]>
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 | { |
6252352d | 50 | q->gso_skb = skb; |
53e91503 | 51 | q->qstats.requeues++; |
bbd8a0d3 | 52 | q->q.qlen++; /* it's still part of the queue */ |
37437bb2 | 53 | __netif_schedule(q); |
6252352d | 54 | |
c716a81a JHS |
55 | return 0; |
56 | } | |
57 | ||
55a93b3e ED |
58 | static void try_bulk_dequeue_skb(struct Qdisc *q, |
59 | struct sk_buff *skb, | |
60 | const struct netdev_queue *txq) | |
5772e9a3 | 61 | { |
55a93b3e | 62 | int bytelimit = qdisc_avail_bulklimit(txq) - skb->len; |
5772e9a3 JDB |
63 | |
64 | while (bytelimit > 0) { | |
55a93b3e | 65 | struct sk_buff *nskb = q->dequeue(q); |
5772e9a3 | 66 | |
55a93b3e | 67 | if (!nskb) |
5772e9a3 JDB |
68 | break; |
69 | ||
55a93b3e ED |
70 | bytelimit -= nskb->len; /* covers GSO len */ |
71 | skb->next = nskb; | |
72 | skb = nskb; | |
5772e9a3 | 73 | } |
55a93b3e | 74 | skb->next = NULL; |
5772e9a3 JDB |
75 | } |
76 | ||
77 | /* Note that dequeue_skb can possibly return a SKB list (via skb->next). | |
78 | * A requeued skb (via q->gso_skb) can also be a SKB list. | |
79 | */ | |
55a93b3e | 80 | static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate) |
c716a81a | 81 | { |
554794de | 82 | struct sk_buff *skb = q->gso_skb; |
1abbe139 | 83 | const struct netdev_queue *txq = q->dev_queue; |
554794de | 84 | |
55a93b3e | 85 | *validate = true; |
ebf05982 | 86 | if (unlikely(skb)) { |
ebf05982 | 87 | /* check the reason of requeuing without tx lock first */ |
10c51b56 | 88 | txq = skb_get_tx_queue(txq->dev, skb); |
73466498 | 89 | if (!netif_xmit_frozen_or_stopped(txq)) { |
6252352d | 90 | q->gso_skb = NULL; |
bbd8a0d3 KK |
91 | q->q.qlen--; |
92 | } else | |
ebf05982 | 93 | skb = NULL; |
55a93b3e ED |
94 | /* skb in gso_skb were already validated */ |
95 | *validate = false; | |
ebf05982 | 96 | } else { |
5772e9a3 JDB |
97 | if (!(q->flags & TCQ_F_ONETXQUEUE) || |
98 | !netif_xmit_frozen_or_stopped(txq)) { | |
1abbe139 | 99 | skb = q->dequeue(q); |
5772e9a3 | 100 | if (skb && qdisc_may_bulk(q)) |
55a93b3e | 101 | try_bulk_dequeue_skb(q, skb, txq); |
50cbe9ab | 102 | } |
ebf05982 | 103 | } |
c716a81a JHS |
104 | return skb; |
105 | } | |
106 | ||
6c1361a6 | 107 | static inline int handle_dev_cpu_collision(struct sk_buff *skb, |
970565bb | 108 | struct netdev_queue *dev_queue, |
6c1361a6 | 109 | struct Qdisc *q) |
c716a81a | 110 | { |
6c1361a6 | 111 | int ret; |
c716a81a | 112 | |
c773e847 | 113 | if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) { |
6c1361a6 KK |
114 | /* |
115 | * Same CPU holding the lock. It may be a transient | |
116 | * configuration error, when hard_start_xmit() recurses. We | |
117 | * detect it by checking xmit owner and drop the packet when | |
118 | * deadloop is detected. Return OK to try the next skb. | |
119 | */ | |
10770bc2 | 120 | kfree_skb_list(skb); |
e87cc472 JP |
121 | net_warn_ratelimited("Dead loop on netdevice %s, fix it urgently!\n", |
122 | dev_queue->dev->name); | |
6c1361a6 KK |
123 | ret = qdisc_qlen(q); |
124 | } else { | |
125 | /* | |
126 | * Another cpu is holding lock, requeue & delay xmits for | |
127 | * some time. | |
128 | */ | |
d6d9ca0f | 129 | __this_cpu_inc(softnet_data.cpu_collision); |
37437bb2 | 130 | ret = dev_requeue_skb(skb, q); |
c716a81a JHS |
131 | } |
132 | ||
6c1361a6 | 133 | return ret; |
c716a81a JHS |
134 | } |
135 | ||
10297b99 | 136 | /* |
10770bc2 JDB |
137 | * Transmit possibly several skbs, and handle the return status as |
138 | * required. Holding the __QDISC___STATE_RUNNING bit guarantees that | |
139 | * only one CPU can execute this function. | |
6c1361a6 KK |
140 | * |
141 | * Returns to the caller: | |
142 | * 0 - queue is empty or throttled. | |
143 | * >0 - queue is not empty. | |
6c1361a6 | 144 | */ |
bbd8a0d3 KK |
145 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
146 | struct net_device *dev, struct netdev_queue *txq, | |
55a93b3e | 147 | spinlock_t *root_lock, bool validate) |
1da177e4 | 148 | { |
5f1a485d | 149 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
150 | |
151 | /* And release qdisc */ | |
152 | spin_unlock(root_lock); | |
c716a81a | 153 | |
55a93b3e ED |
154 | /* Note that we validate skb (GSO, checksum, ...) outside of locks */ |
155 | if (validate) | |
156 | skb = validate_xmit_skb_list(skb, dev); | |
572a9d7b | 157 | |
55a93b3e ED |
158 | if (skb) { |
159 | HARD_TX_LOCK(dev, txq, smp_processor_id()); | |
160 | if (!netif_xmit_frozen_or_stopped(txq)) | |
161 | skb = dev_hard_start_xmit(skb, dev, txq, &ret); | |
c716a81a | 162 | |
55a93b3e ED |
163 | HARD_TX_UNLOCK(dev, txq); |
164 | } | |
7698b4fc | 165 | spin_lock(root_lock); |
c716a81a | 166 | |
9a1654ba JP |
167 | if (dev_xmit_complete(ret)) { |
168 | /* Driver sent out skb successfully or skb was consumed */ | |
6c1361a6 | 169 | ret = qdisc_qlen(q); |
9a1654ba | 170 | } else if (ret == NETDEV_TX_LOCKED) { |
6c1361a6 | 171 | /* Driver try lock failed */ |
eb6aafe3 | 172 | ret = handle_dev_cpu_collision(skb, txq, q); |
9a1654ba | 173 | } else { |
6c1361a6 | 174 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
e87cc472 JP |
175 | if (unlikely(ret != NETDEV_TX_BUSY)) |
176 | net_warn_ratelimited("BUG %s code %d qlen %d\n", | |
177 | dev->name, ret, q->q.qlen); | |
6c1361a6 | 178 | |
37437bb2 | 179 | ret = dev_requeue_skb(skb, q); |
6c1361a6 | 180 | } |
c716a81a | 181 | |
73466498 | 182 | if (ret && netif_xmit_frozen_or_stopped(txq)) |
37437bb2 DM |
183 | ret = 0; |
184 | ||
6c1361a6 | 185 | return ret; |
1da177e4 LT |
186 | } |
187 | ||
bbd8a0d3 KK |
188 | /* |
189 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. | |
190 | * | |
9bf2b8c2 | 191 | * __QDISC___STATE_RUNNING guarantees only one CPU can process |
bbd8a0d3 KK |
192 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for |
193 | * this queue. | |
194 | * | |
195 | * netif_tx_lock serializes accesses to device driver. | |
196 | * | |
197 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, | |
198 | * if one is grabbed, another must be free. | |
199 | * | |
200 | * Note, that this procedure can be called by a watchdog timer | |
201 | * | |
202 | * Returns to the caller: | |
203 | * 0 - queue is empty or throttled. | |
204 | * >0 - queue is not empty. | |
205 | * | |
206 | */ | |
207 | static inline int qdisc_restart(struct Qdisc *q) | |
208 | { | |
209 | struct netdev_queue *txq; | |
210 | struct net_device *dev; | |
211 | spinlock_t *root_lock; | |
212 | struct sk_buff *skb; | |
55a93b3e | 213 | bool validate; |
bbd8a0d3 KK |
214 | |
215 | /* Dequeue packet */ | |
55a93b3e | 216 | skb = dequeue_skb(q, &validate); |
bbd8a0d3 KK |
217 | if (unlikely(!skb)) |
218 | return 0; | |
10c51b56 | 219 | |
bbd8a0d3 KK |
220 | root_lock = qdisc_lock(q); |
221 | dev = qdisc_dev(q); | |
10c51b56 | 222 | txq = skb_get_tx_queue(dev, skb); |
bbd8a0d3 | 223 | |
55a93b3e | 224 | return sch_direct_xmit(skb, q, dev, txq, root_lock, validate); |
bbd8a0d3 KK |
225 | } |
226 | ||
37437bb2 | 227 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 228 | { |
d5b8aa1d | 229 | int quota = weight_p; |
2ba2506c | 230 | |
37437bb2 | 231 | while (qdisc_restart(q)) { |
2ba2506c | 232 | /* |
d5b8aa1d | 233 | * Ordered by possible occurrence: Postpone processing if |
234 | * 1. we've exceeded packet quota | |
235 | * 2. another process needs the CPU; | |
2ba2506c | 236 | */ |
f0c50c7c | 237 | if (--quota <= 0 || need_resched()) { |
37437bb2 | 238 | __netif_schedule(q); |
d90df3ad | 239 | break; |
2ba2506c HX |
240 | } |
241 | } | |
48d83325 | 242 | |
bc135b23 | 243 | qdisc_run_end(q); |
48d83325 HX |
244 | } |
245 | ||
9d21493b ED |
246 | unsigned long dev_trans_start(struct net_device *dev) |
247 | { | |
07ce76aa | 248 | unsigned long val, res; |
9d21493b ED |
249 | unsigned int i; |
250 | ||
07ce76aa | 251 | if (is_vlan_dev(dev)) |
252 | dev = vlan_dev_real_dev(dev); | |
253 | res = dev->trans_start; | |
9d21493b ED |
254 | for (i = 0; i < dev->num_tx_queues; i++) { |
255 | val = netdev_get_tx_queue(dev, i)->trans_start; | |
256 | if (val && time_after(val, res)) | |
257 | res = val; | |
258 | } | |
259 | dev->trans_start = res; | |
07ce76aa | 260 | |
9d21493b ED |
261 | return res; |
262 | } | |
263 | EXPORT_SYMBOL(dev_trans_start); | |
264 | ||
1da177e4 LT |
265 | static void dev_watchdog(unsigned long arg) |
266 | { | |
267 | struct net_device *dev = (struct net_device *)arg; | |
268 | ||
932ff279 | 269 | netif_tx_lock(dev); |
e8a0464c | 270 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
271 | if (netif_device_present(dev) && |
272 | netif_running(dev) && | |
273 | netif_carrier_ok(dev)) { | |
9d21493b | 274 | int some_queue_timedout = 0; |
e8a0464c | 275 | unsigned int i; |
9d21493b | 276 | unsigned long trans_start; |
e8a0464c DM |
277 | |
278 | for (i = 0; i < dev->num_tx_queues; i++) { | |
279 | struct netdev_queue *txq; | |
280 | ||
281 | txq = netdev_get_tx_queue(dev, i); | |
9d21493b ED |
282 | /* |
283 | * old device drivers set dev->trans_start | |
284 | */ | |
285 | trans_start = txq->trans_start ? : dev->trans_start; | |
73466498 | 286 | if (netif_xmit_stopped(txq) && |
9d21493b ED |
287 | time_after(jiffies, (trans_start + |
288 | dev->watchdog_timeo))) { | |
289 | some_queue_timedout = 1; | |
ccf5ff69 | 290 | txq->trans_timeout++; |
e8a0464c DM |
291 | break; |
292 | } | |
293 | } | |
338f7566 | 294 | |
9d21493b | 295 | if (some_queue_timedout) { |
9d21493b | 296 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 297 | dev->name, netdev_drivername(dev), i); |
d314774c | 298 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 299 | } |
e8a0464c DM |
300 | if (!mod_timer(&dev->watchdog_timer, |
301 | round_jiffies(jiffies + | |
302 | dev->watchdog_timeo))) | |
1da177e4 LT |
303 | dev_hold(dev); |
304 | } | |
305 | } | |
932ff279 | 306 | netif_tx_unlock(dev); |
1da177e4 LT |
307 | |
308 | dev_put(dev); | |
309 | } | |
310 | ||
1da177e4 LT |
311 | void __netdev_watchdog_up(struct net_device *dev) |
312 | { | |
d314774c | 313 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
314 | if (dev->watchdog_timeo <= 0) |
315 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
316 | if (!mod_timer(&dev->watchdog_timer, |
317 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
318 | dev_hold(dev); |
319 | } | |
320 | } | |
321 | ||
322 | static void dev_watchdog_up(struct net_device *dev) | |
323 | { | |
1da177e4 | 324 | __netdev_watchdog_up(dev); |
1da177e4 LT |
325 | } |
326 | ||
327 | static void dev_watchdog_down(struct net_device *dev) | |
328 | { | |
932ff279 | 329 | netif_tx_lock_bh(dev); |
1da177e4 | 330 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 331 | dev_put(dev); |
932ff279 | 332 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
333 | } |
334 | ||
bea3348e SH |
335 | /** |
336 | * netif_carrier_on - set carrier | |
337 | * @dev: network device | |
338 | * | |
339 | * Device has detected that carrier. | |
340 | */ | |
0a242efc DV |
341 | void netif_carrier_on(struct net_device *dev) |
342 | { | |
bfaae0f0 | 343 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
344 | if (dev->reg_state == NETREG_UNINITIALIZED) |
345 | return; | |
2d3b479d | 346 | atomic_inc(&dev->carrier_changes); |
0a242efc | 347 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
348 | if (netif_running(dev)) |
349 | __netdev_watchdog_up(dev); | |
350 | } | |
0a242efc | 351 | } |
62e3ba1b | 352 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 353 | |
bea3348e SH |
354 | /** |
355 | * netif_carrier_off - clear carrier | |
356 | * @dev: network device | |
357 | * | |
358 | * Device has detected loss of carrier. | |
359 | */ | |
0a242efc DV |
360 | void netif_carrier_off(struct net_device *dev) |
361 | { | |
b4730016 DM |
362 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
363 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
364 | return; | |
2d3b479d | 365 | atomic_inc(&dev->carrier_changes); |
0a242efc | 366 | linkwatch_fire_event(dev); |
b4730016 | 367 | } |
0a242efc | 368 | } |
62e3ba1b | 369 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 370 | |
1da177e4 LT |
371 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
372 | under all circumstances. It is difficult to invent anything faster or | |
373 | cheaper. | |
374 | */ | |
375 | ||
82d567c2 | 376 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc) |
1da177e4 LT |
377 | { |
378 | kfree_skb(skb); | |
379 | return NET_XMIT_CN; | |
380 | } | |
381 | ||
82d567c2 | 382 | static struct sk_buff *noop_dequeue(struct Qdisc *qdisc) |
1da177e4 LT |
383 | { |
384 | return NULL; | |
385 | } | |
386 | ||
20fea08b | 387 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
388 | .id = "noop", |
389 | .priv_size = 0, | |
390 | .enqueue = noop_enqueue, | |
391 | .dequeue = noop_dequeue, | |
99c0db26 | 392 | .peek = noop_dequeue, |
1da177e4 LT |
393 | .owner = THIS_MODULE, |
394 | }; | |
395 | ||
7698b4fc | 396 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 397 | .qdisc = &noop_qdisc, |
9f3ffae0 | 398 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
399 | }; |
400 | ||
1da177e4 LT |
401 | struct Qdisc noop_qdisc = { |
402 | .enqueue = noop_enqueue, | |
403 | .dequeue = noop_dequeue, | |
404 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 405 | .ops = &noop_qdisc_ops, |
1da177e4 | 406 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 407 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 408 | .dev_queue = &noop_netdev_queue, |
7b5edbc4 | 409 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
1da177e4 | 410 | }; |
62e3ba1b | 411 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 412 | |
20fea08b | 413 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
414 | .id = "noqueue", |
415 | .priv_size = 0, | |
416 | .enqueue = noop_enqueue, | |
417 | .dequeue = noop_dequeue, | |
99c0db26 | 418 | .peek = noop_dequeue, |
1da177e4 LT |
419 | .owner = THIS_MODULE, |
420 | }; | |
421 | ||
30ee42be DM |
422 | static struct Qdisc noqueue_qdisc; |
423 | static struct netdev_queue noqueue_netdev_queue = { | |
424 | .qdisc = &noqueue_qdisc, | |
9f3ffae0 | 425 | .qdisc_sleeping = &noqueue_qdisc, |
30ee42be DM |
426 | }; |
427 | ||
1da177e4 LT |
428 | static struct Qdisc noqueue_qdisc = { |
429 | .enqueue = NULL, | |
430 | .dequeue = noop_dequeue, | |
431 | .flags = TCQ_F_BUILTIN, | |
432 | .ops = &noqueue_qdisc_ops, | |
433 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
30ee42be DM |
434 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
435 | .dev_queue = &noqueue_netdev_queue, | |
7b5edbc4 | 436 | .busylock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock), |
1da177e4 LT |
437 | }; |
438 | ||
439 | ||
cc7ec456 ED |
440 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
441 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
442 | }; | |
d3678b46 DM |
443 | |
444 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
445 | generic prio+fifo combination. | |
446 | */ | |
447 | ||
448 | #define PFIFO_FAST_BANDS 3 | |
449 | ||
fd3ae5e8 KK |
450 | /* |
451 | * Private data for a pfifo_fast scheduler containing: | |
452 | * - queues for the three band | |
453 | * - bitmap indicating which of the bands contain skbs | |
454 | */ | |
455 | struct pfifo_fast_priv { | |
456 | u32 bitmap; | |
457 | struct sk_buff_head q[PFIFO_FAST_BANDS]; | |
458 | }; | |
459 | ||
460 | /* | |
461 | * Convert a bitmap to the first band number where an skb is queued, where: | |
462 | * bitmap=0 means there are no skbs on any band. | |
463 | * bitmap=1 means there is an skb on band 0. | |
464 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
465 | */ | |
466 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
467 | ||
468 | static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv, | |
469 | int band) | |
d3678b46 | 470 | { |
fd3ae5e8 | 471 | return priv->q + band; |
d3678b46 DM |
472 | } |
473 | ||
cc7ec456 | 474 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc) |
321090e7 | 475 | { |
a453e068 KK |
476 | if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) { |
477 | int band = prio2band[skb->priority & TC_PRIO_MAX]; | |
478 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
479 | struct sk_buff_head *list = band2list(priv, band); | |
1da177e4 | 480 | |
fd3ae5e8 | 481 | priv->bitmap |= (1 << band); |
d3678b46 | 482 | qdisc->q.qlen++; |
821d24ae | 483 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 484 | } |
821d24ae TG |
485 | |
486 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
487 | } |
488 | ||
cc7ec456 | 489 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc) |
1da177e4 | 490 | { |
fd3ae5e8 KK |
491 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
492 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 493 | |
fd3ae5e8 KK |
494 | if (likely(band >= 0)) { |
495 | struct sk_buff_head *list = band2list(priv, band); | |
496 | struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list); | |
497 | ||
498 | qdisc->q.qlen--; | |
499 | if (skb_queue_empty(list)) | |
500 | priv->bitmap &= ~(1 << band); | |
501 | ||
502 | return skb; | |
d3678b46 | 503 | } |
f87a9c3d | 504 | |
1da177e4 LT |
505 | return NULL; |
506 | } | |
507 | ||
cc7ec456 | 508 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 509 | { |
fd3ae5e8 KK |
510 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
511 | int band = bitmap2band[priv->bitmap]; | |
512 | ||
513 | if (band >= 0) { | |
514 | struct sk_buff_head *list = band2list(priv, band); | |
99c0db26 | 515 | |
fd3ae5e8 | 516 | return skb_peek(list); |
99c0db26 JP |
517 | } |
518 | ||
519 | return NULL; | |
520 | } | |
521 | ||
cc7ec456 | 522 | static void pfifo_fast_reset(struct Qdisc *qdisc) |
1da177e4 | 523 | { |
d3678b46 | 524 | int prio; |
fd3ae5e8 | 525 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
526 | |
527 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 528 | __qdisc_reset_queue(qdisc, band2list(priv, prio)); |
d3678b46 | 529 | |
fd3ae5e8 | 530 | priv->bitmap = 0; |
821d24ae | 531 | qdisc->qstats.backlog = 0; |
d3678b46 | 532 | qdisc->q.qlen = 0; |
1da177e4 LT |
533 | } |
534 | ||
d3678b46 DM |
535 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
536 | { | |
537 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
538 | ||
cc7ec456 | 539 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
1b34ec43 DM |
540 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
541 | goto nla_put_failure; | |
d3678b46 DM |
542 | return skb->len; |
543 | ||
544 | nla_put_failure: | |
545 | return -1; | |
546 | } | |
547 | ||
548 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
549 | { | |
550 | int prio; | |
fd3ae5e8 | 551 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
552 | |
553 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
ab34f648 | 554 | __skb_queue_head_init(band2list(priv, prio)); |
d3678b46 | 555 | |
23624935 ED |
556 | /* Can by-pass the queue discipline */ |
557 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
558 | return 0; |
559 | } | |
560 | ||
6ec1c69a | 561 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 562 | .id = "pfifo_fast", |
fd3ae5e8 | 563 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
564 | .enqueue = pfifo_fast_enqueue, |
565 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 566 | .peek = pfifo_fast_peek, |
d3678b46 DM |
567 | .init = pfifo_fast_init, |
568 | .reset = pfifo_fast_reset, | |
569 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
570 | .owner = THIS_MODULE, |
571 | }; | |
572 | ||
23d3b8bf ED |
573 | static struct lock_class_key qdisc_tx_busylock; |
574 | ||
5ce2d488 | 575 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
d2a7f269 | 576 | const struct Qdisc_ops *ops) |
1da177e4 LT |
577 | { |
578 | void *p; | |
579 | struct Qdisc *sch; | |
d276055c | 580 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 581 | int err = -ENOBUFS; |
23d3b8bf | 582 | struct net_device *dev = dev_queue->dev; |
1da177e4 | 583 | |
f2cd2d3e ED |
584 | p = kzalloc_node(size, GFP_KERNEL, |
585 | netdev_queue_numa_node_read(dev_queue)); | |
586 | ||
1da177e4 | 587 | if (!p) |
3d54b82f | 588 | goto errout; |
3d54b82f | 589 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
590 | /* if we got non aligned memory, ask more and do alignment ourself */ |
591 | if (sch != p) { | |
592 | kfree(p); | |
593 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
594 | netdev_queue_numa_node_read(dev_queue)); | |
595 | if (!p) | |
596 | goto errout; | |
597 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
598 | sch->padded = (char *) sch - (char *) p; | |
599 | } | |
1da177e4 LT |
600 | INIT_LIST_HEAD(&sch->list); |
601 | skb_queue_head_init(&sch->q); | |
23d3b8bf | 602 | |
79640a4c | 603 | spin_lock_init(&sch->busylock); |
23d3b8bf ED |
604 | lockdep_set_class(&sch->busylock, |
605 | dev->qdisc_tx_busylock ?: &qdisc_tx_busylock); | |
606 | ||
1da177e4 LT |
607 | sch->ops = ops; |
608 | sch->enqueue = ops->enqueue; | |
609 | sch->dequeue = ops->dequeue; | |
bb949fbd | 610 | sch->dev_queue = dev_queue; |
23d3b8bf | 611 | dev_hold(dev); |
1da177e4 | 612 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
613 | |
614 | return sch; | |
615 | errout: | |
01e123d7 | 616 | return ERR_PTR(err); |
3d54b82f TG |
617 | } |
618 | ||
3511c913 | 619 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
d2a7f269 | 620 | const struct Qdisc_ops *ops, |
621 | unsigned int parentid) | |
3d54b82f TG |
622 | { |
623 | struct Qdisc *sch; | |
10297b99 | 624 | |
6da7c8fc | 625 | if (!try_module_get(ops->owner)) |
626 | goto errout; | |
627 | ||
5ce2d488 | 628 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
629 | if (IS_ERR(sch)) |
630 | goto errout; | |
9f9afec4 | 631 | sch->parent = parentid; |
3d54b82f | 632 | |
1da177e4 LT |
633 | if (!ops->init || ops->init(sch, NULL) == 0) |
634 | return sch; | |
635 | ||
0fbbeb1b | 636 | qdisc_destroy(sch); |
3d54b82f | 637 | errout: |
1da177e4 LT |
638 | return NULL; |
639 | } | |
62e3ba1b | 640 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 641 | |
5fb66229 | 642 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
643 | |
644 | void qdisc_reset(struct Qdisc *qdisc) | |
645 | { | |
20fea08b | 646 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
647 | |
648 | if (ops->reset) | |
649 | ops->reset(qdisc); | |
67305ebc | 650 | |
bbd8a0d3 | 651 | if (qdisc->gso_skb) { |
3f3c7eec | 652 | kfree_skb_list(qdisc->gso_skb); |
bbd8a0d3 KK |
653 | qdisc->gso_skb = NULL; |
654 | qdisc->q.qlen = 0; | |
655 | } | |
1da177e4 | 656 | } |
62e3ba1b | 657 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 658 | |
5d944c64 ED |
659 | static void qdisc_rcu_free(struct rcu_head *head) |
660 | { | |
661 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
662 | ||
22e0f8b9 JF |
663 | if (qdisc_is_percpu_stats(qdisc)) |
664 | free_percpu(qdisc->cpu_bstats); | |
665 | ||
5d944c64 ED |
666 | kfree((char *) qdisc - qdisc->padded); |
667 | } | |
668 | ||
1e0d5a57 | 669 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 670 | { |
8a34c5dc DM |
671 | const struct Qdisc_ops *ops = qdisc->ops; |
672 | ||
1e0d5a57 DM |
673 | if (qdisc->flags & TCQ_F_BUILTIN || |
674 | !atomic_dec_and_test(&qdisc->refcnt)) | |
675 | return; | |
676 | ||
3a682fbd | 677 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
678 | qdisc_list_del(qdisc); |
679 | ||
a2da570d | 680 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 681 | #endif |
8a34c5dc DM |
682 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
683 | if (ops->reset) | |
684 | ops->reset(qdisc); | |
685 | if (ops->destroy) | |
686 | ops->destroy(qdisc); | |
687 | ||
688 | module_put(ops->owner); | |
689 | dev_put(qdisc_dev(qdisc)); | |
690 | ||
3f3c7eec | 691 | kfree_skb_list(qdisc->gso_skb); |
5d944c64 ED |
692 | /* |
693 | * gen_estimator est_timer() might access qdisc->q.lock, | |
694 | * wait a RCU grace period before freeing qdisc. | |
695 | */ | |
696 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 697 | } |
62e3ba1b | 698 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 699 | |
589983cd PM |
700 | /* Attach toplevel qdisc to device queue. */ |
701 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
702 | struct Qdisc *qdisc) | |
703 | { | |
704 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
705 | spinlock_t *root_lock; | |
706 | ||
707 | root_lock = qdisc_lock(oqdisc); | |
708 | spin_lock_bh(root_lock); | |
709 | ||
710 | /* Prune old scheduler */ | |
711 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
712 | qdisc_reset(oqdisc); | |
713 | ||
714 | /* ... and graft new one */ | |
715 | if (qdisc == NULL) | |
716 | qdisc = &noop_qdisc; | |
717 | dev_queue->qdisc_sleeping = qdisc; | |
718 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
719 | ||
720 | spin_unlock_bh(root_lock); | |
721 | ||
722 | return oqdisc; | |
723 | } | |
b8970f0b | 724 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 725 | |
e8a0464c DM |
726 | static void attach_one_default_qdisc(struct net_device *dev, |
727 | struct netdev_queue *dev_queue, | |
728 | void *_unused) | |
729 | { | |
cc7ec456 | 730 | struct Qdisc *qdisc = &noqueue_qdisc; |
e8a0464c DM |
731 | |
732 | if (dev->tx_queue_len) { | |
3511c913 | 733 | qdisc = qdisc_create_dflt(dev_queue, |
6da7c8fc | 734 | default_qdisc_ops, TC_H_ROOT); |
e8a0464c | 735 | if (!qdisc) { |
cc7ec456 | 736 | netdev_info(dev, "activation failed\n"); |
e8a0464c DM |
737 | return; |
738 | } | |
1abbe139 ED |
739 | if (!netif_is_multiqueue(dev)) |
740 | qdisc->flags |= TCQ_F_ONETXQUEUE; | |
e8a0464c DM |
741 | } |
742 | dev_queue->qdisc_sleeping = qdisc; | |
743 | } | |
744 | ||
6ec1c69a DM |
745 | static void attach_default_qdiscs(struct net_device *dev) |
746 | { | |
747 | struct netdev_queue *txq; | |
748 | struct Qdisc *qdisc; | |
749 | ||
750 | txq = netdev_get_tx_queue(dev, 0); | |
751 | ||
752 | if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) { | |
753 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); | |
754 | dev->qdisc = txq->qdisc_sleeping; | |
755 | atomic_inc(&dev->qdisc->refcnt); | |
756 | } else { | |
3511c913 | 757 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT); |
6ec1c69a | 758 | if (qdisc) { |
6ec1c69a | 759 | dev->qdisc = qdisc; |
e57a784d | 760 | qdisc->ops->attach(qdisc); |
6ec1c69a DM |
761 | } |
762 | } | |
763 | } | |
764 | ||
e8a0464c DM |
765 | static void transition_one_qdisc(struct net_device *dev, |
766 | struct netdev_queue *dev_queue, | |
767 | void *_need_watchdog) | |
768 | { | |
83874000 | 769 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
770 | int *need_watchdog_p = _need_watchdog; |
771 | ||
a9312ae8 DM |
772 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
773 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
774 | ||
83874000 | 775 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
9d21493b ED |
776 | if (need_watchdog_p && new_qdisc != &noqueue_qdisc) { |
777 | dev_queue->trans_start = 0; | |
e8a0464c | 778 | *need_watchdog_p = 1; |
9d21493b | 779 | } |
e8a0464c DM |
780 | } |
781 | ||
1da177e4 LT |
782 | void dev_activate(struct net_device *dev) |
783 | { | |
e8a0464c | 784 | int need_watchdog; |
b0e1e646 | 785 | |
1da177e4 | 786 | /* No queueing discipline is attached to device; |
6da7c8fc | 787 | * create default one for devices, which need queueing |
788 | * and noqueue_qdisc for virtual interfaces | |
1da177e4 LT |
789 | */ |
790 | ||
6ec1c69a DM |
791 | if (dev->qdisc == &noop_qdisc) |
792 | attach_default_qdiscs(dev); | |
af356afa | 793 | |
cacaddf5 TC |
794 | if (!netif_carrier_ok(dev)) |
795 | /* Delay activation until next carrier-on event */ | |
796 | return; | |
797 | ||
e8a0464c DM |
798 | need_watchdog = 0; |
799 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
24824a09 ED |
800 | if (dev_ingress_queue(dev)) |
801 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
802 | |
803 | if (need_watchdog) { | |
1da177e4 LT |
804 | dev->trans_start = jiffies; |
805 | dev_watchdog_up(dev); | |
806 | } | |
b0e1e646 | 807 | } |
b8970f0b | 808 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 809 | |
e8a0464c DM |
810 | static void dev_deactivate_queue(struct net_device *dev, |
811 | struct netdev_queue *dev_queue, | |
812 | void *_qdisc_default) | |
b0e1e646 | 813 | { |
e8a0464c | 814 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 815 | struct Qdisc *qdisc; |
970565bb | 816 | |
46e5da40 | 817 | qdisc = rtnl_dereference(dev_queue->qdisc); |
b0e1e646 | 818 | if (qdisc) { |
83874000 DM |
819 | spin_lock_bh(qdisc_lock(qdisc)); |
820 | ||
a9312ae8 DM |
821 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
822 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
823 | ||
f7a54c13 | 824 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 825 | qdisc_reset(qdisc); |
d3b753db | 826 | |
83874000 | 827 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 828 | } |
1da177e4 LT |
829 | } |
830 | ||
4335cd2d | 831 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
832 | { |
833 | unsigned int i; | |
834 | ||
835 | for (i = 0; i < dev->num_tx_queues; i++) { | |
836 | struct netdev_queue *dev_queue; | |
7698b4fc | 837 | spinlock_t *root_lock; |
e2627c8c | 838 | struct Qdisc *q; |
e8a0464c DM |
839 | int val; |
840 | ||
841 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 842 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 843 | root_lock = qdisc_lock(q); |
e8a0464c | 844 | |
4335cd2d | 845 | spin_lock_bh(root_lock); |
e8a0464c | 846 | |
bc135b23 | 847 | val = (qdisc_is_running(q) || |
b9a3b110 | 848 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 849 | |
4335cd2d | 850 | spin_unlock_bh(root_lock); |
e8a0464c DM |
851 | |
852 | if (val) | |
853 | return true; | |
854 | } | |
855 | return false; | |
856 | } | |
857 | ||
3137663d ED |
858 | /** |
859 | * dev_deactivate_many - deactivate transmissions on several devices | |
860 | * @head: list of devices to deactivate | |
861 | * | |
862 | * This function returns only when all outstanding transmissions | |
863 | * have completed, unless all devices are in dismantle phase. | |
864 | */ | |
44345724 | 865 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 866 | { |
44345724 | 867 | struct net_device *dev; |
3137663d | 868 | bool sync_needed = false; |
41a23b07 | 869 | |
5cde2829 | 870 | list_for_each_entry(dev, head, close_list) { |
44345724 OP |
871 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, |
872 | &noop_qdisc); | |
873 | if (dev_ingress_queue(dev)) | |
874 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
875 | &noop_qdisc); | |
876 | ||
877 | dev_watchdog_down(dev); | |
3137663d | 878 | sync_needed |= !dev->dismantle; |
44345724 | 879 | } |
1da177e4 | 880 | |
3137663d ED |
881 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
882 | * This is avoided if all devices are in dismantle phase : | |
883 | * Caller will call synchronize_net() for us | |
884 | */ | |
885 | if (sync_needed) | |
886 | synchronize_net(); | |
1da177e4 | 887 | |
d4828d85 | 888 | /* Wait for outstanding qdisc_run calls. */ |
5cde2829 | 889 | list_for_each_entry(dev, head, close_list) |
44345724 OP |
890 | while (some_qdisc_is_busy(dev)) |
891 | yield(); | |
892 | } | |
893 | ||
894 | void dev_deactivate(struct net_device *dev) | |
895 | { | |
896 | LIST_HEAD(single); | |
897 | ||
5cde2829 | 898 | list_add(&dev->close_list, &single); |
44345724 | 899 | dev_deactivate_many(&single); |
5f04d506 | 900 | list_del(&single); |
1da177e4 | 901 | } |
b8970f0b | 902 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 903 | |
b0e1e646 DM |
904 | static void dev_init_scheduler_queue(struct net_device *dev, |
905 | struct netdev_queue *dev_queue, | |
e8a0464c | 906 | void *_qdisc) |
b0e1e646 | 907 | { |
e8a0464c DM |
908 | struct Qdisc *qdisc = _qdisc; |
909 | ||
46e5da40 | 910 | rcu_assign_pointer(dev_queue->qdisc, qdisc); |
b0e1e646 | 911 | dev_queue->qdisc_sleeping = qdisc; |
b0e1e646 DM |
912 | } |
913 | ||
1da177e4 LT |
914 | void dev_init_scheduler(struct net_device *dev) |
915 | { | |
af356afa | 916 | dev->qdisc = &noop_qdisc; |
e8a0464c | 917 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
918 | if (dev_ingress_queue(dev)) |
919 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 920 | |
b24b8a24 | 921 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
922 | } |
923 | ||
e8a0464c DM |
924 | static void shutdown_scheduler_queue(struct net_device *dev, |
925 | struct netdev_queue *dev_queue, | |
926 | void *_qdisc_default) | |
1da177e4 | 927 | { |
b0e1e646 | 928 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 929 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
930 | |
931 | if (qdisc) { | |
f7a54c13 | 932 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 933 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 934 | |
1da177e4 | 935 | qdisc_destroy(qdisc); |
10297b99 | 936 | } |
b0e1e646 DM |
937 | } |
938 | ||
939 | void dev_shutdown(struct net_device *dev) | |
940 | { | |
e8a0464c | 941 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
942 | if (dev_ingress_queue(dev)) |
943 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
af356afa PM |
944 | qdisc_destroy(dev->qdisc); |
945 | dev->qdisc = &noop_qdisc; | |
946 | ||
547b792c | 947 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 948 | } |
292f1c7f | 949 | |
01cb71d2 | 950 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
3e1e3aae ED |
951 | const struct tc_ratespec *conf, |
952 | u64 rate64) | |
292f1c7f | 953 | { |
01cb71d2 ED |
954 | memset(r, 0, sizeof(*r)); |
955 | r->overhead = conf->overhead; | |
3e1e3aae | 956 | r->rate_bytes_ps = max_t(u64, conf->rate, rate64); |
8a8e3d84 | 957 | r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK); |
292f1c7f JP |
958 | r->mult = 1; |
959 | /* | |
130d3d68 ED |
960 | * The deal here is to replace a divide by a reciprocal one |
961 | * in fast path (a reciprocal divide is a multiply and a shift) | |
962 | * | |
963 | * Normal formula would be : | |
964 | * time_in_ns = (NSEC_PER_SEC * len) / rate_bps | |
965 | * | |
966 | * We compute mult/shift to use instead : | |
967 | * time_in_ns = (len * mult) >> shift; | |
968 | * | |
969 | * We try to get the highest possible mult value for accuracy, | |
970 | * but have to make sure no overflows will ever happen. | |
292f1c7f | 971 | */ |
130d3d68 ED |
972 | if (r->rate_bytes_ps > 0) { |
973 | u64 factor = NSEC_PER_SEC; | |
974 | ||
975 | for (;;) { | |
976 | r->mult = div64_u64(factor, r->rate_bytes_ps); | |
977 | if (r->mult & (1U << 31) || factor & (1ULL << 63)) | |
292f1c7f | 978 | break; |
130d3d68 ED |
979 | factor <<= 1; |
980 | r->shift++; | |
292f1c7f | 981 | } |
292f1c7f JP |
982 | } |
983 | } | |
984 | EXPORT_SYMBOL(psched_ratecfg_precompute); |