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 SKBFL_ZEROCOPY_ENABLE 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)->flags |= SKBFL_ZEROCOPY_ENABLE;
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 bool xenvif_handle_tx_interrupt(struct xenvif_queue *queue)
81 {
82 	bool rc;
83 
84 	rc = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
85 	if (rc)
86 		napi_schedule(&queue->napi);
87 	return rc;
88 }
89 
90 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
91 {
92 	struct xenvif_queue *queue = dev_id;
93 	int old;
94 
95 	old = atomic_fetch_or(NETBK_TX_EOI, &queue->eoi_pending);
96 	WARN(old & NETBK_TX_EOI, "Interrupt while EOI pending\n");
97 
98 	if (!xenvif_handle_tx_interrupt(queue)) {
99 		atomic_andnot(NETBK_TX_EOI, &queue->eoi_pending);
100 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
101 	}
102 
103 	return IRQ_HANDLED;
104 }
105 
106 static int xenvif_poll(struct napi_struct *napi, int budget)
107 {
108 	struct xenvif_queue *queue =
109 		container_of(napi, struct xenvif_queue, napi);
110 	int work_done;
111 
112 	/* This vif is rogue, we pretend we've there is nothing to do
113 	 * for this vif to deschedule it from NAPI. But this interface
114 	 * will be turned off in thread context later.
115 	 */
116 	if (unlikely(queue->vif->disabled)) {
117 		napi_complete(napi);
118 		return 0;
119 	}
120 
121 	work_done = xenvif_tx_action(queue, budget);
122 
123 	if (work_done < budget) {
124 		napi_complete_done(napi, work_done);
125 		/* If the queue is rate-limited, it shall be
126 		 * rescheduled in the timer callback.
127 		 */
128 		if (likely(!queue->rate_limited))
129 			xenvif_napi_schedule_or_enable_events(queue);
130 	}
131 
132 	return work_done;
133 }
134 
135 static bool xenvif_handle_rx_interrupt(struct xenvif_queue *queue)
136 {
137 	bool rc;
138 
139 	rc = xenvif_have_rx_work(queue, false);
140 	if (rc)
141 		xenvif_kick_thread(queue);
142 	return rc;
143 }
144 
145 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
146 {
147 	struct xenvif_queue *queue = dev_id;
148 	int old;
149 
150 	old = atomic_fetch_or(NETBK_RX_EOI, &queue->eoi_pending);
151 	WARN(old & NETBK_RX_EOI, "Interrupt while EOI pending\n");
152 
153 	if (!xenvif_handle_rx_interrupt(queue)) {
154 		atomic_andnot(NETBK_RX_EOI, &queue->eoi_pending);
155 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
156 	}
157 
158 	return IRQ_HANDLED;
159 }
160 
161 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
162 {
163 	struct xenvif_queue *queue = dev_id;
164 	int old;
165 	bool has_rx, has_tx;
166 
167 	old = atomic_fetch_or(NETBK_COMMON_EOI, &queue->eoi_pending);
168 	WARN(old, "Interrupt while EOI pending\n");
169 
170 	has_tx = xenvif_handle_tx_interrupt(queue);
171 	has_rx = xenvif_handle_rx_interrupt(queue);
172 
173 	if (!has_rx && !has_tx) {
174 		atomic_andnot(NETBK_COMMON_EOI, &queue->eoi_pending);
175 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
176 	}
177 
178 	return IRQ_HANDLED;
179 }
180 
181 int xenvif_queue_stopped(struct xenvif_queue *queue)
182 {
183 	struct net_device *dev = queue->vif->dev;
184 	unsigned int id = queue->id;
185 	return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
186 }
187 
188 void xenvif_wake_queue(struct xenvif_queue *queue)
189 {
190 	struct net_device *dev = queue->vif->dev;
191 	unsigned int id = queue->id;
192 	netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
193 }
194 
195 static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
196 			       struct net_device *sb_dev)
197 {
198 	struct xenvif *vif = netdev_priv(dev);
199 	unsigned int size = vif->hash.size;
200 	unsigned int num_queues;
201 
202 	/* If queues are not set up internally - always return 0
203 	 * as the packet going to be dropped anyway */
204 	num_queues = READ_ONCE(vif->num_queues);
205 	if (num_queues < 1)
206 		return 0;
207 
208 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
209 		return netdev_pick_tx(dev, skb, NULL) %
210 		       dev->real_num_tx_queues;
211 
212 	xenvif_set_skb_hash(vif, skb);
213 
214 	if (size == 0)
215 		return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
216 
217 	return vif->hash.mapping[vif->hash.mapping_sel]
218 				[skb_get_hash_raw(skb) % size];
219 }
220 
221 static netdev_tx_t
222 xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
223 {
224 	struct xenvif *vif = netdev_priv(dev);
225 	struct xenvif_queue *queue = NULL;
226 	unsigned int num_queues;
227 	u16 index;
228 	struct xenvif_rx_cb *cb;
229 
230 	BUG_ON(skb->dev != dev);
231 
232 	/* Drop the packet if queues are not set up.
233 	 * This handler should be called inside an RCU read section
234 	 * so we don't need to enter it here explicitly.
235 	 */
236 	num_queues = READ_ONCE(vif->num_queues);
237 	if (num_queues < 1)
238 		goto drop;
239 
240 	/* Obtain the queue to be used to transmit this packet */
241 	index = skb_get_queue_mapping(skb);
242 	if (index >= num_queues) {
243 		pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
244 				    index, vif->dev->name);
245 		index %= num_queues;
246 	}
247 	queue = &vif->queues[index];
248 
249 	/* Drop the packet if queue is not ready */
250 	if (queue->task == NULL ||
251 	    queue->dealloc_task == NULL ||
252 	    !xenvif_schedulable(vif))
253 		goto drop;
254 
255 	if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
256 		struct ethhdr *eth = (struct ethhdr *)skb->data;
257 
258 		if (!xenvif_mcast_match(vif, eth->h_dest))
259 			goto drop;
260 	}
261 
262 	cb = XENVIF_RX_CB(skb);
263 	cb->expires = jiffies + vif->drain_timeout;
264 
265 	/* If there is no hash algorithm configured then make sure there
266 	 * is no hash information in the socket buffer otherwise it
267 	 * would be incorrectly forwarded to the frontend.
268 	 */
269 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
270 		skb_clear_hash(skb);
271 
272 	xenvif_rx_queue_tail(queue, skb);
273 	xenvif_kick_thread(queue);
274 
275 	return NETDEV_TX_OK;
276 
277  drop:
278 	vif->dev->stats.tx_dropped++;
279 	dev_kfree_skb(skb);
280 	return NETDEV_TX_OK;
281 }
282 
283 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
284 {
285 	struct xenvif *vif = netdev_priv(dev);
286 	struct xenvif_queue *queue = NULL;
287 	unsigned int num_queues;
288 	u64 rx_bytes = 0;
289 	u64 rx_packets = 0;
290 	u64 tx_bytes = 0;
291 	u64 tx_packets = 0;
292 	unsigned int index;
293 
294 	rcu_read_lock();
295 	num_queues = READ_ONCE(vif->num_queues);
296 
297 	/* Aggregate tx and rx stats from each queue */
298 	for (index = 0; index < num_queues; ++index) {
299 		queue = &vif->queues[index];
300 		rx_bytes += queue->stats.rx_bytes;
301 		rx_packets += queue->stats.rx_packets;
302 		tx_bytes += queue->stats.tx_bytes;
303 		tx_packets += queue->stats.tx_packets;
304 	}
305 
306 	rcu_read_unlock();
307 
308 	vif->dev->stats.rx_bytes = rx_bytes;
309 	vif->dev->stats.rx_packets = rx_packets;
310 	vif->dev->stats.tx_bytes = tx_bytes;
311 	vif->dev->stats.tx_packets = tx_packets;
312 
313 	return &vif->dev->stats;
314 }
315 
316 static void xenvif_up(struct xenvif *vif)
317 {
318 	struct xenvif_queue *queue = NULL;
319 	unsigned int num_queues = vif->num_queues;
320 	unsigned int queue_index;
321 
322 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
323 		queue = &vif->queues[queue_index];
324 		napi_enable(&queue->napi);
325 		enable_irq(queue->tx_irq);
326 		if (queue->tx_irq != queue->rx_irq)
327 			enable_irq(queue->rx_irq);
328 		xenvif_napi_schedule_or_enable_events(queue);
329 	}
330 }
331 
332 static void xenvif_down(struct xenvif *vif)
333 {
334 	struct xenvif_queue *queue = NULL;
335 	unsigned int num_queues = vif->num_queues;
336 	unsigned int queue_index;
337 
338 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
339 		queue = &vif->queues[queue_index];
340 		disable_irq(queue->tx_irq);
341 		if (queue->tx_irq != queue->rx_irq)
342 			disable_irq(queue->rx_irq);
343 		napi_disable(&queue->napi);
344 		del_timer_sync(&queue->credit_timeout);
345 	}
346 }
347 
348 static int xenvif_open(struct net_device *dev)
349 {
350 	struct xenvif *vif = netdev_priv(dev);
351 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
352 		xenvif_up(vif);
353 	netif_tx_start_all_queues(dev);
354 	return 0;
355 }
356 
357 static int xenvif_close(struct net_device *dev)
358 {
359 	struct xenvif *vif = netdev_priv(dev);
360 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
361 		xenvif_down(vif);
362 	netif_tx_stop_all_queues(dev);
363 	return 0;
364 }
365 
366 static int xenvif_change_mtu(struct net_device *dev, int mtu)
367 {
368 	struct xenvif *vif = netdev_priv(dev);
369 	int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
370 
371 	if (mtu > max)
372 		return -EINVAL;
373 	dev->mtu = mtu;
374 	return 0;
375 }
376 
377 static netdev_features_t xenvif_fix_features(struct net_device *dev,
378 	netdev_features_t features)
379 {
380 	struct xenvif *vif = netdev_priv(dev);
381 
382 	if (!vif->can_sg)
383 		features &= ~NETIF_F_SG;
384 	if (~(vif->gso_mask) & GSO_BIT(TCPV4))
385 		features &= ~NETIF_F_TSO;
386 	if (~(vif->gso_mask) & GSO_BIT(TCPV6))
387 		features &= ~NETIF_F_TSO6;
388 	if (!vif->ip_csum)
389 		features &= ~NETIF_F_IP_CSUM;
390 	if (!vif->ipv6_csum)
391 		features &= ~NETIF_F_IPV6_CSUM;
392 
393 	return features;
394 }
395 
396 static const struct xenvif_stat {
397 	char name[ETH_GSTRING_LEN];
398 	u16 offset;
399 } xenvif_stats[] = {
400 	{
401 		"rx_gso_checksum_fixup",
402 		offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
403 	},
404 	/* If (sent != success + fail), there are probably packets never
405 	 * freed up properly!
406 	 */
407 	{
408 		"tx_zerocopy_sent",
409 		offsetof(struct xenvif_stats, tx_zerocopy_sent),
410 	},
411 	{
412 		"tx_zerocopy_success",
413 		offsetof(struct xenvif_stats, tx_zerocopy_success),
414 	},
415 	{
416 		"tx_zerocopy_fail",
417 		offsetof(struct xenvif_stats, tx_zerocopy_fail)
418 	},
419 	/* Number of packets exceeding MAX_SKB_FRAG slots. You should use
420 	 * a guest with the same MAX_SKB_FRAG
421 	 */
422 	{
423 		"tx_frag_overflow",
424 		offsetof(struct xenvif_stats, tx_frag_overflow)
425 	},
426 };
427 
428 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
429 {
430 	switch (string_set) {
431 	case ETH_SS_STATS:
432 		return ARRAY_SIZE(xenvif_stats);
433 	default:
434 		return -EINVAL;
435 	}
436 }
437 
438 static void xenvif_get_ethtool_stats(struct net_device *dev,
439 				     struct ethtool_stats *stats, u64 * data)
440 {
441 	struct xenvif *vif = netdev_priv(dev);
442 	unsigned int num_queues;
443 	int i;
444 	unsigned int queue_index;
445 
446 	rcu_read_lock();
447 	num_queues = READ_ONCE(vif->num_queues);
448 
449 	for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
450 		unsigned long accum = 0;
451 		for (queue_index = 0; queue_index < num_queues; ++queue_index) {
452 			void *vif_stats = &vif->queues[queue_index].stats;
453 			accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
454 		}
455 		data[i] = accum;
456 	}
457 
458 	rcu_read_unlock();
459 }
460 
461 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
462 {
463 	int i;
464 
465 	switch (stringset) {
466 	case ETH_SS_STATS:
467 		for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
468 			memcpy(data + i * ETH_GSTRING_LEN,
469 			       xenvif_stats[i].name, ETH_GSTRING_LEN);
470 		break;
471 	}
472 }
473 
474 static const struct ethtool_ops xenvif_ethtool_ops = {
475 	.get_link	= ethtool_op_get_link,
476 
477 	.get_sset_count = xenvif_get_sset_count,
478 	.get_ethtool_stats = xenvif_get_ethtool_stats,
479 	.get_strings = xenvif_get_strings,
480 };
481 
482 static const struct net_device_ops xenvif_netdev_ops = {
483 	.ndo_select_queue = xenvif_select_queue,
484 	.ndo_start_xmit	= xenvif_start_xmit,
485 	.ndo_get_stats	= xenvif_get_stats,
486 	.ndo_open	= xenvif_open,
487 	.ndo_stop	= xenvif_close,
488 	.ndo_change_mtu	= xenvif_change_mtu,
489 	.ndo_fix_features = xenvif_fix_features,
490 	.ndo_set_mac_address = eth_mac_addr,
491 	.ndo_validate_addr   = eth_validate_addr,
492 };
493 
494 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
495 			    unsigned int handle)
496 {
497 	int err;
498 	struct net_device *dev;
499 	struct xenvif *vif;
500 	char name[IFNAMSIZ] = {};
501 
502 	snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
503 	/* Allocate a netdev with the max. supported number of queues.
504 	 * When the guest selects the desired number, it will be updated
505 	 * via netif_set_real_num_*_queues().
506 	 */
507 	dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
508 			      ether_setup, xenvif_max_queues);
509 	if (dev == NULL) {
510 		pr_warn("Could not allocate netdev for %s\n", name);
511 		return ERR_PTR(-ENOMEM);
512 	}
513 
514 	SET_NETDEV_DEV(dev, parent);
515 
516 	vif = netdev_priv(dev);
517 
518 	vif->domid  = domid;
519 	vif->handle = handle;
520 	vif->can_sg = 1;
521 	vif->ip_csum = 1;
522 	vif->dev = dev;
523 	vif->disabled = false;
524 	vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
525 	vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
526 
527 	/* Start out with no queues. */
528 	vif->queues = NULL;
529 	vif->num_queues = 0;
530 
531 	vif->xdp_headroom = 0;
532 
533 	spin_lock_init(&vif->lock);
534 	INIT_LIST_HEAD(&vif->fe_mcast_addr);
535 
536 	dev->netdev_ops	= &xenvif_netdev_ops;
537 	dev->hw_features = NETIF_F_SG |
538 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
539 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
540 	dev->features = dev->hw_features | NETIF_F_RXCSUM;
541 	dev->ethtool_ops = &xenvif_ethtool_ops;
542 
543 	dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
544 
545 	dev->min_mtu = ETH_MIN_MTU;
546 	dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
547 
548 	/*
549 	 * Initialise a dummy MAC address. We choose the numerically
550 	 * largest non-broadcast address to prevent the address getting
551 	 * stolen by an Ethernet bridge for STP purposes.
552 	 * (FE:FF:FF:FF:FF:FF)
553 	 */
554 	eth_broadcast_addr(dev->dev_addr);
555 	dev->dev_addr[0] &= ~0x01;
556 
557 	netif_carrier_off(dev);
558 
559 	err = register_netdev(dev);
560 	if (err) {
561 		netdev_warn(dev, "Could not register device: err=%d\n", err);
562 		free_netdev(dev);
563 		return ERR_PTR(err);
564 	}
565 
566 	netdev_dbg(dev, "Successfully created xenvif\n");
567 
568 	__module_get(THIS_MODULE);
569 
570 	return vif;
571 }
572 
573 int xenvif_init_queue(struct xenvif_queue *queue)
574 {
575 	int err, i;
576 
577 	queue->credit_bytes = queue->remaining_credit = ~0UL;
578 	queue->credit_usec  = 0UL;
579 	timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
580 	queue->credit_window_start = get_jiffies_64();
581 
582 	queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
583 
584 	skb_queue_head_init(&queue->rx_queue);
585 	skb_queue_head_init(&queue->tx_queue);
586 
587 	queue->pending_cons = 0;
588 	queue->pending_prod = MAX_PENDING_REQS;
589 	for (i = 0; i < MAX_PENDING_REQS; ++i)
590 		queue->pending_ring[i] = i;
591 
592 	spin_lock_init(&queue->callback_lock);
593 	spin_lock_init(&queue->response_lock);
594 
595 	/* If ballooning is disabled, this will consume real memory, so you
596 	 * better enable it. The long term solution would be to use just a
597 	 * bunch of valid page descriptors, without dependency on ballooning
598 	 */
599 	err = gnttab_alloc_pages(MAX_PENDING_REQS,
600 				 queue->mmap_pages);
601 	if (err) {
602 		netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
603 		return -ENOMEM;
604 	}
605 
606 	for (i = 0; i < MAX_PENDING_REQS; i++) {
607 		queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
608 			{ .callback = xenvif_zerocopy_callback,
609 			  { { .ctx = NULL,
610 			      .desc = i } } };
611 		queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
612 	}
613 
614 	return 0;
615 }
616 
617 void xenvif_carrier_on(struct xenvif *vif)
618 {
619 	rtnl_lock();
620 	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
621 		dev_set_mtu(vif->dev, ETH_DATA_LEN);
622 	netdev_update_features(vif->dev);
623 	set_bit(VIF_STATUS_CONNECTED, &vif->status);
624 	if (netif_running(vif->dev))
625 		xenvif_up(vif);
626 	rtnl_unlock();
627 }
628 
629 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
630 			unsigned int evtchn)
631 {
632 	struct net_device *dev = vif->dev;
633 	struct xenbus_device *xendev = xenvif_to_xenbus_device(vif);
634 	void *addr;
635 	struct xen_netif_ctrl_sring *shared;
636 	RING_IDX rsp_prod, req_prod;
637 	int err;
638 
639 	err = xenbus_map_ring_valloc(xendev, &ring_ref, 1, &addr);
640 	if (err)
641 		goto err;
642 
643 	shared = (struct xen_netif_ctrl_sring *)addr;
644 	rsp_prod = READ_ONCE(shared->rsp_prod);
645 	req_prod = READ_ONCE(shared->req_prod);
646 
647 	BACK_RING_ATTACH(&vif->ctrl, shared, rsp_prod, XEN_PAGE_SIZE);
648 
649 	err = -EIO;
650 	if (req_prod - rsp_prod > RING_SIZE(&vif->ctrl))
651 		goto err_unmap;
652 
653 	err = bind_interdomain_evtchn_to_irq_lateeoi(xendev, evtchn);
654 	if (err < 0)
655 		goto err_unmap;
656 
657 	vif->ctrl_irq = err;
658 
659 	xenvif_init_hash(vif);
660 
661 	err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
662 				   IRQF_ONESHOT, "xen-netback-ctrl", vif);
663 	if (err) {
664 		pr_warn("Could not setup irq handler for %s\n", dev->name);
665 		goto err_deinit;
666 	}
667 
668 	return 0;
669 
670 err_deinit:
671 	xenvif_deinit_hash(vif);
672 	unbind_from_irqhandler(vif->ctrl_irq, vif);
673 	vif->ctrl_irq = 0;
674 
675 err_unmap:
676 	xenbus_unmap_ring_vfree(xendev, vif->ctrl.sring);
677 	vif->ctrl.sring = NULL;
678 
679 err:
680 	return err;
681 }
682 
683 static void xenvif_disconnect_queue(struct xenvif_queue *queue)
684 {
685 	if (queue->task) {
686 		kthread_stop(queue->task);
687 		put_task_struct(queue->task);
688 		queue->task = NULL;
689 	}
690 
691 	if (queue->dealloc_task) {
692 		kthread_stop(queue->dealloc_task);
693 		queue->dealloc_task = NULL;
694 	}
695 
696 	if (queue->napi.poll) {
697 		netif_napi_del(&queue->napi);
698 		queue->napi.poll = NULL;
699 	}
700 
701 	if (queue->tx_irq) {
702 		unbind_from_irqhandler(queue->tx_irq, queue);
703 		if (queue->tx_irq == queue->rx_irq)
704 			queue->rx_irq = 0;
705 		queue->tx_irq = 0;
706 	}
707 
708 	if (queue->rx_irq) {
709 		unbind_from_irqhandler(queue->rx_irq, queue);
710 		queue->rx_irq = 0;
711 	}
712 
713 	xenvif_unmap_frontend_data_rings(queue);
714 }
715 
716 int xenvif_connect_data(struct xenvif_queue *queue,
717 			unsigned long tx_ring_ref,
718 			unsigned long rx_ring_ref,
719 			unsigned int tx_evtchn,
720 			unsigned int rx_evtchn)
721 {
722 	struct xenbus_device *dev = xenvif_to_xenbus_device(queue->vif);
723 	struct task_struct *task;
724 	int err;
725 
726 	BUG_ON(queue->tx_irq);
727 	BUG_ON(queue->task);
728 	BUG_ON(queue->dealloc_task);
729 
730 	err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
731 					     rx_ring_ref);
732 	if (err < 0)
733 		goto err;
734 
735 	init_waitqueue_head(&queue->wq);
736 	init_waitqueue_head(&queue->dealloc_wq);
737 	atomic_set(&queue->inflight_packets, 0);
738 
739 	netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
740 			XENVIF_NAPI_WEIGHT);
741 
742 	queue->stalled = true;
743 
744 	task = kthread_run(xenvif_kthread_guest_rx, queue,
745 			   "%s-guest-rx", queue->name);
746 	if (IS_ERR(task))
747 		goto kthread_err;
748 	queue->task = task;
749 	/*
750 	 * Take a reference to the task in order to prevent it from being freed
751 	 * if the thread function returns before kthread_stop is called.
752 	 */
753 	get_task_struct(task);
754 
755 	task = kthread_run(xenvif_dealloc_kthread, queue,
756 			   "%s-dealloc", queue->name);
757 	if (IS_ERR(task))
758 		goto kthread_err;
759 	queue->dealloc_task = task;
760 
761 	if (tx_evtchn == rx_evtchn) {
762 		/* feature-split-event-channels == 0 */
763 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
764 			dev, tx_evtchn, xenvif_interrupt, 0,
765 			queue->name, queue);
766 		if (err < 0)
767 			goto err;
768 		queue->tx_irq = queue->rx_irq = err;
769 		disable_irq(queue->tx_irq);
770 	} else {
771 		/* feature-split-event-channels == 1 */
772 		snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
773 			 "%s-tx", queue->name);
774 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
775 			dev, tx_evtchn, xenvif_tx_interrupt, 0,
776 			queue->tx_irq_name, queue);
777 		if (err < 0)
778 			goto err;
779 		queue->tx_irq = err;
780 		disable_irq(queue->tx_irq);
781 
782 		snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
783 			 "%s-rx", queue->name);
784 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
785 			dev, rx_evtchn, xenvif_rx_interrupt, 0,
786 			queue->rx_irq_name, queue);
787 		if (err < 0)
788 			goto err;
789 		queue->rx_irq = err;
790 		disable_irq(queue->rx_irq);
791 	}
792 
793 	return 0;
794 
795 kthread_err:
796 	pr_warn("Could not allocate kthread for %s\n", queue->name);
797 	err = PTR_ERR(task);
798 err:
799 	xenvif_disconnect_queue(queue);
800 	return err;
801 }
802 
803 void xenvif_carrier_off(struct xenvif *vif)
804 {
805 	struct net_device *dev = vif->dev;
806 
807 	rtnl_lock();
808 	if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
809 		netif_carrier_off(dev); /* discard queued packets */
810 		if (netif_running(dev))
811 			xenvif_down(vif);
812 	}
813 	rtnl_unlock();
814 }
815 
816 void xenvif_disconnect_data(struct xenvif *vif)
817 {
818 	struct xenvif_queue *queue = NULL;
819 	unsigned int num_queues = vif->num_queues;
820 	unsigned int queue_index;
821 
822 	xenvif_carrier_off(vif);
823 
824 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
825 		queue = &vif->queues[queue_index];
826 
827 		xenvif_disconnect_queue(queue);
828 	}
829 
830 	xenvif_mcast_addr_list_free(vif);
831 }
832 
833 void xenvif_disconnect_ctrl(struct xenvif *vif)
834 {
835 	if (vif->ctrl_irq) {
836 		xenvif_deinit_hash(vif);
837 		unbind_from_irqhandler(vif->ctrl_irq, vif);
838 		vif->ctrl_irq = 0;
839 	}
840 
841 	if (vif->ctrl.sring) {
842 		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
843 					vif->ctrl.sring);
844 		vif->ctrl.sring = NULL;
845 	}
846 }
847 
848 /* Reverse the relevant parts of xenvif_init_queue().
849  * Used for queue teardown from xenvif_free(), and on the
850  * error handling paths in xenbus.c:connect().
851  */
852 void xenvif_deinit_queue(struct xenvif_queue *queue)
853 {
854 	gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
855 }
856 
857 void xenvif_free(struct xenvif *vif)
858 {
859 	struct xenvif_queue *queues = vif->queues;
860 	unsigned int num_queues = vif->num_queues;
861 	unsigned int queue_index;
862 
863 	unregister_netdev(vif->dev);
864 	free_netdev(vif->dev);
865 
866 	for (queue_index = 0; queue_index < num_queues; ++queue_index)
867 		xenvif_deinit_queue(&queues[queue_index]);
868 	vfree(queues);
869 
870 	module_put(THIS_MODULE);
871 }
872