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 /* Main transmission queue. */
35 /* Modifications to data participating in scheduling must be protected with
36 * qdisc_lock(qdisc) spinlock.
38 * The idea is the following:
39 * - enqueue, dequeue are serialized via qdisc root lock
40 * - ingress filtering is also serialized via qdisc root lock
41 * - updates to tree and tree walking are only done under the rtnl mutex.
44 static inline int dev_requeue_skb(struct sk_buff
*skb
, struct Qdisc
*q
)
49 q
->q
.qlen
++; /* it's still part of the queue */
55 static inline struct sk_buff
*dequeue_skb(struct Qdisc
*q
)
57 struct sk_buff
*skb
= q
->gso_skb
;
58 const struct netdev_queue
*txq
= q
->dev_queue
;
61 /* check the reason of requeuing without tx lock first */
62 txq
= netdev_get_tx_queue(txq
->dev
, skb_get_queue_mapping(skb
));
63 if (!netif_xmit_frozen_or_stopped(txq
)) {
69 if (!(q
->flags
& TCQ_F_ONETXQUEUE
) || !netif_xmit_frozen_or_stopped(txq
))
76 static inline int handle_dev_cpu_collision(struct sk_buff
*skb
,
77 struct netdev_queue
*dev_queue
,
82 if (unlikely(dev_queue
->xmit_lock_owner
== smp_processor_id())) {
84 * Same CPU holding the lock. It may be a transient
85 * configuration error, when hard_start_xmit() recurses. We
86 * detect it by checking xmit owner and drop the packet when
87 * deadloop is detected. Return OK to try the next skb.
90 net_warn_ratelimited("Dead loop on netdevice %s, fix it urgently!\n",
91 dev_queue
->dev
->name
);
95 * Another cpu is holding lock, requeue & delay xmits for
98 __this_cpu_inc(softnet_data
.cpu_collision
);
99 ret
= dev_requeue_skb(skb
, q
);
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
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_xmit_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 */
134 } else if (ret
== NETDEV_TX_LOCKED
) {
135 /* Driver try lock failed */
136 ret
= handle_dev_cpu_collision(skb
, txq
, q
);
138 /* Driver returned NETDEV_TX_BUSY - requeue skb */
139 if (unlikely(ret
!= NETDEV_TX_BUSY
))
140 net_warn_ratelimited("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_xmit_frozen_or_stopped(txq
))
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
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
;
179 skb
= dequeue_skb(q
);
182 WARN_ON_ONCE(skb_dst_is_noref(skb
));
183 root_lock
= qdisc_lock(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()) {
209 unsigned long dev_trans_start(struct net_device
*dev
)
211 unsigned long val
, res
;
214 if (is_vlan_dev(dev
))
215 dev
= vlan_dev_real_dev(dev
);
216 res
= dev
->trans_start
;
217 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
218 val
= netdev_get_tx_queue(dev
, i
)->trans_start
;
219 if (val
&& time_after(val
, res
))
222 dev
->trans_start
= res
;
226 EXPORT_SYMBOL(dev_trans_start
);
228 static void dev_watchdog(unsigned long arg
)
230 struct net_device
*dev
= (struct net_device
*)arg
;
233 if (!qdisc_tx_is_noop(dev
)) {
234 if (netif_device_present(dev
) &&
235 netif_running(dev
) &&
236 netif_carrier_ok(dev
)) {
237 int some_queue_timedout
= 0;
239 unsigned long trans_start
;
241 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
242 struct netdev_queue
*txq
;
244 txq
= netdev_get_tx_queue(dev
, i
);
246 * old device drivers set dev->trans_start
248 trans_start
= txq
->trans_start
? : dev
->trans_start
;
249 if (netif_xmit_stopped(txq
) &&
250 time_after(jiffies
, (trans_start
+
251 dev
->watchdog_timeo
))) {
252 some_queue_timedout
= 1;
253 txq
->trans_timeout
++;
258 if (some_queue_timedout
) {
259 WARN_ONCE(1, KERN_INFO
"NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
260 dev
->name
, netdev_drivername(dev
), i
);
261 dev
->netdev_ops
->ndo_tx_timeout(dev
);
263 if (!mod_timer(&dev
->watchdog_timer
,
264 round_jiffies(jiffies
+
265 dev
->watchdog_timeo
)))
269 netif_tx_unlock(dev
);
274 void __netdev_watchdog_up(struct net_device
*dev
)
276 if (dev
->netdev_ops
->ndo_tx_timeout
) {
277 if (dev
->watchdog_timeo
<= 0)
278 dev
->watchdog_timeo
= 5*HZ
;
279 if (!mod_timer(&dev
->watchdog_timer
,
280 round_jiffies(jiffies
+ dev
->watchdog_timeo
)))
285 static void dev_watchdog_up(struct net_device
*dev
)
287 __netdev_watchdog_up(dev
);
290 static void dev_watchdog_down(struct net_device
*dev
)
292 netif_tx_lock_bh(dev
);
293 if (del_timer(&dev
->watchdog_timer
))
295 netif_tx_unlock_bh(dev
);
299 * netif_carrier_on - set carrier
300 * @dev: network device
302 * Device has detected that carrier.
304 void netif_carrier_on(struct net_device
*dev
)
306 if (test_and_clear_bit(__LINK_STATE_NOCARRIER
, &dev
->state
)) {
307 if (dev
->reg_state
== NETREG_UNINITIALIZED
)
309 linkwatch_fire_event(dev
);
310 if (netif_running(dev
))
311 __netdev_watchdog_up(dev
);
314 EXPORT_SYMBOL(netif_carrier_on
);
317 * netif_carrier_off - clear carrier
318 * @dev: network device
320 * Device has detected loss of carrier.
322 void netif_carrier_off(struct net_device
*dev
)
324 if (!test_and_set_bit(__LINK_STATE_NOCARRIER
, &dev
->state
)) {
325 if (dev
->reg_state
== NETREG_UNINITIALIZED
)
327 linkwatch_fire_event(dev
);
330 EXPORT_SYMBOL(netif_carrier_off
);
332 /* "NOOP" scheduler: the best scheduler, recommended for all interfaces
333 under all circumstances. It is difficult to invent anything faster or
337 static int noop_enqueue(struct sk_buff
*skb
, struct Qdisc
* qdisc
)
343 static struct sk_buff
*noop_dequeue(struct Qdisc
* qdisc
)
348 struct Qdisc_ops noop_qdisc_ops __read_mostly
= {
351 .enqueue
= noop_enqueue
,
352 .dequeue
= noop_dequeue
,
353 .peek
= noop_dequeue
,
354 .owner
= THIS_MODULE
,
357 static struct netdev_queue noop_netdev_queue
= {
358 .qdisc
= &noop_qdisc
,
359 .qdisc_sleeping
= &noop_qdisc
,
362 struct Qdisc noop_qdisc
= {
363 .enqueue
= noop_enqueue
,
364 .dequeue
= noop_dequeue
,
365 .flags
= TCQ_F_BUILTIN
,
366 .ops
= &noop_qdisc_ops
,
367 .list
= LIST_HEAD_INIT(noop_qdisc
.list
),
368 .q
.lock
= __SPIN_LOCK_UNLOCKED(noop_qdisc
.q
.lock
),
369 .dev_queue
= &noop_netdev_queue
,
370 .busylock
= __SPIN_LOCK_UNLOCKED(noop_qdisc
.busylock
),
372 EXPORT_SYMBOL(noop_qdisc
);
374 static struct Qdisc_ops noqueue_qdisc_ops __read_mostly
= {
377 .enqueue
= noop_enqueue
,
378 .dequeue
= noop_dequeue
,
379 .peek
= noop_dequeue
,
380 .owner
= THIS_MODULE
,
383 static struct Qdisc noqueue_qdisc
;
384 static struct netdev_queue noqueue_netdev_queue
= {
385 .qdisc
= &noqueue_qdisc
,
386 .qdisc_sleeping
= &noqueue_qdisc
,
389 static struct Qdisc noqueue_qdisc
= {
391 .dequeue
= noop_dequeue
,
392 .flags
= TCQ_F_BUILTIN
,
393 .ops
= &noqueue_qdisc_ops
,
394 .list
= LIST_HEAD_INIT(noqueue_qdisc
.list
),
395 .q
.lock
= __SPIN_LOCK_UNLOCKED(noqueue_qdisc
.q
.lock
),
396 .dev_queue
= &noqueue_netdev_queue
,
397 .busylock
= __SPIN_LOCK_UNLOCKED(noqueue_qdisc
.busylock
),
401 static const u8 prio2band
[TC_PRIO_MAX
+ 1] = {
402 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
405 /* 3-band FIFO queue: old style, but should be a bit faster than
406 generic prio+fifo combination.
409 #define PFIFO_FAST_BANDS 3
412 * Private data for a pfifo_fast scheduler containing:
413 * - queues for the three band
414 * - bitmap indicating which of the bands contain skbs
416 struct pfifo_fast_priv
{
418 struct sk_buff_head q
[PFIFO_FAST_BANDS
];
422 * Convert a bitmap to the first band number where an skb is queued, where:
423 * bitmap=0 means there are no skbs on any band.
424 * bitmap=1 means there is an skb on band 0.
425 * bitmap=7 means there are skbs on all 3 bands, etc.
427 static const int bitmap2band
[] = {-1, 0, 1, 0, 2, 0, 1, 0};
429 static inline struct sk_buff_head
*band2list(struct pfifo_fast_priv
*priv
,
432 return priv
->q
+ band
;
435 static int pfifo_fast_enqueue(struct sk_buff
*skb
, struct Qdisc
*qdisc
)
437 if (skb_queue_len(&qdisc
->q
) < qdisc_dev(qdisc
)->tx_queue_len
) {
438 int band
= prio2band
[skb
->priority
& TC_PRIO_MAX
];
439 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
440 struct sk_buff_head
*list
= band2list(priv
, band
);
442 priv
->bitmap
|= (1 << band
);
444 return __qdisc_enqueue_tail(skb
, qdisc
, list
);
447 return qdisc_drop(skb
, qdisc
);
450 static struct sk_buff
*pfifo_fast_dequeue(struct Qdisc
*qdisc
)
452 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
453 int band
= bitmap2band
[priv
->bitmap
];
455 if (likely(band
>= 0)) {
456 struct sk_buff_head
*list
= band2list(priv
, band
);
457 struct sk_buff
*skb
= __qdisc_dequeue_head(qdisc
, list
);
460 if (skb_queue_empty(list
))
461 priv
->bitmap
&= ~(1 << band
);
469 static struct sk_buff
*pfifo_fast_peek(struct Qdisc
*qdisc
)
471 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
472 int band
= bitmap2band
[priv
->bitmap
];
475 struct sk_buff_head
*list
= band2list(priv
, band
);
477 return skb_peek(list
);
483 static void pfifo_fast_reset(struct Qdisc
*qdisc
)
486 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
488 for (prio
= 0; prio
< PFIFO_FAST_BANDS
; prio
++)
489 __qdisc_reset_queue(qdisc
, band2list(priv
, prio
));
492 qdisc
->qstats
.backlog
= 0;
496 static int pfifo_fast_dump(struct Qdisc
*qdisc
, struct sk_buff
*skb
)
498 struct tc_prio_qopt opt
= { .bands
= PFIFO_FAST_BANDS
};
500 memcpy(&opt
.priomap
, prio2band
, TC_PRIO_MAX
+ 1);
501 if (nla_put(skb
, TCA_OPTIONS
, sizeof(opt
), &opt
))
502 goto nla_put_failure
;
509 static int pfifo_fast_init(struct Qdisc
*qdisc
, struct nlattr
*opt
)
512 struct pfifo_fast_priv
*priv
= qdisc_priv(qdisc
);
514 for (prio
= 0; prio
< PFIFO_FAST_BANDS
; prio
++)
515 skb_queue_head_init(band2list(priv
, prio
));
517 /* Can by-pass the queue discipline */
518 qdisc
->flags
|= TCQ_F_CAN_BYPASS
;
522 struct Qdisc_ops pfifo_fast_ops __read_mostly
= {
524 .priv_size
= sizeof(struct pfifo_fast_priv
),
525 .enqueue
= pfifo_fast_enqueue
,
526 .dequeue
= pfifo_fast_dequeue
,
527 .peek
= pfifo_fast_peek
,
528 .init
= pfifo_fast_init
,
529 .reset
= pfifo_fast_reset
,
530 .dump
= pfifo_fast_dump
,
531 .owner
= THIS_MODULE
,
533 EXPORT_SYMBOL(pfifo_fast_ops
);
535 static struct lock_class_key qdisc_tx_busylock
;
537 struct Qdisc
*qdisc_alloc(struct netdev_queue
*dev_queue
,
538 struct Qdisc_ops
*ops
)
542 unsigned int size
= QDISC_ALIGN(sizeof(*sch
)) + ops
->priv_size
;
544 struct net_device
*dev
= dev_queue
->dev
;
546 p
= kzalloc_node(size
, GFP_KERNEL
,
547 netdev_queue_numa_node_read(dev_queue
));
551 sch
= (struct Qdisc
*) QDISC_ALIGN((unsigned long) p
);
552 /* if we got non aligned memory, ask more and do alignment ourself */
555 p
= kzalloc_node(size
+ QDISC_ALIGNTO
- 1, GFP_KERNEL
,
556 netdev_queue_numa_node_read(dev_queue
));
559 sch
= (struct Qdisc
*) QDISC_ALIGN((unsigned long) p
);
560 sch
->padded
= (char *) sch
- (char *) p
;
562 INIT_LIST_HEAD(&sch
->list
);
563 skb_queue_head_init(&sch
->q
);
565 spin_lock_init(&sch
->busylock
);
566 lockdep_set_class(&sch
->busylock
,
567 dev
->qdisc_tx_busylock
?: &qdisc_tx_busylock
);
570 sch
->enqueue
= ops
->enqueue
;
571 sch
->dequeue
= ops
->dequeue
;
572 sch
->dev_queue
= dev_queue
;
574 atomic_set(&sch
->refcnt
, 1);
581 struct Qdisc
*qdisc_create_dflt(struct netdev_queue
*dev_queue
,
582 struct Qdisc_ops
*ops
, unsigned int parentid
)
586 sch
= qdisc_alloc(dev_queue
, ops
);
589 sch
->parent
= parentid
;
591 if (!ops
->init
|| ops
->init(sch
, NULL
) == 0)
598 EXPORT_SYMBOL(qdisc_create_dflt
);
600 /* Under qdisc_lock(qdisc) and BH! */
602 void qdisc_reset(struct Qdisc
*qdisc
)
604 const struct Qdisc_ops
*ops
= qdisc
->ops
;
609 if (qdisc
->gso_skb
) {
610 kfree_skb(qdisc
->gso_skb
);
611 qdisc
->gso_skb
= NULL
;
615 EXPORT_SYMBOL(qdisc_reset
);
617 static void qdisc_rcu_free(struct rcu_head
*head
)
619 struct Qdisc
*qdisc
= container_of(head
, struct Qdisc
, rcu_head
);
621 kfree((char *) qdisc
- qdisc
->padded
);
624 void qdisc_destroy(struct Qdisc
*qdisc
)
626 const struct Qdisc_ops
*ops
= qdisc
->ops
;
628 if (qdisc
->flags
& TCQ_F_BUILTIN
||
629 !atomic_dec_and_test(&qdisc
->refcnt
))
632 #ifdef CONFIG_NET_SCHED
633 qdisc_list_del(qdisc
);
635 qdisc_put_stab(rtnl_dereference(qdisc
->stab
));
637 gen_kill_estimator(&qdisc
->bstats
, &qdisc
->rate_est
);
643 module_put(ops
->owner
);
644 dev_put(qdisc_dev(qdisc
));
646 kfree_skb(qdisc
->gso_skb
);
648 * gen_estimator est_timer() might access qdisc->q.lock,
649 * wait a RCU grace period before freeing qdisc.
651 call_rcu(&qdisc
->rcu_head
, qdisc_rcu_free
);
653 EXPORT_SYMBOL(qdisc_destroy
);
655 /* Attach toplevel qdisc to device queue. */
656 struct Qdisc
*dev_graft_qdisc(struct netdev_queue
*dev_queue
,
659 struct Qdisc
*oqdisc
= dev_queue
->qdisc_sleeping
;
660 spinlock_t
*root_lock
;
662 root_lock
= qdisc_lock(oqdisc
);
663 spin_lock_bh(root_lock
);
665 /* Prune old scheduler */
666 if (oqdisc
&& atomic_read(&oqdisc
->refcnt
) <= 1)
669 /* ... and graft new one */
672 dev_queue
->qdisc_sleeping
= qdisc
;
673 rcu_assign_pointer(dev_queue
->qdisc
, &noop_qdisc
);
675 spin_unlock_bh(root_lock
);
679 EXPORT_SYMBOL(dev_graft_qdisc
);
681 static void attach_one_default_qdisc(struct net_device
*dev
,
682 struct netdev_queue
*dev_queue
,
685 struct Qdisc
*qdisc
= &noqueue_qdisc
;
687 if (dev
->tx_queue_len
) {
688 qdisc
= qdisc_create_dflt(dev_queue
,
689 &pfifo_fast_ops
, TC_H_ROOT
);
691 netdev_info(dev
, "activation failed\n");
694 if (!netif_is_multiqueue(dev
))
695 qdisc
->flags
|= TCQ_F_ONETXQUEUE
;
697 dev_queue
->qdisc_sleeping
= qdisc
;
700 static void attach_default_qdiscs(struct net_device
*dev
)
702 struct netdev_queue
*txq
;
705 txq
= netdev_get_tx_queue(dev
, 0);
707 if (!netif_is_multiqueue(dev
) || dev
->tx_queue_len
== 0) {
708 netdev_for_each_tx_queue(dev
, attach_one_default_qdisc
, NULL
);
709 dev
->qdisc
= txq
->qdisc_sleeping
;
710 atomic_inc(&dev
->qdisc
->refcnt
);
712 qdisc
= qdisc_create_dflt(txq
, &mq_qdisc_ops
, TC_H_ROOT
);
714 qdisc
->ops
->attach(qdisc
);
720 static void transition_one_qdisc(struct net_device
*dev
,
721 struct netdev_queue
*dev_queue
,
722 void *_need_watchdog
)
724 struct Qdisc
*new_qdisc
= dev_queue
->qdisc_sleeping
;
725 int *need_watchdog_p
= _need_watchdog
;
727 if (!(new_qdisc
->flags
& TCQ_F_BUILTIN
))
728 clear_bit(__QDISC_STATE_DEACTIVATED
, &new_qdisc
->state
);
730 rcu_assign_pointer(dev_queue
->qdisc
, new_qdisc
);
731 if (need_watchdog_p
&& new_qdisc
!= &noqueue_qdisc
) {
732 dev_queue
->trans_start
= 0;
733 *need_watchdog_p
= 1;
737 void dev_activate(struct net_device
*dev
)
741 /* No queueing discipline is attached to device;
742 create default one i.e. pfifo_fast for devices,
743 which need queueing and noqueue_qdisc for
747 if (dev
->qdisc
== &noop_qdisc
)
748 attach_default_qdiscs(dev
);
750 if (!netif_carrier_ok(dev
))
751 /* Delay activation until next carrier-on event */
755 netdev_for_each_tx_queue(dev
, transition_one_qdisc
, &need_watchdog
);
756 if (dev_ingress_queue(dev
))
757 transition_one_qdisc(dev
, dev_ingress_queue(dev
), NULL
);
760 dev
->trans_start
= jiffies
;
761 dev_watchdog_up(dev
);
764 EXPORT_SYMBOL(dev_activate
);
766 static void dev_deactivate_queue(struct net_device
*dev
,
767 struct netdev_queue
*dev_queue
,
768 void *_qdisc_default
)
770 struct Qdisc
*qdisc_default
= _qdisc_default
;
773 qdisc
= dev_queue
->qdisc
;
775 spin_lock_bh(qdisc_lock(qdisc
));
777 if (!(qdisc
->flags
& TCQ_F_BUILTIN
))
778 set_bit(__QDISC_STATE_DEACTIVATED
, &qdisc
->state
);
780 rcu_assign_pointer(dev_queue
->qdisc
, qdisc_default
);
783 spin_unlock_bh(qdisc_lock(qdisc
));
787 static bool some_qdisc_is_busy(struct net_device
*dev
)
791 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
792 struct netdev_queue
*dev_queue
;
793 spinlock_t
*root_lock
;
797 dev_queue
= netdev_get_tx_queue(dev
, i
);
798 q
= dev_queue
->qdisc_sleeping
;
799 root_lock
= qdisc_lock(q
);
801 spin_lock_bh(root_lock
);
803 val
= (qdisc_is_running(q
) ||
804 test_bit(__QDISC_STATE_SCHED
, &q
->state
));
806 spin_unlock_bh(root_lock
);
815 * dev_deactivate_many - deactivate transmissions on several devices
816 * @head: list of devices to deactivate
818 * This function returns only when all outstanding transmissions
819 * have completed, unless all devices are in dismantle phase.
821 void dev_deactivate_many(struct list_head
*head
)
823 struct net_device
*dev
;
824 bool sync_needed
= false;
826 list_for_each_entry(dev
, head
, unreg_list
) {
827 netdev_for_each_tx_queue(dev
, dev_deactivate_queue
,
829 if (dev_ingress_queue(dev
))
830 dev_deactivate_queue(dev
, dev_ingress_queue(dev
),
833 dev_watchdog_down(dev
);
834 sync_needed
|= !dev
->dismantle
;
837 /* Wait for outstanding qdisc-less dev_queue_xmit calls.
838 * This is avoided if all devices are in dismantle phase :
839 * Caller will call synchronize_net() for us
844 /* Wait for outstanding qdisc_run calls. */
845 list_for_each_entry(dev
, head
, unreg_list
)
846 while (some_qdisc_is_busy(dev
))
850 void dev_deactivate(struct net_device
*dev
)
854 list_add(&dev
->unreg_list
, &single
);
855 dev_deactivate_many(&single
);
858 EXPORT_SYMBOL(dev_deactivate
);
860 static void dev_init_scheduler_queue(struct net_device
*dev
,
861 struct netdev_queue
*dev_queue
,
864 struct Qdisc
*qdisc
= _qdisc
;
866 dev_queue
->qdisc
= qdisc
;
867 dev_queue
->qdisc_sleeping
= qdisc
;
870 void dev_init_scheduler(struct net_device
*dev
)
872 dev
->qdisc
= &noop_qdisc
;
873 netdev_for_each_tx_queue(dev
, dev_init_scheduler_queue
, &noop_qdisc
);
874 if (dev_ingress_queue(dev
))
875 dev_init_scheduler_queue(dev
, dev_ingress_queue(dev
), &noop_qdisc
);
877 setup_timer(&dev
->watchdog_timer
, dev_watchdog
, (unsigned long)dev
);
880 static void shutdown_scheduler_queue(struct net_device
*dev
,
881 struct netdev_queue
*dev_queue
,
882 void *_qdisc_default
)
884 struct Qdisc
*qdisc
= dev_queue
->qdisc_sleeping
;
885 struct Qdisc
*qdisc_default
= _qdisc_default
;
888 rcu_assign_pointer(dev_queue
->qdisc
, qdisc_default
);
889 dev_queue
->qdisc_sleeping
= qdisc_default
;
891 qdisc_destroy(qdisc
);
895 void dev_shutdown(struct net_device
*dev
)
897 netdev_for_each_tx_queue(dev
, shutdown_scheduler_queue
, &noop_qdisc
);
898 if (dev_ingress_queue(dev
))
899 shutdown_scheduler_queue(dev
, dev_ingress_queue(dev
), &noop_qdisc
);
900 qdisc_destroy(dev
->qdisc
);
901 dev
->qdisc
= &noop_qdisc
;
903 WARN_ON(timer_pending(&dev
->watchdog_timer
));
906 void psched_ratecfg_precompute(struct psched_ratecfg
*r
,
907 const struct tc_ratespec
*conf
)
909 memset(r
, 0, sizeof(*r
));
910 r
->overhead
= conf
->overhead
;
911 r
->rate_bytes_ps
= conf
->rate
;
912 r
->linklayer
= (conf
->linklayer
& TC_LINKLAYER_MASK
);
915 * The deal here is to replace a divide by a reciprocal one
916 * in fast path (a reciprocal divide is a multiply and a shift)
918 * Normal formula would be :
919 * time_in_ns = (NSEC_PER_SEC * len) / rate_bps
921 * We compute mult/shift to use instead :
922 * time_in_ns = (len * mult) >> shift;
924 * We try to get the highest possible mult value for accuracy,
925 * but have to make sure no overflows will ever happen.
927 if (r
->rate_bytes_ps
> 0) {
928 u64 factor
= NSEC_PER_SEC
;
931 r
->mult
= div64_u64(factor
, r
->rate_bytes_ps
);
932 if (r
->mult
& (1U << 31) || factor
& (1ULL << 63))
939 EXPORT_SYMBOL(psched_ratecfg_precompute
);