1 /*
2  * Network-device interface management.
3  *
4  * Copyright (c) 2004-2005, Keir Fraser
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License version 2
8  * as published by the Free Software Foundation; or, when distributed
9  * separately from the Linux kernel or incorporated into other
10  * software packages, subject to the following license:
11  *
12  * Permission is hereby granted, free of charge, to any person obtaining a copy
13  * of this source file (the "Software"), to deal in the Software without
14  * restriction, including without limitation the rights to use, copy, modify,
15  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
16  * and to permit persons to whom the Software is furnished to do so, subject to
17  * the following conditions:
18  *
19  * The above copyright notice and this permission notice shall be included in
20  * all copies or substantial portions of the Software.
21  *
22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
25  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
28  * IN THE SOFTWARE.
29  */
30 
31 #include "common.h"
32 
33 #include <linux/kthread.h>
34 #include <linux/sched/task.h>
35 #include <linux/ethtool.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/if_vlan.h>
38 #include <linux/vmalloc.h>
39 
40 #include <xen/events.h>
41 #include <asm/xen/hypercall.h>
42 #include <xen/balloon.h>
43 
44 #define XENVIF_QUEUE_LENGTH 32
45 #define XENVIF_NAPI_WEIGHT  64
46 
47 /* Number of bytes allowed on the internal guest Rx queue. */
48 #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
49 
50 /* This function is used to set SKBTX_DEV_ZEROCOPY as well as
51  * increasing the inflight counter. We need to increase the inflight
52  * counter because core driver calls into xenvif_zerocopy_callback
53  * which calls xenvif_skb_zerocopy_complete.
54  */
55 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
56 				 struct sk_buff *skb)
57 {
58 	skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
59 	atomic_inc(&queue->inflight_packets);
60 }
61 
62 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
63 {
64 	atomic_dec(&queue->inflight_packets);
65 
66 	/* Wake the dealloc thread _after_ decrementing inflight_packets so
67 	 * that if kthread_stop() has already been called, the dealloc thread
68 	 * does not wait forever with nothing to wake it.
69 	 */
70 	wake_up(&queue->dealloc_wq);
71 }
72 
73 int xenvif_schedulable(struct xenvif *vif)
74 {
75 	return netif_running(vif->dev) &&
76 		test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
77 		!vif->disabled;
78 }
79 
80 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
81 {
82 	struct xenvif_queue *queue = dev_id;
83 
84 	if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
85 		napi_schedule(&queue->napi);
86 
87 	return IRQ_HANDLED;
88 }
89 
90 static int xenvif_poll(struct napi_struct *napi, int budget)
91 {
92 	struct xenvif_queue *queue =
93 		container_of(napi, struct xenvif_queue, napi);
94 	int work_done;
95 
96 	/* This vif is rogue, we pretend we've there is nothing to do
97 	 * for this vif to deschedule it from NAPI. But this interface
98 	 * will be turned off in thread context later.
99 	 */
100 	if (unlikely(queue->vif->disabled)) {
101 		napi_complete(napi);
102 		return 0;
103 	}
104 
105 	work_done = xenvif_tx_action(queue, budget);
106 
107 	if (work_done < budget) {
108 		napi_complete_done(napi, work_done);
109 		/* If the queue is rate-limited, it shall be
110 		 * rescheduled in the timer callback.
111 		 */
112 		if (likely(!queue->rate_limited))
113 			xenvif_napi_schedule_or_enable_events(queue);
114 	}
115 
116 	return work_done;
117 }
118 
119 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
120 {
121 	struct xenvif_queue *queue = dev_id;
122 
123 	xenvif_kick_thread(queue);
124 
125 	return IRQ_HANDLED;
126 }
127 
128 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
129 {
130 	xenvif_tx_interrupt(irq, dev_id);
131 	xenvif_rx_interrupt(irq, dev_id);
132 
133 	return IRQ_HANDLED;
134 }
135 
136 int xenvif_queue_stopped(struct xenvif_queue *queue)
137 {
138 	struct net_device *dev = queue->vif->dev;
139 	unsigned int id = queue->id;
140 	return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
141 }
142 
143 void xenvif_wake_queue(struct xenvif_queue *queue)
144 {
145 	struct net_device *dev = queue->vif->dev;
146 	unsigned int id = queue->id;
147 	netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
148 }
149 
150 static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
151 			       struct net_device *sb_dev,
152 			       select_queue_fallback_t fallback)
153 {
154 	struct xenvif *vif = netdev_priv(dev);
155 	unsigned int size = vif->hash.size;
156 
157 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
158 		return fallback(dev, skb, NULL) % dev->real_num_tx_queues;
159 
160 	xenvif_set_skb_hash(vif, skb);
161 
162 	if (size == 0)
163 		return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
164 
165 	return vif->hash.mapping[vif->hash.mapping_sel]
166 				[skb_get_hash_raw(skb) % size];
167 }
168 
169 static netdev_tx_t
170 xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
171 {
172 	struct xenvif *vif = netdev_priv(dev);
173 	struct xenvif_queue *queue = NULL;
174 	unsigned int num_queues;
175 	u16 index;
176 	struct xenvif_rx_cb *cb;
177 
178 	BUG_ON(skb->dev != dev);
179 
180 	/* Drop the packet if queues are not set up.
181 	 * This handler should be called inside an RCU read section
182 	 * so we don't need to enter it here explicitly.
183 	 */
184 	num_queues = READ_ONCE(vif->num_queues);
185 	if (num_queues < 1)
186 		goto drop;
187 
188 	/* Obtain the queue to be used to transmit this packet */
189 	index = skb_get_queue_mapping(skb);
190 	if (index >= num_queues) {
191 		pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
192 				    index, vif->dev->name);
193 		index %= num_queues;
194 	}
195 	queue = &vif->queues[index];
196 
197 	/* Drop the packet if queue is not ready */
198 	if (queue->task == NULL ||
199 	    queue->dealloc_task == NULL ||
200 	    !xenvif_schedulable(vif))
201 		goto drop;
202 
203 	if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
204 		struct ethhdr *eth = (struct ethhdr *)skb->data;
205 
206 		if (!xenvif_mcast_match(vif, eth->h_dest))
207 			goto drop;
208 	}
209 
210 	cb = XENVIF_RX_CB(skb);
211 	cb->expires = jiffies + vif->drain_timeout;
212 
213 	/* If there is no hash algorithm configured then make sure there
214 	 * is no hash information in the socket buffer otherwise it
215 	 * would be incorrectly forwarded to the frontend.
216 	 */
217 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
218 		skb_clear_hash(skb);
219 
220 	xenvif_rx_queue_tail(queue, skb);
221 	xenvif_kick_thread(queue);
222 
223 	return NETDEV_TX_OK;
224 
225  drop:
226 	vif->dev->stats.tx_dropped++;
227 	dev_kfree_skb(skb);
228 	return NETDEV_TX_OK;
229 }
230 
231 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
232 {
233 	struct xenvif *vif = netdev_priv(dev);
234 	struct xenvif_queue *queue = NULL;
235 	unsigned int num_queues;
236 	u64 rx_bytes = 0;
237 	u64 rx_packets = 0;
238 	u64 tx_bytes = 0;
239 	u64 tx_packets = 0;
240 	unsigned int index;
241 
242 	rcu_read_lock();
243 	num_queues = READ_ONCE(vif->num_queues);
244 
245 	/* Aggregate tx and rx stats from each queue */
246 	for (index = 0; index < num_queues; ++index) {
247 		queue = &vif->queues[index];
248 		rx_bytes += queue->stats.rx_bytes;
249 		rx_packets += queue->stats.rx_packets;
250 		tx_bytes += queue->stats.tx_bytes;
251 		tx_packets += queue->stats.tx_packets;
252 	}
253 
254 	rcu_read_unlock();
255 
256 	vif->dev->stats.rx_bytes = rx_bytes;
257 	vif->dev->stats.rx_packets = rx_packets;
258 	vif->dev->stats.tx_bytes = tx_bytes;
259 	vif->dev->stats.tx_packets = tx_packets;
260 
261 	return &vif->dev->stats;
262 }
263 
264 static void xenvif_up(struct xenvif *vif)
265 {
266 	struct xenvif_queue *queue = NULL;
267 	unsigned int num_queues = vif->num_queues;
268 	unsigned int queue_index;
269 
270 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
271 		queue = &vif->queues[queue_index];
272 		napi_enable(&queue->napi);
273 		enable_irq(queue->tx_irq);
274 		if (queue->tx_irq != queue->rx_irq)
275 			enable_irq(queue->rx_irq);
276 		xenvif_napi_schedule_or_enable_events(queue);
277 	}
278 }
279 
280 static void xenvif_down(struct xenvif *vif)
281 {
282 	struct xenvif_queue *queue = NULL;
283 	unsigned int num_queues = vif->num_queues;
284 	unsigned int queue_index;
285 
286 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
287 		queue = &vif->queues[queue_index];
288 		disable_irq(queue->tx_irq);
289 		if (queue->tx_irq != queue->rx_irq)
290 			disable_irq(queue->rx_irq);
291 		napi_disable(&queue->napi);
292 		del_timer_sync(&queue->credit_timeout);
293 	}
294 }
295 
296 static int xenvif_open(struct net_device *dev)
297 {
298 	struct xenvif *vif = netdev_priv(dev);
299 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
300 		xenvif_up(vif);
301 	netif_tx_start_all_queues(dev);
302 	return 0;
303 }
304 
305 static int xenvif_close(struct net_device *dev)
306 {
307 	struct xenvif *vif = netdev_priv(dev);
308 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
309 		xenvif_down(vif);
310 	netif_tx_stop_all_queues(dev);
311 	return 0;
312 }
313 
314 static int xenvif_change_mtu(struct net_device *dev, int mtu)
315 {
316 	struct xenvif *vif = netdev_priv(dev);
317 	int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
318 
319 	if (mtu > max)
320 		return -EINVAL;
321 	dev->mtu = mtu;
322 	return 0;
323 }
324 
325 static netdev_features_t xenvif_fix_features(struct net_device *dev,
326 	netdev_features_t features)
327 {
328 	struct xenvif *vif = netdev_priv(dev);
329 
330 	if (!vif->can_sg)
331 		features &= ~NETIF_F_SG;
332 	if (~(vif->gso_mask) & GSO_BIT(TCPV4))
333 		features &= ~NETIF_F_TSO;
334 	if (~(vif->gso_mask) & GSO_BIT(TCPV6))
335 		features &= ~NETIF_F_TSO6;
336 	if (!vif->ip_csum)
337 		features &= ~NETIF_F_IP_CSUM;
338 	if (!vif->ipv6_csum)
339 		features &= ~NETIF_F_IPV6_CSUM;
340 
341 	return features;
342 }
343 
344 static const struct xenvif_stat {
345 	char name[ETH_GSTRING_LEN];
346 	u16 offset;
347 } xenvif_stats[] = {
348 	{
349 		"rx_gso_checksum_fixup",
350 		offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
351 	},
352 	/* If (sent != success + fail), there are probably packets never
353 	 * freed up properly!
354 	 */
355 	{
356 		"tx_zerocopy_sent",
357 		offsetof(struct xenvif_stats, tx_zerocopy_sent),
358 	},
359 	{
360 		"tx_zerocopy_success",
361 		offsetof(struct xenvif_stats, tx_zerocopy_success),
362 	},
363 	{
364 		"tx_zerocopy_fail",
365 		offsetof(struct xenvif_stats, tx_zerocopy_fail)
366 	},
367 	/* Number of packets exceeding MAX_SKB_FRAG slots. You should use
368 	 * a guest with the same MAX_SKB_FRAG
369 	 */
370 	{
371 		"tx_frag_overflow",
372 		offsetof(struct xenvif_stats, tx_frag_overflow)
373 	},
374 };
375 
376 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
377 {
378 	switch (string_set) {
379 	case ETH_SS_STATS:
380 		return ARRAY_SIZE(xenvif_stats);
381 	default:
382 		return -EINVAL;
383 	}
384 }
385 
386 static void xenvif_get_ethtool_stats(struct net_device *dev,
387 				     struct ethtool_stats *stats, u64 * data)
388 {
389 	struct xenvif *vif = netdev_priv(dev);
390 	unsigned int num_queues;
391 	int i;
392 	unsigned int queue_index;
393 
394 	rcu_read_lock();
395 	num_queues = READ_ONCE(vif->num_queues);
396 
397 	for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
398 		unsigned long accum = 0;
399 		for (queue_index = 0; queue_index < num_queues; ++queue_index) {
400 			void *vif_stats = &vif->queues[queue_index].stats;
401 			accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
402 		}
403 		data[i] = accum;
404 	}
405 
406 	rcu_read_unlock();
407 }
408 
409 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
410 {
411 	int i;
412 
413 	switch (stringset) {
414 	case ETH_SS_STATS:
415 		for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
416 			memcpy(data + i * ETH_GSTRING_LEN,
417 			       xenvif_stats[i].name, ETH_GSTRING_LEN);
418 		break;
419 	}
420 }
421 
422 static const struct ethtool_ops xenvif_ethtool_ops = {
423 	.get_link	= ethtool_op_get_link,
424 
425 	.get_sset_count = xenvif_get_sset_count,
426 	.get_ethtool_stats = xenvif_get_ethtool_stats,
427 	.get_strings = xenvif_get_strings,
428 };
429 
430 static const struct net_device_ops xenvif_netdev_ops = {
431 	.ndo_select_queue = xenvif_select_queue,
432 	.ndo_start_xmit	= xenvif_start_xmit,
433 	.ndo_get_stats	= xenvif_get_stats,
434 	.ndo_open	= xenvif_open,
435 	.ndo_stop	= xenvif_close,
436 	.ndo_change_mtu	= xenvif_change_mtu,
437 	.ndo_fix_features = xenvif_fix_features,
438 	.ndo_set_mac_address = eth_mac_addr,
439 	.ndo_validate_addr   = eth_validate_addr,
440 };
441 
442 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
443 			    unsigned int handle)
444 {
445 	int err;
446 	struct net_device *dev;
447 	struct xenvif *vif;
448 	char name[IFNAMSIZ] = {};
449 
450 	snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
451 	/* Allocate a netdev with the max. supported number of queues.
452 	 * When the guest selects the desired number, it will be updated
453 	 * via netif_set_real_num_*_queues().
454 	 */
455 	dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
456 			      ether_setup, xenvif_max_queues);
457 	if (dev == NULL) {
458 		pr_warn("Could not allocate netdev for %s\n", name);
459 		return ERR_PTR(-ENOMEM);
460 	}
461 
462 	SET_NETDEV_DEV(dev, parent);
463 
464 	vif = netdev_priv(dev);
465 
466 	vif->domid  = domid;
467 	vif->handle = handle;
468 	vif->can_sg = 1;
469 	vif->ip_csum = 1;
470 	vif->dev = dev;
471 	vif->disabled = false;
472 	vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
473 	vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
474 
475 	/* Start out with no queues. */
476 	vif->queues = NULL;
477 	vif->num_queues = 0;
478 
479 	spin_lock_init(&vif->lock);
480 	INIT_LIST_HEAD(&vif->fe_mcast_addr);
481 
482 	dev->netdev_ops	= &xenvif_netdev_ops;
483 	dev->hw_features = NETIF_F_SG |
484 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
485 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
486 	dev->features = dev->hw_features | NETIF_F_RXCSUM;
487 	dev->ethtool_ops = &xenvif_ethtool_ops;
488 
489 	dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
490 
491 	dev->min_mtu = ETH_MIN_MTU;
492 	dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
493 
494 	/*
495 	 * Initialise a dummy MAC address. We choose the numerically
496 	 * largest non-broadcast address to prevent the address getting
497 	 * stolen by an Ethernet bridge for STP purposes.
498 	 * (FE:FF:FF:FF:FF:FF)
499 	 */
500 	eth_broadcast_addr(dev->dev_addr);
501 	dev->dev_addr[0] &= ~0x01;
502 
503 	netif_carrier_off(dev);
504 
505 	err = register_netdev(dev);
506 	if (err) {
507 		netdev_warn(dev, "Could not register device: err=%d\n", err);
508 		free_netdev(dev);
509 		return ERR_PTR(err);
510 	}
511 
512 	netdev_dbg(dev, "Successfully created xenvif\n");
513 
514 	__module_get(THIS_MODULE);
515 
516 	return vif;
517 }
518 
519 int xenvif_init_queue(struct xenvif_queue *queue)
520 {
521 	int err, i;
522 
523 	queue->credit_bytes = queue->remaining_credit = ~0UL;
524 	queue->credit_usec  = 0UL;
525 	timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
526 	queue->credit_window_start = get_jiffies_64();
527 
528 	queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
529 
530 	skb_queue_head_init(&queue->rx_queue);
531 	skb_queue_head_init(&queue->tx_queue);
532 
533 	queue->pending_cons = 0;
534 	queue->pending_prod = MAX_PENDING_REQS;
535 	for (i = 0; i < MAX_PENDING_REQS; ++i)
536 		queue->pending_ring[i] = i;
537 
538 	spin_lock_init(&queue->callback_lock);
539 	spin_lock_init(&queue->response_lock);
540 
541 	/* If ballooning is disabled, this will consume real memory, so you
542 	 * better enable it. The long term solution would be to use just a
543 	 * bunch of valid page descriptors, without dependency on ballooning
544 	 */
545 	err = gnttab_alloc_pages(MAX_PENDING_REQS,
546 				 queue->mmap_pages);
547 	if (err) {
548 		netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
549 		return -ENOMEM;
550 	}
551 
552 	for (i = 0; i < MAX_PENDING_REQS; i++) {
553 		queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
554 			{ .callback = xenvif_zerocopy_callback,
555 			  { { .ctx = NULL,
556 			      .desc = i } } };
557 		queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
558 	}
559 
560 	return 0;
561 }
562 
563 void xenvif_carrier_on(struct xenvif *vif)
564 {
565 	rtnl_lock();
566 	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
567 		dev_set_mtu(vif->dev, ETH_DATA_LEN);
568 	netdev_update_features(vif->dev);
569 	set_bit(VIF_STATUS_CONNECTED, &vif->status);
570 	if (netif_running(vif->dev))
571 		xenvif_up(vif);
572 	rtnl_unlock();
573 }
574 
575 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
576 			unsigned int evtchn)
577 {
578 	struct net_device *dev = vif->dev;
579 	void *addr;
580 	struct xen_netif_ctrl_sring *shared;
581 	int err;
582 
583 	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
584 				     &ring_ref, 1, &addr);
585 	if (err)
586 		goto err;
587 
588 	shared = (struct xen_netif_ctrl_sring *)addr;
589 	BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
590 
591 	err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
592 	if (err < 0)
593 		goto err_unmap;
594 
595 	vif->ctrl_irq = err;
596 
597 	xenvif_init_hash(vif);
598 
599 	err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
600 				   IRQF_ONESHOT, "xen-netback-ctrl", vif);
601 	if (err) {
602 		pr_warn("Could not setup irq handler for %s\n", dev->name);
603 		goto err_deinit;
604 	}
605 
606 	return 0;
607 
608 err_deinit:
609 	xenvif_deinit_hash(vif);
610 	unbind_from_irqhandler(vif->ctrl_irq, vif);
611 	vif->ctrl_irq = 0;
612 
613 err_unmap:
614 	xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
615 				vif->ctrl.sring);
616 	vif->ctrl.sring = NULL;
617 
618 err:
619 	return err;
620 }
621 
622 int xenvif_connect_data(struct xenvif_queue *queue,
623 			unsigned long tx_ring_ref,
624 			unsigned long rx_ring_ref,
625 			unsigned int tx_evtchn,
626 			unsigned int rx_evtchn)
627 {
628 	struct task_struct *task;
629 	int err = -ENOMEM;
630 
631 	BUG_ON(queue->tx_irq);
632 	BUG_ON(queue->task);
633 	BUG_ON(queue->dealloc_task);
634 
635 	err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
636 					     rx_ring_ref);
637 	if (err < 0)
638 		goto err;
639 
640 	init_waitqueue_head(&queue->wq);
641 	init_waitqueue_head(&queue->dealloc_wq);
642 	atomic_set(&queue->inflight_packets, 0);
643 
644 	netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
645 			XENVIF_NAPI_WEIGHT);
646 
647 	if (tx_evtchn == rx_evtchn) {
648 		/* feature-split-event-channels == 0 */
649 		err = bind_interdomain_evtchn_to_irqhandler(
650 			queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
651 			queue->name, queue);
652 		if (err < 0)
653 			goto err_unmap;
654 		queue->tx_irq = queue->rx_irq = err;
655 		disable_irq(queue->tx_irq);
656 	} else {
657 		/* feature-split-event-channels == 1 */
658 		snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
659 			 "%s-tx", queue->name);
660 		err = bind_interdomain_evtchn_to_irqhandler(
661 			queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
662 			queue->tx_irq_name, queue);
663 		if (err < 0)
664 			goto err_unmap;
665 		queue->tx_irq = err;
666 		disable_irq(queue->tx_irq);
667 
668 		snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
669 			 "%s-rx", queue->name);
670 		err = bind_interdomain_evtchn_to_irqhandler(
671 			queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
672 			queue->rx_irq_name, queue);
673 		if (err < 0)
674 			goto err_tx_unbind;
675 		queue->rx_irq = err;
676 		disable_irq(queue->rx_irq);
677 	}
678 
679 	queue->stalled = true;
680 
681 	task = kthread_create(xenvif_kthread_guest_rx,
682 			      (void *)queue, "%s-guest-rx", queue->name);
683 	if (IS_ERR(task)) {
684 		pr_warn("Could not allocate kthread for %s\n", queue->name);
685 		err = PTR_ERR(task);
686 		goto err_rx_unbind;
687 	}
688 	queue->task = task;
689 	get_task_struct(task);
690 
691 	task = kthread_create(xenvif_dealloc_kthread,
692 			      (void *)queue, "%s-dealloc", queue->name);
693 	if (IS_ERR(task)) {
694 		pr_warn("Could not allocate kthread for %s\n", queue->name);
695 		err = PTR_ERR(task);
696 		goto err_rx_unbind;
697 	}
698 	queue->dealloc_task = task;
699 
700 	wake_up_process(queue->task);
701 	wake_up_process(queue->dealloc_task);
702 
703 	return 0;
704 
705 err_rx_unbind:
706 	unbind_from_irqhandler(queue->rx_irq, queue);
707 	queue->rx_irq = 0;
708 err_tx_unbind:
709 	unbind_from_irqhandler(queue->tx_irq, queue);
710 	queue->tx_irq = 0;
711 err_unmap:
712 	xenvif_unmap_frontend_data_rings(queue);
713 	netif_napi_del(&queue->napi);
714 err:
715 	module_put(THIS_MODULE);
716 	return err;
717 }
718 
719 void xenvif_carrier_off(struct xenvif *vif)
720 {
721 	struct net_device *dev = vif->dev;
722 
723 	rtnl_lock();
724 	if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
725 		netif_carrier_off(dev); /* discard queued packets */
726 		if (netif_running(dev))
727 			xenvif_down(vif);
728 	}
729 	rtnl_unlock();
730 }
731 
732 void xenvif_disconnect_data(struct xenvif *vif)
733 {
734 	struct xenvif_queue *queue = NULL;
735 	unsigned int num_queues = vif->num_queues;
736 	unsigned int queue_index;
737 
738 	xenvif_carrier_off(vif);
739 
740 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
741 		queue = &vif->queues[queue_index];
742 
743 		netif_napi_del(&queue->napi);
744 
745 		if (queue->task) {
746 			kthread_stop(queue->task);
747 			put_task_struct(queue->task);
748 			queue->task = NULL;
749 		}
750 
751 		if (queue->dealloc_task) {
752 			kthread_stop(queue->dealloc_task);
753 			queue->dealloc_task = NULL;
754 		}
755 
756 		if (queue->tx_irq) {
757 			if (queue->tx_irq == queue->rx_irq)
758 				unbind_from_irqhandler(queue->tx_irq, queue);
759 			else {
760 				unbind_from_irqhandler(queue->tx_irq, queue);
761 				unbind_from_irqhandler(queue->rx_irq, queue);
762 			}
763 			queue->tx_irq = 0;
764 		}
765 
766 		xenvif_unmap_frontend_data_rings(queue);
767 	}
768 
769 	xenvif_mcast_addr_list_free(vif);
770 }
771 
772 void xenvif_disconnect_ctrl(struct xenvif *vif)
773 {
774 	if (vif->ctrl_irq) {
775 		xenvif_deinit_hash(vif);
776 		unbind_from_irqhandler(vif->ctrl_irq, vif);
777 		vif->ctrl_irq = 0;
778 	}
779 
780 	if (vif->ctrl.sring) {
781 		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
782 					vif->ctrl.sring);
783 		vif->ctrl.sring = NULL;
784 	}
785 }
786 
787 /* Reverse the relevant parts of xenvif_init_queue().
788  * Used for queue teardown from xenvif_free(), and on the
789  * error handling paths in xenbus.c:connect().
790  */
791 void xenvif_deinit_queue(struct xenvif_queue *queue)
792 {
793 	gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
794 }
795 
796 void xenvif_free(struct xenvif *vif)
797 {
798 	struct xenvif_queue *queues = vif->queues;
799 	unsigned int num_queues = vif->num_queues;
800 	unsigned int queue_index;
801 
802 	unregister_netdev(vif->dev);
803 	free_netdev(vif->dev);
804 
805 	for (queue_index = 0; queue_index < num_queues; ++queue_index)
806 		xenvif_deinit_queue(&queues[queue_index]);
807 	vfree(queues);
808 
809 	module_put(THIS_MODULE);
810 }
811