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_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
72static int 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 bool xenvif_handle_tx_interrupt(struct xenvif_queue *queue)
80{
81	bool rc;
82
83	rc = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
84	if (rc)
85		napi_schedule(&queue->napi);
86	return rc;
87}
88
89static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
90{
91	struct xenvif_queue *queue = dev_id;
92	int old;
93
94	old = atomic_fetch_or(NETBK_TX_EOI, &queue->eoi_pending);
95	WARN(old & NETBK_TX_EOI, "Interrupt while EOI pending\n");
96
97	if (!xenvif_handle_tx_interrupt(queue)) {
98		atomic_andnot(NETBK_TX_EOI, &queue->eoi_pending);
99		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
100	}
101
102	return IRQ_HANDLED;
103}
104
105static int xenvif_poll(struct napi_struct *napi, int budget)
106{
107	struct xenvif_queue *queue =
108		container_of(napi, struct xenvif_queue, napi);
109	int work_done;
110
111	/* This vif is rogue, we pretend we've there is nothing to do
112	 * for this vif to deschedule it from NAPI. But this interface
113	 * will be turned off in thread context later.
114	 */
115	if (unlikely(queue->vif->disabled)) {
116		napi_complete(napi);
117		return 0;
118	}
119
120	work_done = xenvif_tx_action(queue, budget);
121
122	if (work_done < budget) {
123		napi_complete_done(napi, work_done);
124		/* If the queue is rate-limited, it shall be
125		 * rescheduled in the timer callback.
126		 */
127		if (likely(!queue->rate_limited))
128			xenvif_napi_schedule_or_enable_events(queue);
129	}
130
131	return work_done;
132}
133
134static bool xenvif_handle_rx_interrupt(struct xenvif_queue *queue)
135{
136	bool rc;
137
138	rc = xenvif_have_rx_work(queue, false);
139	if (rc)
140		xenvif_kick_thread(queue);
141	return rc;
142}
143
144static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
145{
146	struct xenvif_queue *queue = dev_id;
147	int old;
148
149	old = atomic_fetch_or(NETBK_RX_EOI, &queue->eoi_pending);
150	WARN(old & NETBK_RX_EOI, "Interrupt while EOI pending\n");
151
152	if (!xenvif_handle_rx_interrupt(queue)) {
153		atomic_andnot(NETBK_RX_EOI, &queue->eoi_pending);
154		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
155	}
156
157	return IRQ_HANDLED;
158}
159
160irqreturn_t xenvif_interrupt(int irq, void *dev_id)
161{
162	struct xenvif_queue *queue = dev_id;
163	int old;
164	bool has_rx, has_tx;
165
166	old = atomic_fetch_or(NETBK_COMMON_EOI, &queue->eoi_pending);
167	WARN(old, "Interrupt while EOI pending\n");
168
169	has_tx = xenvif_handle_tx_interrupt(queue);
170	has_rx = xenvif_handle_rx_interrupt(queue);
171
172	if (!has_rx && !has_tx) {
173		atomic_andnot(NETBK_COMMON_EOI, &queue->eoi_pending);
174		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
175	}
176
177	return IRQ_HANDLED;
178}
179
180static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
181			       struct net_device *sb_dev)
182{
183	struct xenvif *vif = netdev_priv(dev);
184	unsigned int size = vif->hash.size;
185	unsigned int num_queues;
186
187	/* If queues are not set up internally - always return 0
188	 * as the packet going to be dropped anyway */
189	num_queues = READ_ONCE(vif->num_queues);
190	if (num_queues < 1)
191		return 0;
192
193	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
194		return netdev_pick_tx(dev, skb, NULL) %
195		       dev->real_num_tx_queues;
196
197	xenvif_set_skb_hash(vif, skb);
198
199	if (size == 0)
200		return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
201
202	return vif->hash.mapping[vif->hash.mapping_sel]
203				[skb_get_hash_raw(skb) % size];
204}
205
206static netdev_tx_t
207xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
208{
209	struct xenvif *vif = netdev_priv(dev);
210	struct xenvif_queue *queue = NULL;
211	unsigned int num_queues;
212	u16 index;
213	struct xenvif_rx_cb *cb;
214
215	BUG_ON(skb->dev != dev);
216
217	/* Drop the packet if queues are not set up.
218	 * This handler should be called inside an RCU read section
219	 * so we don't need to enter it here explicitly.
220	 */
221	num_queues = READ_ONCE(vif->num_queues);
222	if (num_queues < 1)
223		goto drop;
224
225	/* Obtain the queue to be used to transmit this packet */
226	index = skb_get_queue_mapping(skb);
227	if (index >= num_queues) {
228		pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
229				    index, vif->dev->name);
230		index %= num_queues;
231	}
232	queue = &vif->queues[index];
233
234	/* Drop the packet if queue is not ready */
235	if (queue->task == NULL ||
236	    queue->dealloc_task == NULL ||
237	    !xenvif_schedulable(vif))
238		goto drop;
239
240	if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
241		struct ethhdr *eth = (struct ethhdr *)skb->data;
242
243		if (!xenvif_mcast_match(vif, eth->h_dest))
244			goto drop;
245	}
246
247	cb = XENVIF_RX_CB(skb);
248	cb->expires = jiffies + vif->drain_timeout;
249
250	/* If there is no hash algorithm configured then make sure there
251	 * is no hash information in the socket buffer otherwise it
252	 * would be incorrectly forwarded to the frontend.
253	 */
254	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
255		skb_clear_hash(skb);
256
257	if (!xenvif_rx_queue_tail(queue, skb))
258		goto drop;
259
260	xenvif_kick_thread(queue);
261
262	return NETDEV_TX_OK;
263
264 drop:
265	vif->dev->stats.tx_dropped++;
266	dev_kfree_skb_any(skb);
267	return NETDEV_TX_OK;
268}
269
270static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
271{
272	struct xenvif *vif = netdev_priv(dev);
273	struct xenvif_queue *queue = NULL;
274	unsigned int num_queues;
275	u64 rx_bytes = 0;
276	u64 rx_packets = 0;
277	u64 tx_bytes = 0;
278	u64 tx_packets = 0;
279	unsigned int index;
280
281	rcu_read_lock();
282	num_queues = READ_ONCE(vif->num_queues);
283
284	/* Aggregate tx and rx stats from each queue */
285	for (index = 0; index < num_queues; ++index) {
286		queue = &vif->queues[index];
287		rx_bytes += queue->stats.rx_bytes;
288		rx_packets += queue->stats.rx_packets;
289		tx_bytes += queue->stats.tx_bytes;
290		tx_packets += queue->stats.tx_packets;
291	}
292
293	rcu_read_unlock();
294
295	vif->dev->stats.rx_bytes = rx_bytes;
296	vif->dev->stats.rx_packets = rx_packets;
297	vif->dev->stats.tx_bytes = tx_bytes;
298	vif->dev->stats.tx_packets = tx_packets;
299
300	return &vif->dev->stats;
301}
302
303static void xenvif_up(struct xenvif *vif)
304{
305	struct xenvif_queue *queue = NULL;
306	unsigned int num_queues = vif->num_queues;
307	unsigned int queue_index;
308
309	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
310		queue = &vif->queues[queue_index];
311		napi_enable(&queue->napi);
312		enable_irq(queue->tx_irq);
313		if (queue->tx_irq != queue->rx_irq)
314			enable_irq(queue->rx_irq);
315		xenvif_napi_schedule_or_enable_events(queue);
316	}
317}
318
319static void xenvif_down(struct xenvif *vif)
320{
321	struct xenvif_queue *queue = NULL;
322	unsigned int num_queues = vif->num_queues;
323	unsigned int queue_index;
324
325	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
326		queue = &vif->queues[queue_index];
327		disable_irq(queue->tx_irq);
328		if (queue->tx_irq != queue->rx_irq)
329			disable_irq(queue->rx_irq);
330		napi_disable(&queue->napi);
331		del_timer_sync(&queue->credit_timeout);
332	}
333}
334
335static int xenvif_open(struct net_device *dev)
336{
337	struct xenvif *vif = netdev_priv(dev);
338	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
339		xenvif_up(vif);
340	netif_tx_start_all_queues(dev);
341	return 0;
342}
343
344static int xenvif_close(struct net_device *dev)
345{
346	struct xenvif *vif = netdev_priv(dev);
347	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
348		xenvif_down(vif);
349	netif_tx_stop_all_queues(dev);
350	return 0;
351}
352
353static int xenvif_change_mtu(struct net_device *dev, int mtu)
354{
355	struct xenvif *vif = netdev_priv(dev);
356	int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
357
358	if (mtu > max)
359		return -EINVAL;
360	dev->mtu = mtu;
361	return 0;
362}
363
364static netdev_features_t xenvif_fix_features(struct net_device *dev,
365	netdev_features_t features)
366{
367	struct xenvif *vif = netdev_priv(dev);
368
369	if (!vif->can_sg)
370		features &= ~NETIF_F_SG;
371	if (~(vif->gso_mask) & GSO_BIT(TCPV4))
372		features &= ~NETIF_F_TSO;
373	if (~(vif->gso_mask) & GSO_BIT(TCPV6))
374		features &= ~NETIF_F_TSO6;
375	if (!vif->ip_csum)
376		features &= ~NETIF_F_IP_CSUM;
377	if (!vif->ipv6_csum)
378		features &= ~NETIF_F_IPV6_CSUM;
379
380	return features;
381}
382
383static const struct xenvif_stat {
384	char name[ETH_GSTRING_LEN];
385	u16 offset;
386} xenvif_stats[] = {
387	{
388		"rx_gso_checksum_fixup",
389		offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
390	},
391	/* If (sent != success + fail), there are probably packets never
392	 * freed up properly!
393	 */
394	{
395		"tx_zerocopy_sent",
396		offsetof(struct xenvif_stats, tx_zerocopy_sent),
397	},
398	{
399		"tx_zerocopy_success",
400		offsetof(struct xenvif_stats, tx_zerocopy_success),
401	},
402	{
403		"tx_zerocopy_fail",
404		offsetof(struct xenvif_stats, tx_zerocopy_fail)
405	},
406	/* Number of packets exceeding MAX_SKB_FRAG slots. You should use
407	 * a guest with the same MAX_SKB_FRAG
408	 */
409	{
410		"tx_frag_overflow",
411		offsetof(struct xenvif_stats, tx_frag_overflow)
412	},
413};
414
415static int xenvif_get_sset_count(struct net_device *dev, int string_set)
416{
417	switch (string_set) {
418	case ETH_SS_STATS:
419		return ARRAY_SIZE(xenvif_stats);
420	default:
421		return -EINVAL;
422	}
423}
424
425static void xenvif_get_ethtool_stats(struct net_device *dev,
426				     struct ethtool_stats *stats, u64 * data)
427{
428	struct xenvif *vif = netdev_priv(dev);
429	unsigned int num_queues;
430	int i;
431	unsigned int queue_index;
432
433	rcu_read_lock();
434	num_queues = READ_ONCE(vif->num_queues);
435
436	for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
437		unsigned long accum = 0;
438		for (queue_index = 0; queue_index < num_queues; ++queue_index) {
439			void *vif_stats = &vif->queues[queue_index].stats;
440			accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
441		}
442		data[i] = accum;
443	}
444
445	rcu_read_unlock();
446}
447
448static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
449{
450	int i;
451
452	switch (stringset) {
453	case ETH_SS_STATS:
454		for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
455			memcpy(data + i * ETH_GSTRING_LEN,
456			       xenvif_stats[i].name, ETH_GSTRING_LEN);
457		break;
458	}
459}
460
461static const struct ethtool_ops xenvif_ethtool_ops = {
462	.get_link	= ethtool_op_get_link,
463
464	.get_sset_count = xenvif_get_sset_count,
465	.get_ethtool_stats = xenvif_get_ethtool_stats,
466	.get_strings = xenvif_get_strings,
467};
468
469static const struct net_device_ops xenvif_netdev_ops = {
470	.ndo_select_queue = xenvif_select_queue,
471	.ndo_start_xmit	= xenvif_start_xmit,
472	.ndo_get_stats	= xenvif_get_stats,
473	.ndo_open	= xenvif_open,
474	.ndo_stop	= xenvif_close,
475	.ndo_change_mtu	= xenvif_change_mtu,
476	.ndo_fix_features = xenvif_fix_features,
477	.ndo_set_mac_address = eth_mac_addr,
478	.ndo_validate_addr   = eth_validate_addr,
479};
480
481struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
482			    unsigned int handle)
483{
484	int err;
485	struct net_device *dev;
486	struct xenvif *vif;
487	char name[IFNAMSIZ] = {};
488
489	snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
490	/* Allocate a netdev with the max. supported number of queues.
491	 * When the guest selects the desired number, it will be updated
492	 * via netif_set_real_num_*_queues().
493	 */
494	dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
495			      ether_setup, xenvif_max_queues);
496	if (dev == NULL) {
497		pr_warn("Could not allocate netdev for %s\n", name);
498		return ERR_PTR(-ENOMEM);
499	}
500
501	SET_NETDEV_DEV(dev, parent);
502
503	vif = netdev_priv(dev);
504
505	vif->domid  = domid;
506	vif->handle = handle;
507	vif->can_sg = 1;
508	vif->ip_csum = 1;
509	vif->dev = dev;
510	vif->disabled = false;
511	vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
512	vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
513
514	/* Start out with no queues. */
515	vif->queues = NULL;
516	vif->num_queues = 0;
517
518	vif->xdp_headroom = 0;
519
520	spin_lock_init(&vif->lock);
521	INIT_LIST_HEAD(&vif->fe_mcast_addr);
522
523	dev->netdev_ops	= &xenvif_netdev_ops;
524	dev->hw_features = NETIF_F_SG |
525		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
526		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
527	dev->features = dev->hw_features | NETIF_F_RXCSUM;
528	dev->ethtool_ops = &xenvif_ethtool_ops;
529
530	dev->min_mtu = ETH_MIN_MTU;
531	dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
532
533	/*
534	 * Initialise a dummy MAC address. We choose the numerically
535	 * largest non-broadcast address to prevent the address getting
536	 * stolen by an Ethernet bridge for STP purposes.
537	 * (FE:FF:FF:FF:FF:FF)
538	 */
539	eth_broadcast_addr(dev->dev_addr);
540	dev->dev_addr[0] &= ~0x01;
541
542	netif_carrier_off(dev);
543
544	err = register_netdev(dev);
545	if (err) {
546		netdev_warn(dev, "Could not register device: err=%d\n", err);
547		free_netdev(dev);
548		return ERR_PTR(err);
549	}
550
551	netdev_dbg(dev, "Successfully created xenvif\n");
552
553	__module_get(THIS_MODULE);
554
555	return vif;
556}
557
558int xenvif_init_queue(struct xenvif_queue *queue)
559{
560	int err, i;
561
562	queue->credit_bytes = queue->remaining_credit = ~0UL;
563	queue->credit_usec  = 0UL;
564	timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
565	queue->credit_window_start = get_jiffies_64();
566
567	queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
568
569	skb_queue_head_init(&queue->rx_queue);
570	skb_queue_head_init(&queue->tx_queue);
571
572	queue->pending_cons = 0;
573	queue->pending_prod = MAX_PENDING_REQS;
574	for (i = 0; i < MAX_PENDING_REQS; ++i)
575		queue->pending_ring[i] = i;
576
577	spin_lock_init(&queue->callback_lock);
578	spin_lock_init(&queue->response_lock);
579
580	/* If ballooning is disabled, this will consume real memory, so you
581	 * better enable it. The long term solution would be to use just a
582	 * bunch of valid page descriptors, without dependency on ballooning
583	 */
584	err = gnttab_alloc_pages(MAX_PENDING_REQS,
585				 queue->mmap_pages);
586	if (err) {
587		netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
588		return -ENOMEM;
589	}
590
591	for (i = 0; i < MAX_PENDING_REQS; i++) {
592		queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
593			{ .callback = xenvif_zerocopy_callback,
594			  { { .ctx = NULL,
595			      .desc = i } } };
596		queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
597	}
598
599	return 0;
600}
601
602void xenvif_carrier_on(struct xenvif *vif)
603{
604	rtnl_lock();
605	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
606		dev_set_mtu(vif->dev, ETH_DATA_LEN);
607	netdev_update_features(vif->dev);
608	set_bit(VIF_STATUS_CONNECTED, &vif->status);
609	if (netif_running(vif->dev))
610		xenvif_up(vif);
611	rtnl_unlock();
612}
613
614int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
615			unsigned int evtchn)
616{
617	struct net_device *dev = vif->dev;
618	void *addr;
619	struct xen_netif_ctrl_sring *shared;
620	RING_IDX rsp_prod, req_prod;
621	int err;
622
623	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
624				     &ring_ref, 1, &addr);
625	if (err)
626		goto err;
627
628	shared = (struct xen_netif_ctrl_sring *)addr;
629	rsp_prod = READ_ONCE(shared->rsp_prod);
630	req_prod = READ_ONCE(shared->req_prod);
631
632	BACK_RING_ATTACH(&vif->ctrl, shared, rsp_prod, XEN_PAGE_SIZE);
633
634	err = -EIO;
635	if (req_prod - rsp_prod > RING_SIZE(&vif->ctrl))
636		goto err_unmap;
637
638	err = bind_interdomain_evtchn_to_irq_lateeoi(vif->domid, evtchn);
639	if (err < 0)
640		goto err_unmap;
641
642	vif->ctrl_irq = err;
643
644	xenvif_init_hash(vif);
645
646	err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
647				   IRQF_ONESHOT, "xen-netback-ctrl", vif);
648	if (err) {
649		pr_warn("Could not setup irq handler for %s\n", dev->name);
650		goto err_deinit;
651	}
652
653	return 0;
654
655err_deinit:
656	xenvif_deinit_hash(vif);
657	unbind_from_irqhandler(vif->ctrl_irq, vif);
658	vif->ctrl_irq = 0;
659
660err_unmap:
661	xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
662				vif->ctrl.sring);
663	vif->ctrl.sring = NULL;
664
665err:
666	return err;
667}
668
669static void xenvif_disconnect_queue(struct xenvif_queue *queue)
670{
671	if (queue->task) {
672		kthread_stop(queue->task);
673		put_task_struct(queue->task);
674		queue->task = NULL;
675	}
676
677	if (queue->dealloc_task) {
678		kthread_stop(queue->dealloc_task);
679		queue->dealloc_task = NULL;
680	}
681
682	if (queue->napi.poll) {
683		netif_napi_del(&queue->napi);
684		queue->napi.poll = NULL;
685	}
686
687	if (queue->tx_irq) {
688		unbind_from_irqhandler(queue->tx_irq, queue);
689		if (queue->tx_irq == queue->rx_irq)
690			queue->rx_irq = 0;
691		queue->tx_irq = 0;
692	}
693
694	if (queue->rx_irq) {
695		unbind_from_irqhandler(queue->rx_irq, queue);
696		queue->rx_irq = 0;
697	}
698
699	xenvif_unmap_frontend_data_rings(queue);
700}
701
702int xenvif_connect_data(struct xenvif_queue *queue,
703			unsigned long tx_ring_ref,
704			unsigned long rx_ring_ref,
705			unsigned int tx_evtchn,
706			unsigned int rx_evtchn)
707{
708	struct task_struct *task;
709	int err;
710
711	BUG_ON(queue->tx_irq);
712	BUG_ON(queue->task);
713	BUG_ON(queue->dealloc_task);
714
715	err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
716					     rx_ring_ref);
717	if (err < 0)
718		goto err;
719
720	init_waitqueue_head(&queue->wq);
721	init_waitqueue_head(&queue->dealloc_wq);
722	atomic_set(&queue->inflight_packets, 0);
723
724	netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
725			XENVIF_NAPI_WEIGHT);
726
727	queue->stalled = true;
728
729	task = kthread_run(xenvif_kthread_guest_rx, queue,
730			   "%s-guest-rx", queue->name);
731	if (IS_ERR(task))
732		goto kthread_err;
733	queue->task = task;
734	/*
735	 * Take a reference to the task in order to prevent it from being freed
736	 * if the thread function returns before kthread_stop is called.
737	 */
738	get_task_struct(task);
739
740	task = kthread_run(xenvif_dealloc_kthread, queue,
741			   "%s-dealloc", queue->name);
742	if (IS_ERR(task))
743		goto kthread_err;
744	queue->dealloc_task = task;
745
746	if (tx_evtchn == rx_evtchn) {
747		/* feature-split-event-channels == 0 */
748		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
749			queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
750			queue->name, queue);
751		if (err < 0)
752			goto err;
753		queue->tx_irq = queue->rx_irq = err;
754		disable_irq(queue->tx_irq);
755	} else {
756		/* feature-split-event-channels == 1 */
757		snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
758			 "%s-tx", queue->name);
759		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
760			queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
761			queue->tx_irq_name, queue);
762		if (err < 0)
763			goto err;
764		queue->tx_irq = err;
765		disable_irq(queue->tx_irq);
766
767		snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
768			 "%s-rx", queue->name);
769		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
770			queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
771			queue->rx_irq_name, queue);
772		if (err < 0)
773			goto err;
774		queue->rx_irq = err;
775		disable_irq(queue->rx_irq);
776	}
777
778	return 0;
779
780kthread_err:
781	pr_warn("Could not allocate kthread for %s\n", queue->name);
782	err = PTR_ERR(task);
783err:
784	xenvif_disconnect_queue(queue);
785	return err;
786}
787
788void xenvif_carrier_off(struct xenvif *vif)
789{
790	struct net_device *dev = vif->dev;
791
792	rtnl_lock();
793	if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
794		netif_carrier_off(dev); /* discard queued packets */
795		if (netif_running(dev))
796			xenvif_down(vif);
797	}
798	rtnl_unlock();
799}
800
801void xenvif_disconnect_data(struct xenvif *vif)
802{
803	struct xenvif_queue *queue = NULL;
804	unsigned int num_queues = vif->num_queues;
805	unsigned int queue_index;
806
807	xenvif_carrier_off(vif);
808
809	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
810		queue = &vif->queues[queue_index];
811
812		xenvif_disconnect_queue(queue);
813	}
814
815	xenvif_mcast_addr_list_free(vif);
816}
817
818void xenvif_disconnect_ctrl(struct xenvif *vif)
819{
820	if (vif->ctrl_irq) {
821		xenvif_deinit_hash(vif);
822		unbind_from_irqhandler(vif->ctrl_irq, vif);
823		vif->ctrl_irq = 0;
824	}
825
826	if (vif->ctrl.sring) {
827		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
828					vif->ctrl.sring);
829		vif->ctrl.sring = NULL;
830	}
831}
832
833/* Reverse the relevant parts of xenvif_init_queue().
834 * Used for queue teardown from xenvif_free(), and on the
835 * error handling paths in xenbus.c:connect().
836 */
837void xenvif_deinit_queue(struct xenvif_queue *queue)
838{
839	gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
840}
841
842void xenvif_free(struct xenvif *vif)
843{
844	struct xenvif_queue *queues = vif->queues;
845	unsigned int num_queues = vif->num_queues;
846	unsigned int queue_index;
847
848	unregister_netdev(vif->dev);
849	free_netdev(vif->dev);
850
851	for (queue_index = 0; queue_index < num_queues; ++queue_index)
852		xenvif_deinit_queue(&queues[queue_index]);
853	vfree(queues);
854
855	module_put(THIS_MODULE);
856}
857