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