Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * net/sched/sch_generic.c Generic packet scheduler routines. | |
3 | * | |
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. | |
8 | * | |
9 | * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> | |
10 | * Jamal Hadi Salim, <hadi@cyberus.ca> 990601 | |
11 | * - Ingress support | |
12 | */ | |
13 | ||
1da177e4 | 14 | #include <linux/bitops.h> |
1da177e4 LT |
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> | |
1da177e4 | 20 | #include <linux/errno.h> |
1da177e4 LT |
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> | |
5a0e3ad6 | 27 | #include <linux/slab.h> |
07ce76aa | 28 | #include <linux/if_vlan.h> |
292f1c7f | 29 | #include <net/sch_generic.h> |
1da177e4 | 30 | #include <net/pkt_sched.h> |
7fee226a | 31 | #include <net/dst.h> |
1da177e4 | 32 | |
34aedd3f | 33 | /* Qdisc to use by default */ |
34 | const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops; | |
35 | EXPORT_SYMBOL(default_qdisc_ops); | |
36 | ||
1da177e4 LT |
37 | /* Main transmission queue. */ |
38 | ||
0463d4ae | 39 | /* Modifications to data participating in scheduling must be protected with |
5fb66229 | 40 | * qdisc_lock(qdisc) spinlock. |
0463d4ae PM |
41 | * |
42 | * The idea is the following: | |
c7e4f3bb DM |
43 | * - enqueue, dequeue are serialized via qdisc root lock |
44 | * - ingress filtering is also serialized via qdisc root lock | |
0463d4ae | 45 | * - updates to tree and tree walking are only done under the rtnl mutex. |
1da177e4 | 46 | */ |
1da177e4 | 47 | |
37437bb2 | 48 | static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q) |
c716a81a | 49 | { |
7fee226a | 50 | skb_dst_force(skb); |
6252352d | 51 | q->gso_skb = skb; |
53e91503 | 52 | q->qstats.requeues++; |
bbd8a0d3 | 53 | q->q.qlen++; /* it's still part of the queue */ |
37437bb2 | 54 | __netif_schedule(q); |
6252352d | 55 | |
c716a81a JHS |
56 | return 0; |
57 | } | |
58 | ||
5772e9a3 JDB |
59 | static struct sk_buff *try_bulk_dequeue_skb(struct Qdisc *q, |
60 | struct sk_buff *head_skb, | |
61 | int bytelimit) | |
62 | { | |
63 | struct sk_buff *skb, *tail_skb = head_skb; | |
64 | ||
65 | while (bytelimit > 0) { | |
5772e9a3 JDB |
66 | skb = q->dequeue(q); |
67 | if (!skb) | |
68 | break; | |
69 | ||
70 | bytelimit -= skb->len; /* covers GSO len */ | |
71 | skb = validate_xmit_skb(skb, qdisc_dev(q)); | |
72 | if (!skb) | |
73 | break; | |
74 | ||
808e7ac0 JDB |
75 | while (tail_skb->next) /* GSO list goto tail */ |
76 | tail_skb = tail_skb->next; | |
77 | ||
5772e9a3 JDB |
78 | tail_skb->next = skb; |
79 | tail_skb = skb; | |
80 | } | |
81 | ||
82 | return head_skb; | |
83 | } | |
84 | ||
85 | /* Note that dequeue_skb can possibly return a SKB list (via skb->next). | |
86 | * A requeued skb (via q->gso_skb) can also be a SKB list. | |
87 | */ | |
d3b753db | 88 | static inline struct sk_buff *dequeue_skb(struct Qdisc *q) |
c716a81a | 89 | { |
554794de | 90 | struct sk_buff *skb = q->gso_skb; |
1abbe139 | 91 | const struct netdev_queue *txq = q->dev_queue; |
554794de | 92 | |
ebf05982 | 93 | if (unlikely(skb)) { |
ebf05982 | 94 | /* check the reason of requeuing without tx lock first */ |
10c51b56 | 95 | txq = skb_get_tx_queue(txq->dev, skb); |
73466498 | 96 | if (!netif_xmit_frozen_or_stopped(txq)) { |
6252352d | 97 | q->gso_skb = NULL; |
bbd8a0d3 KK |
98 | q->q.qlen--; |
99 | } else | |
ebf05982 JP |
100 | skb = NULL; |
101 | } else { | |
5772e9a3 JDB |
102 | if (!(q->flags & TCQ_F_ONETXQUEUE) || |
103 | !netif_xmit_frozen_or_stopped(txq)) { | |
104 | int bytelimit = qdisc_avail_bulklimit(txq); | |
105 | ||
1abbe139 | 106 | skb = q->dequeue(q); |
5772e9a3 JDB |
107 | if (skb) { |
108 | bytelimit -= skb->len; | |
50cbe9ab | 109 | skb = validate_xmit_skb(skb, qdisc_dev(q)); |
5772e9a3 JDB |
110 | } |
111 | if (skb && qdisc_may_bulk(q)) | |
112 | skb = try_bulk_dequeue_skb(q, skb, bytelimit); | |
50cbe9ab | 113 | } |
ebf05982 | 114 | } |
c716a81a JHS |
115 | |
116 | return skb; | |
117 | } | |
118 | ||
6c1361a6 | 119 | static inline int handle_dev_cpu_collision(struct sk_buff *skb, |
970565bb | 120 | struct netdev_queue *dev_queue, |
6c1361a6 | 121 | struct Qdisc *q) |
c716a81a | 122 | { |
6c1361a6 | 123 | int ret; |
c716a81a | 124 | |
c773e847 | 125 | if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) { |
6c1361a6 KK |
126 | /* |
127 | * Same CPU holding the lock. It may be a transient | |
128 | * configuration error, when hard_start_xmit() recurses. We | |
129 | * detect it by checking xmit owner and drop the packet when | |
130 | * deadloop is detected. Return OK to try the next skb. | |
131 | */ | |
10770bc2 | 132 | kfree_skb_list(skb); |
e87cc472 JP |
133 | net_warn_ratelimited("Dead loop on netdevice %s, fix it urgently!\n", |
134 | dev_queue->dev->name); | |
6c1361a6 KK |
135 | ret = qdisc_qlen(q); |
136 | } else { | |
137 | /* | |
138 | * Another cpu is holding lock, requeue & delay xmits for | |
139 | * some time. | |
140 | */ | |
d6d9ca0f | 141 | __this_cpu_inc(softnet_data.cpu_collision); |
37437bb2 | 142 | ret = dev_requeue_skb(skb, q); |
c716a81a JHS |
143 | } |
144 | ||
6c1361a6 | 145 | return ret; |
c716a81a JHS |
146 | } |
147 | ||
10297b99 | 148 | /* |
10770bc2 JDB |
149 | * Transmit possibly several skbs, and handle the return status as |
150 | * required. Holding the __QDISC___STATE_RUNNING bit guarantees that | |
151 | * only one CPU can execute this function. | |
6c1361a6 KK |
152 | * |
153 | * Returns to the caller: | |
154 | * 0 - queue is empty or throttled. | |
155 | * >0 - queue is not empty. | |
6c1361a6 | 156 | */ |
bbd8a0d3 KK |
157 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
158 | struct net_device *dev, struct netdev_queue *txq, | |
159 | spinlock_t *root_lock) | |
1da177e4 | 160 | { |
5f1a485d | 161 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
162 | |
163 | /* And release qdisc */ | |
164 | spin_unlock(root_lock); | |
c716a81a | 165 | |
c773e847 | 166 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
73466498 | 167 | if (!netif_xmit_frozen_or_stopped(txq)) |
ce93718f | 168 | skb = dev_hard_start_xmit(skb, dev, txq, &ret); |
572a9d7b | 169 | |
c773e847 | 170 | HARD_TX_UNLOCK(dev, txq); |
c716a81a | 171 | |
7698b4fc | 172 | spin_lock(root_lock); |
c716a81a | 173 | |
9a1654ba JP |
174 | if (dev_xmit_complete(ret)) { |
175 | /* Driver sent out skb successfully or skb was consumed */ | |
6c1361a6 | 176 | ret = qdisc_qlen(q); |
9a1654ba | 177 | } else if (ret == NETDEV_TX_LOCKED) { |
6c1361a6 | 178 | /* Driver try lock failed */ |
eb6aafe3 | 179 | ret = handle_dev_cpu_collision(skb, txq, q); |
9a1654ba | 180 | } else { |
6c1361a6 | 181 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
e87cc472 JP |
182 | if (unlikely(ret != NETDEV_TX_BUSY)) |
183 | net_warn_ratelimited("BUG %s code %d qlen %d\n", | |
184 | dev->name, ret, q->q.qlen); | |
6c1361a6 | 185 | |
37437bb2 | 186 | ret = dev_requeue_skb(skb, q); |
6c1361a6 | 187 | } |
c716a81a | 188 | |
73466498 | 189 | if (ret && netif_xmit_frozen_or_stopped(txq)) |
37437bb2 DM |
190 | ret = 0; |
191 | ||
6c1361a6 | 192 | return ret; |
1da177e4 LT |
193 | } |
194 | ||
bbd8a0d3 KK |
195 | /* |
196 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. | |
197 | * | |
9bf2b8c2 | 198 | * __QDISC___STATE_RUNNING guarantees only one CPU can process |
bbd8a0d3 KK |
199 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for |
200 | * this queue. | |
201 | * | |
202 | * netif_tx_lock serializes accesses to device driver. | |
203 | * | |
204 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, | |
205 | * if one is grabbed, another must be free. | |
206 | * | |
207 | * Note, that this procedure can be called by a watchdog timer | |
208 | * | |
209 | * Returns to the caller: | |
210 | * 0 - queue is empty or throttled. | |
211 | * >0 - queue is not empty. | |
212 | * | |
213 | */ | |
214 | static inline int qdisc_restart(struct Qdisc *q) | |
215 | { | |
216 | struct netdev_queue *txq; | |
217 | struct net_device *dev; | |
218 | spinlock_t *root_lock; | |
219 | struct sk_buff *skb; | |
220 | ||
221 | /* Dequeue packet */ | |
222 | skb = dequeue_skb(q); | |
223 | if (unlikely(!skb)) | |
224 | return 0; | |
10c51b56 | 225 | |
7fee226a | 226 | WARN_ON_ONCE(skb_dst_is_noref(skb)); |
10c51b56 | 227 | |
bbd8a0d3 KK |
228 | root_lock = qdisc_lock(q); |
229 | dev = qdisc_dev(q); | |
10c51b56 | 230 | txq = skb_get_tx_queue(dev, skb); |
bbd8a0d3 KK |
231 | |
232 | return sch_direct_xmit(skb, q, dev, txq, root_lock); | |
233 | } | |
234 | ||
37437bb2 | 235 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 236 | { |
d5b8aa1d | 237 | int quota = weight_p; |
2ba2506c | 238 | |
37437bb2 | 239 | while (qdisc_restart(q)) { |
2ba2506c | 240 | /* |
d5b8aa1d | 241 | * Ordered by possible occurrence: Postpone processing if |
242 | * 1. we've exceeded packet quota | |
243 | * 2. another process needs the CPU; | |
2ba2506c | 244 | */ |
f0c50c7c | 245 | if (--quota <= 0 || need_resched()) { |
37437bb2 | 246 | __netif_schedule(q); |
d90df3ad | 247 | break; |
2ba2506c HX |
248 | } |
249 | } | |
48d83325 | 250 | |
bc135b23 | 251 | qdisc_run_end(q); |
48d83325 HX |
252 | } |
253 | ||
9d21493b ED |
254 | unsigned long dev_trans_start(struct net_device *dev) |
255 | { | |
07ce76aa | 256 | unsigned long val, res; |
9d21493b ED |
257 | unsigned int i; |
258 | ||
07ce76aa | 259 | if (is_vlan_dev(dev)) |
260 | dev = vlan_dev_real_dev(dev); | |
261 | res = dev->trans_start; | |
9d21493b ED |
262 | for (i = 0; i < dev->num_tx_queues; i++) { |
263 | val = netdev_get_tx_queue(dev, i)->trans_start; | |
264 | if (val && time_after(val, res)) | |
265 | res = val; | |
266 | } | |
267 | dev->trans_start = res; | |
07ce76aa | 268 | |
9d21493b ED |
269 | return res; |
270 | } | |
271 | EXPORT_SYMBOL(dev_trans_start); | |
272 | ||
1da177e4 LT |
273 | static void dev_watchdog(unsigned long arg) |
274 | { | |
275 | struct net_device *dev = (struct net_device *)arg; | |
276 | ||
932ff279 | 277 | netif_tx_lock(dev); |
e8a0464c | 278 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
279 | if (netif_device_present(dev) && |
280 | netif_running(dev) && | |
281 | netif_carrier_ok(dev)) { | |
9d21493b | 282 | int some_queue_timedout = 0; |
e8a0464c | 283 | unsigned int i; |
9d21493b | 284 | unsigned long trans_start; |
e8a0464c DM |
285 | |
286 | for (i = 0; i < dev->num_tx_queues; i++) { | |
287 | struct netdev_queue *txq; | |
288 | ||
289 | txq = netdev_get_tx_queue(dev, i); | |
9d21493b ED |
290 | /* |
291 | * old device drivers set dev->trans_start | |
292 | */ | |
293 | trans_start = txq->trans_start ? : dev->trans_start; | |
73466498 | 294 | if (netif_xmit_stopped(txq) && |
9d21493b ED |
295 | time_after(jiffies, (trans_start + |
296 | dev->watchdog_timeo))) { | |
297 | some_queue_timedout = 1; | |
ccf5ff69 | 298 | txq->trans_timeout++; |
e8a0464c DM |
299 | break; |
300 | } | |
301 | } | |
338f7566 | 302 | |
9d21493b | 303 | if (some_queue_timedout) { |
9d21493b | 304 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 305 | dev->name, netdev_drivername(dev), i); |
d314774c | 306 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 307 | } |
e8a0464c DM |
308 | if (!mod_timer(&dev->watchdog_timer, |
309 | round_jiffies(jiffies + | |
310 | dev->watchdog_timeo))) | |
1da177e4 LT |
311 | dev_hold(dev); |
312 | } | |
313 | } | |
932ff279 | 314 | netif_tx_unlock(dev); |
1da177e4 LT |
315 | |
316 | dev_put(dev); | |
317 | } | |
318 | ||
1da177e4 LT |
319 | void __netdev_watchdog_up(struct net_device *dev) |
320 | { | |
d314774c | 321 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
322 | if (dev->watchdog_timeo <= 0) |
323 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
324 | if (!mod_timer(&dev->watchdog_timer, |
325 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
326 | dev_hold(dev); |
327 | } | |
328 | } | |
329 | ||
330 | static void dev_watchdog_up(struct net_device *dev) | |
331 | { | |
1da177e4 | 332 | __netdev_watchdog_up(dev); |
1da177e4 LT |
333 | } |
334 | ||
335 | static void dev_watchdog_down(struct net_device *dev) | |
336 | { | |
932ff279 | 337 | netif_tx_lock_bh(dev); |
1da177e4 | 338 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 339 | dev_put(dev); |
932ff279 | 340 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
341 | } |
342 | ||
bea3348e SH |
343 | /** |
344 | * netif_carrier_on - set carrier | |
345 | * @dev: network device | |
346 | * | |
347 | * Device has detected that carrier. | |
348 | */ | |
0a242efc DV |
349 | void netif_carrier_on(struct net_device *dev) |
350 | { | |
bfaae0f0 | 351 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
352 | if (dev->reg_state == NETREG_UNINITIALIZED) |
353 | return; | |
2d3b479d | 354 | atomic_inc(&dev->carrier_changes); |
0a242efc | 355 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
356 | if (netif_running(dev)) |
357 | __netdev_watchdog_up(dev); | |
358 | } | |
0a242efc | 359 | } |
62e3ba1b | 360 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 361 | |
bea3348e SH |
362 | /** |
363 | * netif_carrier_off - clear carrier | |
364 | * @dev: network device | |
365 | * | |
366 | * Device has detected loss of carrier. | |
367 | */ | |
0a242efc DV |
368 | void netif_carrier_off(struct net_device *dev) |
369 | { | |
b4730016 DM |
370 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
371 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
372 | return; | |
2d3b479d | 373 | atomic_inc(&dev->carrier_changes); |
0a242efc | 374 | linkwatch_fire_event(dev); |
b4730016 | 375 | } |
0a242efc | 376 | } |
62e3ba1b | 377 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 378 | |
1da177e4 LT |
379 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
380 | under all circumstances. It is difficult to invent anything faster or | |
381 | cheaper. | |
382 | */ | |
383 | ||
82d567c2 | 384 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc) |
1da177e4 LT |
385 | { |
386 | kfree_skb(skb); | |
387 | return NET_XMIT_CN; | |
388 | } | |
389 | ||
82d567c2 | 390 | static struct sk_buff *noop_dequeue(struct Qdisc *qdisc) |
1da177e4 LT |
391 | { |
392 | return NULL; | |
393 | } | |
394 | ||
20fea08b | 395 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
396 | .id = "noop", |
397 | .priv_size = 0, | |
398 | .enqueue = noop_enqueue, | |
399 | .dequeue = noop_dequeue, | |
99c0db26 | 400 | .peek = noop_dequeue, |
1da177e4 LT |
401 | .owner = THIS_MODULE, |
402 | }; | |
403 | ||
7698b4fc | 404 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 405 | .qdisc = &noop_qdisc, |
9f3ffae0 | 406 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
407 | }; |
408 | ||
1da177e4 LT |
409 | struct Qdisc noop_qdisc = { |
410 | .enqueue = noop_enqueue, | |
411 | .dequeue = noop_dequeue, | |
412 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 413 | .ops = &noop_qdisc_ops, |
1da177e4 | 414 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 415 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 416 | .dev_queue = &noop_netdev_queue, |
7b5edbc4 | 417 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
1da177e4 | 418 | }; |
62e3ba1b | 419 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 420 | |
20fea08b | 421 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
422 | .id = "noqueue", |
423 | .priv_size = 0, | |
424 | .enqueue = noop_enqueue, | |
425 | .dequeue = noop_dequeue, | |
99c0db26 | 426 | .peek = noop_dequeue, |
1da177e4 LT |
427 | .owner = THIS_MODULE, |
428 | }; | |
429 | ||
30ee42be DM |
430 | static struct Qdisc noqueue_qdisc; |
431 | static struct netdev_queue noqueue_netdev_queue = { | |
432 | .qdisc = &noqueue_qdisc, | |
9f3ffae0 | 433 | .qdisc_sleeping = &noqueue_qdisc, |
30ee42be DM |
434 | }; |
435 | ||
1da177e4 LT |
436 | static struct Qdisc noqueue_qdisc = { |
437 | .enqueue = NULL, | |
438 | .dequeue = noop_dequeue, | |
439 | .flags = TCQ_F_BUILTIN, | |
440 | .ops = &noqueue_qdisc_ops, | |
441 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
30ee42be DM |
442 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
443 | .dev_queue = &noqueue_netdev_queue, | |
7b5edbc4 | 444 | .busylock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock), |
1da177e4 LT |
445 | }; |
446 | ||
447 | ||
cc7ec456 ED |
448 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
449 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
450 | }; | |
d3678b46 DM |
451 | |
452 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
453 | generic prio+fifo combination. | |
454 | */ | |
455 | ||
456 | #define PFIFO_FAST_BANDS 3 | |
457 | ||
fd3ae5e8 KK |
458 | /* |
459 | * Private data for a pfifo_fast scheduler containing: | |
460 | * - queues for the three band | |
461 | * - bitmap indicating which of the bands contain skbs | |
462 | */ | |
463 | struct pfifo_fast_priv { | |
464 | u32 bitmap; | |
465 | struct sk_buff_head q[PFIFO_FAST_BANDS]; | |
466 | }; | |
467 | ||
468 | /* | |
469 | * Convert a bitmap to the first band number where an skb is queued, where: | |
470 | * bitmap=0 means there are no skbs on any band. | |
471 | * bitmap=1 means there is an skb on band 0. | |
472 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
473 | */ | |
474 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
475 | ||
476 | static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv, | |
477 | int band) | |
d3678b46 | 478 | { |
fd3ae5e8 | 479 | return priv->q + band; |
d3678b46 DM |
480 | } |
481 | ||
cc7ec456 | 482 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc) |
321090e7 | 483 | { |
a453e068 KK |
484 | if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) { |
485 | int band = prio2band[skb->priority & TC_PRIO_MAX]; | |
486 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
487 | struct sk_buff_head *list = band2list(priv, band); | |
1da177e4 | 488 | |
fd3ae5e8 | 489 | priv->bitmap |= (1 << band); |
d3678b46 | 490 | qdisc->q.qlen++; |
821d24ae | 491 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 492 | } |
821d24ae TG |
493 | |
494 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
495 | } |
496 | ||
cc7ec456 | 497 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc) |
1da177e4 | 498 | { |
fd3ae5e8 KK |
499 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
500 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 501 | |
fd3ae5e8 KK |
502 | if (likely(band >= 0)) { |
503 | struct sk_buff_head *list = band2list(priv, band); | |
504 | struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list); | |
505 | ||
506 | qdisc->q.qlen--; | |
507 | if (skb_queue_empty(list)) | |
508 | priv->bitmap &= ~(1 << band); | |
509 | ||
510 | return skb; | |
d3678b46 | 511 | } |
f87a9c3d | 512 | |
1da177e4 LT |
513 | return NULL; |
514 | } | |
515 | ||
cc7ec456 | 516 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 517 | { |
fd3ae5e8 KK |
518 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
519 | int band = bitmap2band[priv->bitmap]; | |
520 | ||
521 | if (band >= 0) { | |
522 | struct sk_buff_head *list = band2list(priv, band); | |
99c0db26 | 523 | |
fd3ae5e8 | 524 | return skb_peek(list); |
99c0db26 JP |
525 | } |
526 | ||
527 | return NULL; | |
528 | } | |
529 | ||
cc7ec456 | 530 | static void pfifo_fast_reset(struct Qdisc *qdisc) |
1da177e4 | 531 | { |
d3678b46 | 532 | int prio; |
fd3ae5e8 | 533 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
534 | |
535 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 536 | __qdisc_reset_queue(qdisc, band2list(priv, prio)); |
d3678b46 | 537 | |
fd3ae5e8 | 538 | priv->bitmap = 0; |
821d24ae | 539 | qdisc->qstats.backlog = 0; |
d3678b46 | 540 | qdisc->q.qlen = 0; |
1da177e4 LT |
541 | } |
542 | ||
d3678b46 DM |
543 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
544 | { | |
545 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
546 | ||
cc7ec456 | 547 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
1b34ec43 DM |
548 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
549 | goto nla_put_failure; | |
d3678b46 DM |
550 | return skb->len; |
551 | ||
552 | nla_put_failure: | |
553 | return -1; | |
554 | } | |
555 | ||
556 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
557 | { | |
558 | int prio; | |
fd3ae5e8 | 559 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
560 | |
561 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
ab34f648 | 562 | __skb_queue_head_init(band2list(priv, prio)); |
d3678b46 | 563 | |
23624935 ED |
564 | /* Can by-pass the queue discipline */ |
565 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
566 | return 0; |
567 | } | |
568 | ||
6ec1c69a | 569 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 570 | .id = "pfifo_fast", |
fd3ae5e8 | 571 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
572 | .enqueue = pfifo_fast_enqueue, |
573 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 574 | .peek = pfifo_fast_peek, |
d3678b46 DM |
575 | .init = pfifo_fast_init, |
576 | .reset = pfifo_fast_reset, | |
577 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
578 | .owner = THIS_MODULE, |
579 | }; | |
580 | ||
23d3b8bf ED |
581 | static struct lock_class_key qdisc_tx_busylock; |
582 | ||
5ce2d488 | 583 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
d2a7f269 | 584 | const struct Qdisc_ops *ops) |
1da177e4 LT |
585 | { |
586 | void *p; | |
587 | struct Qdisc *sch; | |
d276055c | 588 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 589 | int err = -ENOBUFS; |
23d3b8bf | 590 | struct net_device *dev = dev_queue->dev; |
1da177e4 | 591 | |
f2cd2d3e ED |
592 | p = kzalloc_node(size, GFP_KERNEL, |
593 | netdev_queue_numa_node_read(dev_queue)); | |
594 | ||
1da177e4 | 595 | if (!p) |
3d54b82f | 596 | goto errout; |
3d54b82f | 597 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
598 | /* if we got non aligned memory, ask more and do alignment ourself */ |
599 | if (sch != p) { | |
600 | kfree(p); | |
601 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
602 | netdev_queue_numa_node_read(dev_queue)); | |
603 | if (!p) | |
604 | goto errout; | |
605 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
606 | sch->padded = (char *) sch - (char *) p; | |
607 | } | |
1da177e4 LT |
608 | INIT_LIST_HEAD(&sch->list); |
609 | skb_queue_head_init(&sch->q); | |
23d3b8bf | 610 | |
79640a4c | 611 | spin_lock_init(&sch->busylock); |
23d3b8bf ED |
612 | lockdep_set_class(&sch->busylock, |
613 | dev->qdisc_tx_busylock ?: &qdisc_tx_busylock); | |
614 | ||
1da177e4 LT |
615 | sch->ops = ops; |
616 | sch->enqueue = ops->enqueue; | |
617 | sch->dequeue = ops->dequeue; | |
bb949fbd | 618 | sch->dev_queue = dev_queue; |
23d3b8bf | 619 | dev_hold(dev); |
1da177e4 | 620 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
621 | |
622 | return sch; | |
623 | errout: | |
01e123d7 | 624 | return ERR_PTR(err); |
3d54b82f TG |
625 | } |
626 | ||
3511c913 | 627 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
d2a7f269 | 628 | const struct Qdisc_ops *ops, |
629 | unsigned int parentid) | |
3d54b82f TG |
630 | { |
631 | struct Qdisc *sch; | |
10297b99 | 632 | |
6da7c8fc | 633 | if (!try_module_get(ops->owner)) |
634 | goto errout; | |
635 | ||
5ce2d488 | 636 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
637 | if (IS_ERR(sch)) |
638 | goto errout; | |
9f9afec4 | 639 | sch->parent = parentid; |
3d54b82f | 640 | |
1da177e4 LT |
641 | if (!ops->init || ops->init(sch, NULL) == 0) |
642 | return sch; | |
643 | ||
0fbbeb1b | 644 | qdisc_destroy(sch); |
3d54b82f | 645 | errout: |
1da177e4 LT |
646 | return NULL; |
647 | } | |
62e3ba1b | 648 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 649 | |
5fb66229 | 650 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
651 | |
652 | void qdisc_reset(struct Qdisc *qdisc) | |
653 | { | |
20fea08b | 654 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
655 | |
656 | if (ops->reset) | |
657 | ops->reset(qdisc); | |
67305ebc | 658 | |
bbd8a0d3 | 659 | if (qdisc->gso_skb) { |
3f3c7eec | 660 | kfree_skb_list(qdisc->gso_skb); |
bbd8a0d3 KK |
661 | qdisc->gso_skb = NULL; |
662 | qdisc->q.qlen = 0; | |
663 | } | |
1da177e4 | 664 | } |
62e3ba1b | 665 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 666 | |
5d944c64 ED |
667 | static void qdisc_rcu_free(struct rcu_head *head) |
668 | { | |
669 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
670 | ||
22e0f8b9 JF |
671 | if (qdisc_is_percpu_stats(qdisc)) |
672 | free_percpu(qdisc->cpu_bstats); | |
673 | ||
5d944c64 ED |
674 | kfree((char *) qdisc - qdisc->padded); |
675 | } | |
676 | ||
1e0d5a57 | 677 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 678 | { |
8a34c5dc DM |
679 | const struct Qdisc_ops *ops = qdisc->ops; |
680 | ||
1e0d5a57 DM |
681 | if (qdisc->flags & TCQ_F_BUILTIN || |
682 | !atomic_dec_and_test(&qdisc->refcnt)) | |
683 | return; | |
684 | ||
3a682fbd | 685 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
686 | qdisc_list_del(qdisc); |
687 | ||
a2da570d | 688 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 689 | #endif |
8a34c5dc DM |
690 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
691 | if (ops->reset) | |
692 | ops->reset(qdisc); | |
693 | if (ops->destroy) | |
694 | ops->destroy(qdisc); | |
695 | ||
696 | module_put(ops->owner); | |
697 | dev_put(qdisc_dev(qdisc)); | |
698 | ||
3f3c7eec | 699 | kfree_skb_list(qdisc->gso_skb); |
5d944c64 ED |
700 | /* |
701 | * gen_estimator est_timer() might access qdisc->q.lock, | |
702 | * wait a RCU grace period before freeing qdisc. | |
703 | */ | |
704 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 705 | } |
62e3ba1b | 706 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 707 | |
589983cd PM |
708 | /* Attach toplevel qdisc to device queue. */ |
709 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
710 | struct Qdisc *qdisc) | |
711 | { | |
712 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
713 | spinlock_t *root_lock; | |
714 | ||
715 | root_lock = qdisc_lock(oqdisc); | |
716 | spin_lock_bh(root_lock); | |
717 | ||
718 | /* Prune old scheduler */ | |
719 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
720 | qdisc_reset(oqdisc); | |
721 | ||
722 | /* ... and graft new one */ | |
723 | if (qdisc == NULL) | |
724 | qdisc = &noop_qdisc; | |
725 | dev_queue->qdisc_sleeping = qdisc; | |
726 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
727 | ||
728 | spin_unlock_bh(root_lock); | |
729 | ||
730 | return oqdisc; | |
731 | } | |
b8970f0b | 732 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 733 | |
e8a0464c DM |
734 | static void attach_one_default_qdisc(struct net_device *dev, |
735 | struct netdev_queue *dev_queue, | |
736 | void *_unused) | |
737 | { | |
cc7ec456 | 738 | struct Qdisc *qdisc = &noqueue_qdisc; |
e8a0464c DM |
739 | |
740 | if (dev->tx_queue_len) { | |
3511c913 | 741 | qdisc = qdisc_create_dflt(dev_queue, |
6da7c8fc | 742 | default_qdisc_ops, TC_H_ROOT); |
e8a0464c | 743 | if (!qdisc) { |
cc7ec456 | 744 | netdev_info(dev, "activation failed\n"); |
e8a0464c DM |
745 | return; |
746 | } | |
1abbe139 ED |
747 | if (!netif_is_multiqueue(dev)) |
748 | qdisc->flags |= TCQ_F_ONETXQUEUE; | |
e8a0464c DM |
749 | } |
750 | dev_queue->qdisc_sleeping = qdisc; | |
751 | } | |
752 | ||
6ec1c69a DM |
753 | static void attach_default_qdiscs(struct net_device *dev) |
754 | { | |
755 | struct netdev_queue *txq; | |
756 | struct Qdisc *qdisc; | |
757 | ||
758 | txq = netdev_get_tx_queue(dev, 0); | |
759 | ||
760 | if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) { | |
761 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); | |
762 | dev->qdisc = txq->qdisc_sleeping; | |
763 | atomic_inc(&dev->qdisc->refcnt); | |
764 | } else { | |
3511c913 | 765 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT); |
6ec1c69a | 766 | if (qdisc) { |
6ec1c69a | 767 | dev->qdisc = qdisc; |
e57a784d | 768 | qdisc->ops->attach(qdisc); |
6ec1c69a DM |
769 | } |
770 | } | |
771 | } | |
772 | ||
e8a0464c DM |
773 | static void transition_one_qdisc(struct net_device *dev, |
774 | struct netdev_queue *dev_queue, | |
775 | void *_need_watchdog) | |
776 | { | |
83874000 | 777 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
778 | int *need_watchdog_p = _need_watchdog; |
779 | ||
a9312ae8 DM |
780 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
781 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
782 | ||
83874000 | 783 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
9d21493b ED |
784 | if (need_watchdog_p && new_qdisc != &noqueue_qdisc) { |
785 | dev_queue->trans_start = 0; | |
e8a0464c | 786 | *need_watchdog_p = 1; |
9d21493b | 787 | } |
e8a0464c DM |
788 | } |
789 | ||
1da177e4 LT |
790 | void dev_activate(struct net_device *dev) |
791 | { | |
e8a0464c | 792 | int need_watchdog; |
b0e1e646 | 793 | |
1da177e4 | 794 | /* No queueing discipline is attached to device; |
6da7c8fc | 795 | * create default one for devices, which need queueing |
796 | * and noqueue_qdisc for virtual interfaces | |
1da177e4 LT |
797 | */ |
798 | ||
6ec1c69a DM |
799 | if (dev->qdisc == &noop_qdisc) |
800 | attach_default_qdiscs(dev); | |
af356afa | 801 | |
cacaddf5 TC |
802 | if (!netif_carrier_ok(dev)) |
803 | /* Delay activation until next carrier-on event */ | |
804 | return; | |
805 | ||
e8a0464c DM |
806 | need_watchdog = 0; |
807 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
24824a09 ED |
808 | if (dev_ingress_queue(dev)) |
809 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
810 | |
811 | if (need_watchdog) { | |
1da177e4 LT |
812 | dev->trans_start = jiffies; |
813 | dev_watchdog_up(dev); | |
814 | } | |
b0e1e646 | 815 | } |
b8970f0b | 816 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 817 | |
e8a0464c DM |
818 | static void dev_deactivate_queue(struct net_device *dev, |
819 | struct netdev_queue *dev_queue, | |
820 | void *_qdisc_default) | |
b0e1e646 | 821 | { |
e8a0464c | 822 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 823 | struct Qdisc *qdisc; |
970565bb | 824 | |
46e5da40 | 825 | qdisc = rtnl_dereference(dev_queue->qdisc); |
b0e1e646 | 826 | if (qdisc) { |
83874000 DM |
827 | spin_lock_bh(qdisc_lock(qdisc)); |
828 | ||
a9312ae8 DM |
829 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
830 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
831 | ||
f7a54c13 | 832 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 833 | qdisc_reset(qdisc); |
d3b753db | 834 | |
83874000 | 835 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 836 | } |
1da177e4 LT |
837 | } |
838 | ||
4335cd2d | 839 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
840 | { |
841 | unsigned int i; | |
842 | ||
843 | for (i = 0; i < dev->num_tx_queues; i++) { | |
844 | struct netdev_queue *dev_queue; | |
7698b4fc | 845 | spinlock_t *root_lock; |
e2627c8c | 846 | struct Qdisc *q; |
e8a0464c DM |
847 | int val; |
848 | ||
849 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 850 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 851 | root_lock = qdisc_lock(q); |
e8a0464c | 852 | |
4335cd2d | 853 | spin_lock_bh(root_lock); |
e8a0464c | 854 | |
bc135b23 | 855 | val = (qdisc_is_running(q) || |
b9a3b110 | 856 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 857 | |
4335cd2d | 858 | spin_unlock_bh(root_lock); |
e8a0464c DM |
859 | |
860 | if (val) | |
861 | return true; | |
862 | } | |
863 | return false; | |
864 | } | |
865 | ||
3137663d ED |
866 | /** |
867 | * dev_deactivate_many - deactivate transmissions on several devices | |
868 | * @head: list of devices to deactivate | |
869 | * | |
870 | * This function returns only when all outstanding transmissions | |
871 | * have completed, unless all devices are in dismantle phase. | |
872 | */ | |
44345724 | 873 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 874 | { |
44345724 | 875 | struct net_device *dev; |
3137663d | 876 | bool sync_needed = false; |
41a23b07 | 877 | |
5cde2829 | 878 | list_for_each_entry(dev, head, close_list) { |
44345724 OP |
879 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, |
880 | &noop_qdisc); | |
881 | if (dev_ingress_queue(dev)) | |
882 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
883 | &noop_qdisc); | |
884 | ||
885 | dev_watchdog_down(dev); | |
3137663d | 886 | sync_needed |= !dev->dismantle; |
44345724 | 887 | } |
1da177e4 | 888 | |
3137663d ED |
889 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
890 | * This is avoided if all devices are in dismantle phase : | |
891 | * Caller will call synchronize_net() for us | |
892 | */ | |
893 | if (sync_needed) | |
894 | synchronize_net(); | |
1da177e4 | 895 | |
d4828d85 | 896 | /* Wait for outstanding qdisc_run calls. */ |
5cde2829 | 897 | list_for_each_entry(dev, head, close_list) |
44345724 OP |
898 | while (some_qdisc_is_busy(dev)) |
899 | yield(); | |
900 | } | |
901 | ||
902 | void dev_deactivate(struct net_device *dev) | |
903 | { | |
904 | LIST_HEAD(single); | |
905 | ||
5cde2829 | 906 | list_add(&dev->close_list, &single); |
44345724 | 907 | dev_deactivate_many(&single); |
5f04d506 | 908 | list_del(&single); |
1da177e4 | 909 | } |
b8970f0b | 910 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 911 | |
b0e1e646 DM |
912 | static void dev_init_scheduler_queue(struct net_device *dev, |
913 | struct netdev_queue *dev_queue, | |
e8a0464c | 914 | void *_qdisc) |
b0e1e646 | 915 | { |
e8a0464c DM |
916 | struct Qdisc *qdisc = _qdisc; |
917 | ||
46e5da40 | 918 | rcu_assign_pointer(dev_queue->qdisc, qdisc); |
b0e1e646 | 919 | dev_queue->qdisc_sleeping = qdisc; |
b0e1e646 DM |
920 | } |
921 | ||
1da177e4 LT |
922 | void dev_init_scheduler(struct net_device *dev) |
923 | { | |
af356afa | 924 | dev->qdisc = &noop_qdisc; |
e8a0464c | 925 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
926 | if (dev_ingress_queue(dev)) |
927 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 928 | |
b24b8a24 | 929 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
930 | } |
931 | ||
e8a0464c DM |
932 | static void shutdown_scheduler_queue(struct net_device *dev, |
933 | struct netdev_queue *dev_queue, | |
934 | void *_qdisc_default) | |
1da177e4 | 935 | { |
b0e1e646 | 936 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 937 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
938 | |
939 | if (qdisc) { | |
f7a54c13 | 940 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 941 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 942 | |
1da177e4 | 943 | qdisc_destroy(qdisc); |
10297b99 | 944 | } |
b0e1e646 DM |
945 | } |
946 | ||
947 | void dev_shutdown(struct net_device *dev) | |
948 | { | |
e8a0464c | 949 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
950 | if (dev_ingress_queue(dev)) |
951 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
af356afa PM |
952 | qdisc_destroy(dev->qdisc); |
953 | dev->qdisc = &noop_qdisc; | |
954 | ||
547b792c | 955 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 956 | } |
292f1c7f | 957 | |
01cb71d2 | 958 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
3e1e3aae ED |
959 | const struct tc_ratespec *conf, |
960 | u64 rate64) | |
292f1c7f | 961 | { |
01cb71d2 ED |
962 | memset(r, 0, sizeof(*r)); |
963 | r->overhead = conf->overhead; | |
3e1e3aae | 964 | r->rate_bytes_ps = max_t(u64, conf->rate, rate64); |
8a8e3d84 | 965 | r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK); |
292f1c7f JP |
966 | r->mult = 1; |
967 | /* | |
130d3d68 ED |
968 | * The deal here is to replace a divide by a reciprocal one |
969 | * in fast path (a reciprocal divide is a multiply and a shift) | |
970 | * | |
971 | * Normal formula would be : | |
972 | * time_in_ns = (NSEC_PER_SEC * len) / rate_bps | |
973 | * | |
974 | * We compute mult/shift to use instead : | |
975 | * time_in_ns = (len * mult) >> shift; | |
976 | * | |
977 | * We try to get the highest possible mult value for accuracy, | |
978 | * but have to make sure no overflows will ever happen. | |
292f1c7f | 979 | */ |
130d3d68 ED |
980 | if (r->rate_bytes_ps > 0) { |
981 | u64 factor = NSEC_PER_SEC; | |
982 | ||
983 | for (;;) { | |
984 | r->mult = div64_u64(factor, r->rate_bytes_ps); | |
985 | if (r->mult & (1U << 31) || factor & (1ULL << 63)) | |
292f1c7f | 986 | break; |
130d3d68 ED |
987 | factor <<= 1; |
988 | r->shift++; | |
292f1c7f | 989 | } |
292f1c7f JP |
990 | } |
991 | } | |
992 | EXPORT_SYMBOL(psched_ratecfg_precompute); |