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
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 <linux/if_vlan.h>
29 #include <net/sch_generic.h>
30 #include <net/pkt_sched.h>
33 /* Qdisc to use by default */
34 const struct Qdisc_ops
*default_qdisc_ops
= &pfifo_fast_ops
;
35 EXPORT_SYMBOL(default_qdisc_ops
);
37 /* Main transmission queue. */
39 /* Modifications to data participating in scheduling must be protected with
40 * qdisc_lock(qdisc) spinlock.
42 * The idea is the following:
43 * - enqueue, dequeue are serialized via qdisc root lock
44 * - ingress filtering is also serialized via qdisc root lock
45 * - updates to tree and tree walking are only done under the rtnl mutex.
48 static inline int dev_requeue_skb(struct sk_buff
*skb
, struct Qdisc
*q
)
53 q
->q
.qlen
++; /* it's still part of the queue */
59 static inline struct sk_buff
*dequeue_skb(struct Qdisc
*q
)
61 struct sk_buff
*skb
= q
->gso_skb
;
62 const struct netdev_queue
*txq
= q
->dev_queue
;
65 /* check the reason of requeuing without tx lock first */
66 txq
= skb_get_tx_queue(txq
->dev
, skb
);
67 if (!netif_xmit_frozen_or_stopped(txq
)) {
73 if (!(q
->flags
& TCQ_F_ONETXQUEUE
) || !netif_xmit_frozen_or_stopped(txq
))
80 static inline int handle_dev_cpu_collision(struct sk_buff
*skb
,
81 struct netdev_queue
*dev_queue
,
86 if (unlikely(dev_queue
->xmit_lock_owner
== smp_processor_id())) {
88 * Same CPU holding the lock. It may be a transient
89 * configuration error, when hard_start_xmit() recurses. We
90 * detect it by checking xmit owner and drop the packet when
91 * deadloop is detected. Return OK to try the next skb.
94 net_warn_ratelimited("Dead loop on netdevice %s, fix it urgently!\n",
95 dev_queue
->dev
->name
);
99 * Another cpu is holding lock, requeue & delay xmits for
102 __this_cpu_inc(softnet_data
.cpu_collision
);
103 ret
= dev_requeue_skb(skb
, q
);
110 * Transmit one skb, and handle the return status as required. Holding the
111 * __QDISC___STATE_RUNNING bit guarantees that only one CPU can execute this
114 * Returns to the caller:
115 * 0 - queue is empty or throttled.
116 * >0 - queue is not empty.
118 int sch_direct_xmit(struct sk_buff
*skb
, struct Qdisc
*q
,
119 struct net_device
*dev
, struct netdev_queue
*txq
,
120 spinlock_t
*root_lock
)
122 int ret
= NETDEV_TX_BUSY
;
124 /* And release qdisc */
125 spin_unlock(root_lock
);
127 HARD_TX_LOCK(dev
, txq
, smp_processor_id());
128 if (!netif_xmit_frozen_or_stopped(txq
))
129 ret
= dev_hard_start_xmit(skb
, dev
, txq
);
131 HARD_TX_UNLOCK(dev
, txq
);
133 spin_lock(root_lock
);
135 if (dev_xmit_complete(ret
)) {
136 /* Driver sent out skb successfully or skb was consumed */
138 } else if (ret
== NETDEV_TX_LOCKED
) {
139 /* Driver try lock failed */
140 ret
= handle_dev_cpu_collision(skb
, txq
, q
);
142 /* Driver returned NETDEV_TX_BUSY - requeue skb */
143 if (unlikely(ret
!= NETDEV_TX_BUSY
))
144 net_warn_ratelimited("BUG %s code %d qlen %d\n",
145 dev
->name
, ret
, q
->q
.qlen
);
147 ret
= dev_requeue_skb(skb
, q
);
150 if (ret
&& netif_xmit_frozen_or_stopped(txq
))
157 * NOTE: Called under qdisc_lock(q) with locally disabled BH.
159 * __QDISC___STATE_RUNNING guarantees only one CPU can process
160 * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
163 * netif_tx_lock serializes accesses to device driver.
165 * qdisc_lock(q) and netif_tx_lock are mutually exclusive,
166 * if one is grabbed, another must be free.
168 * Note, that this procedure can be called by a watchdog timer
170 * Returns to the caller:
171 * 0 - queue is empty or throttled.
172 * >0 - queue is not empty.
175 static inline int qdisc_restart(struct Qdisc
*q
)
177 struct netdev_queue
*txq
;
178 struct net_device
*dev
;
179 spinlock_t
*root_lock
;
183 skb
= dequeue_skb(q
);
187 WARN_ON_ONCE(skb_dst_is_noref(skb
));
189 root_lock
= qdisc_lock(q
);
191 txq
= skb_get_tx_queue(dev
, skb
);
193 return sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
);
196 void __qdisc_run(struct Qdisc
*q
)
198 int quota
= weight_p
;
200 while (qdisc_restart(q
)) {
202 * Ordered by possible occurrence: Postpone processing if
203 * 1. we've exceeded packet quota
204 * 2. another process needs the CPU;
206 if (--quota
<= 0 || need_resched()) {
215 unsigned long dev_trans_start(struct net_device
*dev
)
217 unsigned long val
, res
;
220 if (is_vlan_dev(dev
))
221 dev
= vlan_dev_real_dev(dev
);
222 res
= dev
->trans_start
;
223 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
224 val
= netdev_get_tx_queue(dev
, i
)->trans_start
;
225 if (val
&& time_after(val
, res
))
228 dev
->trans_start
= res
;
232 EXPORT_SYMBOL(dev_trans_start
);
234 static void dev_watchdog(unsigned long arg
)
236 struct net_device
*dev
= (struct net_device
*)arg
;
239 if (!qdisc_tx_is_noop(dev
)) {
240 if (netif_device_present(dev
) &&
241 netif_running(dev
) &&
242 netif_carrier_ok(dev
)) {
243 int some_queue_timedout
= 0;
245 unsigned long trans_start
;
247 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
248 struct netdev_queue
*txq
;
250 txq
= netdev_get_tx_queue(dev
, i
);
252 * old device drivers set dev->trans_start
254 trans_start
= txq
->trans_start
? : dev
->trans_start
;
255 if (netif_xmit_stopped(txq
) &&
256 time_after(jiffies
, (trans_start
+
257 dev
->watchdog_timeo
))) {
258 some_queue_timedout
= 1;
259 txq
->trans_timeout
++;
264 if (some_queue_timedout
) {
265 WARN_ONCE(1, KERN_INFO
"NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
266 dev
->name
, netdev_drivername(dev
), i
);
267 dev
->netdev_ops
->ndo_tx_timeout(dev
);
269 if (!mod_timer(&dev
->watchdog_timer
,
270 round_jiffies(jiffies
+
271 dev
->watchdog_timeo
)))
275 netif_tx_unlock(dev
);
280 void __netdev_watchdog_up(struct net_device
*dev
)
282 if (dev
->netdev_ops
->ndo_tx_timeout
) {
283 if (dev
->watchdog_timeo
<= 0)
284 dev
->watchdog_timeo
= 5*HZ
;
285 if (!mod_timer(&dev
->watchdog_timer
,
286 round_jiffies(jiffies
+ dev
->watchdog_timeo
)))
291 static void dev_watchdog_up(struct net_device
*dev
)
293 __netdev_watchdog_up(dev
);
296 static void dev_watchdog_down(struct net_device
*dev
)
298 netif_tx_lock_bh(dev
);
299 if (del_timer(&dev
->watchdog_timer
))
301 netif_tx_unlock_bh(dev
);
305 * netif_carrier_on - set carrier
306 * @dev: network device
308 * Device has detected that carrier.
310 void netif_carrier_on(struct net_device
*dev
)
312 if (test_and_clear_bit(__LINK_STATE_NOCARRIER
, &dev
->state
)) {
313 if (dev
->reg_state
== NETREG_UNINITIALIZED
)
315 atomic_inc(&dev
->carrier_changes
);
316 linkwatch_fire_event(dev
);
317 if (netif_running(dev
))
318 __netdev_watchdog_up(dev
);
321 EXPORT_SYMBOL(netif_carrier_on
);
324 * netif_carrier_off - clear carrier
325 * @dev: network device
327 * Device has detected loss of carrier.
329 void netif_carrier_off(struct net_device
*dev
)
331 if (!test_and_set_bit(__LINK_STATE_NOCARRIER
, &dev
->state
)) {
332 if (dev
->reg_state
== NETREG_UNINITIALIZED
)
334 atomic_inc(&dev
->carrier_changes
);
335 linkwatch_fire_event(dev
);
338 EXPORT_SYMBOL(netif_carrier_off
);
340 /* "NOOP" scheduler: the best scheduler, recommended for all interfaces
341 under all circumstances. It is difficult to invent anything faster or
345 static int noop_enqueue(struct sk_buff
*skb
, struct Qdisc
*qdisc
)
351 static struct sk_buff
*noop_dequeue(struct Qdisc
*qdisc
)
356 struct Qdisc_ops noop_qdisc_ops __read_mostly
= {
359 .enqueue
= noop_enqueue
,
360 .dequeue
= noop_dequeue
,
361 .peek
= noop_dequeue
,
362 .owner
= THIS_MODULE
,
365 static struct netdev_queue noop_netdev_queue
= {
366 .qdisc
= &noop_qdisc
,
367 .qdisc_sleeping
= &noop_qdisc
,
370 struct Qdisc noop_qdisc
= {
371 .enqueue
= noop_enqueue
,
372 .dequeue
= noop_dequeue
,
373 .flags
= TCQ_F_BUILTIN
,
374 .ops
= &noop_qdisc_ops
,
375 .list
= LIST_HEAD_INIT(noop_qdisc
.list
),
376 .q
.lock
= __SPIN_LOCK_UNLOCKED(noop_qdisc
.q
.lock
),
377 .dev_queue
= &noop_netdev_queue
,
378 .busylock
= __SPIN_LOCK_UNLOCKED(noop_qdisc
.busylock
),
380 EXPORT_SYMBOL(noop_qdisc
);
382 static struct Qdisc_ops noqueue_qdisc_ops __read_mostly
= {
385 .enqueue
= noop_enqueue
,
386 .dequeue
= noop_dequeue
,
387 .peek
= noop_dequeue
,
388 .owner
= THIS_MODULE
,
391 static struct Qdisc noqueue_qdisc
;
392 static struct netdev_queue noqueue_netdev_queue
= {
393 .qdisc
= &noqueue_qdisc
,
394 .qdisc_sleeping
= &noqueue_qdisc
,
397 static struct Qdisc noqueue_qdisc
= {
399 .dequeue
= noop_dequeue
,
400 .flags
= TCQ_F_BUILTIN
,
401 .ops
= &noqueue_qdisc_ops
,
402 .list
= LIST_HEAD_INIT(noqueue_qdisc
.list
),
403 .q
.lock
= __SPIN_LOCK_UNLOCKED(noqueue_qdisc
.q
.lock
),
404 .dev_queue
= &noqueue_netdev_queue
,
405 .busylock
= __SPIN_LOCK_UNLOCKED(noqueue_qdisc
.busylock
),
409 static const u8 prio2band
[TC_PRIO_MAX
+ 1] = {
410 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
413 /* 3-band FIFO queue: old style, but should be a bit faster than
414 generic prio+fifo combination.
417 #define PFIFO_FAST_BANDS 3
420 * Private data for a pfifo_fast scheduler containing:
421 * - queues for the three band
422 * - bitmap indicating which of the bands contain skbs
424 struct pfifo_fast_priv
{
426 struct sk_buff_head q
[PFIFO_FAST_BANDS
];
430 * Convert a bitmap to the first band number where an skb is queued, where:
431 * bitmap=0 means there are no skbs on any band.
432 * bitmap=1 means there is an skb on band 0.
433 * bitmap=7 means there are skbs on all 3 bands, etc.
435 static const int bitmap2band
[] = {-1, 0, 1, 0, 2, 0, 1, 0};
437 static inline struct sk_buff_head
*band2list(struct pfifo_fast_priv
*priv
,
440 return priv
->q
+ band
;
443 static int pfifo_fast_enqueue(struct sk_buff
*skb
, struct Qdisc
*qdisc
)
445 if (skb_queue_len(&qdisc
->q
) < qdisc_dev(qdisc
)->tx_queue_len
) {
446 int band
= prio2band
[skb
->priority
& TC_PRIO_MAX
];
447 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
448 struct sk_buff_head
*list
= band2list(priv
, band
);
450 priv
->bitmap
|= (1 << band
);
452 return __qdisc_enqueue_tail(skb
, qdisc
, list
);
455 return qdisc_drop(skb
, qdisc
);
458 static struct sk_buff
*pfifo_fast_dequeue(struct Qdisc
*qdisc
)
460 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
461 int band
= bitmap2band
[priv
->bitmap
];
463 if (likely(band
>= 0)) {
464 struct sk_buff_head
*list
= band2list(priv
, band
);
465 struct sk_buff
*skb
= __qdisc_dequeue_head(qdisc
, list
);
468 if (skb_queue_empty(list
))
469 priv
->bitmap
&= ~(1 << band
);
477 static struct sk_buff
*pfifo_fast_peek(struct Qdisc
*qdisc
)
479 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
480 int band
= bitmap2band
[priv
->bitmap
];
483 struct sk_buff_head
*list
= band2list(priv
, band
);
485 return skb_peek(list
);
491 static void pfifo_fast_reset(struct Qdisc
*qdisc
)
494 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
496 for (prio
= 0; prio
< PFIFO_FAST_BANDS
; prio
++)
497 __qdisc_reset_queue(qdisc
, band2list(priv
, prio
));
500 qdisc
->qstats
.backlog
= 0;
504 static int pfifo_fast_dump(struct Qdisc
*qdisc
, struct sk_buff
*skb
)
506 struct tc_prio_qopt opt
= { .bands
= PFIFO_FAST_BANDS
};
508 memcpy(&opt
.priomap
, prio2band
, TC_PRIO_MAX
+ 1);
509 if (nla_put(skb
, TCA_OPTIONS
, sizeof(opt
), &opt
))
510 goto nla_put_failure
;
517 static int pfifo_fast_init(struct Qdisc
*qdisc
, struct nlattr
*opt
)
520 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
522 for (prio
= 0; prio
< PFIFO_FAST_BANDS
; prio
++)
523 skb_queue_head_init(band2list(priv
, prio
));
525 /* Can by-pass the queue discipline */
526 qdisc
->flags
|= TCQ_F_CAN_BYPASS
;
530 struct Qdisc_ops pfifo_fast_ops __read_mostly
= {
532 .priv_size
= sizeof(struct pfifo_fast_priv
),
533 .enqueue
= pfifo_fast_enqueue
,
534 .dequeue
= pfifo_fast_dequeue
,
535 .peek
= pfifo_fast_peek
,
536 .init
= pfifo_fast_init
,
537 .reset
= pfifo_fast_reset
,
538 .dump
= pfifo_fast_dump
,
539 .owner
= THIS_MODULE
,
542 static struct lock_class_key qdisc_tx_busylock
;
544 struct Qdisc
*qdisc_alloc(struct netdev_queue
*dev_queue
,
545 const struct Qdisc_ops
*ops
)
549 unsigned int size
= QDISC_ALIGN(sizeof(*sch
)) + ops
->priv_size
;
551 struct net_device
*dev
= dev_queue
->dev
;
553 p
= kzalloc_node(size
, GFP_KERNEL
,
554 netdev_queue_numa_node_read(dev_queue
));
558 sch
= (struct Qdisc
*) QDISC_ALIGN((unsigned long) p
);
559 /* if we got non aligned memory, ask more and do alignment ourself */
562 p
= kzalloc_node(size
+ QDISC_ALIGNTO
- 1, GFP_KERNEL
,
563 netdev_queue_numa_node_read(dev_queue
));
566 sch
= (struct Qdisc
*) QDISC_ALIGN((unsigned long) p
);
567 sch
->padded
= (char *) sch
- (char *) p
;
569 INIT_LIST_HEAD(&sch
->list
);
570 skb_queue_head_init(&sch
->q
);
572 spin_lock_init(&sch
->busylock
);
573 lockdep_set_class(&sch
->busylock
,
574 dev
->qdisc_tx_busylock
?: &qdisc_tx_busylock
);
577 sch
->enqueue
= ops
->enqueue
;
578 sch
->dequeue
= ops
->dequeue
;
579 sch
->dev_queue
= dev_queue
;
581 atomic_set(&sch
->refcnt
, 1);
588 struct Qdisc
*qdisc_create_dflt(struct netdev_queue
*dev_queue
,
589 const struct Qdisc_ops
*ops
,
590 unsigned int parentid
)
594 if (!try_module_get(ops
->owner
))
597 sch
= qdisc_alloc(dev_queue
, ops
);
600 sch
->parent
= parentid
;
602 if (!ops
->init
|| ops
->init(sch
, NULL
) == 0)
609 EXPORT_SYMBOL(qdisc_create_dflt
);
611 /* Under qdisc_lock(qdisc) and BH! */
613 void qdisc_reset(struct Qdisc
*qdisc
)
615 const struct Qdisc_ops
*ops
= qdisc
->ops
;
620 if (qdisc
->gso_skb
) {
621 kfree_skb(qdisc
->gso_skb
);
622 qdisc
->gso_skb
= NULL
;
626 EXPORT_SYMBOL(qdisc_reset
);
628 static void qdisc_rcu_free(struct rcu_head
*head
)
630 struct Qdisc
*qdisc
= container_of(head
, struct Qdisc
, rcu_head
);
632 kfree((char *) qdisc
- qdisc
->padded
);
635 void qdisc_destroy(struct Qdisc
*qdisc
)
637 const struct Qdisc_ops
*ops
= qdisc
->ops
;
639 if (qdisc
->flags
& TCQ_F_BUILTIN
||
640 !atomic_dec_and_test(&qdisc
->refcnt
))
643 #ifdef CONFIG_NET_SCHED
644 qdisc_list_del(qdisc
);
646 qdisc_put_stab(rtnl_dereference(qdisc
->stab
));
648 gen_kill_estimator(&qdisc
->bstats
, &qdisc
->rate_est
);
654 module_put(ops
->owner
);
655 dev_put(qdisc_dev(qdisc
));
657 kfree_skb(qdisc
->gso_skb
);
659 * gen_estimator est_timer() might access qdisc->q.lock,
660 * wait a RCU grace period before freeing qdisc.
662 call_rcu(&qdisc
->rcu_head
, qdisc_rcu_free
);
664 EXPORT_SYMBOL(qdisc_destroy
);
666 /* Attach toplevel qdisc to device queue. */
667 struct Qdisc
*dev_graft_qdisc(struct netdev_queue
*dev_queue
,
670 struct Qdisc
*oqdisc
= dev_queue
->qdisc_sleeping
;
671 spinlock_t
*root_lock
;
673 root_lock
= qdisc_lock(oqdisc
);
674 spin_lock_bh(root_lock
);
676 /* Prune old scheduler */
677 if (oqdisc
&& atomic_read(&oqdisc
->refcnt
) <= 1)
680 /* ... and graft new one */
683 dev_queue
->qdisc_sleeping
= qdisc
;
684 rcu_assign_pointer(dev_queue
->qdisc
, &noop_qdisc
);
686 spin_unlock_bh(root_lock
);
690 EXPORT_SYMBOL(dev_graft_qdisc
);
692 static void attach_one_default_qdisc(struct net_device
*dev
,
693 struct netdev_queue
*dev_queue
,
696 struct Qdisc
*qdisc
= &noqueue_qdisc
;
698 if (dev
->tx_queue_len
) {
699 qdisc
= qdisc_create_dflt(dev_queue
,
700 default_qdisc_ops
, TC_H_ROOT
);
702 netdev_info(dev
, "activation failed\n");
705 if (!netif_is_multiqueue(dev
))
706 qdisc
->flags
|= TCQ_F_ONETXQUEUE
;
708 dev_queue
->qdisc_sleeping
= qdisc
;
711 static void attach_default_qdiscs(struct net_device
*dev
)
713 struct netdev_queue
*txq
;
716 txq
= netdev_get_tx_queue(dev
, 0);
718 if (!netif_is_multiqueue(dev
) || dev
->tx_queue_len
== 0) {
719 netdev_for_each_tx_queue(dev
, attach_one_default_qdisc
, NULL
);
720 dev
->qdisc
= txq
->qdisc_sleeping
;
721 atomic_inc(&dev
->qdisc
->refcnt
);
723 qdisc
= qdisc_create_dflt(txq
, &mq_qdisc_ops
, TC_H_ROOT
);
726 qdisc
->ops
->attach(qdisc
);
731 static void transition_one_qdisc(struct net_device
*dev
,
732 struct netdev_queue
*dev_queue
,
733 void *_need_watchdog
)
735 struct Qdisc
*new_qdisc
= dev_queue
->qdisc_sleeping
;
736 int *need_watchdog_p
= _need_watchdog
;
738 if (!(new_qdisc
->flags
& TCQ_F_BUILTIN
))
739 clear_bit(__QDISC_STATE_DEACTIVATED
, &new_qdisc
->state
);
741 rcu_assign_pointer(dev_queue
->qdisc
, new_qdisc
);
742 if (need_watchdog_p
&& new_qdisc
!= &noqueue_qdisc
) {
743 dev_queue
->trans_start
= 0;
744 *need_watchdog_p
= 1;
748 void dev_activate(struct net_device
*dev
)
752 /* No queueing discipline is attached to device;
753 * create default one for devices, which need queueing
754 * and noqueue_qdisc for virtual interfaces
757 if (dev
->qdisc
== &noop_qdisc
)
758 attach_default_qdiscs(dev
);
760 if (!netif_carrier_ok(dev
))
761 /* Delay activation until next carrier-on event */
765 netdev_for_each_tx_queue(dev
, transition_one_qdisc
, &need_watchdog
);
766 if (dev_ingress_queue(dev
))
767 transition_one_qdisc(dev
, dev_ingress_queue(dev
), NULL
);
770 dev
->trans_start
= jiffies
;
771 dev_watchdog_up(dev
);
774 EXPORT_SYMBOL(dev_activate
);
776 static void dev_deactivate_queue(struct net_device
*dev
,
777 struct netdev_queue
*dev_queue
,
778 void *_qdisc_default
)
780 struct Qdisc
*qdisc_default
= _qdisc_default
;
783 qdisc
= dev_queue
->qdisc
;
785 spin_lock_bh(qdisc_lock(qdisc
));
787 if (!(qdisc
->flags
& TCQ_F_BUILTIN
))
788 set_bit(__QDISC_STATE_DEACTIVATED
, &qdisc
->state
);
790 rcu_assign_pointer(dev_queue
->qdisc
, qdisc_default
);
793 spin_unlock_bh(qdisc_lock(qdisc
));
797 static bool some_qdisc_is_busy(struct net_device
*dev
)
801 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
802 struct netdev_queue
*dev_queue
;
803 spinlock_t
*root_lock
;
807 dev_queue
= netdev_get_tx_queue(dev
, i
);
808 q
= dev_queue
->qdisc_sleeping
;
809 root_lock
= qdisc_lock(q
);
811 spin_lock_bh(root_lock
);
813 val
= (qdisc_is_running(q
) ||
814 test_bit(__QDISC_STATE_SCHED
, &q
->state
));
816 spin_unlock_bh(root_lock
);
825 * dev_deactivate_many - deactivate transmissions on several devices
826 * @head: list of devices to deactivate
828 * This function returns only when all outstanding transmissions
829 * have completed, unless all devices are in dismantle phase.
831 void dev_deactivate_many(struct list_head
*head
)
833 struct net_device
*dev
;
834 bool sync_needed
= false;
836 list_for_each_entry(dev
, head
, close_list
) {
837 netdev_for_each_tx_queue(dev
, dev_deactivate_queue
,
839 if (dev_ingress_queue(dev
))
840 dev_deactivate_queue(dev
, dev_ingress_queue(dev
),
843 dev_watchdog_down(dev
);
844 sync_needed
|= !dev
->dismantle
;
847 /* Wait for outstanding qdisc-less dev_queue_xmit calls.
848 * This is avoided if all devices are in dismantle phase :
849 * Caller will call synchronize_net() for us
854 /* Wait for outstanding qdisc_run calls. */
855 list_for_each_entry(dev
, head
, close_list
)
856 while (some_qdisc_is_busy(dev
))
860 void dev_deactivate(struct net_device
*dev
)
864 list_add(&dev
->close_list
, &single
);
865 dev_deactivate_many(&single
);
868 EXPORT_SYMBOL(dev_deactivate
);
870 static void dev_init_scheduler_queue(struct net_device
*dev
,
871 struct netdev_queue
*dev_queue
,
874 struct Qdisc
*qdisc
= _qdisc
;
876 dev_queue
->qdisc
= qdisc
;
877 dev_queue
->qdisc_sleeping
= qdisc
;
880 void dev_init_scheduler(struct net_device
*dev
)
882 dev
->qdisc
= &noop_qdisc
;
883 netdev_for_each_tx_queue(dev
, dev_init_scheduler_queue
, &noop_qdisc
);
884 if (dev_ingress_queue(dev
))
885 dev_init_scheduler_queue(dev
, dev_ingress_queue(dev
), &noop_qdisc
);
887 setup_timer(&dev
->watchdog_timer
, dev_watchdog
, (unsigned long)dev
);
890 static void shutdown_scheduler_queue(struct net_device
*dev
,
891 struct netdev_queue
*dev_queue
,
892 void *_qdisc_default
)
894 struct Qdisc
*qdisc
= dev_queue
->qdisc_sleeping
;
895 struct Qdisc
*qdisc_default
= _qdisc_default
;
898 rcu_assign_pointer(dev_queue
->qdisc
, qdisc_default
);
899 dev_queue
->qdisc_sleeping
= qdisc_default
;
901 qdisc_destroy(qdisc
);
905 void dev_shutdown(struct net_device
*dev
)
907 netdev_for_each_tx_queue(dev
, shutdown_scheduler_queue
, &noop_qdisc
);
908 if (dev_ingress_queue(dev
))
909 shutdown_scheduler_queue(dev
, dev_ingress_queue(dev
), &noop_qdisc
);
910 qdisc_destroy(dev
->qdisc
);
911 dev
->qdisc
= &noop_qdisc
;
913 WARN_ON(timer_pending(&dev
->watchdog_timer
));
916 void psched_ratecfg_precompute(struct psched_ratecfg
*r
,
917 const struct tc_ratespec
*conf
,
920 memset(r
, 0, sizeof(*r
));
921 r
->overhead
= conf
->overhead
;
922 r
->rate_bytes_ps
= max_t(u64
, conf
->rate
, rate64
);
923 r
->linklayer
= (conf
->linklayer
& TC_LINKLAYER_MASK
);
926 * The deal here is to replace a divide by a reciprocal one
927 * in fast path (a reciprocal divide is a multiply and a shift)
929 * Normal formula would be :
930 * time_in_ns = (NSEC_PER_SEC * len) / rate_bps
932 * We compute mult/shift to use instead :
933 * time_in_ns = (len * mult) >> shift;
935 * We try to get the highest possible mult value for accuracy,
936 * but have to make sure no overflows will ever happen.
938 if (r
->rate_bytes_ps
> 0) {
939 u64 factor
= NSEC_PER_SEC
;
942 r
->mult
= div64_u64(factor
, r
->rate_bytes_ps
);
943 if (r
->mult
& (1U << 31) || factor
& (1ULL << 63))
950 EXPORT_SYMBOL(psched_ratecfg_precompute
);