1486  * rest of the kernel.
1487  */
1488 static int netfront_resume(struct xenbus_device *dev)
1489 {
1490 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1491 
1492 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1493 
1494 	xennet_disconnect_backend(info);
1495 	return 0;
1496 }
1497 
1498 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1499 {
1500 	char *s, *e, *macstr;
1501 	int i;
1502 
1503 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1504 	if (IS_ERR(macstr))
1505 		return PTR_ERR(macstr);
1506 
1507 	for (i = 0; i < ETH_ALEN; i++) {
1508 		mac[i] = simple_strtoul(s, &e, 16);
1509 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1510 			kfree(macstr);
1511 			return -ENOENT;
1512 		}
1513 		s = e+1;
1514 	}
1515 
1516 	kfree(macstr);
1517 	return 0;
1518 }
1519 
1520 static int setup_netfront_single(struct netfront_info *info)
1521 {
1522 	int err;
1523 
1524 	err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1525 	if (err < 0)
1526 		goto fail;
1527 
1528 	err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1529 					xennet_interrupt,
1530 					0, info->netdev->name, info);
1531 	if (err < 0)
1532 		goto bind_fail;
1533 	info->rx_evtchn = info->tx_evtchn;
1534 	info->rx_irq = info->tx_irq = err;
1535 
1536 	return 0;
1537 
1538 bind_fail:
1539 	xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1540 	info->tx_evtchn = 0;
1541 fail:
1542 	return err;
1543 }
1544 
1545 static int setup_netfront_split(struct netfront_info *info)
1546 {
1547 	int err;
1548 
1549 	err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1550 	if (err < 0)
1551 		goto fail;
1552 	err = xenbus_alloc_evtchn(info->xbdev, &info->rx_evtchn);
1553 	if (err < 0)
1554 		goto alloc_rx_evtchn_fail;
1555 
1556 	snprintf(info->tx_irq_name, sizeof(info->tx_irq_name),
1557 		 "%s-tx", info->netdev->name);
1558 	err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1559 					xennet_tx_interrupt,
1560 					0, info->tx_irq_name, info);
1561 	if (err < 0)
1562 		goto bind_tx_fail;
1563 	info->tx_irq = err;
1564 
1565 	snprintf(info->rx_irq_name, sizeof(info->rx_irq_name),
1566 		 "%s-rx", info->netdev->name);
1567 	err = bind_evtchn_to_irqhandler(info->rx_evtchn,
1568 					xennet_rx_interrupt,
1569 					0, info->rx_irq_name, info);
1570 	if (err < 0)
1571 		goto bind_rx_fail;
1572 	info->rx_irq = err;
1573 
1574 	return 0;
1575 
1576 bind_rx_fail:
1577 	unbind_from_irqhandler(info->tx_irq, info);
1578 	info->tx_irq = 0;
1579 bind_tx_fail:
1580 	xenbus_free_evtchn(info->xbdev, info->rx_evtchn);
1581 	info->rx_evtchn = 0;
1582 alloc_rx_evtchn_fail:
1583 	xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1584 	info->tx_evtchn = 0;
1585 fail:
1586 	return err;
1587 }
1588 
1589 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1590 {
1591 	struct xen_netif_tx_sring *txs;
1592 	struct xen_netif_rx_sring *rxs;
1593 	int err;
1594 	struct net_device *netdev = info->netdev;
1595 	unsigned int feature_split_evtchn;
1596 
1597 	info->tx_ring_ref = GRANT_INVALID_REF;
1598 	info->rx_ring_ref = GRANT_INVALID_REF;
1599 	info->rx.sring = NULL;
1600 	info->tx.sring = NULL;
1601 	netdev->irq = 0;
1602 
1603 	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1604 			   "feature-split-event-channels", "%u",
1605 			   &feature_split_evtchn);
1606 	if (err < 0)
1607 		feature_split_evtchn = 0;
1608 
1609 	err = xen_net_read_mac(dev, netdev->dev_addr);
1610 	if (err) {
1611 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1612 		goto fail;
1613 	}
1614 
1615 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1616 	if (!txs) {
1617 		err = -ENOMEM;
1618 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1619 		goto fail;
1620 	}
1621 	SHARED_RING_INIT(txs);
1622 	FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1623 
1624 	err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1625 	if (err < 0)
1626 		goto grant_tx_ring_fail;
1627 
1628 	info->tx_ring_ref = err;
1629 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1630 	if (!rxs) {
1631 		err = -ENOMEM;
1632 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1633 		goto alloc_rx_ring_fail;
1634 	}
1635 	SHARED_RING_INIT(rxs);
1636 	FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1637 
1638 	err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1639 	if (err < 0)
1640 		goto grant_rx_ring_fail;
1641 	info->rx_ring_ref = err;
1642 
1643 	if (feature_split_evtchn)
1644 		err = setup_netfront_split(info);
1645 	/* setup single event channel if
1646 	 *  a) feature-split-event-channels == 0
1647 	 *  b) feature-split-event-channels == 1 but failed to setup
1648 	 */
1649 	if (!feature_split_evtchn || (feature_split_evtchn && err))
1650 		err = setup_netfront_single(info);
1651 
1652 	if (err)
1653 		goto alloc_evtchn_fail;
1654 
1655 	return 0;
1656 
1657 	/* If we fail to setup netfront, it is safe to just revoke access to
1658 	 * granted pages because backend is not accessing it at this point.
1659 	 */
1660 alloc_evtchn_fail:
1661 	gnttab_end_foreign_access_ref(info->rx_ring_ref, 0);
1662 grant_rx_ring_fail:
1663 	free_page((unsigned long)rxs);
1664 alloc_rx_ring_fail:
1665 	gnttab_end_foreign_access_ref(info->tx_ring_ref, 0);
1666 grant_tx_ring_fail:
1667 	free_page((unsigned long)txs);
1668 fail:
1669 	return err;
1670 }
1671 
1672 /* Common code used when first setting up, and when resuming. */
1673 static int talk_to_netback(struct xenbus_device *dev,
1674 			   struct netfront_info *info)
1675 {
1676 	const char *message;
1677 	struct xenbus_transaction xbt;
1678 	int err;
1679 
1680 	/* Create shared ring, alloc event channel. */
1681 	err = setup_netfront(dev, info);
1682 	if (err)
1683 		goto out;
1684 
1685 again:
1686 	err = xenbus_transaction_start(&xbt);
1687 	if (err) {
1688 		xenbus_dev_fatal(dev, err, "starting transaction");
1689 		goto destroy_ring;
1690 	}
1691 
1692 	err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1693 			    info->tx_ring_ref);
1694 	if (err) {
1695 		message = "writing tx ring-ref";
1696 		goto abort_transaction;
1697 	}
1698 	err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1699 			    info->rx_ring_ref);
1700 	if (err) {
1701 		message = "writing rx ring-ref";
1702 		goto abort_transaction;
1703 	}
1704 
1705 	if (info->tx_evtchn == info->rx_evtchn) {
1706 		err = xenbus_printf(xbt, dev->nodename,
1707 				    "event-channel", "%u", info->tx_evtchn);
1708 		if (err) {
1709 			message = "writing event-channel";
1710 			goto abort_transaction;
1711 		}
1712 	} else {
1713 		err = xenbus_printf(xbt, dev->nodename,
1714 				    "event-channel-tx", "%u", info->tx_evtchn);
1715 		if (err) {
1716 			message = "writing event-channel-tx";
1717 			goto abort_transaction;
1718 		}
1719 		err = xenbus_printf(xbt, dev->nodename,
1720 				    "event-channel-rx", "%u", info->rx_evtchn);
1721 		if (err) {
1722 			message = "writing event-channel-rx";
1723 			goto abort_transaction;
1724 		}
1725 	}
1726 
1727 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1728 			    1);
1729 	if (err) {
1730 		message = "writing request-rx-copy";
1731 		goto abort_transaction;
1732 	}
1733 
1734 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1735 	if (err) {
1736 		message = "writing feature-rx-notify";
1737 		goto abort_transaction;
1738 	}
1739 
1740 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1741 	if (err) {
1742 		message = "writing feature-sg";
1743 		goto abort_transaction;
1744 	}
1745 
1746 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1747 	if (err) {
1748 		message = "writing feature-gso-tcpv4";
1749 		goto abort_transaction;
1750 	}
1751 
1752 	err = xenbus_transaction_end(xbt, 0);
1753 	if (err) {
1754 		if (err == -EAGAIN)
1755 			goto again;
1756 		xenbus_dev_fatal(dev, err, "completing transaction");
1757 		goto destroy_ring;
1758 	}
1759 
1760 	return 0;
1761 
1762  abort_transaction:
1763 	xenbus_transaction_end(xbt, 1);
1764 	xenbus_dev_fatal(dev, err, "%s", message);
1765  destroy_ring:
1766 	xennet_disconnect_backend(info);
1767  out:
1768 	return err;
1769 }
1770 
1771 static int xennet_connect(struct net_device *dev)
1772 {
1773 	struct netfront_info *np = netdev_priv(dev);
1774 	int i, requeue_idx, err;
1775 	struct sk_buff *skb;
1776 	grant_ref_t ref;
1777 	struct xen_netif_rx_request *req;
1778 	unsigned int feature_rx_copy;
1779 
1780 	err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1781 			   "feature-rx-copy", "%u", &feature_rx_copy);
1782 	if (err != 1)
1783 		feature_rx_copy = 0;
1784 
1785 	if (!feature_rx_copy) {
1786 		dev_info(&dev->dev,
1787 			 "backend does not support copying receive path\n");
1788 		return -ENODEV;
1789 	}
1790 
1791 	err = talk_to_netback(np->xbdev, np);
1792 	if (err)
1793 		return err;
1794 
1795 	rtnl_lock();
1796 	netdev_update_features(dev);
1797 	rtnl_unlock();
1798 
1799 	spin_lock_bh(&np->rx_lock);
1800 	spin_lock_irq(&np->tx_lock);
1801 
1802 	/* Step 1: Discard all pending TX packet fragments. */
1803 	xennet_release_tx_bufs(np);
1804 
1805 	/* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1806 	for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1807 		skb_frag_t *frag;
1808 		const struct page *page;
1809 		if (!np->rx_skbs[i])
1810 			continue;
1811 
1812 		skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1813 		ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1814 		req = RING_GET_REQUEST(&np->rx, requeue_idx);
1815 
1816 		frag = &skb_shinfo(skb)->frags[0];
1817 		page = skb_frag_page(frag);
1818 		gnttab_grant_foreign_access_ref(
1819 			ref, np->xbdev->otherend_id,
1820 			pfn_to_mfn(page_to_pfn(page)),
1821 			0);
1822 		req->gref = ref;
1823 		req->id   = requeue_idx;
1824 
1825 		requeue_idx++;
1826 	}
1827 
1828 	np->rx.req_prod_pvt = requeue_idx;
1829 
1830 	/*
1831 	 * Step 3: All public and private state should now be sane.  Get
1832 	 * ready to start sending and receiving packets and give the driver
1833 	 * domain a kick because we've probably just requeued some
1834 	 * packets.
1835 	 */
1836 	netif_carrier_on(np->netdev);
1837 	notify_remote_via_irq(np->tx_irq);
1838 	if (np->tx_irq != np->rx_irq)
1839 		notify_remote_via_irq(np->rx_irq);
1840 	xennet_tx_buf_gc(dev);
1841 	xennet_alloc_rx_buffers(dev);
1842 
1843 	spin_unlock_irq(&np->tx_lock);
1844 	spin_unlock_bh(&np->rx_lock);
1845 
1846 	return 0;
1847 }
1848 
1849 /**
1850  * Callback received when the backend's state changes.
1851  */
1852 static void netback_changed(struct xenbus_device *dev,
1853 			    enum xenbus_state backend_state)
1854 {
1855 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
1856 	struct net_device *netdev = np->netdev;
1857 
1858 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1859 
1860 	switch (backend_state) {
1861 	case XenbusStateInitialising:
1862 	case XenbusStateInitialised:
1863 	case XenbusStateReconfiguring:
1864 	case XenbusStateReconfigured:
1865 	case XenbusStateUnknown:
1866 	case XenbusStateClosed:
1867 		break;
1868 
1869 	case XenbusStateInitWait:
1870 		if (dev->state != XenbusStateInitialising)
1871 			break;
1872 		if (xennet_connect(netdev) != 0)
1873 			break;
1874 		xenbus_switch_state(dev, XenbusStateConnected);
1875 		break;
1876 
1877 	case XenbusStateConnected:
1878 		netdev_notify_peers(netdev);
1879 		break;
1880 
1881 	case XenbusStateClosing:
1882 		xenbus_frontend_closed(dev);
1883 		break;
1884 	}
1885 }
1886 
1887 static const struct xennet_stat {
1888 	char name[ETH_GSTRING_LEN];
1889 	u16 offset;
1890 } xennet_stats[] = {
1891 	{
1892 		"rx_gso_checksum_fixup",
1893 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
1894 	},
1895 };
1896 
1897 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1898 {
1899 	switch (string_set) {
1900 	case ETH_SS_STATS:
1901 		return ARRAY_SIZE(xennet_stats);
1902 	default:
1903 		return -EINVAL;
1904 	}
1905 }
1906 
1907 static void xennet_get_ethtool_stats(struct net_device *dev,
1908 				     struct ethtool_stats *stats, u64 * data)
1909 {
1910 	void *np = netdev_priv(dev);
1911 	int i;
1912 
1913 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1914 		data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1915 }
1916 
1917 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1918 {
1919 	int i;
1920 
1921 	switch (stringset) {
1922 	case ETH_SS_STATS:
1923 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1924 			memcpy(data + i * ETH_GSTRING_LEN,
1925 			       xennet_stats[i].name, ETH_GSTRING_LEN);
1926 		break;
1927 	}
1928 }
1929 
1930 static const struct ethtool_ops xennet_ethtool_ops =
1931 {
1932 	.get_link = ethtool_op_get_link,
1933 
1934 	.get_sset_count = xennet_get_sset_count,
1935 	.get_ethtool_stats = xennet_get_ethtool_stats,
1936 	.get_strings = xennet_get_strings,
1937 };
1938 
1939 #ifdef CONFIG_SYSFS
1940 static ssize_t show_rxbuf_min(struct device *dev,
1941 			      struct device_attribute *attr, char *buf)
1942 {
1943 	struct net_device *netdev = to_net_dev(dev);
1944 	struct netfront_info *info = netdev_priv(netdev);
1945 
1946 	return sprintf(buf, "%u\n", info->rx_min_target);
1947 }
1948 
1949 static ssize_t store_rxbuf_min(struct device *dev,
1950 			       struct device_attribute *attr,
1951 			       const char *buf, size_t len)
1952 {
1953 	struct net_device *netdev = to_net_dev(dev);
1954 	struct netfront_info *np = netdev_priv(netdev);
1955 	char *endp;
1956 	unsigned long target;
1957 
1958 	if (!capable(CAP_NET_ADMIN))
1959 		return -EPERM;
1960 
1961 	target = simple_strtoul(buf, &endp, 0);
1962 	if (endp == buf)
1963 		return -EBADMSG;
1964 
1965 	if (target < RX_MIN_TARGET)
1966 		target = RX_MIN_TARGET;
1967 	if (target > RX_MAX_TARGET)
1968 		target = RX_MAX_TARGET;
1969 
1970 	spin_lock_bh(&np->rx_lock);
1971 	if (target > np->rx_max_target)
1972 		np->rx_max_target = target;
1973 	np->rx_min_target = target;
1974 	if (target > np->rx_target)
1975 		np->rx_target = target;
1976 
1977 	xennet_alloc_rx_buffers(netdev);
1978 
1979 	spin_unlock_bh(&np->rx_lock);
1980 	return len;
1981 }
1982 
1983 static ssize_t show_rxbuf_max(struct device *dev,
1984 			      struct device_attribute *attr, char *buf)
1985 {
1986 	struct net_device *netdev = to_net_dev(dev);
1987 	struct netfront_info *info = netdev_priv(netdev);
1988 
1989 	return sprintf(buf, "%u\n", info->rx_max_target);
1990 }
1991 
1992 static ssize_t store_rxbuf_max(struct device *dev,
1993 			       struct device_attribute *attr,
1994 			       const char *buf, size_t len)
1995 {
1996 	struct net_device *netdev = to_net_dev(dev);
1997 	struct netfront_info *np = netdev_priv(netdev);
1998 	char *endp;
1999 	unsigned long target;
2000 
2001 	if (!capable(CAP_NET_ADMIN))
2002 		return -EPERM;
2003 
2004 	target = simple_strtoul(buf, &endp, 0);
2005 	if (endp == buf)
2006 		return -EBADMSG;
2007 
2008 	if (target < RX_MIN_TARGET)
2009 		target = RX_MIN_TARGET;
2010 	if (target > RX_MAX_TARGET)
2011 		target = RX_MAX_TARGET;
2012 
2013 	spin_lock_bh(&np->rx_lock);
2014 	if (target < np->rx_min_target)
2015 		np->rx_min_target = target;
2016 	np->rx_max_target = target;
2017 	if (target < np->rx_target)
2018 		np->rx_target = target;
2019 
2020 	xennet_alloc_rx_buffers(netdev);
2021 
2022 	spin_unlock_bh(&np->rx_lock);
2023 	return len;
2024 }
2025 
2026 static ssize_t show_rxbuf_cur(struct device *dev,
2027 			      struct device_attribute *attr, char *buf)
2028 {
2029 	struct net_device *netdev = to_net_dev(dev);
2030 	struct netfront_info *info = netdev_priv(netdev);
2031 
2032 	return sprintf(buf, "%u\n", info->rx_target);
2033 }
2034 
2035 static struct device_attribute xennet_attrs[] = {
2036 	__ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
2037 	__ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
2038 	__ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
2039 };
2040 
2041 static int xennet_sysfs_addif(struct net_device *netdev)
2042 {
2043 	int i;
2044 	int err;
2045 
2046 	for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
2047 		err = device_create_file(&netdev->dev,
2048 					   &xennet_attrs[i]);
2049 		if (err)
2050 			goto fail;
2051 	}
2052 	return 0;
2053 
2054  fail:
2055 	while (--i >= 0)
2056 		device_remove_file(&netdev->dev, &xennet_attrs[i]);
2057 	return err;
2058 }
2059 
2060 static void xennet_sysfs_delif(struct net_device *netdev)
2061 {
2062 	int i;
2063 
2064 	for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
2065 		device_remove_file(&netdev->dev, &xennet_attrs[i]);
2066 }
2067 
2068 #endif /* CONFIG_SYSFS */
2069 
2070 static const struct xenbus_device_id netfront_ids[] = {
2071 	{ "vif" },
2072 	{ "" }
2073 };
2074 
2075 
2076 static int xennet_remove(struct xenbus_device *dev)
2077 {
2078 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2079 
2080 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
2081 
2082 	xennet_disconnect_backend(info);
2083 
2084 	xennet_sysfs_delif(info->netdev);
2085 
2086 	unregister_netdev(info->netdev);
2087 
2088 	del_timer_sync(&info->rx_refill_timer);
2089 
2090 	free_percpu(info->stats);
2091 
2092 	free_netdev(info->netdev);
2093 
2094 	return 0;
2095 }
2096 
2097 static DEFINE_XENBUS_DRIVER(netfront, ,
2098 	.probe = netfront_probe,
2099 	.remove = xennet_remove,
2100 	.resume = netfront_resume,
2101 	.otherend_changed = netback_changed,
2102 );
2103 
2104 static int __init netif_init(void)
2105 {
2106 	if (!xen_domain())
2107 		return -ENODEV;
2108 
2109 	if (xen_hvm_domain() && !xen_platform_pci_unplug)
2110 		return -ENODEV;
2111 
2112 	pr_info("Initialising Xen virtual ethernet driver\n");
2113 
2114 	return xenbus_register_frontend(&netfront_driver);
2115 }
2116 module_init(netif_init);
2117 
2118 
2119 static void __exit netif_exit(void)
2120 {
2121 	xenbus_unregister_driver(&netfront_driver);
2122 }
2123 module_exit(netif_exit);
2124 
2125 MODULE_DESCRIPTION("Xen virtual network device frontend");
2126 MODULE_LICENSE("GPL");
2127 MODULE_ALIAS("xen:vif");
2128 MODULE_ALIAS("xennet");
2129