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