• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Network-device interface management.
3  *
4  * Copyright (c) 2004-2005, Keir Fraser
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License version 2
8  * as published by the Free Software Foundation; or, when distributed
9  * separately from the Linux kernel or incorporated into other
10  * software packages, subject to the following license:
11  *
12  * Permission is hereby granted, free of charge, to any person obtaining a copy
13  * of this source file (the "Software"), to deal in the Software without
14  * restriction, including without limitation the rights to use, copy, modify,
15  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
16  * and to permit persons to whom the Software is furnished to do so, subject to
17  * the following conditions:
18  *
19  * The above copyright notice and this permission notice shall be included in
20  * all copies or substantial portions of the Software.
21  *
22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
25  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
28  * IN THE SOFTWARE.
29  */
30 
31 #include "common.h"
32 
33 #include <linux/kthread.h>
34 #include <linux/ethtool.h>
35 #include <linux/rtnetlink.h>
36 #include <linux/if_vlan.h>
37 #include <linux/vmalloc.h>
38 
39 #include <xen/events.h>
40 #include <asm/xen/hypercall.h>
41 #include <xen/balloon.h>
42 
43 #define XENVIF_QUEUE_LENGTH 32
44 #define XENVIF_NAPI_WEIGHT  64
45 
46 /* Number of bytes allowed on the internal guest Rx queue. */
47 #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
48 
49 /* This function is used to set SKBTX_DEV_ZEROCOPY as well as
50  * increasing the inflight counter. We need to increase the inflight
51  * counter because core driver calls into xenvif_zerocopy_callback
52  * which calls xenvif_skb_zerocopy_complete.
53  */
xenvif_skb_zerocopy_prepare(struct xenvif_queue * queue,struct sk_buff * skb)54 void 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 
xenvif_skb_zerocopy_complete(struct xenvif_queue * queue)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 
xenvif_schedulable(struct xenvif * vif)72 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 
xenvif_handle_tx_interrupt(struct xenvif_queue * queue)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 
xenvif_tx_interrupt(int irq,void * dev_id)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 = xenvif_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 
xenvif_poll(struct napi_struct * napi,int budget)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(napi);
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 
xenvif_handle_rx_interrupt(struct xenvif_queue * queue)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 
xenvif_rx_interrupt(int irq,void * dev_id)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 = xenvif_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 
xenvif_interrupt(int irq,void * dev_id)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 = xenvif_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 
xenvif_queue_stopped(struct xenvif_queue * queue)180 int xenvif_queue_stopped(struct xenvif_queue *queue)
181 {
182 	struct net_device *dev = queue->vif->dev;
183 	unsigned int id = queue->id;
184 	return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
185 }
186 
xenvif_wake_queue(struct xenvif_queue * queue)187 void xenvif_wake_queue(struct xenvif_queue *queue)
188 {
189 	struct net_device *dev = queue->vif->dev;
190 	unsigned int id = queue->id;
191 	netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
192 }
193 
xenvif_start_xmit(struct sk_buff * skb,struct net_device * dev)194 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
195 {
196 	struct xenvif *vif = netdev_priv(dev);
197 	struct xenvif_queue *queue = NULL;
198 	unsigned int num_queues = vif->num_queues;
199 	u16 index;
200 	struct xenvif_rx_cb *cb;
201 
202 	BUG_ON(skb->dev != dev);
203 
204 	/* Drop the packet if queues are not set up */
205 	if (num_queues < 1)
206 		goto drop;
207 
208 	/* Obtain the queue to be used to transmit this packet */
209 	index = skb_get_queue_mapping(skb);
210 	if (index >= num_queues) {
211 		pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
212 				    index, vif->dev->name);
213 		index %= num_queues;
214 	}
215 	queue = &vif->queues[index];
216 
217 	/* Drop the packet if queue is not ready */
218 	if (queue->task == NULL ||
219 	    queue->dealloc_task == NULL ||
220 	    !xenvif_schedulable(vif))
221 		goto drop;
222 
223 	if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
224 		struct ethhdr *eth = (struct ethhdr *)skb->data;
225 
226 		if (!xenvif_mcast_match(vif, eth->h_dest))
227 			goto drop;
228 	}
229 
230 	cb = XENVIF_RX_CB(skb);
231 	cb->expires = jiffies + vif->drain_timeout;
232 
233 	xenvif_rx_queue_tail(queue, skb);
234 	xenvif_kick_thread(queue);
235 
236 	return NETDEV_TX_OK;
237 
238  drop:
239 	vif->dev->stats.tx_dropped++;
240 	dev_kfree_skb(skb);
241 	return NETDEV_TX_OK;
242 }
243 
xenvif_get_stats(struct net_device * dev)244 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
245 {
246 	struct xenvif *vif = netdev_priv(dev);
247 	struct xenvif_queue *queue = NULL;
248 	unsigned int num_queues = vif->num_queues;
249 	unsigned long rx_bytes = 0;
250 	unsigned long rx_packets = 0;
251 	unsigned long tx_bytes = 0;
252 	unsigned long tx_packets = 0;
253 	unsigned int index;
254 
255 	if (vif->queues == NULL)
256 		goto out;
257 
258 	/* Aggregate tx and rx stats from each queue */
259 	for (index = 0; index < num_queues; ++index) {
260 		queue = &vif->queues[index];
261 		rx_bytes += queue->stats.rx_bytes;
262 		rx_packets += queue->stats.rx_packets;
263 		tx_bytes += queue->stats.tx_bytes;
264 		tx_packets += queue->stats.tx_packets;
265 	}
266 
267 out:
268 	vif->dev->stats.rx_bytes = rx_bytes;
269 	vif->dev->stats.rx_packets = rx_packets;
270 	vif->dev->stats.tx_bytes = tx_bytes;
271 	vif->dev->stats.tx_packets = tx_packets;
272 
273 	return &vif->dev->stats;
274 }
275 
xenvif_up(struct xenvif * vif)276 static void xenvif_up(struct xenvif *vif)
277 {
278 	struct xenvif_queue *queue = NULL;
279 	unsigned int num_queues = vif->num_queues;
280 	unsigned int queue_index;
281 
282 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
283 		queue = &vif->queues[queue_index];
284 		napi_enable(&queue->napi);
285 		enable_irq(queue->tx_irq);
286 		if (queue->tx_irq != queue->rx_irq)
287 			enable_irq(queue->rx_irq);
288 		xenvif_napi_schedule_or_enable_events(queue);
289 	}
290 }
291 
xenvif_down(struct xenvif * vif)292 static void xenvif_down(struct xenvif *vif)
293 {
294 	struct xenvif_queue *queue = NULL;
295 	unsigned int num_queues = vif->num_queues;
296 	unsigned int queue_index;
297 
298 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
299 		queue = &vif->queues[queue_index];
300 		disable_irq(queue->tx_irq);
301 		if (queue->tx_irq != queue->rx_irq)
302 			disable_irq(queue->rx_irq);
303 		napi_disable(&queue->napi);
304 		del_timer_sync(&queue->credit_timeout);
305 	}
306 }
307 
xenvif_open(struct net_device * dev)308 static int xenvif_open(struct net_device *dev)
309 {
310 	struct xenvif *vif = netdev_priv(dev);
311 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
312 		xenvif_up(vif);
313 	netif_tx_start_all_queues(dev);
314 	return 0;
315 }
316 
xenvif_close(struct net_device * dev)317 static int xenvif_close(struct net_device *dev)
318 {
319 	struct xenvif *vif = netdev_priv(dev);
320 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
321 		xenvif_down(vif);
322 	netif_tx_stop_all_queues(dev);
323 	return 0;
324 }
325 
xenvif_change_mtu(struct net_device * dev,int mtu)326 static int xenvif_change_mtu(struct net_device *dev, int mtu)
327 {
328 	struct xenvif *vif = netdev_priv(dev);
329 	int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
330 
331 	if (mtu > max)
332 		return -EINVAL;
333 	dev->mtu = mtu;
334 	return 0;
335 }
336 
xenvif_fix_features(struct net_device * dev,netdev_features_t features)337 static netdev_features_t xenvif_fix_features(struct net_device *dev,
338 	netdev_features_t features)
339 {
340 	struct xenvif *vif = netdev_priv(dev);
341 
342 	if (!vif->can_sg)
343 		features &= ~NETIF_F_SG;
344 	if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
345 		features &= ~NETIF_F_TSO;
346 	if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
347 		features &= ~NETIF_F_TSO6;
348 	if (!vif->ip_csum)
349 		features &= ~NETIF_F_IP_CSUM;
350 	if (!vif->ipv6_csum)
351 		features &= ~NETIF_F_IPV6_CSUM;
352 
353 	return features;
354 }
355 
356 static const struct xenvif_stat {
357 	char name[ETH_GSTRING_LEN];
358 	u16 offset;
359 } xenvif_stats[] = {
360 	{
361 		"rx_gso_checksum_fixup",
362 		offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
363 	},
364 	/* If (sent != success + fail), there are probably packets never
365 	 * freed up properly!
366 	 */
367 	{
368 		"tx_zerocopy_sent",
369 		offsetof(struct xenvif_stats, tx_zerocopy_sent),
370 	},
371 	{
372 		"tx_zerocopy_success",
373 		offsetof(struct xenvif_stats, tx_zerocopy_success),
374 	},
375 	{
376 		"tx_zerocopy_fail",
377 		offsetof(struct xenvif_stats, tx_zerocopy_fail)
378 	},
379 	/* Number of packets exceeding MAX_SKB_FRAG slots. You should use
380 	 * a guest with the same MAX_SKB_FRAG
381 	 */
382 	{
383 		"tx_frag_overflow",
384 		offsetof(struct xenvif_stats, tx_frag_overflow)
385 	},
386 };
387 
xenvif_get_sset_count(struct net_device * dev,int string_set)388 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
389 {
390 	switch (string_set) {
391 	case ETH_SS_STATS:
392 		return ARRAY_SIZE(xenvif_stats);
393 	default:
394 		return -EINVAL;
395 	}
396 }
397 
xenvif_get_ethtool_stats(struct net_device * dev,struct ethtool_stats * stats,u64 * data)398 static void xenvif_get_ethtool_stats(struct net_device *dev,
399 				     struct ethtool_stats *stats, u64 * data)
400 {
401 	struct xenvif *vif = netdev_priv(dev);
402 	unsigned int num_queues = vif->num_queues;
403 	int i;
404 	unsigned int queue_index;
405 
406 	for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
407 		unsigned long accum = 0;
408 		for (queue_index = 0; queue_index < num_queues; ++queue_index) {
409 			void *vif_stats = &vif->queues[queue_index].stats;
410 			accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
411 		}
412 		data[i] = accum;
413 	}
414 }
415 
xenvif_get_strings(struct net_device * dev,u32 stringset,u8 * data)416 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
417 {
418 	int i;
419 
420 	switch (stringset) {
421 	case ETH_SS_STATS:
422 		for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
423 			memcpy(data + i * ETH_GSTRING_LEN,
424 			       xenvif_stats[i].name, ETH_GSTRING_LEN);
425 		break;
426 	}
427 }
428 
429 static const struct ethtool_ops xenvif_ethtool_ops = {
430 	.get_link	= ethtool_op_get_link,
431 
432 	.get_sset_count = xenvif_get_sset_count,
433 	.get_ethtool_stats = xenvif_get_ethtool_stats,
434 	.get_strings = xenvif_get_strings,
435 };
436 
437 static const struct net_device_ops xenvif_netdev_ops = {
438 	.ndo_start_xmit	= xenvif_start_xmit,
439 	.ndo_get_stats	= xenvif_get_stats,
440 	.ndo_open	= xenvif_open,
441 	.ndo_stop	= xenvif_close,
442 	.ndo_change_mtu	= xenvif_change_mtu,
443 	.ndo_fix_features = xenvif_fix_features,
444 	.ndo_set_mac_address = eth_mac_addr,
445 	.ndo_validate_addr   = eth_validate_addr,
446 };
447 
xenvif_alloc(struct device * parent,domid_t domid,unsigned int handle)448 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
449 			    unsigned int handle)
450 {
451 	int err;
452 	struct net_device *dev;
453 	struct xenvif *vif;
454 	char name[IFNAMSIZ] = {};
455 
456 	snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
457 	/* Allocate a netdev with the max. supported number of queues.
458 	 * When the guest selects the desired number, it will be updated
459 	 * via netif_set_real_num_*_queues().
460 	 */
461 	dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
462 			      ether_setup, xenvif_max_queues);
463 	if (dev == NULL) {
464 		pr_warn("Could not allocate netdev for %s\n", name);
465 		return ERR_PTR(-ENOMEM);
466 	}
467 
468 	SET_NETDEV_DEV(dev, parent);
469 
470 	vif = netdev_priv(dev);
471 
472 	vif->domid  = domid;
473 	vif->handle = handle;
474 	vif->can_sg = 1;
475 	vif->ip_csum = 1;
476 	vif->dev = dev;
477 	vif->disabled = false;
478 	vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
479 	vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
480 
481 	/* Start out with no queues. */
482 	vif->queues = NULL;
483 	vif->num_queues = 0;
484 
485 	spin_lock_init(&vif->lock);
486 	INIT_LIST_HEAD(&vif->fe_mcast_addr);
487 
488 	dev->netdev_ops	= &xenvif_netdev_ops;
489 	dev->hw_features = NETIF_F_SG |
490 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
491 		NETIF_F_TSO | NETIF_F_TSO6;
492 	dev->features = dev->hw_features | NETIF_F_RXCSUM;
493 	dev->ethtool_ops = &xenvif_ethtool_ops;
494 
495 	dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
496 
497 	/*
498 	 * Initialise a dummy MAC address. We choose the numerically
499 	 * largest non-broadcast address to prevent the address getting
500 	 * stolen by an Ethernet bridge for STP purposes.
501 	 * (FE:FF:FF:FF:FF:FF)
502 	 */
503 	eth_broadcast_addr(dev->dev_addr);
504 	dev->dev_addr[0] &= ~0x01;
505 
506 	netif_carrier_off(dev);
507 
508 	err = register_netdev(dev);
509 	if (err) {
510 		netdev_warn(dev, "Could not register device: err=%d\n", err);
511 		free_netdev(dev);
512 		return ERR_PTR(err);
513 	}
514 
515 	netdev_dbg(dev, "Successfully created xenvif\n");
516 
517 	__module_get(THIS_MODULE);
518 
519 	return vif;
520 }
521 
xenvif_init_queue(struct xenvif_queue * queue)522 int xenvif_init_queue(struct xenvif_queue *queue)
523 {
524 	int err, i;
525 
526 	queue->credit_bytes = queue->remaining_credit = ~0UL;
527 	queue->credit_usec  = 0UL;
528 	init_timer(&queue->credit_timeout);
529 	queue->credit_timeout.function = xenvif_tx_credit_callback;
530 	queue->credit_window_start = get_jiffies_64();
531 
532 	queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
533 
534 	skb_queue_head_init(&queue->rx_queue);
535 	skb_queue_head_init(&queue->tx_queue);
536 
537 	queue->pending_cons = 0;
538 	queue->pending_prod = MAX_PENDING_REQS;
539 	for (i = 0; i < MAX_PENDING_REQS; ++i)
540 		queue->pending_ring[i] = i;
541 
542 	spin_lock_init(&queue->callback_lock);
543 	spin_lock_init(&queue->response_lock);
544 
545 	/* If ballooning is disabled, this will consume real memory, so you
546 	 * better enable it. The long term solution would be to use just a
547 	 * bunch of valid page descriptors, without dependency on ballooning
548 	 */
549 	err = gnttab_alloc_pages(MAX_PENDING_REQS,
550 				 queue->mmap_pages);
551 	if (err) {
552 		netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
553 		return -ENOMEM;
554 	}
555 
556 	for (i = 0; i < MAX_PENDING_REQS; i++) {
557 		queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
558 			{ .callback = xenvif_zerocopy_callback,
559 			  .ctx = NULL,
560 			  .desc = i };
561 		queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
562 	}
563 
564 	return 0;
565 }
566 
xenvif_carrier_on(struct xenvif * vif)567 void xenvif_carrier_on(struct xenvif *vif)
568 {
569 	rtnl_lock();
570 	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
571 		dev_set_mtu(vif->dev, ETH_DATA_LEN);
572 	netdev_update_features(vif->dev);
573 	set_bit(VIF_STATUS_CONNECTED, &vif->status);
574 	if (netif_running(vif->dev))
575 		xenvif_up(vif);
576 	rtnl_unlock();
577 }
578 
xenvif_connect(struct xenvif_queue * queue,unsigned long tx_ring_ref,unsigned long rx_ring_ref,unsigned int tx_evtchn,unsigned int rx_evtchn)579 int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
580 		   unsigned long rx_ring_ref, unsigned int tx_evtchn,
581 		   unsigned int rx_evtchn)
582 {
583 	struct task_struct *task;
584 	int err = -ENOMEM;
585 
586 	BUG_ON(queue->tx_irq);
587 	BUG_ON(queue->task);
588 	BUG_ON(queue->dealloc_task);
589 
590 	err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
591 	if (err < 0)
592 		goto err;
593 
594 	init_waitqueue_head(&queue->wq);
595 	init_waitqueue_head(&queue->dealloc_wq);
596 	atomic_set(&queue->inflight_packets, 0);
597 
598 	netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
599 			XENVIF_NAPI_WEIGHT);
600 
601 	if (tx_evtchn == rx_evtchn) {
602 		/* feature-split-event-channels == 0 */
603 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
604 			queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
605 			queue->name, queue);
606 		if (err < 0)
607 			goto err_unmap;
608 		queue->tx_irq = queue->rx_irq = err;
609 		disable_irq(queue->tx_irq);
610 	} else {
611 		/* feature-split-event-channels == 1 */
612 		snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
613 			 "%s-tx", queue->name);
614 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
615 			queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
616 			queue->tx_irq_name, queue);
617 		if (err < 0)
618 			goto err_unmap;
619 		queue->tx_irq = err;
620 		disable_irq(queue->tx_irq);
621 
622 		snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
623 			 "%s-rx", queue->name);
624 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
625 			queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
626 			queue->rx_irq_name, queue);
627 		if (err < 0)
628 			goto err_tx_unbind;
629 		queue->rx_irq = err;
630 		disable_irq(queue->rx_irq);
631 	}
632 
633 	queue->stalled = true;
634 
635 	task = kthread_create(xenvif_kthread_guest_rx,
636 			      (void *)queue, "%s-guest-rx", queue->name);
637 	if (IS_ERR(task)) {
638 		pr_warn("Could not allocate kthread for %s\n", queue->name);
639 		err = PTR_ERR(task);
640 		goto err_rx_unbind;
641 	}
642 	queue->task = task;
643 	get_task_struct(task);
644 
645 	task = kthread_create(xenvif_dealloc_kthread,
646 			      (void *)queue, "%s-dealloc", queue->name);
647 	if (IS_ERR(task)) {
648 		pr_warn("Could not allocate kthread for %s\n", queue->name);
649 		err = PTR_ERR(task);
650 		goto err_rx_unbind;
651 	}
652 	queue->dealloc_task = task;
653 
654 	wake_up_process(queue->task);
655 	wake_up_process(queue->dealloc_task);
656 
657 	return 0;
658 
659 err_rx_unbind:
660 	unbind_from_irqhandler(queue->rx_irq, queue);
661 	queue->rx_irq = 0;
662 err_tx_unbind:
663 	unbind_from_irqhandler(queue->tx_irq, queue);
664 	queue->tx_irq = 0;
665 err_unmap:
666 	xenvif_unmap_frontend_rings(queue);
667 	netif_napi_del(&queue->napi);
668 err:
669 	return err;
670 }
671 
xenvif_carrier_off(struct xenvif * vif)672 void xenvif_carrier_off(struct xenvif *vif)
673 {
674 	struct net_device *dev = vif->dev;
675 
676 	rtnl_lock();
677 	if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
678 		netif_carrier_off(dev); /* discard queued packets */
679 		if (netif_running(dev))
680 			xenvif_down(vif);
681 	}
682 	rtnl_unlock();
683 }
684 
xenvif_disconnect(struct xenvif * vif)685 void xenvif_disconnect(struct xenvif *vif)
686 {
687 	struct xenvif_queue *queue = NULL;
688 	unsigned int num_queues = vif->num_queues;
689 	unsigned int queue_index;
690 
691 	xenvif_carrier_off(vif);
692 
693 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
694 		queue = &vif->queues[queue_index];
695 
696 		netif_napi_del(&queue->napi);
697 
698 		if (queue->task) {
699 			kthread_stop(queue->task);
700 			put_task_struct(queue->task);
701 			queue->task = NULL;
702 		}
703 
704 		if (queue->dealloc_task) {
705 			kthread_stop(queue->dealloc_task);
706 			queue->dealloc_task = NULL;
707 		}
708 
709 		if (queue->tx_irq) {
710 			if (queue->tx_irq == queue->rx_irq)
711 				unbind_from_irqhandler(queue->tx_irq, queue);
712 			else {
713 				unbind_from_irqhandler(queue->tx_irq, queue);
714 				unbind_from_irqhandler(queue->rx_irq, queue);
715 			}
716 			queue->tx_irq = 0;
717 		}
718 
719 		xenvif_unmap_frontend_rings(queue);
720 	}
721 
722 	xenvif_mcast_addr_list_free(vif);
723 }
724 
725 /* Reverse the relevant parts of xenvif_init_queue().
726  * Used for queue teardown from xenvif_free(), and on the
727  * error handling paths in xenbus.c:connect().
728  */
xenvif_deinit_queue(struct xenvif_queue * queue)729 void xenvif_deinit_queue(struct xenvif_queue *queue)
730 {
731 	gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
732 }
733 
xenvif_free(struct xenvif * vif)734 void xenvif_free(struct xenvif *vif)
735 {
736 	struct xenvif_queue *queue = NULL;
737 	unsigned int num_queues = vif->num_queues;
738 	unsigned int queue_index;
739 
740 	unregister_netdev(vif->dev);
741 
742 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
743 		queue = &vif->queues[queue_index];
744 		xenvif_deinit_queue(queue);
745 	}
746 
747 	vfree(vif->queues);
748 	vif->queues = NULL;
749 	vif->num_queues = 0;
750 
751 	free_netdev(vif->dev);
752 
753 	module_put(THIS_MODULE);
754 }
755