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 /* Number of bytes allowed on the internal guest Rx queue. */
45 #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
46 
47 /* This function is used to set SKBFL_ZEROCOPY_ENABLE as well as
48  * increasing the inflight counter. We need to increase the inflight
49  * counter because core driver calls into xenvif_zerocopy_callback
50  * which calls xenvif_skb_zerocopy_complete.
51  */
52 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
53 				 struct sk_buff *skb)
54 {
55 	skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_ENABLE;
56 	atomic_inc(&queue->inflight_packets);
57 }
58 
59 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
60 {
61 	atomic_dec(&queue->inflight_packets);
62 
63 	/* Wake the dealloc thread _after_ decrementing inflight_packets so
64 	 * that if kthread_stop() has already been called, the dealloc thread
65 	 * does not wait forever with nothing to wake it.
66 	 */
67 	wake_up(&queue->dealloc_wq);
68 }
69 
70 static int xenvif_schedulable(struct xenvif *vif)
71 {
72 	return netif_running(vif->dev) &&
73 		test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
74 		!vif->disabled;
75 }
76 
77 static bool xenvif_handle_tx_interrupt(struct xenvif_queue *queue)
78 {
79 	bool rc;
80 
81 	rc = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
82 	if (rc)
83 		napi_schedule(&queue->napi);
84 	return rc;
85 }
86 
87 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
88 {
89 	struct xenvif_queue *queue = dev_id;
90 	int old;
91 
92 	old = atomic_fetch_or(NETBK_TX_EOI, &queue->eoi_pending);
93 	WARN(old & NETBK_TX_EOI, "Interrupt while EOI pending\n");
94 
95 	if (!xenvif_handle_tx_interrupt(queue)) {
96 		atomic_andnot(NETBK_TX_EOI, &queue->eoi_pending);
97 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
98 	}
99 
100 	return IRQ_HANDLED;
101 }
102 
103 static int xenvif_poll(struct napi_struct *napi, int budget)
104 {
105 	struct xenvif_queue *queue =
106 		container_of(napi, struct xenvif_queue, napi);
107 	int work_done;
108 
109 	/* This vif is rogue, we pretend we've there is nothing to do
110 	 * for this vif to deschedule it from NAPI. But this interface
111 	 * will be turned off in thread context later.
112 	 */
113 	if (unlikely(queue->vif->disabled)) {
114 		napi_complete(napi);
115 		return 0;
116 	}
117 
118 	work_done = xenvif_tx_action(queue, budget);
119 
120 	if (work_done < budget) {
121 		napi_complete_done(napi, work_done);
122 		/* If the queue is rate-limited, it shall be
123 		 * rescheduled in the timer callback.
124 		 */
125 		if (likely(!queue->rate_limited))
126 			xenvif_napi_schedule_or_enable_events(queue);
127 	}
128 
129 	return work_done;
130 }
131 
132 static bool xenvif_handle_rx_interrupt(struct xenvif_queue *queue)
133 {
134 	bool rc;
135 
136 	rc = xenvif_have_rx_work(queue, false);
137 	if (rc)
138 		xenvif_kick_thread(queue);
139 	return rc;
140 }
141 
142 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
143 {
144 	struct xenvif_queue *queue = dev_id;
145 	int old;
146 
147 	old = atomic_fetch_or(NETBK_RX_EOI, &queue->eoi_pending);
148 	WARN(old & NETBK_RX_EOI, "Interrupt while EOI pending\n");
149 
150 	if (!xenvif_handle_rx_interrupt(queue)) {
151 		atomic_andnot(NETBK_RX_EOI, &queue->eoi_pending);
152 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
153 	}
154 
155 	return IRQ_HANDLED;
156 }
157 
158 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
159 {
160 	struct xenvif_queue *queue = dev_id;
161 	int old;
162 	bool has_rx, has_tx;
163 
164 	old = atomic_fetch_or(NETBK_COMMON_EOI, &queue->eoi_pending);
165 	WARN(old, "Interrupt while EOI pending\n");
166 
167 	has_tx = xenvif_handle_tx_interrupt(queue);
168 	has_rx = xenvif_handle_rx_interrupt(queue);
169 
170 	if (!has_rx && !has_tx) {
171 		atomic_andnot(NETBK_COMMON_EOI, &queue->eoi_pending);
172 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
173 	}
174 
175 	return IRQ_HANDLED;
176 }
177 
178 static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
179 			       struct net_device *sb_dev)
180 {
181 	struct xenvif *vif = netdev_priv(dev);
182 	unsigned int size = vif->hash.size;
183 	unsigned int num_queues;
184 
185 	/* If queues are not set up internally - always return 0
186 	 * as the packet going to be dropped anyway */
187 	num_queues = READ_ONCE(vif->num_queues);
188 	if (num_queues < 1)
189 		return 0;
190 
191 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
192 		return netdev_pick_tx(dev, skb, NULL) %
193 		       dev->real_num_tx_queues;
194 
195 	xenvif_set_skb_hash(vif, skb);
196 
197 	if (size == 0)
198 		return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
199 
200 	return vif->hash.mapping[vif->hash.mapping_sel]
201 				[skb_get_hash_raw(skb) % size];
202 }
203 
204 static netdev_tx_t
205 xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
206 {
207 	struct xenvif *vif = netdev_priv(dev);
208 	struct xenvif_queue *queue = NULL;
209 	unsigned int num_queues;
210 	u16 index;
211 	struct xenvif_rx_cb *cb;
212 
213 	BUG_ON(skb->dev != dev);
214 
215 	/* Drop the packet if queues are not set up.
216 	 * This handler should be called inside an RCU read section
217 	 * so we don't need to enter it here explicitly.
218 	 */
219 	num_queues = READ_ONCE(vif->num_queues);
220 	if (num_queues < 1)
221 		goto drop;
222 
223 	/* Obtain the queue to be used to transmit this packet */
224 	index = skb_get_queue_mapping(skb);
225 	if (index >= num_queues) {
226 		pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
227 				    index, vif->dev->name);
228 		index %= num_queues;
229 	}
230 	queue = &vif->queues[index];
231 
232 	/* Drop the packet if queue is not ready */
233 	if (queue->task == NULL ||
234 	    queue->dealloc_task == NULL ||
235 	    !xenvif_schedulable(vif))
236 		goto drop;
237 
238 	if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
239 		struct ethhdr *eth = (struct ethhdr *)skb->data;
240 
241 		if (!xenvif_mcast_match(vif, eth->h_dest))
242 			goto drop;
243 	}
244 
245 	cb = XENVIF_RX_CB(skb);
246 	cb->expires = jiffies + vif->drain_timeout;
247 
248 	/* If there is no hash algorithm configured then make sure there
249 	 * is no hash information in the socket buffer otherwise it
250 	 * would be incorrectly forwarded to the frontend.
251 	 */
252 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
253 		skb_clear_hash(skb);
254 
255 	if (!xenvif_rx_queue_tail(queue, skb))
256 		goto drop;
257 
258 	xenvif_kick_thread(queue);
259 
260 	return NETDEV_TX_OK;
261 
262  drop:
263 	vif->dev->stats.tx_dropped++;
264 	dev_kfree_skb_any(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->min_mtu = ETH_MIN_MTU;
532 	dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
533 
534 	/*
535 	 * Initialise a dummy MAC address. We choose the numerically
536 	 * largest non-broadcast address to prevent the address getting
537 	 * stolen by an Ethernet bridge for STP purposes.
538 	 * (FE:FF:FF:FF:FF:FF)
539 	 */
540 	eth_hw_addr_set(dev, dummy_addr);
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 
558 int 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_msgzc)
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 
602 void 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 
614 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
615 			unsigned int evtchn)
616 {
617 	struct net_device *dev = vif->dev;
618 	struct xenbus_device *xendev = xenvif_to_xenbus_device(vif);
619 	void *addr;
620 	struct xen_netif_ctrl_sring *shared;
621 	RING_IDX rsp_prod, req_prod;
622 	int err;
623 
624 	err = xenbus_map_ring_valloc(xendev, &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(xendev, 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 
655 err_deinit:
656 	xenvif_deinit_hash(vif);
657 	unbind_from_irqhandler(vif->ctrl_irq, vif);
658 	vif->ctrl_irq = 0;
659 
660 err_unmap:
661 	xenbus_unmap_ring_vfree(xendev, vif->ctrl.sring);
662 	vif->ctrl.sring = NULL;
663 
664 err:
665 	return err;
666 }
667 
668 static void xenvif_disconnect_queue(struct xenvif_queue *queue)
669 {
670 	if (queue->task) {
671 		kthread_stop_put(queue->task);
672 		queue->task = NULL;
673 	}
674 
675 	if (queue->dealloc_task) {
676 		kthread_stop(queue->dealloc_task);
677 		queue->dealloc_task = NULL;
678 	}
679 
680 	if (queue->napi.poll) {
681 		netif_napi_del(&queue->napi);
682 		queue->napi.poll = NULL;
683 	}
684 
685 	if (queue->tx_irq) {
686 		unbind_from_irqhandler(queue->tx_irq, queue);
687 		if (queue->tx_irq == queue->rx_irq)
688 			queue->rx_irq = 0;
689 		queue->tx_irq = 0;
690 	}
691 
692 	if (queue->rx_irq) {
693 		unbind_from_irqhandler(queue->rx_irq, queue);
694 		queue->rx_irq = 0;
695 	}
696 
697 	xenvif_unmap_frontend_data_rings(queue);
698 }
699 
700 int xenvif_connect_data(struct xenvif_queue *queue,
701 			unsigned long tx_ring_ref,
702 			unsigned long rx_ring_ref,
703 			unsigned int tx_evtchn,
704 			unsigned int rx_evtchn)
705 {
706 	struct xenbus_device *dev = xenvif_to_xenbus_device(queue->vif);
707 	struct task_struct *task;
708 	int err;
709 
710 	BUG_ON(queue->tx_irq);
711 	BUG_ON(queue->task);
712 	BUG_ON(queue->dealloc_task);
713 
714 	err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
715 					     rx_ring_ref);
716 	if (err < 0)
717 		goto err;
718 
719 	init_waitqueue_head(&queue->wq);
720 	init_waitqueue_head(&queue->dealloc_wq);
721 	atomic_set(&queue->inflight_packets, 0);
722 
723 	netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll);
724 
725 	queue->stalled = true;
726 
727 	task = kthread_run(xenvif_kthread_guest_rx, queue,
728 			   "%s-guest-rx", queue->name);
729 	if (IS_ERR(task))
730 		goto kthread_err;
731 	queue->task = task;
732 	/*
733 	 * Take a reference to the task in order to prevent it from being freed
734 	 * if the thread function returns before kthread_stop is called.
735 	 */
736 	get_task_struct(task);
737 
738 	task = kthread_run(xenvif_dealloc_kthread, queue,
739 			   "%s-dealloc", queue->name);
740 	if (IS_ERR(task))
741 		goto kthread_err;
742 	queue->dealloc_task = task;
743 
744 	if (tx_evtchn == rx_evtchn) {
745 		/* feature-split-event-channels == 0 */
746 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
747 			dev, tx_evtchn, xenvif_interrupt, 0,
748 			queue->name, queue);
749 		if (err < 0)
750 			goto err;
751 		queue->tx_irq = queue->rx_irq = err;
752 		disable_irq(queue->tx_irq);
753 	} else {
754 		/* feature-split-event-channels == 1 */
755 		snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
756 			 "%s-tx", queue->name);
757 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
758 			dev, tx_evtchn, xenvif_tx_interrupt, 0,
759 			queue->tx_irq_name, queue);
760 		if (err < 0)
761 			goto err;
762 		queue->tx_irq = err;
763 		disable_irq(queue->tx_irq);
764 
765 		snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
766 			 "%s-rx", queue->name);
767 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
768 			dev, rx_evtchn, xenvif_rx_interrupt, 0,
769 			queue->rx_irq_name, queue);
770 		if (err < 0)
771 			goto err;
772 		queue->rx_irq = err;
773 		disable_irq(queue->rx_irq);
774 	}
775 
776 	return 0;
777 
778 kthread_err:
779 	pr_warn("Could not allocate kthread for %s\n", queue->name);
780 	err = PTR_ERR(task);
781 err:
782 	xenvif_disconnect_queue(queue);
783 	return err;
784 }
785 
786 void xenvif_carrier_off(struct xenvif *vif)
787 {
788 	struct net_device *dev = vif->dev;
789 
790 	rtnl_lock();
791 	if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
792 		netif_carrier_off(dev); /* discard queued packets */
793 		if (netif_running(dev))
794 			xenvif_down(vif);
795 	}
796 	rtnl_unlock();
797 }
798 
799 void xenvif_disconnect_data(struct xenvif *vif)
800 {
801 	struct xenvif_queue *queue = NULL;
802 	unsigned int num_queues = vif->num_queues;
803 	unsigned int queue_index;
804 
805 	xenvif_carrier_off(vif);
806 
807 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
808 		queue = &vif->queues[queue_index];
809 
810 		xenvif_disconnect_queue(queue);
811 	}
812 
813 	xenvif_mcast_addr_list_free(vif);
814 }
815 
816 void xenvif_disconnect_ctrl(struct xenvif *vif)
817 {
818 	if (vif->ctrl_irq) {
819 		xenvif_deinit_hash(vif);
820 		unbind_from_irqhandler(vif->ctrl_irq, vif);
821 		vif->ctrl_irq = 0;
822 	}
823 
824 	if (vif->ctrl.sring) {
825 		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
826 					vif->ctrl.sring);
827 		vif->ctrl.sring = NULL;
828 	}
829 }
830 
831 /* Reverse the relevant parts of xenvif_init_queue().
832  * Used for queue teardown from xenvif_free(), and on the
833  * error handling paths in xenbus.c:connect().
834  */
835 void xenvif_deinit_queue(struct xenvif_queue *queue)
836 {
837 	gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
838 }
839 
840 void xenvif_free(struct xenvif *vif)
841 {
842 	struct xenvif_queue *queues = vif->queues;
843 	unsigned int num_queues = vif->num_queues;
844 	unsigned int queue_index;
845 
846 	unregister_netdev(vif->dev);
847 	free_netdev(vif->dev);
848 
849 	for (queue_index = 0; queue_index < num_queues; ++queue_index)
850 		xenvif_deinit_queue(&queues[queue_index]);
851 	vfree(queues);
852 
853 	module_put(THIS_MODULE);
854 }
855