]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - drivers/net/xen-netback/interface.c
xen-netback: Factor queue-specific data into queue struct
[mirror_ubuntu-jammy-kernel.git] / drivers / net / xen-netback / interface.c
CommitLineData
f942dc25
IC
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
b3f980bd 33#include <linux/kthread.h>
f942dc25
IC
34#include <linux/ethtool.h>
35#include <linux/rtnetlink.h>
36#include <linux/if_vlan.h>
37
38#include <xen/events.h>
39#include <asm/xen/hypercall.h>
f53c3fe8 40#include <xen/balloon.h>
f942dc25
IC
41
42#define XENVIF_QUEUE_LENGTH 32
b3f980bd 43#define XENVIF_NAPI_WEIGHT 64
f942dc25 44
e9ce7cb6
WL
45static inline void xenvif_stop_queue(struct xenvif_queue *queue)
46{
47 struct net_device *dev = queue->vif->dev;
48
49 if (!queue->vif->can_queue)
50 return;
51
52 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
53}
54
f942dc25
IC
55int xenvif_schedulable(struct xenvif *vif)
56{
57 return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
58}
59
e1f00a69 60static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
f942dc25 61{
e9ce7cb6 62 struct xenvif_queue *queue = dev_id;
f942dc25 63
e9ce7cb6
WL
64 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
65 napi_schedule(&queue->napi);
f942dc25 66
e1f00a69
WL
67 return IRQ_HANDLED;
68}
69
e9ce7cb6 70int xenvif_poll(struct napi_struct *napi, int budget)
b3f980bd 71{
e9ce7cb6
WL
72 struct xenvif_queue *queue =
73 container_of(napi, struct xenvif_queue, napi);
b3f980bd
WL
74 int work_done;
75
e9d8b2c2
WL
76 /* This vif is rogue, we pretend we've there is nothing to do
77 * for this vif to deschedule it from NAPI. But this interface
78 * will be turned off in thread context later.
79 */
e9ce7cb6 80 if (unlikely(queue->vif->disabled)) {
e9d8b2c2
WL
81 napi_complete(napi);
82 return 0;
83 }
84
e9ce7cb6 85 work_done = xenvif_tx_action(queue, budget);
b3f980bd
WL
86
87 if (work_done < budget) {
0d08fceb 88 napi_complete(napi);
e9ce7cb6 89 xenvif_napi_schedule_or_enable_events(queue);
b3f980bd
WL
90 }
91
92 return work_done;
93}
94
e1f00a69
WL
95static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
96{
e9ce7cb6 97 struct xenvif_queue *queue = dev_id;
e1f00a69 98
e9ce7cb6 99 xenvif_kick_thread(queue);
f942dc25
IC
100
101 return IRQ_HANDLED;
102}
103
e1f00a69
WL
104static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
105{
106 xenvif_tx_interrupt(irq, dev_id);
107 xenvif_rx_interrupt(irq, dev_id);
108
109 return IRQ_HANDLED;
110}
111
e9ce7cb6
WL
112int xenvif_queue_stopped(struct xenvif_queue *queue)
113{
114 struct net_device *dev = queue->vif->dev;
115 unsigned int id = queue->id;
116 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
117}
118
119void xenvif_wake_queue(struct xenvif_queue *queue)
120{
121 struct net_device *dev = queue->vif->dev;
122 unsigned int id = queue->id;
123 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
124}
125
126/* Callback to wake the queue and drain it on timeout */
127static void xenvif_wake_queue_callback(unsigned long data)
09350788 128{
e9ce7cb6
WL
129 struct xenvif_queue *queue = (struct xenvif_queue *)data;
130
131 if (xenvif_queue_stopped(queue)) {
132 netdev_err(queue->vif->dev, "draining TX queue\n");
133 queue->rx_queue_purge = true;
134 xenvif_kick_thread(queue);
135 xenvif_wake_queue(queue);
136 }
137}
09350788 138
e9ce7cb6
WL
139static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
140 void *accel_priv, select_queue_fallback_t fallback)
141{
142 struct xenvif *vif = netdev_priv(dev);
143 unsigned int num_queues = dev->real_num_tx_queues;
144 u32 hash;
145 u16 queue_index;
146
147 /* First, check if there is only one queue to optimise the
148 * single-queue or old frontend scenario.
149 */
150 if (num_queues == 1) {
151 queue_index = 0;
152 } else {
153 /* Use skb_get_hash to obtain an L4 hash if available */
154 hash = skb_get_hash(skb);
155 queue_index = hash % num_queues;
09350788 156 }
e9ce7cb6
WL
157
158 return queue_index;
09350788
ZK
159}
160
f942dc25
IC
161static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
162{
163 struct xenvif *vif = netdev_priv(dev);
e9ce7cb6
WL
164 struct xenvif_queue *queue = NULL;
165 unsigned int num_queues = dev->real_num_tx_queues;
166 u16 index;
ca2f09f2 167 int min_slots_needed;
f942dc25
IC
168
169 BUG_ON(skb->dev != dev);
170
e9ce7cb6
WL
171 /* Drop the packet if queues are not set up */
172 if (num_queues < 1)
173 goto drop;
174
175 /* Obtain the queue to be used to transmit this packet */
176 index = skb_get_queue_mapping(skb);
177 if (index >= num_queues) {
178 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
179 index, vif->dev->name);
180 index %= num_queues;
181 }
182 queue = &vif->queues[index];
183
184 /* Drop the packet if queue is not ready */
185 if (queue->task == NULL ||
186 queue->dealloc_task == NULL ||
f53c3fe8 187 !xenvif_schedulable(vif))
f942dc25
IC
188 goto drop;
189
ca2f09f2
PD
190 /* At best we'll need one slot for the header and one for each
191 * frag.
192 */
193 min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
f942dc25 194
ca2f09f2
PD
195 /* If the skb is GSO then we'll also need an extra slot for the
196 * metadata.
197 */
836fbaf4 198 if (skb_is_gso(skb))
ca2f09f2 199 min_slots_needed++;
f942dc25 200
ca2f09f2
PD
201 /* If the skb can't possibly fit in the remaining slots
202 * then turn off the queue to give the ring a chance to
203 * drain.
204 */
e9ce7cb6
WL
205 if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) {
206 queue->wake_queue.function = xenvif_wake_queue_callback;
207 queue->wake_queue.data = (unsigned long)queue;
208 xenvif_stop_queue(queue);
209 mod_timer(&queue->wake_queue,
09350788
ZK
210 jiffies + rx_drain_timeout_jiffies);
211 }
f942dc25 212
e9ce7cb6
WL
213 skb_queue_tail(&queue->rx_queue, skb);
214 xenvif_kick_thread(queue);
f942dc25
IC
215
216 return NETDEV_TX_OK;
217
218 drop:
219 vif->dev->stats.tx_dropped++;
220 dev_kfree_skb(skb);
221 return NETDEV_TX_OK;
222}
223
f942dc25
IC
224static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
225{
226 struct xenvif *vif = netdev_priv(dev);
e9ce7cb6
WL
227 struct xenvif_queue *queue = NULL;
228 unsigned int num_queues = dev->real_num_tx_queues;
229 unsigned long rx_bytes = 0;
230 unsigned long rx_packets = 0;
231 unsigned long tx_bytes = 0;
232 unsigned long tx_packets = 0;
233 unsigned int index;
234
235 if (vif->queues == NULL)
236 goto out;
237
238 /* Aggregate tx and rx stats from each queue */
239 for (index = 0; index < num_queues; ++index) {
240 queue = &vif->queues[index];
241 rx_bytes += queue->stats.rx_bytes;
242 rx_packets += queue->stats.rx_packets;
243 tx_bytes += queue->stats.tx_bytes;
244 tx_packets += queue->stats.tx_packets;
245 }
246
247out:
248 vif->dev->stats.rx_bytes = rx_bytes;
249 vif->dev->stats.rx_packets = rx_packets;
250 vif->dev->stats.tx_bytes = tx_bytes;
251 vif->dev->stats.tx_packets = tx_packets;
252
f942dc25
IC
253 return &vif->dev->stats;
254}
255
256static void xenvif_up(struct xenvif *vif)
257{
e9ce7cb6
WL
258 struct xenvif_queue *queue = NULL;
259 unsigned int num_queues = vif->dev->real_num_tx_queues;
260 unsigned int queue_index;
261
262 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
263 queue = &vif->queues[queue_index];
264 napi_enable(&queue->napi);
265 enable_irq(queue->tx_irq);
266 if (queue->tx_irq != queue->rx_irq)
267 enable_irq(queue->rx_irq);
268 xenvif_napi_schedule_or_enable_events(queue);
269 }
f942dc25
IC
270}
271
272static void xenvif_down(struct xenvif *vif)
273{
e9ce7cb6
WL
274 struct xenvif_queue *queue = NULL;
275 unsigned int num_queues = vif->dev->real_num_tx_queues;
276 unsigned int queue_index;
277
278 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
279 queue = &vif->queues[queue_index];
280 napi_disable(&queue->napi);
281 disable_irq(queue->tx_irq);
282 if (queue->tx_irq != queue->rx_irq)
283 disable_irq(queue->rx_irq);
284 del_timer_sync(&queue->credit_timeout);
285 }
f942dc25
IC
286}
287
288static int xenvif_open(struct net_device *dev)
289{
290 struct xenvif *vif = netdev_priv(dev);
291 if (netif_carrier_ok(dev))
292 xenvif_up(vif);
e9ce7cb6 293 netif_tx_start_all_queues(dev);
f942dc25
IC
294 return 0;
295}
296
297static int xenvif_close(struct net_device *dev)
298{
299 struct xenvif *vif = netdev_priv(dev);
300 if (netif_carrier_ok(dev))
301 xenvif_down(vif);
e9ce7cb6 302 netif_tx_stop_all_queues(dev);
f942dc25
IC
303 return 0;
304}
305
306static int xenvif_change_mtu(struct net_device *dev, int mtu)
307{
308 struct xenvif *vif = netdev_priv(dev);
309 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
310
311 if (mtu > max)
312 return -EINVAL;
313 dev->mtu = mtu;
314 return 0;
315}
316
c8f44aff
MM
317static netdev_features_t xenvif_fix_features(struct net_device *dev,
318 netdev_features_t features)
f942dc25
IC
319{
320 struct xenvif *vif = netdev_priv(dev);
f942dc25 321
47103041
MM
322 if (!vif->can_sg)
323 features &= ~NETIF_F_SG;
82cada22 324 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
47103041 325 features &= ~NETIF_F_TSO;
82cada22
PD
326 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
327 features &= ~NETIF_F_TSO6;
146c8a77 328 if (!vif->ip_csum)
47103041 329 features &= ~NETIF_F_IP_CSUM;
146c8a77
PD
330 if (!vif->ipv6_csum)
331 features &= ~NETIF_F_IPV6_CSUM;
f942dc25 332
47103041 333 return features;
f942dc25
IC
334}
335
336static const struct xenvif_stat {
337 char name[ETH_GSTRING_LEN];
338 u16 offset;
339} xenvif_stats[] = {
340 {
341 "rx_gso_checksum_fixup",
e9ce7cb6 342 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
f942dc25 343 },
1bb332af
ZK
344 /* If (sent != success + fail), there are probably packets never
345 * freed up properly!
346 */
347 {
348 "tx_zerocopy_sent",
e9ce7cb6 349 offsetof(struct xenvif_stats, tx_zerocopy_sent),
1bb332af
ZK
350 },
351 {
352 "tx_zerocopy_success",
e9ce7cb6 353 offsetof(struct xenvif_stats, tx_zerocopy_success),
1bb332af
ZK
354 },
355 {
356 "tx_zerocopy_fail",
e9ce7cb6 357 offsetof(struct xenvif_stats, tx_zerocopy_fail)
1bb332af 358 },
e3377f36
ZK
359 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
360 * a guest with the same MAX_SKB_FRAG
361 */
362 {
363 "tx_frag_overflow",
e9ce7cb6 364 offsetof(struct xenvif_stats, tx_frag_overflow)
e3377f36 365 },
f942dc25
IC
366};
367
368static int xenvif_get_sset_count(struct net_device *dev, int string_set)
369{
370 switch (string_set) {
371 case ETH_SS_STATS:
372 return ARRAY_SIZE(xenvif_stats);
373 default:
374 return -EINVAL;
375 }
376}
377
378static void xenvif_get_ethtool_stats(struct net_device *dev,
379 struct ethtool_stats *stats, u64 * data)
380{
e9ce7cb6
WL
381 struct xenvif *vif = netdev_priv(dev);
382 unsigned int num_queues = dev->real_num_tx_queues;
f942dc25 383 int i;
e9ce7cb6
WL
384 unsigned int queue_index;
385 struct xenvif_stats *vif_stats;
386
387 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
388 unsigned long accum = 0;
389 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
390 vif_stats = &vif->queues[queue_index].stats;
391 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
392 }
393 data[i] = accum;
394 }
f942dc25
IC
395}
396
397static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
398{
399 int i;
400
401 switch (stringset) {
402 case ETH_SS_STATS:
403 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
404 memcpy(data + i * ETH_GSTRING_LEN,
405 xenvif_stats[i].name, ETH_GSTRING_LEN);
406 break;
407 }
408}
409
813abbba 410static const struct ethtool_ops xenvif_ethtool_ops = {
f942dc25
IC
411 .get_link = ethtool_op_get_link,
412
413 .get_sset_count = xenvif_get_sset_count,
414 .get_ethtool_stats = xenvif_get_ethtool_stats,
415 .get_strings = xenvif_get_strings,
416};
417
813abbba 418static const struct net_device_ops xenvif_netdev_ops = {
f942dc25
IC
419 .ndo_start_xmit = xenvif_start_xmit,
420 .ndo_get_stats = xenvif_get_stats,
421 .ndo_open = xenvif_open,
422 .ndo_stop = xenvif_close,
423 .ndo_change_mtu = xenvif_change_mtu,
47103041 424 .ndo_fix_features = xenvif_fix_features,
4a633a60
MW
425 .ndo_set_mac_address = eth_mac_addr,
426 .ndo_validate_addr = eth_validate_addr,
e9ce7cb6 427 .ndo_select_queue = xenvif_select_queue,
f942dc25
IC
428};
429
430struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
431 unsigned int handle)
432{
433 int err;
434 struct net_device *dev;
435 struct xenvif *vif;
436 char name[IFNAMSIZ] = {};
437
438 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
e9ce7cb6 439 dev = alloc_netdev_mq(sizeof(struct xenvif), name, ether_setup, 1);
f942dc25 440 if (dev == NULL) {
b3f980bd 441 pr_warn("Could not allocate netdev for %s\n", name);
f942dc25
IC
442 return ERR_PTR(-ENOMEM);
443 }
444
445 SET_NETDEV_DEV(dev, parent);
446
447 vif = netdev_priv(dev);
ac3d5ac2 448
f942dc25
IC
449 vif->domid = domid;
450 vif->handle = handle;
f942dc25 451 vif->can_sg = 1;
146c8a77 452 vif->ip_csum = 1;
f942dc25 453 vif->dev = dev;
e9d8b2c2
WL
454 vif->disabled = false;
455
e9ce7cb6
WL
456 /* Start out with no queues. The call below does not require
457 * rtnl_lock() as it happens before register_netdev().
458 */
459 vif->queues = NULL;
460 netif_set_real_num_tx_queues(dev, 0);
09350788 461
f942dc25 462 dev->netdev_ops = &xenvif_netdev_ops;
146c8a77
PD
463 dev->hw_features = NETIF_F_SG |
464 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
82cada22 465 NETIF_F_TSO | NETIF_F_TSO6;
7365bcfa 466 dev->features = dev->hw_features | NETIF_F_RXCSUM;
7ad24ea4 467 dev->ethtool_ops = &xenvif_ethtool_ops;
f942dc25
IC
468
469 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
470
471 /*
472 * Initialise a dummy MAC address. We choose the numerically
473 * largest non-broadcast address to prevent the address getting
474 * stolen by an Ethernet bridge for STP purposes.
475 * (FE:FF:FF:FF:FF:FF)
476 */
477 memset(dev->dev_addr, 0xFF, ETH_ALEN);
478 dev->dev_addr[0] &= ~0x01;
479
480 netif_carrier_off(dev);
481
482 err = register_netdev(dev);
483 if (err) {
484 netdev_warn(dev, "Could not register device: err=%d\n", err);
485 free_netdev(dev);
486 return ERR_PTR(err);
487 }
488
489 netdev_dbg(dev, "Successfully created xenvif\n");
279f438e
PD
490
491 __module_get(THIS_MODULE);
492
f942dc25
IC
493 return vif;
494}
495
e9ce7cb6
WL
496int xenvif_init_queue(struct xenvif_queue *queue)
497{
498 int err, i;
499
500 queue->credit_bytes = queue->remaining_credit = ~0UL;
501 queue->credit_usec = 0UL;
502 init_timer(&queue->credit_timeout);
503 queue->credit_window_start = get_jiffies_64();
504
505 skb_queue_head_init(&queue->rx_queue);
506 skb_queue_head_init(&queue->tx_queue);
507
508 queue->pending_cons = 0;
509 queue->pending_prod = MAX_PENDING_REQS;
510 for (i = 0; i < MAX_PENDING_REQS; ++i)
511 queue->pending_ring[i] = i;
512
513 spin_lock_init(&queue->callback_lock);
514 spin_lock_init(&queue->response_lock);
515
516 /* If ballooning is disabled, this will consume real memory, so you
517 * better enable it. The long term solution would be to use just a
518 * bunch of valid page descriptors, without dependency on ballooning
519 */
520 err = alloc_xenballooned_pages(MAX_PENDING_REQS,
521 queue->mmap_pages,
522 false);
523 if (err) {
524 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
525 return -ENOMEM;
526 }
527
528 for (i = 0; i < MAX_PENDING_REQS; i++) {
529 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
530 { .callback = xenvif_zerocopy_callback,
531 .ctx = NULL,
532 .desc = i };
533 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
534 }
535
536 init_timer(&queue->wake_queue);
537
538 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
539 XENVIF_NAPI_WEIGHT);
540
541 return 0;
542}
543
544void xenvif_carrier_on(struct xenvif *vif)
545{
546 rtnl_lock();
547 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
548 dev_set_mtu(vif->dev, ETH_DATA_LEN);
549 netdev_update_features(vif->dev);
550 netif_carrier_on(vif->dev);
551 if (netif_running(vif->dev))
552 xenvif_up(vif);
553 rtnl_unlock();
554}
555
556int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
e1f00a69
WL
557 unsigned long rx_ring_ref, unsigned int tx_evtchn,
558 unsigned int rx_evtchn)
f942dc25 559{
67fa3660 560 struct task_struct *task;
f942dc25
IC
561 int err = -ENOMEM;
562
e9ce7cb6
WL
563 BUG_ON(queue->tx_irq);
564 BUG_ON(queue->task);
565 BUG_ON(queue->dealloc_task);
f942dc25 566
e9ce7cb6 567 err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
f942dc25
IC
568 if (err < 0)
569 goto err;
570
e9ce7cb6
WL
571 init_waitqueue_head(&queue->wq);
572 init_waitqueue_head(&queue->dealloc_wq);
ca2f09f2 573
e1f00a69
WL
574 if (tx_evtchn == rx_evtchn) {
575 /* feature-split-event-channels == 0 */
576 err = bind_interdomain_evtchn_to_irqhandler(
e9ce7cb6
WL
577 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
578 queue->name, queue);
e1f00a69
WL
579 if (err < 0)
580 goto err_unmap;
e9ce7cb6
WL
581 queue->tx_irq = queue->rx_irq = err;
582 disable_irq(queue->tx_irq);
e1f00a69
WL
583 } else {
584 /* feature-split-event-channels == 1 */
e9ce7cb6
WL
585 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
586 "%s-tx", queue->name);
e1f00a69 587 err = bind_interdomain_evtchn_to_irqhandler(
e9ce7cb6
WL
588 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
589 queue->tx_irq_name, queue);
e1f00a69
WL
590 if (err < 0)
591 goto err_unmap;
e9ce7cb6
WL
592 queue->tx_irq = err;
593 disable_irq(queue->tx_irq);
e1f00a69 594
e9ce7cb6
WL
595 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
596 "%s-rx", queue->name);
e1f00a69 597 err = bind_interdomain_evtchn_to_irqhandler(
e9ce7cb6
WL
598 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
599 queue->rx_irq_name, queue);
e1f00a69
WL
600 if (err < 0)
601 goto err_tx_unbind;
e9ce7cb6
WL
602 queue->rx_irq = err;
603 disable_irq(queue->rx_irq);
e1f00a69 604 }
f942dc25 605
121fa4b7 606 task = kthread_create(xenvif_kthread_guest_rx,
e9ce7cb6 607 (void *)queue, "%s-guest-rx", queue->name);
67fa3660 608 if (IS_ERR(task)) {
e9ce7cb6 609 pr_warn("Could not allocate kthread for %s\n", queue->name);
67fa3660 610 err = PTR_ERR(task);
b3f980bd
WL
611 goto err_rx_unbind;
612 }
e9ce7cb6 613 queue->task = task;
67fa3660 614
f53c3fe8 615 task = kthread_create(xenvif_dealloc_kthread,
e9ce7cb6 616 (void *)queue, "%s-dealloc", queue->name);
f53c3fe8 617 if (IS_ERR(task)) {
e9ce7cb6 618 pr_warn("Could not allocate kthread for %s\n", queue->name);
f53c3fe8
ZK
619 err = PTR_ERR(task);
620 goto err_rx_unbind;
621 }
e9ce7cb6 622 queue->dealloc_task = task;
f53c3fe8 623
e9ce7cb6
WL
624 wake_up_process(queue->task);
625 wake_up_process(queue->dealloc_task);
b3f980bd 626
f942dc25 627 return 0;
b3f980bd
WL
628
629err_rx_unbind:
e9ce7cb6
WL
630 unbind_from_irqhandler(queue->rx_irq, queue);
631 queue->rx_irq = 0;
e1f00a69 632err_tx_unbind:
e9ce7cb6
WL
633 unbind_from_irqhandler(queue->tx_irq, queue);
634 queue->tx_irq = 0;
f942dc25 635err_unmap:
e9ce7cb6 636 xenvif_unmap_frontend_rings(queue);
f942dc25 637err:
b103f358 638 module_put(THIS_MODULE);
f942dc25
IC
639 return err;
640}
641
48856286 642void xenvif_carrier_off(struct xenvif *vif)
f942dc25
IC
643{
644 struct net_device *dev = vif->dev;
48856286
IC
645
646 rtnl_lock();
647 netif_carrier_off(dev); /* discard queued packets */
648 if (netif_running(dev))
649 xenvif_down(vif);
650 rtnl_unlock();
48856286
IC
651}
652
e9ce7cb6
WL
653static void xenvif_wait_unmap_timeout(struct xenvif_queue *queue,
654 unsigned int worst_case_skb_lifetime)
655{
656 int i, unmap_timeout = 0;
657
658 for (i = 0; i < MAX_PENDING_REQS; ++i) {
659 if (queue->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
660 unmap_timeout++;
661 schedule_timeout(msecs_to_jiffies(1000));
662 if (unmap_timeout > worst_case_skb_lifetime &&
663 net_ratelimit())
664 netdev_err(queue->vif->dev,
665 "Page still granted! Index: %x\n",
666 i);
667 i = -1;
668 }
669 }
670}
671
48856286
IC
672void xenvif_disconnect(struct xenvif *vif)
673{
e9ce7cb6
WL
674 struct xenvif_queue *queue = NULL;
675 unsigned int num_queues = vif->dev->real_num_tx_queues;
676 unsigned int queue_index;
677
48856286
IC
678 if (netif_carrier_ok(vif->dev))
679 xenvif_carrier_off(vif);
f942dc25 680
e9ce7cb6
WL
681 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
682 queue = &vif->queues[queue_index];
db739ef3 683
e9ce7cb6
WL
684 if (queue->task) {
685 del_timer_sync(&queue->wake_queue);
686 kthread_stop(queue->task);
687 queue->task = NULL;
688 }
f53c3fe8 689
e9ce7cb6
WL
690 if (queue->dealloc_task) {
691 kthread_stop(queue->dealloc_task);
692 queue->dealloc_task = NULL;
693 }
694
695 if (queue->tx_irq) {
696 if (queue->tx_irq == queue->rx_irq)
697 unbind_from_irqhandler(queue->tx_irq, queue);
698 else {
699 unbind_from_irqhandler(queue->tx_irq, queue);
700 unbind_from_irqhandler(queue->rx_irq, queue);
701 }
702 queue->tx_irq = 0;
e1f00a69 703 }
f942dc25 704
e9ce7cb6
WL
705 xenvif_unmap_frontend_rings(queue);
706 }
279f438e
PD
707}
708
709void xenvif_free(struct xenvif *vif)
710{
e9ce7cb6
WL
711 struct xenvif_queue *queue = NULL;
712 unsigned int num_queues = vif->dev->real_num_tx_queues;
713 unsigned int queue_index;
0e59a4a5
ZK
714 /* Here we want to avoid timeout messages if an skb can be legitimately
715 * stuck somewhere else. Realistically this could be an another vif's
09350788
ZK
716 * internal or QDisc queue. That another vif also has this
717 * rx_drain_timeout_msecs timeout, but the timer only ditches the
718 * internal queue. After that, the QDisc queue can put in worst case
719 * XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS skbs into that another vif's
720 * internal queue, so we need several rounds of such timeouts until we
721 * can be sure that no another vif should have skb's from us. We are
0e59a4a5 722 * not sending more skb's, so newly stuck packets are not interesting
09350788
ZK
723 * for us here.
724 */
725 unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000) *
726 DIV_ROUND_UP(XENVIF_QUEUE_LENGTH, (XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS));
f53c3fe8 727
e9ce7cb6 728 unregister_netdev(vif->dev);
f53c3fe8 729
e9ce7cb6
WL
730 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
731 queue = &vif->queues[queue_index];
f53c3fe8 732
e9ce7cb6
WL
733 xenvif_wait_unmap_timeout(queue, worst_case_skb_lifetime);
734 free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages);
b3f980bd 735
e9ce7cb6
WL
736 netif_napi_del(&queue->napi);
737 }
738
739 /* Free the array of queues. The call below does not require
740 * rtnl_lock() because it happens after unregister_netdev().
741 */
742 netif_set_real_num_tx_queues(vif->dev, 0);
743 vfree(vif->queues);
744 vif->queues = NULL;
f942dc25 745
f942dc25 746 free_netdev(vif->dev);
b103f358 747
279f438e 748 module_put(THIS_MODULE);
f942dc25 749}