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, | |
b8358d70 JDB |
60 | const struct netdev_queue *txq, |
61 | int *packets) | |
5772e9a3 | 62 | { |
55a93b3e | 63 | int bytelimit = qdisc_avail_bulklimit(txq) - skb->len; |
5772e9a3 JDB |
64 | |
65 | while (bytelimit > 0) { | |
55a93b3e | 66 | struct sk_buff *nskb = q->dequeue(q); |
5772e9a3 | 67 | |
55a93b3e | 68 | if (!nskb) |
5772e9a3 JDB |
69 | break; |
70 | ||
55a93b3e ED |
71 | bytelimit -= nskb->len; /* covers GSO len */ |
72 | skb->next = nskb; | |
73 | skb = nskb; | |
b8358d70 | 74 | (*packets)++; /* GSO counts as one pkt */ |
5772e9a3 | 75 | } |
55a93b3e | 76 | skb->next = NULL; |
5772e9a3 JDB |
77 | } |
78 | ||
79 | /* Note that dequeue_skb can possibly return a SKB list (via skb->next). | |
80 | * A requeued skb (via q->gso_skb) can also be a SKB list. | |
81 | */ | |
b8358d70 JDB |
82 | static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate, |
83 | int *packets) | |
c716a81a | 84 | { |
554794de | 85 | struct sk_buff *skb = q->gso_skb; |
1abbe139 | 86 | const struct netdev_queue *txq = q->dev_queue; |
554794de | 87 | |
b8358d70 | 88 | *packets = 1; |
55a93b3e | 89 | *validate = true; |
ebf05982 | 90 | if (unlikely(skb)) { |
ebf05982 | 91 | /* check the reason of requeuing without tx lock first */ |
10c51b56 | 92 | txq = skb_get_tx_queue(txq->dev, skb); |
73466498 | 93 | if (!netif_xmit_frozen_or_stopped(txq)) { |
6252352d | 94 | q->gso_skb = NULL; |
bbd8a0d3 KK |
95 | q->q.qlen--; |
96 | } else | |
ebf05982 | 97 | skb = NULL; |
55a93b3e ED |
98 | /* skb in gso_skb were already validated */ |
99 | *validate = false; | |
ebf05982 | 100 | } else { |
5772e9a3 JDB |
101 | if (!(q->flags & TCQ_F_ONETXQUEUE) || |
102 | !netif_xmit_frozen_or_stopped(txq)) { | |
1abbe139 | 103 | skb = q->dequeue(q); |
5772e9a3 | 104 | if (skb && qdisc_may_bulk(q)) |
b8358d70 | 105 | try_bulk_dequeue_skb(q, skb, txq, packets); |
50cbe9ab | 106 | } |
ebf05982 | 107 | } |
c716a81a JHS |
108 | return skb; |
109 | } | |
110 | ||
6c1361a6 | 111 | static inline int handle_dev_cpu_collision(struct sk_buff *skb, |
970565bb | 112 | struct netdev_queue *dev_queue, |
6c1361a6 | 113 | struct Qdisc *q) |
c716a81a | 114 | { |
6c1361a6 | 115 | int ret; |
c716a81a | 116 | |
c773e847 | 117 | if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) { |
6c1361a6 KK |
118 | /* |
119 | * Same CPU holding the lock. It may be a transient | |
120 | * configuration error, when hard_start_xmit() recurses. We | |
121 | * detect it by checking xmit owner and drop the packet when | |
122 | * deadloop is detected. Return OK to try the next skb. | |
123 | */ | |
10770bc2 | 124 | kfree_skb_list(skb); |
e87cc472 JP |
125 | net_warn_ratelimited("Dead loop on netdevice %s, fix it urgently!\n", |
126 | dev_queue->dev->name); | |
6c1361a6 KK |
127 | ret = qdisc_qlen(q); |
128 | } else { | |
129 | /* | |
130 | * Another cpu is holding lock, requeue & delay xmits for | |
131 | * some time. | |
132 | */ | |
d6d9ca0f | 133 | __this_cpu_inc(softnet_data.cpu_collision); |
37437bb2 | 134 | ret = dev_requeue_skb(skb, q); |
c716a81a JHS |
135 | } |
136 | ||
6c1361a6 | 137 | return ret; |
c716a81a JHS |
138 | } |
139 | ||
10297b99 | 140 | /* |
10770bc2 JDB |
141 | * Transmit possibly several skbs, and handle the return status as |
142 | * required. Holding the __QDISC___STATE_RUNNING bit guarantees that | |
143 | * only one CPU can execute this function. | |
6c1361a6 KK |
144 | * |
145 | * Returns to the caller: | |
146 | * 0 - queue is empty or throttled. | |
147 | * >0 - queue is not empty. | |
6c1361a6 | 148 | */ |
bbd8a0d3 KK |
149 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
150 | struct net_device *dev, struct netdev_queue *txq, | |
55a93b3e | 151 | spinlock_t *root_lock, bool validate) |
1da177e4 | 152 | { |
5f1a485d | 153 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
154 | |
155 | /* And release qdisc */ | |
156 | spin_unlock(root_lock); | |
c716a81a | 157 | |
55a93b3e ED |
158 | /* Note that we validate skb (GSO, checksum, ...) outside of locks */ |
159 | if (validate) | |
160 | skb = validate_xmit_skb_list(skb, dev); | |
572a9d7b | 161 | |
3dcd493f | 162 | if (likely(skb)) { |
55a93b3e ED |
163 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
164 | if (!netif_xmit_frozen_or_stopped(txq)) | |
165 | skb = dev_hard_start_xmit(skb, dev, txq, &ret); | |
c716a81a | 166 | |
55a93b3e | 167 | HARD_TX_UNLOCK(dev, txq); |
3dcd493f LP |
168 | } else { |
169 | spin_lock(root_lock); | |
170 | return qdisc_qlen(q); | |
55a93b3e | 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 | */ | |
b8358d70 | 214 | static inline int qdisc_restart(struct Qdisc *q, int *packets) |
bbd8a0d3 KK |
215 | { |
216 | struct netdev_queue *txq; | |
217 | struct net_device *dev; | |
218 | spinlock_t *root_lock; | |
219 | struct sk_buff *skb; | |
55a93b3e | 220 | bool validate; |
bbd8a0d3 KK |
221 | |
222 | /* Dequeue packet */ | |
b8358d70 | 223 | skb = dequeue_skb(q, &validate, packets); |
bbd8a0d3 KK |
224 | if (unlikely(!skb)) |
225 | return 0; | |
10c51b56 | 226 | |
bbd8a0d3 KK |
227 | root_lock = qdisc_lock(q); |
228 | dev = qdisc_dev(q); | |
10c51b56 | 229 | txq = skb_get_tx_queue(dev, skb); |
bbd8a0d3 | 230 | |
55a93b3e | 231 | return sch_direct_xmit(skb, q, dev, txq, root_lock, validate); |
bbd8a0d3 KK |
232 | } |
233 | ||
37437bb2 | 234 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 235 | { |
d5b8aa1d | 236 | int quota = weight_p; |
b8358d70 | 237 | int packets; |
2ba2506c | 238 | |
b8358d70 | 239 | while (qdisc_restart(q, &packets)) { |
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 | */ |
b8358d70 JDB |
245 | quota -= packets; |
246 | if (quota <= 0 || need_resched()) { | |
37437bb2 | 247 | __netif_schedule(q); |
d90df3ad | 248 | break; |
2ba2506c HX |
249 | } |
250 | } | |
48d83325 | 251 | |
bc135b23 | 252 | qdisc_run_end(q); |
48d83325 HX |
253 | } |
254 | ||
9d21493b ED |
255 | unsigned long dev_trans_start(struct net_device *dev) |
256 | { | |
07ce76aa | 257 | unsigned long val, res; |
9d21493b ED |
258 | unsigned int i; |
259 | ||
07ce76aa | 260 | if (is_vlan_dev(dev)) |
261 | dev = vlan_dev_real_dev(dev); | |
262 | res = dev->trans_start; | |
9d21493b ED |
263 | for (i = 0; i < dev->num_tx_queues; i++) { |
264 | val = netdev_get_tx_queue(dev, i)->trans_start; | |
265 | if (val && time_after(val, res)) | |
266 | res = val; | |
267 | } | |
268 | dev->trans_start = res; | |
07ce76aa | 269 | |
9d21493b ED |
270 | return res; |
271 | } | |
272 | EXPORT_SYMBOL(dev_trans_start); | |
273 | ||
1da177e4 LT |
274 | static void dev_watchdog(unsigned long arg) |
275 | { | |
276 | struct net_device *dev = (struct net_device *)arg; | |
277 | ||
932ff279 | 278 | netif_tx_lock(dev); |
e8a0464c | 279 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
280 | if (netif_device_present(dev) && |
281 | netif_running(dev) && | |
282 | netif_carrier_ok(dev)) { | |
9d21493b | 283 | int some_queue_timedout = 0; |
e8a0464c | 284 | unsigned int i; |
9d21493b | 285 | unsigned long trans_start; |
e8a0464c DM |
286 | |
287 | for (i = 0; i < dev->num_tx_queues; i++) { | |
288 | struct netdev_queue *txq; | |
289 | ||
290 | txq = netdev_get_tx_queue(dev, i); | |
9d21493b ED |
291 | /* |
292 | * old device drivers set dev->trans_start | |
293 | */ | |
294 | trans_start = txq->trans_start ? : dev->trans_start; | |
73466498 | 295 | if (netif_xmit_stopped(txq) && |
9d21493b ED |
296 | time_after(jiffies, (trans_start + |
297 | dev->watchdog_timeo))) { | |
298 | some_queue_timedout = 1; | |
ccf5ff69 | 299 | txq->trans_timeout++; |
e8a0464c DM |
300 | break; |
301 | } | |
302 | } | |
338f7566 | 303 | |
9d21493b | 304 | if (some_queue_timedout) { |
9d21493b | 305 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 306 | dev->name, netdev_drivername(dev), i); |
d314774c | 307 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 308 | } |
e8a0464c DM |
309 | if (!mod_timer(&dev->watchdog_timer, |
310 | round_jiffies(jiffies + | |
311 | dev->watchdog_timeo))) | |
1da177e4 LT |
312 | dev_hold(dev); |
313 | } | |
314 | } | |
932ff279 | 315 | netif_tx_unlock(dev); |
1da177e4 LT |
316 | |
317 | dev_put(dev); | |
318 | } | |
319 | ||
1da177e4 LT |
320 | void __netdev_watchdog_up(struct net_device *dev) |
321 | { | |
d314774c | 322 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
323 | if (dev->watchdog_timeo <= 0) |
324 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
325 | if (!mod_timer(&dev->watchdog_timer, |
326 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
327 | dev_hold(dev); |
328 | } | |
329 | } | |
330 | ||
331 | static void dev_watchdog_up(struct net_device *dev) | |
332 | { | |
1da177e4 | 333 | __netdev_watchdog_up(dev); |
1da177e4 LT |
334 | } |
335 | ||
336 | static void dev_watchdog_down(struct net_device *dev) | |
337 | { | |
932ff279 | 338 | netif_tx_lock_bh(dev); |
1da177e4 | 339 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 340 | dev_put(dev); |
932ff279 | 341 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
342 | } |
343 | ||
bea3348e SH |
344 | /** |
345 | * netif_carrier_on - set carrier | |
346 | * @dev: network device | |
347 | * | |
348 | * Device has detected that carrier. | |
349 | */ | |
0a242efc DV |
350 | void netif_carrier_on(struct net_device *dev) |
351 | { | |
bfaae0f0 | 352 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
353 | if (dev->reg_state == NETREG_UNINITIALIZED) |
354 | return; | |
2d3b479d | 355 | atomic_inc(&dev->carrier_changes); |
0a242efc | 356 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
357 | if (netif_running(dev)) |
358 | __netdev_watchdog_up(dev); | |
359 | } | |
0a242efc | 360 | } |
62e3ba1b | 361 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 362 | |
bea3348e SH |
363 | /** |
364 | * netif_carrier_off - clear carrier | |
365 | * @dev: network device | |
366 | * | |
367 | * Device has detected loss of carrier. | |
368 | */ | |
0a242efc DV |
369 | void netif_carrier_off(struct net_device *dev) |
370 | { | |
b4730016 DM |
371 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
372 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
373 | return; | |
2d3b479d | 374 | atomic_inc(&dev->carrier_changes); |
0a242efc | 375 | linkwatch_fire_event(dev); |
b4730016 | 376 | } |
0a242efc | 377 | } |
62e3ba1b | 378 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 379 | |
1da177e4 LT |
380 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
381 | under all circumstances. It is difficult to invent anything faster or | |
382 | cheaper. | |
383 | */ | |
384 | ||
82d567c2 | 385 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc) |
1da177e4 LT |
386 | { |
387 | kfree_skb(skb); | |
388 | return NET_XMIT_CN; | |
389 | } | |
390 | ||
82d567c2 | 391 | static struct sk_buff *noop_dequeue(struct Qdisc *qdisc) |
1da177e4 LT |
392 | { |
393 | return NULL; | |
394 | } | |
395 | ||
20fea08b | 396 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
397 | .id = "noop", |
398 | .priv_size = 0, | |
399 | .enqueue = noop_enqueue, | |
400 | .dequeue = noop_dequeue, | |
99c0db26 | 401 | .peek = noop_dequeue, |
1da177e4 LT |
402 | .owner = THIS_MODULE, |
403 | }; | |
404 | ||
7698b4fc | 405 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 406 | .qdisc = &noop_qdisc, |
9f3ffae0 | 407 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
408 | }; |
409 | ||
1da177e4 LT |
410 | struct Qdisc noop_qdisc = { |
411 | .enqueue = noop_enqueue, | |
412 | .dequeue = noop_dequeue, | |
413 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 414 | .ops = &noop_qdisc_ops, |
1da177e4 | 415 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 416 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 417 | .dev_queue = &noop_netdev_queue, |
7b5edbc4 | 418 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
1da177e4 | 419 | }; |
62e3ba1b | 420 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 421 | |
d66d6c31 PS |
422 | static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt) |
423 | { | |
424 | /* register_qdisc() assigns a default of noop_enqueue if unset, | |
425 | * but __dev_queue_xmit() treats noqueue only as such | |
426 | * if this is NULL - so clear it here. */ | |
427 | qdisc->enqueue = NULL; | |
428 | return 0; | |
429 | } | |
430 | ||
431 | struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { | |
1da177e4 LT |
432 | .id = "noqueue", |
433 | .priv_size = 0, | |
d66d6c31 | 434 | .init = noqueue_init, |
1da177e4 LT |
435 | .enqueue = noop_enqueue, |
436 | .dequeue = noop_dequeue, | |
99c0db26 | 437 | .peek = noop_dequeue, |
1da177e4 LT |
438 | .owner = THIS_MODULE, |
439 | }; | |
440 | ||
cc7ec456 ED |
441 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
442 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
443 | }; | |
d3678b46 DM |
444 | |
445 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
446 | generic prio+fifo combination. | |
447 | */ | |
448 | ||
449 | #define PFIFO_FAST_BANDS 3 | |
450 | ||
fd3ae5e8 KK |
451 | /* |
452 | * Private data for a pfifo_fast scheduler containing: | |
453 | * - queues for the three band | |
454 | * - bitmap indicating which of the bands contain skbs | |
455 | */ | |
456 | struct pfifo_fast_priv { | |
457 | u32 bitmap; | |
458 | struct sk_buff_head q[PFIFO_FAST_BANDS]; | |
459 | }; | |
460 | ||
461 | /* | |
462 | * Convert a bitmap to the first band number where an skb is queued, where: | |
463 | * bitmap=0 means there are no skbs on any band. | |
464 | * bitmap=1 means there is an skb on band 0. | |
465 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
466 | */ | |
467 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
468 | ||
469 | static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv, | |
470 | int band) | |
d3678b46 | 471 | { |
fd3ae5e8 | 472 | return priv->q + band; |
d3678b46 DM |
473 | } |
474 | ||
cc7ec456 | 475 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc) |
321090e7 | 476 | { |
a453e068 KK |
477 | if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) { |
478 | int band = prio2band[skb->priority & TC_PRIO_MAX]; | |
479 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
480 | struct sk_buff_head *list = band2list(priv, band); | |
1da177e4 | 481 | |
fd3ae5e8 | 482 | priv->bitmap |= (1 << band); |
d3678b46 | 483 | qdisc->q.qlen++; |
821d24ae | 484 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 485 | } |
821d24ae TG |
486 | |
487 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
488 | } |
489 | ||
cc7ec456 | 490 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc) |
1da177e4 | 491 | { |
fd3ae5e8 KK |
492 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
493 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 494 | |
fd3ae5e8 KK |
495 | if (likely(band >= 0)) { |
496 | struct sk_buff_head *list = band2list(priv, band); | |
497 | struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list); | |
498 | ||
499 | qdisc->q.qlen--; | |
500 | if (skb_queue_empty(list)) | |
501 | priv->bitmap &= ~(1 << band); | |
502 | ||
503 | return skb; | |
d3678b46 | 504 | } |
f87a9c3d | 505 | |
1da177e4 LT |
506 | return NULL; |
507 | } | |
508 | ||
cc7ec456 | 509 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 510 | { |
fd3ae5e8 KK |
511 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
512 | int band = bitmap2band[priv->bitmap]; | |
513 | ||
514 | if (band >= 0) { | |
515 | struct sk_buff_head *list = band2list(priv, band); | |
99c0db26 | 516 | |
fd3ae5e8 | 517 | return skb_peek(list); |
99c0db26 JP |
518 | } |
519 | ||
520 | return NULL; | |
521 | } | |
522 | ||
cc7ec456 | 523 | static void pfifo_fast_reset(struct Qdisc *qdisc) |
1da177e4 | 524 | { |
d3678b46 | 525 | int prio; |
fd3ae5e8 | 526 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
527 | |
528 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 529 | __qdisc_reset_queue(qdisc, band2list(priv, prio)); |
d3678b46 | 530 | |
fd3ae5e8 | 531 | priv->bitmap = 0; |
821d24ae | 532 | qdisc->qstats.backlog = 0; |
d3678b46 | 533 | qdisc->q.qlen = 0; |
1da177e4 LT |
534 | } |
535 | ||
d3678b46 DM |
536 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
537 | { | |
538 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
539 | ||
cc7ec456 | 540 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
1b34ec43 DM |
541 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
542 | goto nla_put_failure; | |
d3678b46 DM |
543 | return skb->len; |
544 | ||
545 | nla_put_failure: | |
546 | return -1; | |
547 | } | |
548 | ||
549 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
550 | { | |
551 | int prio; | |
fd3ae5e8 | 552 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
553 | |
554 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
ab34f648 | 555 | __skb_queue_head_init(band2list(priv, prio)); |
d3678b46 | 556 | |
23624935 ED |
557 | /* Can by-pass the queue discipline */ |
558 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
559 | return 0; |
560 | } | |
561 | ||
6ec1c69a | 562 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 563 | .id = "pfifo_fast", |
fd3ae5e8 | 564 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
565 | .enqueue = pfifo_fast_enqueue, |
566 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 567 | .peek = pfifo_fast_peek, |
d3678b46 DM |
568 | .init = pfifo_fast_init, |
569 | .reset = pfifo_fast_reset, | |
570 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
571 | .owner = THIS_MODULE, |
572 | }; | |
1f27cde3 | 573 | EXPORT_SYMBOL(pfifo_fast_ops); |
1da177e4 | 574 | |
23d3b8bf ED |
575 | static struct lock_class_key qdisc_tx_busylock; |
576 | ||
5ce2d488 | 577 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
d2a7f269 | 578 | const struct Qdisc_ops *ops) |
1da177e4 LT |
579 | { |
580 | void *p; | |
581 | struct Qdisc *sch; | |
d276055c | 582 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 583 | int err = -ENOBUFS; |
23d3b8bf | 584 | struct net_device *dev = dev_queue->dev; |
1da177e4 | 585 | |
f2cd2d3e ED |
586 | p = kzalloc_node(size, GFP_KERNEL, |
587 | netdev_queue_numa_node_read(dev_queue)); | |
588 | ||
1da177e4 | 589 | if (!p) |
3d54b82f | 590 | goto errout; |
3d54b82f | 591 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
592 | /* if we got non aligned memory, ask more and do alignment ourself */ |
593 | if (sch != p) { | |
594 | kfree(p); | |
595 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
596 | netdev_queue_numa_node_read(dev_queue)); | |
597 | if (!p) | |
598 | goto errout; | |
599 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
600 | sch->padded = (char *) sch - (char *) p; | |
601 | } | |
1da177e4 LT |
602 | INIT_LIST_HEAD(&sch->list); |
603 | skb_queue_head_init(&sch->q); | |
23d3b8bf | 604 | |
79640a4c | 605 | spin_lock_init(&sch->busylock); |
23d3b8bf ED |
606 | lockdep_set_class(&sch->busylock, |
607 | dev->qdisc_tx_busylock ?: &qdisc_tx_busylock); | |
608 | ||
1da177e4 LT |
609 | sch->ops = ops; |
610 | sch->enqueue = ops->enqueue; | |
611 | sch->dequeue = ops->dequeue; | |
bb949fbd | 612 | sch->dev_queue = dev_queue; |
23d3b8bf | 613 | dev_hold(dev); |
1da177e4 | 614 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
615 | |
616 | return sch; | |
617 | errout: | |
01e123d7 | 618 | return ERR_PTR(err); |
3d54b82f TG |
619 | } |
620 | ||
3511c913 | 621 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
d2a7f269 | 622 | const struct Qdisc_ops *ops, |
623 | unsigned int parentid) | |
3d54b82f TG |
624 | { |
625 | struct Qdisc *sch; | |
10297b99 | 626 | |
6da7c8fc | 627 | if (!try_module_get(ops->owner)) |
628 | goto errout; | |
629 | ||
5ce2d488 | 630 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
631 | if (IS_ERR(sch)) |
632 | goto errout; | |
9f9afec4 | 633 | sch->parent = parentid; |
3d54b82f | 634 | |
1da177e4 LT |
635 | if (!ops->init || ops->init(sch, NULL) == 0) |
636 | return sch; | |
637 | ||
0fbbeb1b | 638 | qdisc_destroy(sch); |
3d54b82f | 639 | errout: |
1da177e4 LT |
640 | return NULL; |
641 | } | |
62e3ba1b | 642 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 643 | |
5fb66229 | 644 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
645 | |
646 | void qdisc_reset(struct Qdisc *qdisc) | |
647 | { | |
20fea08b | 648 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
649 | |
650 | if (ops->reset) | |
651 | ops->reset(qdisc); | |
67305ebc | 652 | |
bbd8a0d3 | 653 | if (qdisc->gso_skb) { |
3f3c7eec | 654 | kfree_skb_list(qdisc->gso_skb); |
bbd8a0d3 KK |
655 | qdisc->gso_skb = NULL; |
656 | qdisc->q.qlen = 0; | |
657 | } | |
1da177e4 | 658 | } |
62e3ba1b | 659 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 660 | |
5d944c64 ED |
661 | static void qdisc_rcu_free(struct rcu_head *head) |
662 | { | |
663 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
664 | ||
73c20a8b | 665 | if (qdisc_is_percpu_stats(qdisc)) { |
22e0f8b9 | 666 | free_percpu(qdisc->cpu_bstats); |
73c20a8b JF |
667 | free_percpu(qdisc->cpu_qstats); |
668 | } | |
22e0f8b9 | 669 | |
5d944c64 ED |
670 | kfree((char *) qdisc - qdisc->padded); |
671 | } | |
672 | ||
1e0d5a57 | 673 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 674 | { |
8a34c5dc DM |
675 | const struct Qdisc_ops *ops = qdisc->ops; |
676 | ||
1e0d5a57 DM |
677 | if (qdisc->flags & TCQ_F_BUILTIN || |
678 | !atomic_dec_and_test(&qdisc->refcnt)) | |
679 | return; | |
680 | ||
3a682fbd | 681 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
682 | qdisc_list_del(qdisc); |
683 | ||
a2da570d | 684 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 685 | #endif |
8a34c5dc DM |
686 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
687 | if (ops->reset) | |
688 | ops->reset(qdisc); | |
689 | if (ops->destroy) | |
690 | ops->destroy(qdisc); | |
691 | ||
692 | module_put(ops->owner); | |
693 | dev_put(qdisc_dev(qdisc)); | |
694 | ||
3f3c7eec | 695 | kfree_skb_list(qdisc->gso_skb); |
5d944c64 ED |
696 | /* |
697 | * gen_estimator est_timer() might access qdisc->q.lock, | |
698 | * wait a RCU grace period before freeing qdisc. | |
699 | */ | |
700 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 701 | } |
62e3ba1b | 702 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 703 | |
589983cd PM |
704 | /* Attach toplevel qdisc to device queue. */ |
705 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
706 | struct Qdisc *qdisc) | |
707 | { | |
708 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
709 | spinlock_t *root_lock; | |
710 | ||
711 | root_lock = qdisc_lock(oqdisc); | |
712 | spin_lock_bh(root_lock); | |
713 | ||
714 | /* Prune old scheduler */ | |
715 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
716 | qdisc_reset(oqdisc); | |
717 | ||
718 | /* ... and graft new one */ | |
719 | if (qdisc == NULL) | |
720 | qdisc = &noop_qdisc; | |
721 | dev_queue->qdisc_sleeping = qdisc; | |
722 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
723 | ||
724 | spin_unlock_bh(root_lock); | |
725 | ||
726 | return oqdisc; | |
727 | } | |
b8970f0b | 728 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 729 | |
e8a0464c DM |
730 | static void attach_one_default_qdisc(struct net_device *dev, |
731 | struct netdev_queue *dev_queue, | |
732 | void *_unused) | |
733 | { | |
3e692f21 PS |
734 | struct Qdisc *qdisc; |
735 | const struct Qdisc_ops *ops = default_qdisc_ops; | |
e8a0464c | 736 | |
3e692f21 PS |
737 | if (dev->priv_flags & IFF_NO_QUEUE) |
738 | ops = &noqueue_qdisc_ops; | |
739 | ||
740 | qdisc = qdisc_create_dflt(dev_queue, ops, TC_H_ROOT); | |
741 | if (!qdisc) { | |
742 | netdev_info(dev, "activation failed\n"); | |
743 | return; | |
e8a0464c | 744 | } |
3e692f21 | 745 | if (!netif_is_multiqueue(dev)) |
4eaf3b84 | 746 | qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT; |
e8a0464c DM |
747 | dev_queue->qdisc_sleeping = qdisc; |
748 | } | |
749 | ||
6ec1c69a DM |
750 | static void attach_default_qdiscs(struct net_device *dev) |
751 | { | |
752 | struct netdev_queue *txq; | |
753 | struct Qdisc *qdisc; | |
754 | ||
755 | txq = netdev_get_tx_queue(dev, 0); | |
756 | ||
4b469955 | 757 | if (!netif_is_multiqueue(dev) || |
4b469955 | 758 | dev->priv_flags & IFF_NO_QUEUE) { |
6ec1c69a DM |
759 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); |
760 | dev->qdisc = txq->qdisc_sleeping; | |
761 | atomic_inc(&dev->qdisc->refcnt); | |
762 | } else { | |
3511c913 | 763 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT); |
6ec1c69a | 764 | if (qdisc) { |
6ec1c69a | 765 | dev->qdisc = qdisc; |
e57a784d | 766 | qdisc->ops->attach(qdisc); |
6ec1c69a DM |
767 | } |
768 | } | |
769 | } | |
770 | ||
e8a0464c DM |
771 | static void transition_one_qdisc(struct net_device *dev, |
772 | struct netdev_queue *dev_queue, | |
773 | void *_need_watchdog) | |
774 | { | |
83874000 | 775 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
776 | int *need_watchdog_p = _need_watchdog; |
777 | ||
a9312ae8 DM |
778 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
779 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
780 | ||
83874000 | 781 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
3e692f21 | 782 | if (need_watchdog_p) { |
9d21493b | 783 | dev_queue->trans_start = 0; |
e8a0464c | 784 | *need_watchdog_p = 1; |
9d21493b | 785 | } |
e8a0464c DM |
786 | } |
787 | ||
1da177e4 LT |
788 | void dev_activate(struct net_device *dev) |
789 | { | |
e8a0464c | 790 | int need_watchdog; |
b0e1e646 | 791 | |
1da177e4 | 792 | /* No queueing discipline is attached to device; |
6da7c8fc | 793 | * create default one for devices, which need queueing |
794 | * and noqueue_qdisc for virtual interfaces | |
1da177e4 LT |
795 | */ |
796 | ||
6ec1c69a DM |
797 | if (dev->qdisc == &noop_qdisc) |
798 | attach_default_qdiscs(dev); | |
af356afa | 799 | |
cacaddf5 TC |
800 | if (!netif_carrier_ok(dev)) |
801 | /* Delay activation until next carrier-on event */ | |
802 | return; | |
803 | ||
e8a0464c DM |
804 | need_watchdog = 0; |
805 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
24824a09 ED |
806 | if (dev_ingress_queue(dev)) |
807 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
808 | |
809 | if (need_watchdog) { | |
1da177e4 LT |
810 | dev->trans_start = jiffies; |
811 | dev_watchdog_up(dev); | |
812 | } | |
b0e1e646 | 813 | } |
b8970f0b | 814 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 815 | |
e8a0464c DM |
816 | static void dev_deactivate_queue(struct net_device *dev, |
817 | struct netdev_queue *dev_queue, | |
818 | void *_qdisc_default) | |
b0e1e646 | 819 | { |
e8a0464c | 820 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 821 | struct Qdisc *qdisc; |
970565bb | 822 | |
46e5da40 | 823 | qdisc = rtnl_dereference(dev_queue->qdisc); |
b0e1e646 | 824 | if (qdisc) { |
83874000 DM |
825 | spin_lock_bh(qdisc_lock(qdisc)); |
826 | ||
a9312ae8 DM |
827 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
828 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
829 | ||
f7a54c13 | 830 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 831 | qdisc_reset(qdisc); |
d3b753db | 832 | |
83874000 | 833 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 834 | } |
1da177e4 LT |
835 | } |
836 | ||
4335cd2d | 837 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
838 | { |
839 | unsigned int i; | |
840 | ||
841 | for (i = 0; i < dev->num_tx_queues; i++) { | |
842 | struct netdev_queue *dev_queue; | |
7698b4fc | 843 | spinlock_t *root_lock; |
e2627c8c | 844 | struct Qdisc *q; |
e8a0464c DM |
845 | int val; |
846 | ||
847 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 848 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 849 | root_lock = qdisc_lock(q); |
e8a0464c | 850 | |
4335cd2d | 851 | spin_lock_bh(root_lock); |
e8a0464c | 852 | |
bc135b23 | 853 | val = (qdisc_is_running(q) || |
b9a3b110 | 854 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 855 | |
4335cd2d | 856 | spin_unlock_bh(root_lock); |
e8a0464c DM |
857 | |
858 | if (val) | |
859 | return true; | |
860 | } | |
861 | return false; | |
862 | } | |
863 | ||
3137663d ED |
864 | /** |
865 | * dev_deactivate_many - deactivate transmissions on several devices | |
866 | * @head: list of devices to deactivate | |
867 | * | |
868 | * This function returns only when all outstanding transmissions | |
869 | * have completed, unless all devices are in dismantle phase. | |
870 | */ | |
44345724 | 871 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 872 | { |
44345724 | 873 | struct net_device *dev; |
3137663d | 874 | bool sync_needed = false; |
41a23b07 | 875 | |
5cde2829 | 876 | list_for_each_entry(dev, head, close_list) { |
44345724 OP |
877 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, |
878 | &noop_qdisc); | |
879 | if (dev_ingress_queue(dev)) | |
880 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
881 | &noop_qdisc); | |
882 | ||
883 | dev_watchdog_down(dev); | |
3137663d | 884 | sync_needed |= !dev->dismantle; |
44345724 | 885 | } |
1da177e4 | 886 | |
3137663d ED |
887 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
888 | * This is avoided if all devices are in dismantle phase : | |
889 | * Caller will call synchronize_net() for us | |
890 | */ | |
891 | if (sync_needed) | |
892 | synchronize_net(); | |
1da177e4 | 893 | |
d4828d85 | 894 | /* Wait for outstanding qdisc_run calls. */ |
5cde2829 | 895 | list_for_each_entry(dev, head, close_list) |
44345724 OP |
896 | while (some_qdisc_is_busy(dev)) |
897 | yield(); | |
898 | } | |
899 | ||
900 | void dev_deactivate(struct net_device *dev) | |
901 | { | |
902 | LIST_HEAD(single); | |
903 | ||
5cde2829 | 904 | list_add(&dev->close_list, &single); |
44345724 | 905 | dev_deactivate_many(&single); |
5f04d506 | 906 | list_del(&single); |
1da177e4 | 907 | } |
b8970f0b | 908 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 909 | |
b0e1e646 DM |
910 | static void dev_init_scheduler_queue(struct net_device *dev, |
911 | struct netdev_queue *dev_queue, | |
e8a0464c | 912 | void *_qdisc) |
b0e1e646 | 913 | { |
e8a0464c DM |
914 | struct Qdisc *qdisc = _qdisc; |
915 | ||
46e5da40 | 916 | rcu_assign_pointer(dev_queue->qdisc, qdisc); |
b0e1e646 | 917 | dev_queue->qdisc_sleeping = qdisc; |
b0e1e646 DM |
918 | } |
919 | ||
1da177e4 LT |
920 | void dev_init_scheduler(struct net_device *dev) |
921 | { | |
af356afa | 922 | dev->qdisc = &noop_qdisc; |
e8a0464c | 923 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
924 | if (dev_ingress_queue(dev)) |
925 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 926 | |
b24b8a24 | 927 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
928 | } |
929 | ||
e8a0464c DM |
930 | static void shutdown_scheduler_queue(struct net_device *dev, |
931 | struct netdev_queue *dev_queue, | |
932 | void *_qdisc_default) | |
1da177e4 | 933 | { |
b0e1e646 | 934 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 935 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
936 | |
937 | if (qdisc) { | |
f7a54c13 | 938 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 939 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 940 | |
1da177e4 | 941 | qdisc_destroy(qdisc); |
10297b99 | 942 | } |
b0e1e646 DM |
943 | } |
944 | ||
945 | void dev_shutdown(struct net_device *dev) | |
946 | { | |
e8a0464c | 947 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
948 | if (dev_ingress_queue(dev)) |
949 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
af356afa PM |
950 | qdisc_destroy(dev->qdisc); |
951 | dev->qdisc = &noop_qdisc; | |
952 | ||
547b792c | 953 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 954 | } |
292f1c7f | 955 | |
01cb71d2 | 956 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
3e1e3aae ED |
957 | const struct tc_ratespec *conf, |
958 | u64 rate64) | |
292f1c7f | 959 | { |
01cb71d2 ED |
960 | memset(r, 0, sizeof(*r)); |
961 | r->overhead = conf->overhead; | |
3e1e3aae | 962 | r->rate_bytes_ps = max_t(u64, conf->rate, rate64); |
8a8e3d84 | 963 | r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK); |
292f1c7f JP |
964 | r->mult = 1; |
965 | /* | |
130d3d68 ED |
966 | * The deal here is to replace a divide by a reciprocal one |
967 | * in fast path (a reciprocal divide is a multiply and a shift) | |
968 | * | |
969 | * Normal formula would be : | |
970 | * time_in_ns = (NSEC_PER_SEC * len) / rate_bps | |
971 | * | |
972 | * We compute mult/shift to use instead : | |
973 | * time_in_ns = (len * mult) >> shift; | |
974 | * | |
975 | * We try to get the highest possible mult value for accuracy, | |
976 | * but have to make sure no overflows will ever happen. | |
292f1c7f | 977 | */ |
130d3d68 ED |
978 | if (r->rate_bytes_ps > 0) { |
979 | u64 factor = NSEC_PER_SEC; | |
980 | ||
981 | for (;;) { | |
982 | r->mult = div64_u64(factor, r->rate_bytes_ps); | |
983 | if (r->mult & (1U << 31) || factor & (1ULL << 63)) | |
292f1c7f | 984 | break; |
130d3d68 ED |
985 | factor <<= 1; |
986 | r->shift++; | |
292f1c7f | 987 | } |
292f1c7f JP |
988 | } |
989 | } | |
990 | EXPORT_SYMBOL(psched_ratecfg_precompute); |