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 		xenvif_napi_schedule_or_enable_events(queue);
110 	}
111 
112 	return work_done;
113 }
114 
115 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
116 {
117 	struct xenvif_queue *queue = dev_id;
118 
119 	xenvif_kick_thread(queue);
120 
121 	return IRQ_HANDLED;
122 }
123 
124 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
125 {
126 	xenvif_tx_interrupt(irq, dev_id);
127 	xenvif_rx_interrupt(irq, dev_id);
128 
129 	return IRQ_HANDLED;
130 }
131 
132 int xenvif_queue_stopped(struct xenvif_queue *queue)
133 {
134 	struct net_device *dev = queue->vif->dev;
135 	unsigned int id = queue->id;
136 	return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
137 }
138 
139 void xenvif_wake_queue(struct xenvif_queue *queue)
140 {
141 	struct net_device *dev = queue->vif->dev;
142 	unsigned int id = queue->id;
143 	netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
144 }
145 
146 static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
147 			       void *accel_priv,
148 			       select_queue_fallback_t fallback)
149 {
150 	struct xenvif *vif = netdev_priv(dev);
151 	unsigned int size = vif->hash.size;
152 
153 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
154 		return fallback(dev, skb) % dev->real_num_tx_queues;
155 
156 	xenvif_set_skb_hash(vif, skb);
157 
158 	if (size == 0)
159 		return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
160 
161 	return vif->hash.mapping[skb_get_hash_raw(skb) % size];
162 }
163 
164 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
165 {
166 	struct xenvif *vif = netdev_priv(dev);
167 	struct xenvif_queue *queue = NULL;
168 	unsigned int num_queues;
169 	u16 index;
170 	struct xenvif_rx_cb *cb;
171 
172 	BUG_ON(skb->dev != dev);
173 
174 	/* Drop the packet if queues are not set up.
175 	 * This handler should be called inside an RCU read section
176 	 * so we don't need to enter it here explicitly.
177 	 */
178 	num_queues = READ_ONCE(vif->num_queues);
179 	if (num_queues < 1)
180 		goto drop;
181 
182 	/* Obtain the queue to be used to transmit this packet */
183 	index = skb_get_queue_mapping(skb);
184 	if (index >= num_queues) {
185 		pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
186 				    index, vif->dev->name);
187 		index %= num_queues;
188 	}
189 	queue = &vif->queues[index];
190 
191 	/* Drop the packet if queue is not ready */
192 	if (queue->task == NULL ||
193 	    queue->dealloc_task == NULL ||
194 	    !xenvif_schedulable(vif))
195 		goto drop;
196 
197 	if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
198 		struct ethhdr *eth = (struct ethhdr *)skb->data;
199 
200 		if (!xenvif_mcast_match(vif, eth->h_dest))
201 			goto drop;
202 	}
203 
204 	cb = XENVIF_RX_CB(skb);
205 	cb->expires = jiffies + vif->drain_timeout;
206 
207 	/* If there is no hash algorithm configured then make sure there
208 	 * is no hash information in the socket buffer otherwise it
209 	 * would be incorrectly forwarded to the frontend.
210 	 */
211 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
212 		skb_clear_hash(skb);
213 
214 	xenvif_rx_queue_tail(queue, skb);
215 	xenvif_kick_thread(queue);
216 
217 	return NETDEV_TX_OK;
218 
219  drop:
220 	vif->dev->stats.tx_dropped++;
221 	dev_kfree_skb(skb);
222 	return NETDEV_TX_OK;
223 }
224 
225 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
226 {
227 	struct xenvif *vif = netdev_priv(dev);
228 	struct xenvif_queue *queue = NULL;
229 	unsigned int num_queues;
230 	u64 rx_bytes = 0;
231 	u64 rx_packets = 0;
232 	u64 tx_bytes = 0;
233 	u64 tx_packets = 0;
234 	unsigned int index;
235 
236 	rcu_read_lock();
237 	num_queues = READ_ONCE(vif->num_queues);
238 
239 	/* Aggregate tx and rx stats from each queue */
240 	for (index = 0; index < num_queues; ++index) {
241 		queue = &vif->queues[index];
242 		rx_bytes += queue->stats.rx_bytes;
243 		rx_packets += queue->stats.rx_packets;
244 		tx_bytes += queue->stats.tx_bytes;
245 		tx_packets += queue->stats.tx_packets;
246 	}
247 
248 	rcu_read_unlock();
249 
250 	vif->dev->stats.rx_bytes = rx_bytes;
251 	vif->dev->stats.rx_packets = rx_packets;
252 	vif->dev->stats.tx_bytes = tx_bytes;
253 	vif->dev->stats.tx_packets = tx_packets;
254 
255 	return &vif->dev->stats;
256 }
257 
258 static void xenvif_up(struct xenvif *vif)
259 {
260 	struct xenvif_queue *queue = NULL;
261 	unsigned int num_queues = vif->num_queues;
262 	unsigned int queue_index;
263 
264 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
265 		queue = &vif->queues[queue_index];
266 		napi_enable(&queue->napi);
267 		enable_irq(queue->tx_irq);
268 		if (queue->tx_irq != queue->rx_irq)
269 			enable_irq(queue->rx_irq);
270 		xenvif_napi_schedule_or_enable_events(queue);
271 	}
272 }
273 
274 static void xenvif_down(struct xenvif *vif)
275 {
276 	struct xenvif_queue *queue = NULL;
277 	unsigned int num_queues = vif->num_queues;
278 	unsigned int queue_index;
279 
280 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
281 		queue = &vif->queues[queue_index];
282 		disable_irq(queue->tx_irq);
283 		if (queue->tx_irq != queue->rx_irq)
284 			disable_irq(queue->rx_irq);
285 		napi_disable(&queue->napi);
286 		del_timer_sync(&queue->credit_timeout);
287 	}
288 }
289 
290 static int xenvif_open(struct net_device *dev)
291 {
292 	struct xenvif *vif = netdev_priv(dev);
293 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
294 		xenvif_up(vif);
295 	netif_tx_start_all_queues(dev);
296 	return 0;
297 }
298 
299 static int xenvif_close(struct net_device *dev)
300 {
301 	struct xenvif *vif = netdev_priv(dev);
302 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
303 		xenvif_down(vif);
304 	netif_tx_stop_all_queues(dev);
305 	return 0;
306 }
307 
308 static int xenvif_change_mtu(struct net_device *dev, int mtu)
309 {
310 	struct xenvif *vif = netdev_priv(dev);
311 	int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
312 
313 	if (mtu > max)
314 		return -EINVAL;
315 	dev->mtu = mtu;
316 	return 0;
317 }
318 
319 static netdev_features_t xenvif_fix_features(struct net_device *dev,
320 	netdev_features_t features)
321 {
322 	struct xenvif *vif = netdev_priv(dev);
323 
324 	if (!vif->can_sg)
325 		features &= ~NETIF_F_SG;
326 	if (~(vif->gso_mask) & GSO_BIT(TCPV4))
327 		features &= ~NETIF_F_TSO;
328 	if (~(vif->gso_mask) & GSO_BIT(TCPV6))
329 		features &= ~NETIF_F_TSO6;
330 	if (!vif->ip_csum)
331 		features &= ~NETIF_F_IP_CSUM;
332 	if (!vif->ipv6_csum)
333 		features &= ~NETIF_F_IPV6_CSUM;
334 
335 	return features;
336 }
337 
338 static const struct xenvif_stat {
339 	char name[ETH_GSTRING_LEN];
340 	u16 offset;
341 } xenvif_stats[] = {
342 	{
343 		"rx_gso_checksum_fixup",
344 		offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
345 	},
346 	/* If (sent != success + fail), there are probably packets never
347 	 * freed up properly!
348 	 */
349 	{
350 		"tx_zerocopy_sent",
351 		offsetof(struct xenvif_stats, tx_zerocopy_sent),
352 	},
353 	{
354 		"tx_zerocopy_success",
355 		offsetof(struct xenvif_stats, tx_zerocopy_success),
356 	},
357 	{
358 		"tx_zerocopy_fail",
359 		offsetof(struct xenvif_stats, tx_zerocopy_fail)
360 	},
361 	/* Number of packets exceeding MAX_SKB_FRAG slots. You should use
362 	 * a guest with the same MAX_SKB_FRAG
363 	 */
364 	{
365 		"tx_frag_overflow",
366 		offsetof(struct xenvif_stats, tx_frag_overflow)
367 	},
368 };
369 
370 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
371 {
372 	switch (string_set) {
373 	case ETH_SS_STATS:
374 		return ARRAY_SIZE(xenvif_stats);
375 	default:
376 		return -EINVAL;
377 	}
378 }
379 
380 static void xenvif_get_ethtool_stats(struct net_device *dev,
381 				     struct ethtool_stats *stats, u64 * data)
382 {
383 	struct xenvif *vif = netdev_priv(dev);
384 	unsigned int num_queues;
385 	int i;
386 	unsigned int queue_index;
387 
388 	rcu_read_lock();
389 	num_queues = READ_ONCE(vif->num_queues);
390 
391 	for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
392 		unsigned long accum = 0;
393 		for (queue_index = 0; queue_index < num_queues; ++queue_index) {
394 			void *vif_stats = &vif->queues[queue_index].stats;
395 			accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
396 		}
397 		data[i] = accum;
398 	}
399 
400 	rcu_read_unlock();
401 }
402 
403 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
404 {
405 	int i;
406 
407 	switch (stringset) {
408 	case ETH_SS_STATS:
409 		for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
410 			memcpy(data + i * ETH_GSTRING_LEN,
411 			       xenvif_stats[i].name, ETH_GSTRING_LEN);
412 		break;
413 	}
414 }
415 
416 static const struct ethtool_ops xenvif_ethtool_ops = {
417 	.get_link	= ethtool_op_get_link,
418 
419 	.get_sset_count = xenvif_get_sset_count,
420 	.get_ethtool_stats = xenvif_get_ethtool_stats,
421 	.get_strings = xenvif_get_strings,
422 };
423 
424 static const struct net_device_ops xenvif_netdev_ops = {
425 	.ndo_select_queue = xenvif_select_queue,
426 	.ndo_start_xmit	= xenvif_start_xmit,
427 	.ndo_get_stats	= xenvif_get_stats,
428 	.ndo_open	= xenvif_open,
429 	.ndo_stop	= xenvif_close,
430 	.ndo_change_mtu	= xenvif_change_mtu,
431 	.ndo_fix_features = xenvif_fix_features,
432 	.ndo_set_mac_address = eth_mac_addr,
433 	.ndo_validate_addr   = eth_validate_addr,
434 };
435 
436 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
437 			    unsigned int handle)
438 {
439 	int err;
440 	struct net_device *dev;
441 	struct xenvif *vif;
442 	char name[IFNAMSIZ] = {};
443 
444 	snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
445 	/* Allocate a netdev with the max. supported number of queues.
446 	 * When the guest selects the desired number, it will be updated
447 	 * via netif_set_real_num_*_queues().
448 	 */
449 	dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
450 			      ether_setup, xenvif_max_queues);
451 	if (dev == NULL) {
452 		pr_warn("Could not allocate netdev for %s\n", name);
453 		return ERR_PTR(-ENOMEM);
454 	}
455 
456 	SET_NETDEV_DEV(dev, parent);
457 
458 	vif = netdev_priv(dev);
459 
460 	vif->domid  = domid;
461 	vif->handle = handle;
462 	vif->can_sg = 1;
463 	vif->ip_csum = 1;
464 	vif->dev = dev;
465 	vif->disabled = false;
466 	vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
467 	vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
468 
469 	/* Start out with no queues. */
470 	vif->queues = NULL;
471 	vif->num_queues = 0;
472 
473 	spin_lock_init(&vif->lock);
474 	INIT_LIST_HEAD(&vif->fe_mcast_addr);
475 
476 	dev->netdev_ops	= &xenvif_netdev_ops;
477 	dev->hw_features = NETIF_F_SG |
478 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
479 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
480 	dev->features = dev->hw_features | NETIF_F_RXCSUM;
481 	dev->ethtool_ops = &xenvif_ethtool_ops;
482 
483 	dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
484 
485 	dev->min_mtu = 0;
486 	dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
487 
488 	/*
489 	 * Initialise a dummy MAC address. We choose the numerically
490 	 * largest non-broadcast address to prevent the address getting
491 	 * stolen by an Ethernet bridge for STP purposes.
492 	 * (FE:FF:FF:FF:FF:FF)
493 	 */
494 	eth_broadcast_addr(dev->dev_addr);
495 	dev->dev_addr[0] &= ~0x01;
496 
497 	netif_carrier_off(dev);
498 
499 	err = register_netdev(dev);
500 	if (err) {
501 		netdev_warn(dev, "Could not register device: err=%d\n", err);
502 		free_netdev(dev);
503 		return ERR_PTR(err);
504 	}
505 
506 	netdev_dbg(dev, "Successfully created xenvif\n");
507 
508 	__module_get(THIS_MODULE);
509 
510 	return vif;
511 }
512 
513 int xenvif_init_queue(struct xenvif_queue *queue)
514 {
515 	int err, i;
516 
517 	queue->credit_bytes = queue->remaining_credit = ~0UL;
518 	queue->credit_usec  = 0UL;
519 	init_timer(&queue->credit_timeout);
520 	queue->credit_timeout.function = xenvif_tx_credit_callback;
521 	queue->credit_window_start = get_jiffies_64();
522 
523 	queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
524 
525 	skb_queue_head_init(&queue->rx_queue);
526 	skb_queue_head_init(&queue->tx_queue);
527 
528 	queue->pending_cons = 0;
529 	queue->pending_prod = MAX_PENDING_REQS;
530 	for (i = 0; i < MAX_PENDING_REQS; ++i)
531 		queue->pending_ring[i] = i;
532 
533 	spin_lock_init(&queue->callback_lock);
534 	spin_lock_init(&queue->response_lock);
535 
536 	/* If ballooning is disabled, this will consume real memory, so you
537 	 * better enable it. The long term solution would be to use just a
538 	 * bunch of valid page descriptors, without dependency on ballooning
539 	 */
540 	err = gnttab_alloc_pages(MAX_PENDING_REQS,
541 				 queue->mmap_pages);
542 	if (err) {
543 		netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
544 		return -ENOMEM;
545 	}
546 
547 	for (i = 0; i < MAX_PENDING_REQS; i++) {
548 		queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
549 			{ .callback = xenvif_zerocopy_callback,
550 			  .ctx = NULL,
551 			  .desc = i };
552 		queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
553 	}
554 
555 	return 0;
556 }
557 
558 void xenvif_carrier_on(struct xenvif *vif)
559 {
560 	rtnl_lock();
561 	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
562 		dev_set_mtu(vif->dev, ETH_DATA_LEN);
563 	netdev_update_features(vif->dev);
564 	set_bit(VIF_STATUS_CONNECTED, &vif->status);
565 	if (netif_running(vif->dev))
566 		xenvif_up(vif);
567 	rtnl_unlock();
568 }
569 
570 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
571 			unsigned int evtchn)
572 {
573 	struct net_device *dev = vif->dev;
574 	void *addr;
575 	struct xen_netif_ctrl_sring *shared;
576 	int err;
577 
578 	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
579 				     &ring_ref, 1, &addr);
580 	if (err)
581 		goto err;
582 
583 	shared = (struct xen_netif_ctrl_sring *)addr;
584 	BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
585 
586 	err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
587 	if (err < 0)
588 		goto err_unmap;
589 
590 	vif->ctrl_irq = err;
591 
592 	xenvif_init_hash(vif);
593 
594 	err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
595 				   IRQF_ONESHOT, "xen-netback-ctrl", vif);
596 	if (err) {
597 		pr_warn("Could not setup irq handler for %s\n", dev->name);
598 		goto err_deinit;
599 	}
600 
601 	return 0;
602 
603 err_deinit:
604 	xenvif_deinit_hash(vif);
605 	unbind_from_irqhandler(vif->ctrl_irq, vif);
606 	vif->ctrl_irq = 0;
607 
608 err_unmap:
609 	xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
610 				vif->ctrl.sring);
611 	vif->ctrl.sring = NULL;
612 
613 err:
614 	return err;
615 }
616 
617 int xenvif_connect_data(struct xenvif_queue *queue,
618 			unsigned long tx_ring_ref,
619 			unsigned long rx_ring_ref,
620 			unsigned int tx_evtchn,
621 			unsigned int rx_evtchn)
622 {
623 	struct task_struct *task;
624 	int err = -ENOMEM;
625 
626 	BUG_ON(queue->tx_irq);
627 	BUG_ON(queue->task);
628 	BUG_ON(queue->dealloc_task);
629 
630 	err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
631 					     rx_ring_ref);
632 	if (err < 0)
633 		goto err;
634 
635 	init_waitqueue_head(&queue->wq);
636 	init_waitqueue_head(&queue->dealloc_wq);
637 	atomic_set(&queue->inflight_packets, 0);
638 
639 	netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
640 			XENVIF_NAPI_WEIGHT);
641 
642 	if (tx_evtchn == rx_evtchn) {
643 		/* feature-split-event-channels == 0 */
644 		err = bind_interdomain_evtchn_to_irqhandler(
645 			queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
646 			queue->name, queue);
647 		if (err < 0)
648 			goto err_unmap;
649 		queue->tx_irq = queue->rx_irq = err;
650 		disable_irq(queue->tx_irq);
651 	} else {
652 		/* feature-split-event-channels == 1 */
653 		snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
654 			 "%s-tx", queue->name);
655 		err = bind_interdomain_evtchn_to_irqhandler(
656 			queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
657 			queue->tx_irq_name, queue);
658 		if (err < 0)
659 			goto err_unmap;
660 		queue->tx_irq = err;
661 		disable_irq(queue->tx_irq);
662 
663 		snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
664 			 "%s-rx", queue->name);
665 		err = bind_interdomain_evtchn_to_irqhandler(
666 			queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
667 			queue->rx_irq_name, queue);
668 		if (err < 0)
669 			goto err_tx_unbind;
670 		queue->rx_irq = err;
671 		disable_irq(queue->rx_irq);
672 	}
673 
674 	queue->stalled = true;
675 
676 	task = kthread_create(xenvif_kthread_guest_rx,
677 			      (void *)queue, "%s-guest-rx", queue->name);
678 	if (IS_ERR(task)) {
679 		pr_warn("Could not allocate kthread for %s\n", queue->name);
680 		err = PTR_ERR(task);
681 		goto err_rx_unbind;
682 	}
683 	queue->task = task;
684 	get_task_struct(task);
685 
686 	task = kthread_create(xenvif_dealloc_kthread,
687 			      (void *)queue, "%s-dealloc", queue->name);
688 	if (IS_ERR(task)) {
689 		pr_warn("Could not allocate kthread for %s\n", queue->name);
690 		err = PTR_ERR(task);
691 		goto err_rx_unbind;
692 	}
693 	queue->dealloc_task = task;
694 
695 	wake_up_process(queue->task);
696 	wake_up_process(queue->dealloc_task);
697 
698 	return 0;
699 
700 err_rx_unbind:
701 	unbind_from_irqhandler(queue->rx_irq, queue);
702 	queue->rx_irq = 0;
703 err_tx_unbind:
704 	unbind_from_irqhandler(queue->tx_irq, queue);
705 	queue->tx_irq = 0;
706 err_unmap:
707 	xenvif_unmap_frontend_data_rings(queue);
708 	netif_napi_del(&queue->napi);
709 err:
710 	module_put(THIS_MODULE);
711 	return err;
712 }
713 
714 void xenvif_carrier_off(struct xenvif *vif)
715 {
716 	struct net_device *dev = vif->dev;
717 
718 	rtnl_lock();
719 	if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
720 		netif_carrier_off(dev); /* discard queued packets */
721 		if (netif_running(dev))
722 			xenvif_down(vif);
723 	}
724 	rtnl_unlock();
725 }
726 
727 void xenvif_disconnect_data(struct xenvif *vif)
728 {
729 	struct xenvif_queue *queue = NULL;
730 	unsigned int num_queues = vif->num_queues;
731 	unsigned int queue_index;
732 
733 	xenvif_carrier_off(vif);
734 
735 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
736 		queue = &vif->queues[queue_index];
737 
738 		netif_napi_del(&queue->napi);
739 
740 		if (queue->task) {
741 			kthread_stop(queue->task);
742 			put_task_struct(queue->task);
743 			queue->task = NULL;
744 		}
745 
746 		if (queue->dealloc_task) {
747 			kthread_stop(queue->dealloc_task);
748 			queue->dealloc_task = NULL;
749 		}
750 
751 		if (queue->tx_irq) {
752 			if (queue->tx_irq == queue->rx_irq)
753 				unbind_from_irqhandler(queue->tx_irq, queue);
754 			else {
755 				unbind_from_irqhandler(queue->tx_irq, queue);
756 				unbind_from_irqhandler(queue->rx_irq, queue);
757 			}
758 			queue->tx_irq = 0;
759 		}
760 
761 		xenvif_unmap_frontend_data_rings(queue);
762 	}
763 
764 	xenvif_mcast_addr_list_free(vif);
765 }
766 
767 void xenvif_disconnect_ctrl(struct xenvif *vif)
768 {
769 	if (vif->ctrl_irq) {
770 		xenvif_deinit_hash(vif);
771 		unbind_from_irqhandler(vif->ctrl_irq, vif);
772 		vif->ctrl_irq = 0;
773 	}
774 
775 	if (vif->ctrl.sring) {
776 		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
777 					vif->ctrl.sring);
778 		vif->ctrl.sring = NULL;
779 	}
780 }
781 
782 /* Reverse the relevant parts of xenvif_init_queue().
783  * Used for queue teardown from xenvif_free(), and on the
784  * error handling paths in xenbus.c:connect().
785  */
786 void xenvif_deinit_queue(struct xenvif_queue *queue)
787 {
788 	gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
789 }
790 
791 void xenvif_free(struct xenvif *vif)
792 {
793 	struct xenvif_queue *queues = vif->queues;
794 	unsigned int num_queues = vif->num_queues;
795 	unsigned int queue_index;
796 
797 	unregister_netdev(vif->dev);
798 	free_netdev(vif->dev);
799 
800 	for (queue_index = 0; queue_index < num_queues; ++queue_index)
801 		xenvif_deinit_queue(&queues[queue_index]);
802 	vfree(queues);
803 
804 	module_put(THIS_MODULE);
805 }
806