drm/radeon/kms: Add an MSI quirk for Dell RS690
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sched / sch_generic.c
blob69fca2798804f29f5cd3cbdadd139fdc6801e9d5
1 /*
2 * net/sched/sch_generic.c Generic packet scheduler routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * Jamal Hadi Salim, <hadi@cyberus.ca> 990601
11 * - Ingress support
14 #include <linux/bitops.h>
15 #include <linux/module.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/sched.h>
19 #include <linux/string.h>
20 #include <linux/errno.h>
21 #include <linux/netdevice.h>
22 #include <linux/skbuff.h>
23 #include <linux/rtnetlink.h>
24 #include <linux/init.h>
25 #include <linux/rcupdate.h>
26 #include <linux/list.h>
27 #include <linux/slab.h>
28 #include <net/pkt_sched.h>
29 #include <net/dst.h>
31 /* Main transmission queue. */
33 /* Modifications to data participating in scheduling must be protected with
34 * qdisc_lock(qdisc) spinlock.
36 * The idea is the following:
37 * - enqueue, dequeue are serialized via qdisc root lock
38 * - ingress filtering is also serialized via qdisc root lock
39 * - updates to tree and tree walking are only done under the rtnl mutex.
42 static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
44 skb_dst_force(skb);
45 q->gso_skb = skb;
46 q->qstats.requeues++;
47 q->q.qlen++; /* it's still part of the queue */
48 __netif_schedule(q);
50 return 0;
53 static inline struct sk_buff *dequeue_skb(struct Qdisc *q)
55 struct sk_buff *skb = q->gso_skb;
57 if (unlikely(skb)) {
58 struct net_device *dev = qdisc_dev(q);
59 struct netdev_queue *txq;
61 /* check the reason of requeuing without tx lock first */
62 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
63 if (!netif_tx_queue_frozen_or_stopped(txq)) {
64 q->gso_skb = NULL;
65 q->q.qlen--;
66 } else
67 skb = NULL;
68 } else {
69 skb = q->dequeue(q);
72 return skb;
75 static inline int handle_dev_cpu_collision(struct sk_buff *skb,
76 struct netdev_queue *dev_queue,
77 struct Qdisc *q)
79 int ret;
81 if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) {
83 * Same CPU holding the lock. It may be a transient
84 * configuration error, when hard_start_xmit() recurses. We
85 * detect it by checking xmit owner and drop the packet when
86 * deadloop is detected. Return OK to try the next skb.
88 kfree_skb(skb);
89 if (net_ratelimit())
90 pr_warning("Dead loop on netdevice %s, fix it urgently!\n",
91 dev_queue->dev->name);
92 ret = qdisc_qlen(q);
93 } else {
95 * Another cpu is holding lock, requeue & delay xmits for
96 * some time.
98 __this_cpu_inc(softnet_data.cpu_collision);
99 ret = dev_requeue_skb(skb, q);
102 return ret;
106 * Transmit one skb, and handle the return status as required. Holding the
107 * __QDISC_STATE_RUNNING bit guarantees that only one CPU can execute this
108 * function.
110 * Returns to the caller:
111 * 0 - queue is empty or throttled.
112 * >0 - queue is not empty.
114 int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
115 struct net_device *dev, struct netdev_queue *txq,
116 spinlock_t *root_lock)
118 int ret = NETDEV_TX_BUSY;
120 /* And release qdisc */
121 spin_unlock(root_lock);
123 HARD_TX_LOCK(dev, txq, smp_processor_id());
124 if (!netif_tx_queue_frozen_or_stopped(txq))
125 ret = dev_hard_start_xmit(skb, dev, txq);
127 HARD_TX_UNLOCK(dev, txq);
129 spin_lock(root_lock);
131 if (dev_xmit_complete(ret)) {
132 /* Driver sent out skb successfully or skb was consumed */
133 ret = qdisc_qlen(q);
134 } else if (ret == NETDEV_TX_LOCKED) {
135 /* Driver try lock failed */
136 ret = handle_dev_cpu_collision(skb, txq, q);
137 } else {
138 /* Driver returned NETDEV_TX_BUSY - requeue skb */
139 if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit()))
140 pr_warning("BUG %s code %d qlen %d\n",
141 dev->name, ret, q->q.qlen);
143 ret = dev_requeue_skb(skb, q);
146 if (ret && netif_tx_queue_frozen_or_stopped(txq))
147 ret = 0;
149 return ret;
153 * NOTE: Called under qdisc_lock(q) with locally disabled BH.
155 * __QDISC_STATE_RUNNING guarantees only one CPU can process
156 * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
157 * this queue.
159 * netif_tx_lock serializes accesses to device driver.
161 * qdisc_lock(q) and netif_tx_lock are mutually exclusive,
162 * if one is grabbed, another must be free.
164 * Note, that this procedure can be called by a watchdog timer
166 * Returns to the caller:
167 * 0 - queue is empty or throttled.
168 * >0 - queue is not empty.
171 static inline int qdisc_restart(struct Qdisc *q)
173 struct netdev_queue *txq;
174 struct net_device *dev;
175 spinlock_t *root_lock;
176 struct sk_buff *skb;
178 /* Dequeue packet */
179 skb = dequeue_skb(q);
180 if (unlikely(!skb))
181 return 0;
182 WARN_ON_ONCE(skb_dst_is_noref(skb));
183 root_lock = qdisc_lock(q);
184 dev = qdisc_dev(q);
185 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
187 return sch_direct_xmit(skb, q, dev, txq, root_lock);
190 void __qdisc_run(struct Qdisc *q)
192 int quota = weight_p;
194 while (qdisc_restart(q)) {
196 * Ordered by possible occurrence: Postpone processing if
197 * 1. we've exceeded packet quota
198 * 2. another process needs the CPU;
200 if (--quota <= 0 || need_resched()) {
201 __netif_schedule(q);
202 break;
206 qdisc_run_end(q);
209 unsigned long dev_trans_start(struct net_device *dev)
211 unsigned long val, res = dev->trans_start;
212 unsigned int i;
214 for (i = 0; i < dev->num_tx_queues; i++) {
215 val = netdev_get_tx_queue(dev, i)->trans_start;
216 if (val && time_after(val, res))
217 res = val;
219 dev->trans_start = res;
220 return res;
222 EXPORT_SYMBOL(dev_trans_start);
224 static void dev_watchdog(unsigned long arg)
226 struct net_device *dev = (struct net_device *)arg;
228 netif_tx_lock(dev);
229 if (!qdisc_tx_is_noop(dev)) {
230 if (netif_device_present(dev) &&
231 netif_running(dev) &&
232 netif_carrier_ok(dev)) {
233 int some_queue_timedout = 0;
234 unsigned int i;
235 unsigned long trans_start;
237 for (i = 0; i < dev->num_tx_queues; i++) {
238 struct netdev_queue *txq;
240 txq = netdev_get_tx_queue(dev, i);
242 * old device drivers set dev->trans_start
244 trans_start = txq->trans_start ? : dev->trans_start;
245 if (netif_tx_queue_stopped(txq) &&
246 time_after(jiffies, (trans_start +
247 dev->watchdog_timeo))) {
248 some_queue_timedout = 1;
249 break;
253 if (some_queue_timedout) {
254 WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
255 dev->name, netdev_drivername(dev), i);
256 dev->netdev_ops->ndo_tx_timeout(dev);
258 if (!mod_timer(&dev->watchdog_timer,
259 round_jiffies(jiffies +
260 dev->watchdog_timeo)))
261 dev_hold(dev);
264 netif_tx_unlock(dev);
266 dev_put(dev);
269 void __netdev_watchdog_up(struct net_device *dev)
271 if (dev->netdev_ops->ndo_tx_timeout) {
272 if (dev->watchdog_timeo <= 0)
273 dev->watchdog_timeo = 5*HZ;
274 if (!mod_timer(&dev->watchdog_timer,
275 round_jiffies(jiffies + dev->watchdog_timeo)))
276 dev_hold(dev);
280 static void dev_watchdog_up(struct net_device *dev)
282 __netdev_watchdog_up(dev);
285 static void dev_watchdog_down(struct net_device *dev)
287 netif_tx_lock_bh(dev);
288 if (del_timer(&dev->watchdog_timer))
289 dev_put(dev);
290 netif_tx_unlock_bh(dev);
294 * netif_carrier_on - set carrier
295 * @dev: network device
297 * Device has detected that carrier.
299 void netif_carrier_on(struct net_device *dev)
301 if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
302 if (dev->reg_state == NETREG_UNINITIALIZED)
303 return;
304 linkwatch_fire_event(dev);
305 if (netif_running(dev))
306 __netdev_watchdog_up(dev);
309 EXPORT_SYMBOL(netif_carrier_on);
312 * netif_carrier_off - clear carrier
313 * @dev: network device
315 * Device has detected loss of carrier.
317 void netif_carrier_off(struct net_device *dev)
319 if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
320 if (dev->reg_state == NETREG_UNINITIALIZED)
321 return;
322 linkwatch_fire_event(dev);
325 EXPORT_SYMBOL(netif_carrier_off);
328 * netif_notify_peers - notify network peers about existence of @dev
329 * @dev: network device
331 * Generate traffic such that interested network peers are aware of
332 * @dev, such as by generating a gratuitous ARP. This may be used when
333 * a device wants to inform the rest of the network about some sort of
334 * reconfiguration such as a failover event or virtual machine
335 * migration.
337 void netif_notify_peers(struct net_device *dev)
339 rtnl_lock();
340 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
341 rtnl_unlock();
343 EXPORT_SYMBOL(netif_notify_peers);
345 /* "NOOP" scheduler: the best scheduler, recommended for all interfaces
346 under all circumstances. It is difficult to invent anything faster or
347 cheaper.
350 static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc)
352 kfree_skb(skb);
353 return NET_XMIT_CN;
356 static struct sk_buff *noop_dequeue(struct Qdisc * qdisc)
358 return NULL;
361 struct Qdisc_ops noop_qdisc_ops __read_mostly = {
362 .id = "noop",
363 .priv_size = 0,
364 .enqueue = noop_enqueue,
365 .dequeue = noop_dequeue,
366 .peek = noop_dequeue,
367 .owner = THIS_MODULE,
370 static struct netdev_queue noop_netdev_queue = {
371 .qdisc = &noop_qdisc,
372 .qdisc_sleeping = &noop_qdisc,
375 struct Qdisc noop_qdisc = {
376 .enqueue = noop_enqueue,
377 .dequeue = noop_dequeue,
378 .flags = TCQ_F_BUILTIN,
379 .ops = &noop_qdisc_ops,
380 .list = LIST_HEAD_INIT(noop_qdisc.list),
381 .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
382 .dev_queue = &noop_netdev_queue,
383 .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
385 EXPORT_SYMBOL(noop_qdisc);
387 static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
388 .id = "noqueue",
389 .priv_size = 0,
390 .enqueue = noop_enqueue,
391 .dequeue = noop_dequeue,
392 .peek = noop_dequeue,
393 .owner = THIS_MODULE,
396 static struct Qdisc noqueue_qdisc;
397 static struct netdev_queue noqueue_netdev_queue = {
398 .qdisc = &noqueue_qdisc,
399 .qdisc_sleeping = &noqueue_qdisc,
402 static struct Qdisc noqueue_qdisc = {
403 .enqueue = NULL,
404 .dequeue = noop_dequeue,
405 .flags = TCQ_F_BUILTIN,
406 .ops = &noqueue_qdisc_ops,
407 .list = LIST_HEAD_INIT(noqueue_qdisc.list),
408 .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock),
409 .dev_queue = &noqueue_netdev_queue,
410 .busylock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock),
414 static const u8 prio2band[TC_PRIO_MAX + 1] = {
415 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
418 /* 3-band FIFO queue: old style, but should be a bit faster than
419 generic prio+fifo combination.
422 #define PFIFO_FAST_BANDS 3
425 * Private data for a pfifo_fast scheduler containing:
426 * - queues for the three band
427 * - bitmap indicating which of the bands contain skbs
429 struct pfifo_fast_priv {
430 u32 bitmap;
431 struct sk_buff_head q[PFIFO_FAST_BANDS];
435 * Convert a bitmap to the first band number where an skb is queued, where:
436 * bitmap=0 means there are no skbs on any band.
437 * bitmap=1 means there is an skb on band 0.
438 * bitmap=7 means there are skbs on all 3 bands, etc.
440 static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0};
442 static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv,
443 int band)
445 return priv->q + band;
448 static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc)
450 if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) {
451 int band = prio2band[skb->priority & TC_PRIO_MAX];
452 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
453 struct sk_buff_head *list = band2list(priv, band);
455 priv->bitmap |= (1 << band);
456 qdisc->q.qlen++;
457 return __qdisc_enqueue_tail(skb, qdisc, list);
460 return qdisc_drop(skb, qdisc);
463 static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
465 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
466 int band = bitmap2band[priv->bitmap];
468 if (likely(band >= 0)) {
469 struct sk_buff_head *list = band2list(priv, band);
470 struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list);
472 qdisc->q.qlen--;
473 if (skb_queue_empty(list))
474 priv->bitmap &= ~(1 << band);
476 return skb;
479 return NULL;
482 static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc)
484 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
485 int band = bitmap2band[priv->bitmap];
487 if (band >= 0) {
488 struct sk_buff_head *list = band2list(priv, band);
490 return skb_peek(list);
493 return NULL;
496 static void pfifo_fast_reset(struct Qdisc *qdisc)
498 int prio;
499 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
501 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
502 __qdisc_reset_queue(qdisc, band2list(priv, prio));
504 priv->bitmap = 0;
505 qdisc->qstats.backlog = 0;
506 qdisc->q.qlen = 0;
509 static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
511 struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
513 memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1);
514 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
515 return skb->len;
517 nla_put_failure:
518 return -1;
521 static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
523 int prio;
524 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
526 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
527 skb_queue_head_init(band2list(priv, prio));
529 /* Can by-pass the queue discipline */
530 qdisc->flags |= TCQ_F_CAN_BYPASS;
531 return 0;
534 struct Qdisc_ops pfifo_fast_ops __read_mostly = {
535 .id = "pfifo_fast",
536 .priv_size = sizeof(struct pfifo_fast_priv),
537 .enqueue = pfifo_fast_enqueue,
538 .dequeue = pfifo_fast_dequeue,
539 .peek = pfifo_fast_peek,
540 .init = pfifo_fast_init,
541 .reset = pfifo_fast_reset,
542 .dump = pfifo_fast_dump,
543 .owner = THIS_MODULE,
545 EXPORT_SYMBOL(pfifo_fast_ops);
547 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
548 struct Qdisc_ops *ops)
550 void *p;
551 struct Qdisc *sch;
552 unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size;
553 int err = -ENOBUFS;
555 p = kzalloc_node(size, GFP_KERNEL,
556 netdev_queue_numa_node_read(dev_queue));
558 if (!p)
559 goto errout;
560 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
561 /* if we got non aligned memory, ask more and do alignment ourself */
562 if (sch != p) {
563 kfree(p);
564 p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL,
565 netdev_queue_numa_node_read(dev_queue));
566 if (!p)
567 goto errout;
568 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
569 sch->padded = (char *) sch - (char *) p;
571 INIT_LIST_HEAD(&sch->list);
572 skb_queue_head_init(&sch->q);
573 spin_lock_init(&sch->busylock);
574 sch->ops = ops;
575 sch->enqueue = ops->enqueue;
576 sch->dequeue = ops->dequeue;
577 sch->dev_queue = dev_queue;
578 dev_hold(qdisc_dev(sch));
579 atomic_set(&sch->refcnt, 1);
581 return sch;
582 errout:
583 return ERR_PTR(err);
586 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
587 struct Qdisc_ops *ops, unsigned int parentid)
589 struct Qdisc *sch;
591 sch = qdisc_alloc(dev_queue, ops);
592 if (IS_ERR(sch))
593 goto errout;
594 sch->parent = parentid;
596 if (!ops->init || ops->init(sch, NULL) == 0)
597 return sch;
599 qdisc_destroy(sch);
600 errout:
601 return NULL;
603 EXPORT_SYMBOL(qdisc_create_dflt);
605 /* Under qdisc_lock(qdisc) and BH! */
607 void qdisc_reset(struct Qdisc *qdisc)
609 const struct Qdisc_ops *ops = qdisc->ops;
611 if (ops->reset)
612 ops->reset(qdisc);
614 if (qdisc->gso_skb) {
615 kfree_skb(qdisc->gso_skb);
616 qdisc->gso_skb = NULL;
617 qdisc->q.qlen = 0;
620 EXPORT_SYMBOL(qdisc_reset);
622 static void qdisc_rcu_free(struct rcu_head *head)
624 struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head);
626 kfree((char *) qdisc - qdisc->padded);
629 void qdisc_destroy(struct Qdisc *qdisc)
631 const struct Qdisc_ops *ops = qdisc->ops;
633 if (qdisc->flags & TCQ_F_BUILTIN ||
634 !atomic_dec_and_test(&qdisc->refcnt))
635 return;
637 #ifdef CONFIG_NET_SCHED
638 qdisc_list_del(qdisc);
640 qdisc_put_stab(rtnl_dereference(qdisc->stab));
641 #endif
642 gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
643 if (ops->reset)
644 ops->reset(qdisc);
645 if (ops->destroy)
646 ops->destroy(qdisc);
648 module_put(ops->owner);
649 dev_put(qdisc_dev(qdisc));
651 kfree_skb(qdisc->gso_skb);
653 * gen_estimator est_timer() might access qdisc->q.lock,
654 * wait a RCU grace period before freeing qdisc.
656 call_rcu(&qdisc->rcu_head, qdisc_rcu_free);
658 EXPORT_SYMBOL(qdisc_destroy);
660 /* Attach toplevel qdisc to device queue. */
661 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
662 struct Qdisc *qdisc)
664 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
665 spinlock_t *root_lock;
667 root_lock = qdisc_lock(oqdisc);
668 spin_lock_bh(root_lock);
670 /* Prune old scheduler */
671 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
672 qdisc_reset(oqdisc);
674 /* ... and graft new one */
675 if (qdisc == NULL)
676 qdisc = &noop_qdisc;
677 dev_queue->qdisc_sleeping = qdisc;
678 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
680 spin_unlock_bh(root_lock);
682 return oqdisc;
684 EXPORT_SYMBOL(dev_graft_qdisc);
686 static void attach_one_default_qdisc(struct net_device *dev,
687 struct netdev_queue *dev_queue,
688 void *_unused)
690 struct Qdisc *qdisc = &noqueue_qdisc;
692 if (dev->tx_queue_len) {
693 qdisc = qdisc_create_dflt(dev_queue,
694 &pfifo_fast_ops, TC_H_ROOT);
695 if (!qdisc) {
696 netdev_info(dev, "activation failed\n");
697 return;
700 dev_queue->qdisc_sleeping = qdisc;
703 static void attach_default_qdiscs(struct net_device *dev)
705 struct netdev_queue *txq;
706 struct Qdisc *qdisc;
708 txq = netdev_get_tx_queue(dev, 0);
710 if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) {
711 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
712 dev->qdisc = txq->qdisc_sleeping;
713 atomic_inc(&dev->qdisc->refcnt);
714 } else {
715 qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT);
716 if (qdisc) {
717 qdisc->ops->attach(qdisc);
718 dev->qdisc = qdisc;
723 static void transition_one_qdisc(struct net_device *dev,
724 struct netdev_queue *dev_queue,
725 void *_need_watchdog)
727 struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
728 int *need_watchdog_p = _need_watchdog;
730 if (!(new_qdisc->flags & TCQ_F_BUILTIN))
731 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
733 rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
734 if (need_watchdog_p && new_qdisc != &noqueue_qdisc) {
735 dev_queue->trans_start = 0;
736 *need_watchdog_p = 1;
740 void dev_activate(struct net_device *dev)
742 int need_watchdog;
744 /* No queueing discipline is attached to device;
745 create default one i.e. pfifo_fast for devices,
746 which need queueing and noqueue_qdisc for
747 virtual interfaces
750 if (dev->qdisc == &noop_qdisc)
751 attach_default_qdiscs(dev);
753 if (!netif_carrier_ok(dev))
754 /* Delay activation until next carrier-on event */
755 return;
757 need_watchdog = 0;
758 netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
759 if (dev_ingress_queue(dev))
760 transition_one_qdisc(dev, dev_ingress_queue(dev), NULL);
762 if (need_watchdog) {
763 dev->trans_start = jiffies;
764 dev_watchdog_up(dev);
767 EXPORT_SYMBOL(dev_activate);
769 static void dev_deactivate_queue(struct net_device *dev,
770 struct netdev_queue *dev_queue,
771 void *_qdisc_default)
773 struct Qdisc *qdisc_default = _qdisc_default;
774 struct Qdisc *qdisc;
776 qdisc = dev_queue->qdisc;
777 if (qdisc) {
778 spin_lock_bh(qdisc_lock(qdisc));
780 if (!(qdisc->flags & TCQ_F_BUILTIN))
781 set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
783 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
784 qdisc_reset(qdisc);
786 spin_unlock_bh(qdisc_lock(qdisc));
790 static bool some_qdisc_is_busy(struct net_device *dev)
792 unsigned int i;
794 for (i = 0; i < dev->num_tx_queues; i++) {
795 struct netdev_queue *dev_queue;
796 spinlock_t *root_lock;
797 struct Qdisc *q;
798 int val;
800 dev_queue = netdev_get_tx_queue(dev, i);
801 q = dev_queue->qdisc_sleeping;
802 root_lock = qdisc_lock(q);
804 spin_lock_bh(root_lock);
806 val = (qdisc_is_running(q) ||
807 test_bit(__QDISC_STATE_SCHED, &q->state));
809 spin_unlock_bh(root_lock);
811 if (val)
812 return true;
814 return false;
818 * dev_deactivate_many - deactivate transmissions on several devices
819 * @head: list of devices to deactivate
821 * This function returns only when all outstanding transmissions
822 * have completed, unless all devices are in dismantle phase.
824 void dev_deactivate_many(struct list_head *head)
826 struct net_device *dev;
827 bool sync_needed = false;
829 list_for_each_entry(dev, head, unreg_list) {
830 netdev_for_each_tx_queue(dev, dev_deactivate_queue,
831 &noop_qdisc);
832 if (dev_ingress_queue(dev))
833 dev_deactivate_queue(dev, dev_ingress_queue(dev),
834 &noop_qdisc);
836 dev_watchdog_down(dev);
837 sync_needed |= !dev->dismantle;
840 /* Wait for outstanding qdisc-less dev_queue_xmit calls.
841 * This is avoided if all devices are in dismantle phase :
842 * Caller will call synchronize_net() for us
844 if (sync_needed)
845 synchronize_net();
847 /* Wait for outstanding qdisc_run calls. */
848 list_for_each_entry(dev, head, unreg_list)
849 while (some_qdisc_is_busy(dev))
850 yield();
853 void dev_deactivate(struct net_device *dev)
855 LIST_HEAD(single);
857 list_add(&dev->unreg_list, &single);
858 dev_deactivate_many(&single);
859 list_del(&single);
861 EXPORT_SYMBOL(dev_deactivate);
863 static void dev_init_scheduler_queue(struct net_device *dev,
864 struct netdev_queue *dev_queue,
865 void *_qdisc)
867 struct Qdisc *qdisc = _qdisc;
869 dev_queue->qdisc = qdisc;
870 dev_queue->qdisc_sleeping = qdisc;
873 void dev_init_scheduler(struct net_device *dev)
875 dev->qdisc = &noop_qdisc;
876 netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
877 if (dev_ingress_queue(dev))
878 dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
880 setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
883 static void shutdown_scheduler_queue(struct net_device *dev,
884 struct netdev_queue *dev_queue,
885 void *_qdisc_default)
887 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
888 struct Qdisc *qdisc_default = _qdisc_default;
890 if (qdisc) {
891 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
892 dev_queue->qdisc_sleeping = qdisc_default;
894 qdisc_destroy(qdisc);
898 void dev_shutdown(struct net_device *dev)
900 netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
901 if (dev_ingress_queue(dev))
902 shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
903 qdisc_destroy(dev->qdisc);
904 dev->qdisc = &noop_qdisc;
906 WARN_ON(timer_pending(&dev->watchdog_timer));