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 			       void *accel_priv,
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) % 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[skb_get_hash_raw(skb) % size];
166 }
167 
168 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
169 {
170 	struct xenvif *vif = netdev_priv(dev);
171 	struct xenvif_queue *queue = NULL;
172 	unsigned int num_queues;
173 	u16 index;
174 	struct xenvif_rx_cb *cb;
175 
176 	BUG_ON(skb->dev != dev);
177 
178 	/* Drop the packet if queues are not set up.
179 	 * This handler should be called inside an RCU read section
180 	 * so we don't need to enter it here explicitly.
181 	 */
182 	num_queues = READ_ONCE(vif->num_queues);
183 	if (num_queues < 1)
184 		goto drop;
185 
186 	/* Obtain the queue to be used to transmit this packet */
187 	index = skb_get_queue_mapping(skb);
188 	if (index >= num_queues) {
189 		pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
190 				    index, vif->dev->name);
191 		index %= num_queues;
192 	}
193 	queue = &vif->queues[index];
194 
195 	/* Drop the packet if queue is not ready */
196 	if (queue->task == NULL ||
197 	    queue->dealloc_task == NULL ||
198 	    !xenvif_schedulable(vif))
199 		goto drop;
200 
201 	if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
202 		struct ethhdr *eth = (struct ethhdr *)skb->data;
203 
204 		if (!xenvif_mcast_match(vif, eth->h_dest))
205 			goto drop;
206 	}
207 
208 	cb = XENVIF_RX_CB(skb);
209 	cb->expires = jiffies + vif->drain_timeout;
210 
211 	/* If there is no hash algorithm configured then make sure there
212 	 * is no hash information in the socket buffer otherwise it
213 	 * would be incorrectly forwarded to the frontend.
214 	 */
215 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
216 		skb_clear_hash(skb);
217 
218 	xenvif_rx_queue_tail(queue, skb);
219 	xenvif_kick_thread(queue);
220 
221 	return NETDEV_TX_OK;
222 
223  drop:
224 	vif->dev->stats.tx_dropped++;
225 	dev_kfree_skb(skb);
226 	return NETDEV_TX_OK;
227 }
228 
229 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
230 {
231 	struct xenvif *vif = netdev_priv(dev);
232 	struct xenvif_queue *queue = NULL;
233 	unsigned int num_queues;
234 	u64 rx_bytes = 0;
235 	u64 rx_packets = 0;
236 	u64 tx_bytes = 0;
237 	u64 tx_packets = 0;
238 	unsigned int index;
239 
240 	rcu_read_lock();
241 	num_queues = READ_ONCE(vif->num_queues);
242 
243 	/* Aggregate tx and rx stats from each queue */
244 	for (index = 0; index < num_queues; ++index) {
245 		queue = &vif->queues[index];
246 		rx_bytes += queue->stats.rx_bytes;
247 		rx_packets += queue->stats.rx_packets;
248 		tx_bytes += queue->stats.tx_bytes;
249 		tx_packets += queue->stats.tx_packets;
250 	}
251 
252 	rcu_read_unlock();
253 
254 	vif->dev->stats.rx_bytes = rx_bytes;
255 	vif->dev->stats.rx_packets = rx_packets;
256 	vif->dev->stats.tx_bytes = tx_bytes;
257 	vif->dev->stats.tx_packets = tx_packets;
258 
259 	return &vif->dev->stats;
260 }
261 
262 static void xenvif_up(struct xenvif *vif)
263 {
264 	struct xenvif_queue *queue = NULL;
265 	unsigned int num_queues = vif->num_queues;
266 	unsigned int queue_index;
267 
268 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
269 		queue = &vif->queues[queue_index];
270 		napi_enable(&queue->napi);
271 		enable_irq(queue->tx_irq);
272 		if (queue->tx_irq != queue->rx_irq)
273 			enable_irq(queue->rx_irq);
274 		xenvif_napi_schedule_or_enable_events(queue);
275 	}
276 }
277 
278 static void xenvif_down(struct xenvif *vif)
279 {
280 	struct xenvif_queue *queue = NULL;
281 	unsigned int num_queues = vif->num_queues;
282 	unsigned int queue_index;
283 
284 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
285 		queue = &vif->queues[queue_index];
286 		disable_irq(queue->tx_irq);
287 		if (queue->tx_irq != queue->rx_irq)
288 			disable_irq(queue->rx_irq);
289 		napi_disable(&queue->napi);
290 		del_timer_sync(&queue->credit_timeout);
291 	}
292 }
293 
294 static int xenvif_open(struct net_device *dev)
295 {
296 	struct xenvif *vif = netdev_priv(dev);
297 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
298 		xenvif_up(vif);
299 	netif_tx_start_all_queues(dev);
300 	return 0;
301 }
302 
303 static int xenvif_close(struct net_device *dev)
304 {
305 	struct xenvif *vif = netdev_priv(dev);
306 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
307 		xenvif_down(vif);
308 	netif_tx_stop_all_queues(dev);
309 	return 0;
310 }
311 
312 static int xenvif_change_mtu(struct net_device *dev, int mtu)
313 {
314 	struct xenvif *vif = netdev_priv(dev);
315 	int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
316 
317 	if (mtu > max)
318 		return -EINVAL;
319 	dev->mtu = mtu;
320 	return 0;
321 }
322 
323 static netdev_features_t xenvif_fix_features(struct net_device *dev,
324 	netdev_features_t features)
325 {
326 	struct xenvif *vif = netdev_priv(dev);
327 
328 	if (!vif->can_sg)
329 		features &= ~NETIF_F_SG;
330 	if (~(vif->gso_mask) & GSO_BIT(TCPV4))
331 		features &= ~NETIF_F_TSO;
332 	if (~(vif->gso_mask) & GSO_BIT(TCPV6))
333 		features &= ~NETIF_F_TSO6;
334 	if (!vif->ip_csum)
335 		features &= ~NETIF_F_IP_CSUM;
336 	if (!vif->ipv6_csum)
337 		features &= ~NETIF_F_IPV6_CSUM;
338 
339 	return features;
340 }
341 
342 static const struct xenvif_stat {
343 	char name[ETH_GSTRING_LEN];
344 	u16 offset;
345 } xenvif_stats[] = {
346 	{
347 		"rx_gso_checksum_fixup",
348 		offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
349 	},
350 	/* If (sent != success + fail), there are probably packets never
351 	 * freed up properly!
352 	 */
353 	{
354 		"tx_zerocopy_sent",
355 		offsetof(struct xenvif_stats, tx_zerocopy_sent),
356 	},
357 	{
358 		"tx_zerocopy_success",
359 		offsetof(struct xenvif_stats, tx_zerocopy_success),
360 	},
361 	{
362 		"tx_zerocopy_fail",
363 		offsetof(struct xenvif_stats, tx_zerocopy_fail)
364 	},
365 	/* Number of packets exceeding MAX_SKB_FRAG slots. You should use
366 	 * a guest with the same MAX_SKB_FRAG
367 	 */
368 	{
369 		"tx_frag_overflow",
370 		offsetof(struct xenvif_stats, tx_frag_overflow)
371 	},
372 };
373 
374 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
375 {
376 	switch (string_set) {
377 	case ETH_SS_STATS:
378 		return ARRAY_SIZE(xenvif_stats);
379 	default:
380 		return -EINVAL;
381 	}
382 }
383 
384 static void xenvif_get_ethtool_stats(struct net_device *dev,
385 				     struct ethtool_stats *stats, u64 * data)
386 {
387 	struct xenvif *vif = netdev_priv(dev);
388 	unsigned int num_queues;
389 	int i;
390 	unsigned int queue_index;
391 
392 	rcu_read_lock();
393 	num_queues = READ_ONCE(vif->num_queues);
394 
395 	for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
396 		unsigned long accum = 0;
397 		for (queue_index = 0; queue_index < num_queues; ++queue_index) {
398 			void *vif_stats = &vif->queues[queue_index].stats;
399 			accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
400 		}
401 		data[i] = accum;
402 	}
403 
404 	rcu_read_unlock();
405 }
406 
407 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
408 {
409 	int i;
410 
411 	switch (stringset) {
412 	case ETH_SS_STATS:
413 		for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
414 			memcpy(data + i * ETH_GSTRING_LEN,
415 			       xenvif_stats[i].name, ETH_GSTRING_LEN);
416 		break;
417 	}
418 }
419 
420 static const struct ethtool_ops xenvif_ethtool_ops = {
421 	.get_link	= ethtool_op_get_link,
422 
423 	.get_sset_count = xenvif_get_sset_count,
424 	.get_ethtool_stats = xenvif_get_ethtool_stats,
425 	.get_strings = xenvif_get_strings,
426 };
427 
428 static const struct net_device_ops xenvif_netdev_ops = {
429 	.ndo_select_queue = xenvif_select_queue,
430 	.ndo_start_xmit	= xenvif_start_xmit,
431 	.ndo_get_stats	= xenvif_get_stats,
432 	.ndo_open	= xenvif_open,
433 	.ndo_stop	= xenvif_close,
434 	.ndo_change_mtu	= xenvif_change_mtu,
435 	.ndo_fix_features = xenvif_fix_features,
436 	.ndo_set_mac_address = eth_mac_addr,
437 	.ndo_validate_addr   = eth_validate_addr,
438 };
439 
440 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
441 			    unsigned int handle)
442 {
443 	int err;
444 	struct net_device *dev;
445 	struct xenvif *vif;
446 	char name[IFNAMSIZ] = {};
447 
448 	snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
449 	/* Allocate a netdev with the max. supported number of queues.
450 	 * When the guest selects the desired number, it will be updated
451 	 * via netif_set_real_num_*_queues().
452 	 */
453 	dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
454 			      ether_setup, xenvif_max_queues);
455 	if (dev == NULL) {
456 		pr_warn("Could not allocate netdev for %s\n", name);
457 		return ERR_PTR(-ENOMEM);
458 	}
459 
460 	SET_NETDEV_DEV(dev, parent);
461 
462 	vif = netdev_priv(dev);
463 
464 	vif->domid  = domid;
465 	vif->handle = handle;
466 	vif->can_sg = 1;
467 	vif->ip_csum = 1;
468 	vif->dev = dev;
469 	vif->disabled = false;
470 	vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
471 	vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
472 
473 	/* Start out with no queues. */
474 	vif->queues = NULL;
475 	vif->num_queues = 0;
476 
477 	spin_lock_init(&vif->lock);
478 	INIT_LIST_HEAD(&vif->fe_mcast_addr);
479 
480 	dev->netdev_ops	= &xenvif_netdev_ops;
481 	dev->hw_features = NETIF_F_SG |
482 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
483 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
484 	dev->features = dev->hw_features | NETIF_F_RXCSUM;
485 	dev->ethtool_ops = &xenvif_ethtool_ops;
486 
487 	dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
488 
489 	dev->min_mtu = 0;
490 	dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
491 
492 	/*
493 	 * Initialise a dummy MAC address. We choose the numerically
494 	 * largest non-broadcast address to prevent the address getting
495 	 * stolen by an Ethernet bridge for STP purposes.
496 	 * (FE:FF:FF:FF:FF:FF)
497 	 */
498 	eth_broadcast_addr(dev->dev_addr);
499 	dev->dev_addr[0] &= ~0x01;
500 
501 	netif_carrier_off(dev);
502 
503 	err = register_netdev(dev);
504 	if (err) {
505 		netdev_warn(dev, "Could not register device: err=%d\n", err);
506 		free_netdev(dev);
507 		return ERR_PTR(err);
508 	}
509 
510 	netdev_dbg(dev, "Successfully created xenvif\n");
511 
512 	__module_get(THIS_MODULE);
513 
514 	return vif;
515 }
516 
517 int xenvif_init_queue(struct xenvif_queue *queue)
518 {
519 	int err, i;
520 
521 	queue->credit_bytes = queue->remaining_credit = ~0UL;
522 	queue->credit_usec  = 0UL;
523 	init_timer(&queue->credit_timeout);
524 	queue->credit_timeout.function = xenvif_tx_credit_callback;
525 	queue->credit_window_start = get_jiffies_64();
526 
527 	queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
528 
529 	skb_queue_head_init(&queue->rx_queue);
530 	skb_queue_head_init(&queue->tx_queue);
531 
532 	queue->pending_cons = 0;
533 	queue->pending_prod = MAX_PENDING_REQS;
534 	for (i = 0; i < MAX_PENDING_REQS; ++i)
535 		queue->pending_ring[i] = i;
536 
537 	spin_lock_init(&queue->callback_lock);
538 	spin_lock_init(&queue->response_lock);
539 
540 	/* If ballooning is disabled, this will consume real memory, so you
541 	 * better enable it. The long term solution would be to use just a
542 	 * bunch of valid page descriptors, without dependency on ballooning
543 	 */
544 	err = gnttab_alloc_pages(MAX_PENDING_REQS,
545 				 queue->mmap_pages);
546 	if (err) {
547 		netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
548 		return -ENOMEM;
549 	}
550 
551 	for (i = 0; i < MAX_PENDING_REQS; i++) {
552 		queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
553 			{ .callback = xenvif_zerocopy_callback,
554 			  { { .ctx = NULL,
555 			      .desc = i } } };
556 		queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
557 	}
558 
559 	return 0;
560 }
561 
562 void xenvif_carrier_on(struct xenvif *vif)
563 {
564 	rtnl_lock();
565 	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
566 		dev_set_mtu(vif->dev, ETH_DATA_LEN);
567 	netdev_update_features(vif->dev);
568 	set_bit(VIF_STATUS_CONNECTED, &vif->status);
569 	if (netif_running(vif->dev))
570 		xenvif_up(vif);
571 	rtnl_unlock();
572 }
573 
574 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
575 			unsigned int evtchn)
576 {
577 	struct net_device *dev = vif->dev;
578 	void *addr;
579 	struct xen_netif_ctrl_sring *shared;
580 	int err;
581 
582 	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
583 				     &ring_ref, 1, &addr);
584 	if (err)
585 		goto err;
586 
587 	shared = (struct xen_netif_ctrl_sring *)addr;
588 	BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
589 
590 	err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
591 	if (err < 0)
592 		goto err_unmap;
593 
594 	vif->ctrl_irq = err;
595 
596 	xenvif_init_hash(vif);
597 
598 	err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
599 				   IRQF_ONESHOT, "xen-netback-ctrl", vif);
600 	if (err) {
601 		pr_warn("Could not setup irq handler for %s\n", dev->name);
602 		goto err_deinit;
603 	}
604 
605 	return 0;
606 
607 err_deinit:
608 	xenvif_deinit_hash(vif);
609 	unbind_from_irqhandler(vif->ctrl_irq, vif);
610 	vif->ctrl_irq = 0;
611 
612 err_unmap:
613 	xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
614 				vif->ctrl.sring);
615 	vif->ctrl.sring = NULL;
616 
617 err:
618 	return err;
619 }
620 
621 int xenvif_connect_data(struct xenvif_queue *queue,
622 			unsigned long tx_ring_ref,
623 			unsigned long rx_ring_ref,
624 			unsigned int tx_evtchn,
625 			unsigned int rx_evtchn)
626 {
627 	struct task_struct *task;
628 	int err = -ENOMEM;
629 
630 	BUG_ON(queue->tx_irq);
631 	BUG_ON(queue->task);
632 	BUG_ON(queue->dealloc_task);
633 
634 	err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
635 					     rx_ring_ref);
636 	if (err < 0)
637 		goto err;
638 
639 	init_waitqueue_head(&queue->wq);
640 	init_waitqueue_head(&queue->dealloc_wq);
641 	atomic_set(&queue->inflight_packets, 0);
642 
643 	netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
644 			XENVIF_NAPI_WEIGHT);
645 
646 	if (tx_evtchn == rx_evtchn) {
647 		/* feature-split-event-channels == 0 */
648 		err = bind_interdomain_evtchn_to_irqhandler(
649 			queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
650 			queue->name, queue);
651 		if (err < 0)
652 			goto err_unmap;
653 		queue->tx_irq = queue->rx_irq = err;
654 		disable_irq(queue->tx_irq);
655 	} else {
656 		/* feature-split-event-channels == 1 */
657 		snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
658 			 "%s-tx", queue->name);
659 		err = bind_interdomain_evtchn_to_irqhandler(
660 			queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
661 			queue->tx_irq_name, queue);
662 		if (err < 0)
663 			goto err_unmap;
664 		queue->tx_irq = err;
665 		disable_irq(queue->tx_irq);
666 
667 		snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
668 			 "%s-rx", queue->name);
669 		err = bind_interdomain_evtchn_to_irqhandler(
670 			queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
671 			queue->rx_irq_name, queue);
672 		if (err < 0)
673 			goto err_tx_unbind;
674 		queue->rx_irq = err;
675 		disable_irq(queue->rx_irq);
676 	}
677 
678 	queue->stalled = true;
679 
680 	task = kthread_create(xenvif_kthread_guest_rx,
681 			      (void *)queue, "%s-guest-rx", queue->name);
682 	if (IS_ERR(task)) {
683 		pr_warn("Could not allocate kthread for %s\n", queue->name);
684 		err = PTR_ERR(task);
685 		goto err_rx_unbind;
686 	}
687 	queue->task = task;
688 	get_task_struct(task);
689 
690 	task = kthread_create(xenvif_dealloc_kthread,
691 			      (void *)queue, "%s-dealloc", queue->name);
692 	if (IS_ERR(task)) {
693 		pr_warn("Could not allocate kthread for %s\n", queue->name);
694 		err = PTR_ERR(task);
695 		goto err_rx_unbind;
696 	}
697 	queue->dealloc_task = task;
698 
699 	wake_up_process(queue->task);
700 	wake_up_process(queue->dealloc_task);
701 
702 	return 0;
703 
704 err_rx_unbind:
705 	unbind_from_irqhandler(queue->rx_irq, queue);
706 	queue->rx_irq = 0;
707 err_tx_unbind:
708 	unbind_from_irqhandler(queue->tx_irq, queue);
709 	queue->tx_irq = 0;
710 err_unmap:
711 	xenvif_unmap_frontend_data_rings(queue);
712 	netif_napi_del(&queue->napi);
713 err:
714 	module_put(THIS_MODULE);
715 	return err;
716 }
717 
718 void xenvif_carrier_off(struct xenvif *vif)
719 {
720 	struct net_device *dev = vif->dev;
721 
722 	rtnl_lock();
723 	if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
724 		netif_carrier_off(dev); /* discard queued packets */
725 		if (netif_running(dev))
726 			xenvif_down(vif);
727 	}
728 	rtnl_unlock();
729 }
730 
731 void xenvif_disconnect_data(struct xenvif *vif)
732 {
733 	struct xenvif_queue *queue = NULL;
734 	unsigned int num_queues = vif->num_queues;
735 	unsigned int queue_index;
736 
737 	xenvif_carrier_off(vif);
738 
739 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
740 		queue = &vif->queues[queue_index];
741 
742 		netif_napi_del(&queue->napi);
743 
744 		if (queue->task) {
745 			kthread_stop(queue->task);
746 			put_task_struct(queue->task);
747 			queue->task = NULL;
748 		}
749 
750 		if (queue->dealloc_task) {
751 			kthread_stop(queue->dealloc_task);
752 			queue->dealloc_task = NULL;
753 		}
754 
755 		if (queue->tx_irq) {
756 			if (queue->tx_irq == queue->rx_irq)
757 				unbind_from_irqhandler(queue->tx_irq, queue);
758 			else {
759 				unbind_from_irqhandler(queue->tx_irq, queue);
760 				unbind_from_irqhandler(queue->rx_irq, queue);
761 			}
762 			queue->tx_irq = 0;
763 		}
764 
765 		xenvif_unmap_frontend_data_rings(queue);
766 	}
767 
768 	xenvif_mcast_addr_list_free(vif);
769 }
770 
771 void xenvif_disconnect_ctrl(struct xenvif *vif)
772 {
773 	if (vif->ctrl_irq) {
774 		xenvif_deinit_hash(vif);
775 		unbind_from_irqhandler(vif->ctrl_irq, vif);
776 		vif->ctrl_irq = 0;
777 	}
778 
779 	if (vif->ctrl.sring) {
780 		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
781 					vif->ctrl.sring);
782 		vif->ctrl.sring = NULL;
783 	}
784 }
785 
786 /* Reverse the relevant parts of xenvif_init_queue().
787  * Used for queue teardown from xenvif_free(), and on the
788  * error handling paths in xenbus.c:connect().
789  */
790 void xenvif_deinit_queue(struct xenvif_queue *queue)
791 {
792 	gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
793 }
794 
795 void xenvif_free(struct xenvif *vif)
796 {
797 	struct xenvif_queue *queues = vif->queues;
798 	unsigned int num_queues = vif->num_queues;
799 	unsigned int queue_index;
800 
801 	unregister_netdev(vif->dev);
802 	free_netdev(vif->dev);
803 
804 	for (queue_index = 0; queue_index < num_queues; ++queue_index)
805 		xenvif_deinit_queue(&queues[queue_index]);
806 	vfree(queues);
807 
808 	module_put(THIS_MODULE);
809 }
810