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 
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 SKBFL_ZEROCOPY_ENABLE 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  */
54 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
55 				 struct sk_buff *skb)
56 {
57 	skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_ENABLE;
58 	atomic_inc(&queue->inflight_packets);
59 }
60 
61 void 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 
72 static 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 
79 static 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 
89 static 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 
105 static 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 
134 static 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 
144 static 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 
160 irqreturn_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 
180 static 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 
206 static netdev_tx_t
207 xenvif_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 	xenvif_rx_queue_tail(queue, skb);
258 	xenvif_kick_thread(queue);
259 
260 	return NETDEV_TX_OK;
261 
262  drop:
263 	vif->dev->stats.tx_dropped++;
264 	dev_kfree_skb(skb);
265 	return NETDEV_TX_OK;
266 }
267 
268 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
269 {
270 	struct xenvif *vif = netdev_priv(dev);
271 	struct xenvif_queue *queue = NULL;
272 	unsigned int num_queues;
273 	u64 rx_bytes = 0;
274 	u64 rx_packets = 0;
275 	u64 tx_bytes = 0;
276 	u64 tx_packets = 0;
277 	unsigned int index;
278 
279 	rcu_read_lock();
280 	num_queues = READ_ONCE(vif->num_queues);
281 
282 	/* Aggregate tx and rx stats from each queue */
283 	for (index = 0; index < num_queues; ++index) {
284 		queue = &vif->queues[index];
285 		rx_bytes += queue->stats.rx_bytes;
286 		rx_packets += queue->stats.rx_packets;
287 		tx_bytes += queue->stats.tx_bytes;
288 		tx_packets += queue->stats.tx_packets;
289 	}
290 
291 	rcu_read_unlock();
292 
293 	vif->dev->stats.rx_bytes = rx_bytes;
294 	vif->dev->stats.rx_packets = rx_packets;
295 	vif->dev->stats.tx_bytes = tx_bytes;
296 	vif->dev->stats.tx_packets = tx_packets;
297 
298 	return &vif->dev->stats;
299 }
300 
301 static void xenvif_up(struct xenvif *vif)
302 {
303 	struct xenvif_queue *queue = NULL;
304 	unsigned int num_queues = vif->num_queues;
305 	unsigned int queue_index;
306 
307 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
308 		queue = &vif->queues[queue_index];
309 		napi_enable(&queue->napi);
310 		enable_irq(queue->tx_irq);
311 		if (queue->tx_irq != queue->rx_irq)
312 			enable_irq(queue->rx_irq);
313 		xenvif_napi_schedule_or_enable_events(queue);
314 	}
315 }
316 
317 static void xenvif_down(struct xenvif *vif)
318 {
319 	struct xenvif_queue *queue = NULL;
320 	unsigned int num_queues = vif->num_queues;
321 	unsigned int queue_index;
322 
323 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
324 		queue = &vif->queues[queue_index];
325 		disable_irq(queue->tx_irq);
326 		if (queue->tx_irq != queue->rx_irq)
327 			disable_irq(queue->rx_irq);
328 		napi_disable(&queue->napi);
329 		del_timer_sync(&queue->credit_timeout);
330 	}
331 }
332 
333 static int xenvif_open(struct net_device *dev)
334 {
335 	struct xenvif *vif = netdev_priv(dev);
336 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
337 		xenvif_up(vif);
338 	netif_tx_start_all_queues(dev);
339 	return 0;
340 }
341 
342 static int xenvif_close(struct net_device *dev)
343 {
344 	struct xenvif *vif = netdev_priv(dev);
345 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
346 		xenvif_down(vif);
347 	netif_tx_stop_all_queues(dev);
348 	return 0;
349 }
350 
351 static int xenvif_change_mtu(struct net_device *dev, int mtu)
352 {
353 	struct xenvif *vif = netdev_priv(dev);
354 	int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
355 
356 	if (mtu > max)
357 		return -EINVAL;
358 	dev->mtu = mtu;
359 	return 0;
360 }
361 
362 static netdev_features_t xenvif_fix_features(struct net_device *dev,
363 	netdev_features_t features)
364 {
365 	struct xenvif *vif = netdev_priv(dev);
366 
367 	if (!vif->can_sg)
368 		features &= ~NETIF_F_SG;
369 	if (~(vif->gso_mask) & GSO_BIT(TCPV4))
370 		features &= ~NETIF_F_TSO;
371 	if (~(vif->gso_mask) & GSO_BIT(TCPV6))
372 		features &= ~NETIF_F_TSO6;
373 	if (!vif->ip_csum)
374 		features &= ~NETIF_F_IP_CSUM;
375 	if (!vif->ipv6_csum)
376 		features &= ~NETIF_F_IPV6_CSUM;
377 
378 	return features;
379 }
380 
381 static const struct xenvif_stat {
382 	char name[ETH_GSTRING_LEN];
383 	u16 offset;
384 } xenvif_stats[] = {
385 	{
386 		"rx_gso_checksum_fixup",
387 		offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
388 	},
389 	/* If (sent != success + fail), there are probably packets never
390 	 * freed up properly!
391 	 */
392 	{
393 		"tx_zerocopy_sent",
394 		offsetof(struct xenvif_stats, tx_zerocopy_sent),
395 	},
396 	{
397 		"tx_zerocopy_success",
398 		offsetof(struct xenvif_stats, tx_zerocopy_success),
399 	},
400 	{
401 		"tx_zerocopy_fail",
402 		offsetof(struct xenvif_stats, tx_zerocopy_fail)
403 	},
404 	/* Number of packets exceeding MAX_SKB_FRAG slots. You should use
405 	 * a guest with the same MAX_SKB_FRAG
406 	 */
407 	{
408 		"tx_frag_overflow",
409 		offsetof(struct xenvif_stats, tx_frag_overflow)
410 	},
411 };
412 
413 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
414 {
415 	switch (string_set) {
416 	case ETH_SS_STATS:
417 		return ARRAY_SIZE(xenvif_stats);
418 	default:
419 		return -EINVAL;
420 	}
421 }
422 
423 static void xenvif_get_ethtool_stats(struct net_device *dev,
424 				     struct ethtool_stats *stats, u64 * data)
425 {
426 	struct xenvif *vif = netdev_priv(dev);
427 	unsigned int num_queues;
428 	int i;
429 	unsigned int queue_index;
430 
431 	rcu_read_lock();
432 	num_queues = READ_ONCE(vif->num_queues);
433 
434 	for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
435 		unsigned long accum = 0;
436 		for (queue_index = 0; queue_index < num_queues; ++queue_index) {
437 			void *vif_stats = &vif->queues[queue_index].stats;
438 			accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
439 		}
440 		data[i] = accum;
441 	}
442 
443 	rcu_read_unlock();
444 }
445 
446 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
447 {
448 	int i;
449 
450 	switch (stringset) {
451 	case ETH_SS_STATS:
452 		for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
453 			memcpy(data + i * ETH_GSTRING_LEN,
454 			       xenvif_stats[i].name, ETH_GSTRING_LEN);
455 		break;
456 	}
457 }
458 
459 static const struct ethtool_ops xenvif_ethtool_ops = {
460 	.get_link	= ethtool_op_get_link,
461 
462 	.get_sset_count = xenvif_get_sset_count,
463 	.get_ethtool_stats = xenvif_get_ethtool_stats,
464 	.get_strings = xenvif_get_strings,
465 };
466 
467 static const struct net_device_ops xenvif_netdev_ops = {
468 	.ndo_select_queue = xenvif_select_queue,
469 	.ndo_start_xmit	= xenvif_start_xmit,
470 	.ndo_get_stats	= xenvif_get_stats,
471 	.ndo_open	= xenvif_open,
472 	.ndo_stop	= xenvif_close,
473 	.ndo_change_mtu	= xenvif_change_mtu,
474 	.ndo_fix_features = xenvif_fix_features,
475 	.ndo_set_mac_address = eth_mac_addr,
476 	.ndo_validate_addr   = eth_validate_addr,
477 };
478 
479 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
480 			    unsigned int handle)
481 {
482 	static const u8 dummy_addr[ETH_ALEN] = {
483 		0xfe, 0xff, 0xff, 0xff, 0xff, 0xff,
484 	};
485 	int err;
486 	struct net_device *dev;
487 	struct xenvif *vif;
488 	char name[IFNAMSIZ] = {};
489 
490 	snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
491 	/* Allocate a netdev with the max. supported number of queues.
492 	 * When the guest selects the desired number, it will be updated
493 	 * via netif_set_real_num_*_queues().
494 	 */
495 	dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
496 			      ether_setup, xenvif_max_queues);
497 	if (dev == NULL) {
498 		pr_warn("Could not allocate netdev for %s\n", name);
499 		return ERR_PTR(-ENOMEM);
500 	}
501 
502 	SET_NETDEV_DEV(dev, parent);
503 
504 	vif = netdev_priv(dev);
505 
506 	vif->domid  = domid;
507 	vif->handle = handle;
508 	vif->can_sg = 1;
509 	vif->ip_csum = 1;
510 	vif->dev = dev;
511 	vif->disabled = false;
512 	vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
513 	vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
514 
515 	/* Start out with no queues. */
516 	vif->queues = NULL;
517 	vif->num_queues = 0;
518 
519 	vif->xdp_headroom = 0;
520 
521 	spin_lock_init(&vif->lock);
522 	INIT_LIST_HEAD(&vif->fe_mcast_addr);
523 
524 	dev->netdev_ops	= &xenvif_netdev_ops;
525 	dev->hw_features = NETIF_F_SG |
526 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
527 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
528 	dev->features = dev->hw_features | NETIF_F_RXCSUM;
529 	dev->ethtool_ops = &xenvif_ethtool_ops;
530 
531 	dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
532 
533 	dev->min_mtu = ETH_MIN_MTU;
534 	dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
535 
536 	/*
537 	 * Initialise a dummy MAC address. We choose the numerically
538 	 * largest non-broadcast address to prevent the address getting
539 	 * stolen by an Ethernet bridge for STP purposes.
540 	 * (FE:FF:FF:FF:FF:FF)
541 	 */
542 	eth_hw_addr_set(dev, dummy_addr);
543 
544 	netif_carrier_off(dev);
545 
546 	err = register_netdev(dev);
547 	if (err) {
548 		netdev_warn(dev, "Could not register device: err=%d\n", err);
549 		free_netdev(dev);
550 		return ERR_PTR(err);
551 	}
552 
553 	netdev_dbg(dev, "Successfully created xenvif\n");
554 
555 	__module_get(THIS_MODULE);
556 
557 	return vif;
558 }
559 
560 int xenvif_init_queue(struct xenvif_queue *queue)
561 {
562 	int err, i;
563 
564 	queue->credit_bytes = queue->remaining_credit = ~0UL;
565 	queue->credit_usec  = 0UL;
566 	timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
567 	queue->credit_window_start = get_jiffies_64();
568 
569 	queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
570 
571 	skb_queue_head_init(&queue->rx_queue);
572 	skb_queue_head_init(&queue->tx_queue);
573 
574 	queue->pending_cons = 0;
575 	queue->pending_prod = MAX_PENDING_REQS;
576 	for (i = 0; i < MAX_PENDING_REQS; ++i)
577 		queue->pending_ring[i] = i;
578 
579 	spin_lock_init(&queue->callback_lock);
580 	spin_lock_init(&queue->response_lock);
581 
582 	/* If ballooning is disabled, this will consume real memory, so you
583 	 * better enable it. The long term solution would be to use just a
584 	 * bunch of valid page descriptors, without dependency on ballooning
585 	 */
586 	err = gnttab_alloc_pages(MAX_PENDING_REQS,
587 				 queue->mmap_pages);
588 	if (err) {
589 		netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
590 		return -ENOMEM;
591 	}
592 
593 	for (i = 0; i < MAX_PENDING_REQS; i++) {
594 		queue->pending_tx_info[i].callback_struct = (struct ubuf_info_msgzc)
595 			{ { .callback = xenvif_zerocopy_callback },
596 			  { { .ctx = NULL,
597 			      .desc = i } } };
598 		queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
599 	}
600 
601 	return 0;
602 }
603 
604 void xenvif_carrier_on(struct xenvif *vif)
605 {
606 	rtnl_lock();
607 	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
608 		dev_set_mtu(vif->dev, ETH_DATA_LEN);
609 	netdev_update_features(vif->dev);
610 	set_bit(VIF_STATUS_CONNECTED, &vif->status);
611 	if (netif_running(vif->dev))
612 		xenvif_up(vif);
613 	rtnl_unlock();
614 }
615 
616 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
617 			unsigned int evtchn)
618 {
619 	struct net_device *dev = vif->dev;
620 	struct xenbus_device *xendev = xenvif_to_xenbus_device(vif);
621 	void *addr;
622 	struct xen_netif_ctrl_sring *shared;
623 	RING_IDX rsp_prod, req_prod;
624 	int err;
625 
626 	err = xenbus_map_ring_valloc(xendev, &ring_ref, 1, &addr);
627 	if (err)
628 		goto err;
629 
630 	shared = (struct xen_netif_ctrl_sring *)addr;
631 	rsp_prod = READ_ONCE(shared->rsp_prod);
632 	req_prod = READ_ONCE(shared->req_prod);
633 
634 	BACK_RING_ATTACH(&vif->ctrl, shared, rsp_prod, XEN_PAGE_SIZE);
635 
636 	err = -EIO;
637 	if (req_prod - rsp_prod > RING_SIZE(&vif->ctrl))
638 		goto err_unmap;
639 
640 	err = bind_interdomain_evtchn_to_irq_lateeoi(xendev, evtchn);
641 	if (err < 0)
642 		goto err_unmap;
643 
644 	vif->ctrl_irq = err;
645 
646 	xenvif_init_hash(vif);
647 
648 	err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
649 				   IRQF_ONESHOT, "xen-netback-ctrl", vif);
650 	if (err) {
651 		pr_warn("Could not setup irq handler for %s\n", dev->name);
652 		goto err_deinit;
653 	}
654 
655 	return 0;
656 
657 err_deinit:
658 	xenvif_deinit_hash(vif);
659 	unbind_from_irqhandler(vif->ctrl_irq, vif);
660 	vif->ctrl_irq = 0;
661 
662 err_unmap:
663 	xenbus_unmap_ring_vfree(xendev, vif->ctrl.sring);
664 	vif->ctrl.sring = NULL;
665 
666 err:
667 	return err;
668 }
669 
670 static void xenvif_disconnect_queue(struct xenvif_queue *queue)
671 {
672 	if (queue->task) {
673 		kthread_stop(queue->task);
674 		put_task_struct(queue->task);
675 		queue->task = NULL;
676 	}
677 
678 	if (queue->dealloc_task) {
679 		kthread_stop(queue->dealloc_task);
680 		queue->dealloc_task = NULL;
681 	}
682 
683 	if (queue->napi.poll) {
684 		netif_napi_del(&queue->napi);
685 		queue->napi.poll = NULL;
686 	}
687 
688 	if (queue->tx_irq) {
689 		unbind_from_irqhandler(queue->tx_irq, queue);
690 		if (queue->tx_irq == queue->rx_irq)
691 			queue->rx_irq = 0;
692 		queue->tx_irq = 0;
693 	}
694 
695 	if (queue->rx_irq) {
696 		unbind_from_irqhandler(queue->rx_irq, queue);
697 		queue->rx_irq = 0;
698 	}
699 
700 	xenvif_unmap_frontend_data_rings(queue);
701 }
702 
703 int xenvif_connect_data(struct xenvif_queue *queue,
704 			unsigned long tx_ring_ref,
705 			unsigned long rx_ring_ref,
706 			unsigned int tx_evtchn,
707 			unsigned int rx_evtchn)
708 {
709 	struct xenbus_device *dev = xenvif_to_xenbus_device(queue->vif);
710 	struct task_struct *task;
711 	int err;
712 
713 	BUG_ON(queue->tx_irq);
714 	BUG_ON(queue->task);
715 	BUG_ON(queue->dealloc_task);
716 
717 	err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
718 					     rx_ring_ref);
719 	if (err < 0)
720 		goto err;
721 
722 	init_waitqueue_head(&queue->wq);
723 	init_waitqueue_head(&queue->dealloc_wq);
724 	atomic_set(&queue->inflight_packets, 0);
725 
726 	netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll);
727 
728 	queue->stalled = true;
729 
730 	task = kthread_run(xenvif_kthread_guest_rx, queue,
731 			   "%s-guest-rx", queue->name);
732 	if (IS_ERR(task))
733 		goto kthread_err;
734 	queue->task = task;
735 	/*
736 	 * Take a reference to the task in order to prevent it from being freed
737 	 * if the thread function returns before kthread_stop is called.
738 	 */
739 	get_task_struct(task);
740 
741 	task = kthread_run(xenvif_dealloc_kthread, queue,
742 			   "%s-dealloc", queue->name);
743 	if (IS_ERR(task))
744 		goto kthread_err;
745 	queue->dealloc_task = task;
746 
747 	if (tx_evtchn == rx_evtchn) {
748 		/* feature-split-event-channels == 0 */
749 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
750 			dev, tx_evtchn, xenvif_interrupt, 0,
751 			queue->name, queue);
752 		if (err < 0)
753 			goto err;
754 		queue->tx_irq = queue->rx_irq = err;
755 		disable_irq(queue->tx_irq);
756 	} else {
757 		/* feature-split-event-channels == 1 */
758 		snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
759 			 "%s-tx", queue->name);
760 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
761 			dev, tx_evtchn, xenvif_tx_interrupt, 0,
762 			queue->tx_irq_name, queue);
763 		if (err < 0)
764 			goto err;
765 		queue->tx_irq = err;
766 		disable_irq(queue->tx_irq);
767 
768 		snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
769 			 "%s-rx", queue->name);
770 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
771 			dev, rx_evtchn, xenvif_rx_interrupt, 0,
772 			queue->rx_irq_name, queue);
773 		if (err < 0)
774 			goto err;
775 		queue->rx_irq = err;
776 		disable_irq(queue->rx_irq);
777 	}
778 
779 	return 0;
780 
781 kthread_err:
782 	pr_warn("Could not allocate kthread for %s\n", queue->name);
783 	err = PTR_ERR(task);
784 err:
785 	xenvif_disconnect_queue(queue);
786 	return err;
787 }
788 
789 void xenvif_carrier_off(struct xenvif *vif)
790 {
791 	struct net_device *dev = vif->dev;
792 
793 	rtnl_lock();
794 	if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
795 		netif_carrier_off(dev); /* discard queued packets */
796 		if (netif_running(dev))
797 			xenvif_down(vif);
798 	}
799 	rtnl_unlock();
800 }
801 
802 void xenvif_disconnect_data(struct xenvif *vif)
803 {
804 	struct xenvif_queue *queue = NULL;
805 	unsigned int num_queues = vif->num_queues;
806 	unsigned int queue_index;
807 
808 	xenvif_carrier_off(vif);
809 
810 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
811 		queue = &vif->queues[queue_index];
812 
813 		xenvif_disconnect_queue(queue);
814 	}
815 
816 	xenvif_mcast_addr_list_free(vif);
817 }
818 
819 void xenvif_disconnect_ctrl(struct xenvif *vif)
820 {
821 	if (vif->ctrl_irq) {
822 		xenvif_deinit_hash(vif);
823 		unbind_from_irqhandler(vif->ctrl_irq, vif);
824 		vif->ctrl_irq = 0;
825 	}
826 
827 	if (vif->ctrl.sring) {
828 		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
829 					vif->ctrl.sring);
830 		vif->ctrl.sring = NULL;
831 	}
832 }
833 
834 /* Reverse the relevant parts of xenvif_init_queue().
835  * Used for queue teardown from xenvif_free(), and on the
836  * error handling paths in xenbus.c:connect().
837  */
838 void xenvif_deinit_queue(struct xenvif_queue *queue)
839 {
840 	gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
841 }
842 
843 void xenvif_free(struct xenvif *vif)
844 {
845 	struct xenvif_queue *queues = vif->queues;
846 	unsigned int num_queues = vif->num_queues;
847 	unsigned int queue_index;
848 
849 	unregister_netdev(vif->dev);
850 	free_netdev(vif->dev);
851 
852 	for (queue_index = 0; queue_index < num_queues; ++queue_index)
853 		xenvif_deinit_queue(&queues[queue_index]);
854 	vfree(queues);
855 
856 	module_put(THIS_MODULE);
857 }
858