]>
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++; |
a27758ff | 52 | qdisc_qstats_backlog_inc(q, skb); |
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 | ||
55a93b3e ED |
59 | static void try_bulk_dequeue_skb(struct Qdisc *q, |
60 | struct sk_buff *skb, | |
b8358d70 JDB |
61 | const struct netdev_queue *txq, |
62 | int *packets) | |
5772e9a3 | 63 | { |
55a93b3e | 64 | int bytelimit = qdisc_avail_bulklimit(txq) - skb->len; |
5772e9a3 JDB |
65 | |
66 | while (bytelimit > 0) { | |
55a93b3e | 67 | struct sk_buff *nskb = q->dequeue(q); |
5772e9a3 | 68 | |
55a93b3e | 69 | if (!nskb) |
5772e9a3 JDB |
70 | break; |
71 | ||
55a93b3e ED |
72 | bytelimit -= nskb->len; /* covers GSO len */ |
73 | skb->next = nskb; | |
74 | skb = nskb; | |
b8358d70 | 75 | (*packets)++; /* GSO counts as one pkt */ |
5772e9a3 | 76 | } |
55a93b3e | 77 | skb->next = NULL; |
5772e9a3 JDB |
78 | } |
79 | ||
4d202a0d ED |
80 | /* This variant of try_bulk_dequeue_skb() makes sure |
81 | * all skbs in the chain are for the same txq | |
82 | */ | |
83 | static void try_bulk_dequeue_skb_slow(struct Qdisc *q, | |
84 | struct sk_buff *skb, | |
85 | int *packets) | |
86 | { | |
87 | int mapping = skb_get_queue_mapping(skb); | |
88 | struct sk_buff *nskb; | |
89 | int cnt = 0; | |
90 | ||
91 | do { | |
92 | nskb = q->dequeue(q); | |
93 | if (!nskb) | |
94 | break; | |
95 | if (unlikely(skb_get_queue_mapping(nskb) != mapping)) { | |
96 | q->skb_bad_txq = nskb; | |
97 | qdisc_qstats_backlog_inc(q, nskb); | |
98 | q->q.qlen++; | |
99 | break; | |
100 | } | |
101 | skb->next = nskb; | |
102 | skb = nskb; | |
103 | } while (++cnt < 8); | |
104 | (*packets) += cnt; | |
105 | skb->next = NULL; | |
106 | } | |
107 | ||
5772e9a3 JDB |
108 | /* Note that dequeue_skb can possibly return a SKB list (via skb->next). |
109 | * A requeued skb (via q->gso_skb) can also be a SKB list. | |
110 | */ | |
b8358d70 JDB |
111 | static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate, |
112 | int *packets) | |
c716a81a | 113 | { |
554794de | 114 | struct sk_buff *skb = q->gso_skb; |
1abbe139 | 115 | const struct netdev_queue *txq = q->dev_queue; |
554794de | 116 | |
b8358d70 | 117 | *packets = 1; |
ebf05982 | 118 | if (unlikely(skb)) { |
4d202a0d ED |
119 | /* skb in gso_skb were already validated */ |
120 | *validate = false; | |
ebf05982 | 121 | /* check the reason of requeuing without tx lock first */ |
10c51b56 | 122 | txq = skb_get_tx_queue(txq->dev, skb); |
73466498 | 123 | if (!netif_xmit_frozen_or_stopped(txq)) { |
6252352d | 124 | q->gso_skb = NULL; |
a27758ff | 125 | qdisc_qstats_backlog_dec(q, skb); |
bbd8a0d3 KK |
126 | q->q.qlen--; |
127 | } else | |
ebf05982 | 128 | skb = NULL; |
4d202a0d ED |
129 | return skb; |
130 | } | |
131 | *validate = true; | |
132 | skb = q->skb_bad_txq; | |
133 | if (unlikely(skb)) { | |
134 | /* check the reason of requeuing without tx lock first */ | |
135 | txq = skb_get_tx_queue(txq->dev, skb); | |
136 | if (!netif_xmit_frozen_or_stopped(txq)) { | |
137 | q->skb_bad_txq = NULL; | |
138 | qdisc_qstats_backlog_dec(q, skb); | |
139 | q->q.qlen--; | |
140 | goto bulk; | |
50cbe9ab | 141 | } |
4d202a0d ED |
142 | return NULL; |
143 | } | |
144 | if (!(q->flags & TCQ_F_ONETXQUEUE) || | |
145 | !netif_xmit_frozen_or_stopped(txq)) | |
146 | skb = q->dequeue(q); | |
147 | if (skb) { | |
148 | bulk: | |
149 | if (qdisc_may_bulk(q)) | |
150 | try_bulk_dequeue_skb(q, skb, txq, packets); | |
151 | else | |
152 | try_bulk_dequeue_skb_slow(q, skb, packets); | |
ebf05982 | 153 | } |
c716a81a JHS |
154 | return skb; |
155 | } | |
156 | ||
10297b99 | 157 | /* |
10770bc2 | 158 | * Transmit possibly several skbs, and handle the return status as |
f9eb8aea | 159 | * required. Owning running seqcount bit guarantees that |
10770bc2 | 160 | * only one CPU can execute this function. |
6c1361a6 KK |
161 | * |
162 | * Returns to the caller: | |
163 | * 0 - queue is empty or throttled. | |
164 | * >0 - queue is not empty. | |
6c1361a6 | 165 | */ |
bbd8a0d3 KK |
166 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
167 | struct net_device *dev, struct netdev_queue *txq, | |
55a93b3e | 168 | spinlock_t *root_lock, bool validate) |
1da177e4 | 169 | { |
5f1a485d | 170 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
171 | |
172 | /* And release qdisc */ | |
173 | spin_unlock(root_lock); | |
c716a81a | 174 | |
55a93b3e ED |
175 | /* Note that we validate skb (GSO, checksum, ...) outside of locks */ |
176 | if (validate) | |
177 | skb = validate_xmit_skb_list(skb, dev); | |
572a9d7b | 178 | |
3dcd493f | 179 | if (likely(skb)) { |
55a93b3e ED |
180 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
181 | if (!netif_xmit_frozen_or_stopped(txq)) | |
182 | skb = dev_hard_start_xmit(skb, dev, txq, &ret); | |
c716a81a | 183 | |
55a93b3e | 184 | HARD_TX_UNLOCK(dev, txq); |
3dcd493f | 185 | } else { |
52fbb290 | 186 | spin_lock(root_lock); |
3dcd493f | 187 | return qdisc_qlen(q); |
55a93b3e | 188 | } |
52fbb290 | 189 | spin_lock(root_lock); |
c716a81a | 190 | |
9a1654ba JP |
191 | if (dev_xmit_complete(ret)) { |
192 | /* Driver sent out skb successfully or skb was consumed */ | |
6c1361a6 | 193 | ret = qdisc_qlen(q); |
9a1654ba | 194 | } else { |
6c1361a6 | 195 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
e87cc472 JP |
196 | if (unlikely(ret != NETDEV_TX_BUSY)) |
197 | net_warn_ratelimited("BUG %s code %d qlen %d\n", | |
198 | dev->name, ret, q->q.qlen); | |
6c1361a6 | 199 | |
37437bb2 | 200 | ret = dev_requeue_skb(skb, q); |
6c1361a6 | 201 | } |
c716a81a | 202 | |
73466498 | 203 | if (ret && netif_xmit_frozen_or_stopped(txq)) |
37437bb2 DM |
204 | ret = 0; |
205 | ||
6c1361a6 | 206 | return ret; |
1da177e4 LT |
207 | } |
208 | ||
bbd8a0d3 KK |
209 | /* |
210 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. | |
211 | * | |
f9eb8aea | 212 | * running seqcount guarantees only one CPU can process |
bbd8a0d3 KK |
213 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for |
214 | * this queue. | |
215 | * | |
216 | * netif_tx_lock serializes accesses to device driver. | |
217 | * | |
218 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, | |
219 | * if one is grabbed, another must be free. | |
220 | * | |
221 | * Note, that this procedure can be called by a watchdog timer | |
222 | * | |
223 | * Returns to the caller: | |
224 | * 0 - queue is empty or throttled. | |
225 | * >0 - queue is not empty. | |
226 | * | |
227 | */ | |
b8358d70 | 228 | static inline int qdisc_restart(struct Qdisc *q, int *packets) |
bbd8a0d3 KK |
229 | { |
230 | struct netdev_queue *txq; | |
231 | struct net_device *dev; | |
232 | spinlock_t *root_lock; | |
233 | struct sk_buff *skb; | |
55a93b3e | 234 | bool validate; |
bbd8a0d3 KK |
235 | |
236 | /* Dequeue packet */ | |
b8358d70 | 237 | skb = dequeue_skb(q, &validate, packets); |
bbd8a0d3 KK |
238 | if (unlikely(!skb)) |
239 | return 0; | |
10c51b56 | 240 | |
bbd8a0d3 KK |
241 | root_lock = qdisc_lock(q); |
242 | dev = qdisc_dev(q); | |
10c51b56 | 243 | txq = skb_get_tx_queue(dev, skb); |
bbd8a0d3 | 244 | |
55a93b3e | 245 | return sch_direct_xmit(skb, q, dev, txq, root_lock, validate); |
bbd8a0d3 KK |
246 | } |
247 | ||
37437bb2 | 248 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 249 | { |
d5b8aa1d | 250 | int quota = weight_p; |
b8358d70 | 251 | int packets; |
2ba2506c | 252 | |
b8358d70 | 253 | while (qdisc_restart(q, &packets)) { |
2ba2506c | 254 | /* |
d5b8aa1d | 255 | * Ordered by possible occurrence: Postpone processing if |
256 | * 1. we've exceeded packet quota | |
257 | * 2. another process needs the CPU; | |
2ba2506c | 258 | */ |
b8358d70 JDB |
259 | quota -= packets; |
260 | if (quota <= 0 || need_resched()) { | |
37437bb2 | 261 | __netif_schedule(q); |
d90df3ad | 262 | break; |
2ba2506c HX |
263 | } |
264 | } | |
48d83325 | 265 | |
bc135b23 | 266 | qdisc_run_end(q); |
48d83325 HX |
267 | } |
268 | ||
9d21493b ED |
269 | unsigned long dev_trans_start(struct net_device *dev) |
270 | { | |
07ce76aa | 271 | unsigned long val, res; |
9d21493b ED |
272 | unsigned int i; |
273 | ||
07ce76aa | 274 | if (is_vlan_dev(dev)) |
275 | dev = vlan_dev_real_dev(dev); | |
9b36627a FW |
276 | res = netdev_get_tx_queue(dev, 0)->trans_start; |
277 | for (i = 1; i < dev->num_tx_queues; i++) { | |
9d21493b ED |
278 | val = netdev_get_tx_queue(dev, i)->trans_start; |
279 | if (val && time_after(val, res)) | |
280 | res = val; | |
281 | } | |
07ce76aa | 282 | |
9d21493b ED |
283 | return res; |
284 | } | |
285 | EXPORT_SYMBOL(dev_trans_start); | |
286 | ||
1da177e4 LT |
287 | static void dev_watchdog(unsigned long arg) |
288 | { | |
289 | struct net_device *dev = (struct net_device *)arg; | |
290 | ||
932ff279 | 291 | netif_tx_lock(dev); |
e8a0464c | 292 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
293 | if (netif_device_present(dev) && |
294 | netif_running(dev) && | |
295 | netif_carrier_ok(dev)) { | |
9d21493b | 296 | int some_queue_timedout = 0; |
e8a0464c | 297 | unsigned int i; |
9d21493b | 298 | unsigned long trans_start; |
e8a0464c DM |
299 | |
300 | for (i = 0; i < dev->num_tx_queues; i++) { | |
301 | struct netdev_queue *txq; | |
302 | ||
303 | txq = netdev_get_tx_queue(dev, i); | |
9b36627a | 304 | trans_start = txq->trans_start; |
73466498 | 305 | if (netif_xmit_stopped(txq) && |
9d21493b ED |
306 | time_after(jiffies, (trans_start + |
307 | dev->watchdog_timeo))) { | |
308 | some_queue_timedout = 1; | |
ccf5ff69 | 309 | txq->trans_timeout++; |
e8a0464c DM |
310 | break; |
311 | } | |
312 | } | |
338f7566 | 313 | |
9d21493b | 314 | if (some_queue_timedout) { |
9d21493b | 315 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 316 | dev->name, netdev_drivername(dev), i); |
d314774c | 317 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 318 | } |
e8a0464c DM |
319 | if (!mod_timer(&dev->watchdog_timer, |
320 | round_jiffies(jiffies + | |
321 | dev->watchdog_timeo))) | |
1da177e4 LT |
322 | dev_hold(dev); |
323 | } | |
324 | } | |
932ff279 | 325 | netif_tx_unlock(dev); |
1da177e4 LT |
326 | |
327 | dev_put(dev); | |
328 | } | |
329 | ||
1da177e4 LT |
330 | void __netdev_watchdog_up(struct net_device *dev) |
331 | { | |
d314774c | 332 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
333 | if (dev->watchdog_timeo <= 0) |
334 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
335 | if (!mod_timer(&dev->watchdog_timer, |
336 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
337 | dev_hold(dev); |
338 | } | |
339 | } | |
340 | ||
341 | static void dev_watchdog_up(struct net_device *dev) | |
342 | { | |
1da177e4 | 343 | __netdev_watchdog_up(dev); |
1da177e4 LT |
344 | } |
345 | ||
346 | static void dev_watchdog_down(struct net_device *dev) | |
347 | { | |
932ff279 | 348 | netif_tx_lock_bh(dev); |
1da177e4 | 349 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 350 | dev_put(dev); |
932ff279 | 351 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
352 | } |
353 | ||
bea3348e SH |
354 | /** |
355 | * netif_carrier_on - set carrier | |
356 | * @dev: network device | |
357 | * | |
358 | * Device has detected that carrier. | |
359 | */ | |
0a242efc DV |
360 | void netif_carrier_on(struct net_device *dev) |
361 | { | |
bfaae0f0 | 362 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
363 | if (dev->reg_state == NETREG_UNINITIALIZED) |
364 | return; | |
2d3b479d | 365 | atomic_inc(&dev->carrier_changes); |
0a242efc | 366 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
367 | if (netif_running(dev)) |
368 | __netdev_watchdog_up(dev); | |
369 | } | |
0a242efc | 370 | } |
62e3ba1b | 371 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 372 | |
bea3348e SH |
373 | /** |
374 | * netif_carrier_off - clear carrier | |
375 | * @dev: network device | |
376 | * | |
377 | * Device has detected loss of carrier. | |
378 | */ | |
0a242efc DV |
379 | void netif_carrier_off(struct net_device *dev) |
380 | { | |
b4730016 DM |
381 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
382 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
383 | return; | |
2d3b479d | 384 | atomic_inc(&dev->carrier_changes); |
0a242efc | 385 | linkwatch_fire_event(dev); |
b4730016 | 386 | } |
0a242efc | 387 | } |
62e3ba1b | 388 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 389 | |
1da177e4 LT |
390 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
391 | under all circumstances. It is difficult to invent anything faster or | |
392 | cheaper. | |
393 | */ | |
394 | ||
520ac30f ED |
395 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc, |
396 | struct sk_buff **to_free) | |
1da177e4 | 397 | { |
520ac30f | 398 | __qdisc_drop(skb, to_free); |
1da177e4 LT |
399 | return NET_XMIT_CN; |
400 | } | |
401 | ||
82d567c2 | 402 | static struct sk_buff *noop_dequeue(struct Qdisc *qdisc) |
1da177e4 LT |
403 | { |
404 | return NULL; | |
405 | } | |
406 | ||
20fea08b | 407 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
408 | .id = "noop", |
409 | .priv_size = 0, | |
410 | .enqueue = noop_enqueue, | |
411 | .dequeue = noop_dequeue, | |
99c0db26 | 412 | .peek = noop_dequeue, |
1da177e4 LT |
413 | .owner = THIS_MODULE, |
414 | }; | |
415 | ||
7698b4fc | 416 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 417 | .qdisc = &noop_qdisc, |
9f3ffae0 | 418 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
419 | }; |
420 | ||
1da177e4 LT |
421 | struct Qdisc noop_qdisc = { |
422 | .enqueue = noop_enqueue, | |
423 | .dequeue = noop_dequeue, | |
424 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 425 | .ops = &noop_qdisc_ops, |
1da177e4 | 426 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 427 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 428 | .dev_queue = &noop_netdev_queue, |
f9eb8aea | 429 | .running = SEQCNT_ZERO(noop_qdisc.running), |
7b5edbc4 | 430 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
1da177e4 | 431 | }; |
62e3ba1b | 432 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 433 | |
d66d6c31 PS |
434 | static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt) |
435 | { | |
436 | /* register_qdisc() assigns a default of noop_enqueue if unset, | |
437 | * but __dev_queue_xmit() treats noqueue only as such | |
438 | * if this is NULL - so clear it here. */ | |
439 | qdisc->enqueue = NULL; | |
440 | return 0; | |
441 | } | |
442 | ||
443 | struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { | |
1da177e4 LT |
444 | .id = "noqueue", |
445 | .priv_size = 0, | |
d66d6c31 | 446 | .init = noqueue_init, |
1da177e4 LT |
447 | .enqueue = noop_enqueue, |
448 | .dequeue = noop_dequeue, | |
99c0db26 | 449 | .peek = noop_dequeue, |
1da177e4 LT |
450 | .owner = THIS_MODULE, |
451 | }; | |
452 | ||
cc7ec456 ED |
453 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
454 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
455 | }; | |
d3678b46 DM |
456 | |
457 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
458 | generic prio+fifo combination. | |
459 | */ | |
460 | ||
461 | #define PFIFO_FAST_BANDS 3 | |
462 | ||
fd3ae5e8 KK |
463 | /* |
464 | * Private data for a pfifo_fast scheduler containing: | |
465 | * - queues for the three band | |
466 | * - bitmap indicating which of the bands contain skbs | |
467 | */ | |
468 | struct pfifo_fast_priv { | |
469 | u32 bitmap; | |
470 | struct sk_buff_head q[PFIFO_FAST_BANDS]; | |
471 | }; | |
472 | ||
473 | /* | |
474 | * Convert a bitmap to the first band number where an skb is queued, where: | |
475 | * bitmap=0 means there are no skbs on any band. | |
476 | * bitmap=1 means there is an skb on band 0. | |
477 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
478 | */ | |
479 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
480 | ||
481 | static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv, | |
482 | int band) | |
d3678b46 | 483 | { |
fd3ae5e8 | 484 | return priv->q + band; |
d3678b46 DM |
485 | } |
486 | ||
520ac30f ED |
487 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc, |
488 | struct sk_buff **to_free) | |
321090e7 | 489 | { |
a453e068 KK |
490 | if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) { |
491 | int band = prio2band[skb->priority & TC_PRIO_MAX]; | |
492 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
493 | struct sk_buff_head *list = band2list(priv, band); | |
1da177e4 | 494 | |
fd3ae5e8 | 495 | priv->bitmap |= (1 << band); |
d3678b46 | 496 | qdisc->q.qlen++; |
821d24ae | 497 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 498 | } |
821d24ae | 499 | |
520ac30f | 500 | return qdisc_drop(skb, qdisc, to_free); |
1da177e4 LT |
501 | } |
502 | ||
cc7ec456 | 503 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc) |
1da177e4 | 504 | { |
fd3ae5e8 KK |
505 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
506 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 507 | |
fd3ae5e8 KK |
508 | if (likely(band >= 0)) { |
509 | struct sk_buff_head *list = band2list(priv, band); | |
510 | struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list); | |
511 | ||
512 | qdisc->q.qlen--; | |
513 | if (skb_queue_empty(list)) | |
514 | priv->bitmap &= ~(1 << band); | |
515 | ||
516 | return skb; | |
d3678b46 | 517 | } |
f87a9c3d | 518 | |
1da177e4 LT |
519 | return NULL; |
520 | } | |
521 | ||
cc7ec456 | 522 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 523 | { |
fd3ae5e8 KK |
524 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
525 | int band = bitmap2band[priv->bitmap]; | |
526 | ||
527 | if (band >= 0) { | |
528 | struct sk_buff_head *list = band2list(priv, band); | |
99c0db26 | 529 | |
fd3ae5e8 | 530 | return skb_peek(list); |
99c0db26 JP |
531 | } |
532 | ||
533 | return NULL; | |
534 | } | |
535 | ||
cc7ec456 | 536 | static void pfifo_fast_reset(struct Qdisc *qdisc) |
1da177e4 | 537 | { |
d3678b46 | 538 | int prio; |
fd3ae5e8 | 539 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
540 | |
541 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
1b5c5493 | 542 | __qdisc_reset_queue(band2list(priv, prio)); |
d3678b46 | 543 | |
fd3ae5e8 | 544 | priv->bitmap = 0; |
821d24ae | 545 | qdisc->qstats.backlog = 0; |
d3678b46 | 546 | qdisc->q.qlen = 0; |
1da177e4 LT |
547 | } |
548 | ||
d3678b46 DM |
549 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
550 | { | |
551 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
552 | ||
cc7ec456 | 553 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
1b34ec43 DM |
554 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
555 | goto nla_put_failure; | |
d3678b46 DM |
556 | return skb->len; |
557 | ||
558 | nla_put_failure: | |
559 | return -1; | |
560 | } | |
561 | ||
562 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
563 | { | |
564 | int prio; | |
fd3ae5e8 | 565 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
566 | |
567 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
ab34f648 | 568 | __skb_queue_head_init(band2list(priv, prio)); |
d3678b46 | 569 | |
23624935 ED |
570 | /* Can by-pass the queue discipline */ |
571 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
572 | return 0; |
573 | } | |
574 | ||
6ec1c69a | 575 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 576 | .id = "pfifo_fast", |
fd3ae5e8 | 577 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
578 | .enqueue = pfifo_fast_enqueue, |
579 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 580 | .peek = pfifo_fast_peek, |
d3678b46 DM |
581 | .init = pfifo_fast_init, |
582 | .reset = pfifo_fast_reset, | |
583 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
584 | .owner = THIS_MODULE, |
585 | }; | |
1f27cde3 | 586 | EXPORT_SYMBOL(pfifo_fast_ops); |
1da177e4 | 587 | |
23d3b8bf | 588 | static struct lock_class_key qdisc_tx_busylock; |
f9eb8aea | 589 | static struct lock_class_key qdisc_running_key; |
23d3b8bf | 590 | |
5ce2d488 | 591 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
d2a7f269 | 592 | const struct Qdisc_ops *ops) |
1da177e4 LT |
593 | { |
594 | void *p; | |
595 | struct Qdisc *sch; | |
d276055c | 596 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 597 | int err = -ENOBUFS; |
23d3b8bf | 598 | struct net_device *dev = dev_queue->dev; |
1da177e4 | 599 | |
f2cd2d3e ED |
600 | p = kzalloc_node(size, GFP_KERNEL, |
601 | netdev_queue_numa_node_read(dev_queue)); | |
602 | ||
1da177e4 | 603 | if (!p) |
3d54b82f | 604 | goto errout; |
3d54b82f | 605 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
606 | /* if we got non aligned memory, ask more and do alignment ourself */ |
607 | if (sch != p) { | |
608 | kfree(p); | |
609 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
610 | netdev_queue_numa_node_read(dev_queue)); | |
611 | if (!p) | |
612 | goto errout; | |
613 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
614 | sch->padded = (char *) sch - (char *) p; | |
615 | } | |
1da177e4 LT |
616 | INIT_LIST_HEAD(&sch->list); |
617 | skb_queue_head_init(&sch->q); | |
23d3b8bf | 618 | |
79640a4c | 619 | spin_lock_init(&sch->busylock); |
23d3b8bf ED |
620 | lockdep_set_class(&sch->busylock, |
621 | dev->qdisc_tx_busylock ?: &qdisc_tx_busylock); | |
622 | ||
f9eb8aea ED |
623 | seqcount_init(&sch->running); |
624 | lockdep_set_class(&sch->running, | |
625 | dev->qdisc_running_key ?: &qdisc_running_key); | |
626 | ||
1da177e4 LT |
627 | sch->ops = ops; |
628 | sch->enqueue = ops->enqueue; | |
629 | sch->dequeue = ops->dequeue; | |
bb949fbd | 630 | sch->dev_queue = dev_queue; |
23d3b8bf | 631 | dev_hold(dev); |
1da177e4 | 632 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
633 | |
634 | return sch; | |
635 | errout: | |
01e123d7 | 636 | return ERR_PTR(err); |
3d54b82f TG |
637 | } |
638 | ||
3511c913 | 639 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
d2a7f269 | 640 | const struct Qdisc_ops *ops, |
641 | unsigned int parentid) | |
3d54b82f TG |
642 | { |
643 | struct Qdisc *sch; | |
10297b99 | 644 | |
6da7c8fc | 645 | if (!try_module_get(ops->owner)) |
166ee5b8 | 646 | return NULL; |
6da7c8fc | 647 | |
5ce2d488 | 648 | sch = qdisc_alloc(dev_queue, ops); |
166ee5b8 ED |
649 | if (IS_ERR(sch)) { |
650 | module_put(ops->owner); | |
651 | return NULL; | |
652 | } | |
9f9afec4 | 653 | sch->parent = parentid; |
3d54b82f | 654 | |
1da177e4 LT |
655 | if (!ops->init || ops->init(sch, NULL) == 0) |
656 | return sch; | |
657 | ||
0fbbeb1b | 658 | qdisc_destroy(sch); |
1da177e4 LT |
659 | return NULL; |
660 | } | |
62e3ba1b | 661 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 662 | |
5fb66229 | 663 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
664 | |
665 | void qdisc_reset(struct Qdisc *qdisc) | |
666 | { | |
20fea08b | 667 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
668 | |
669 | if (ops->reset) | |
670 | ops->reset(qdisc); | |
67305ebc | 671 | |
4d202a0d ED |
672 | kfree_skb(qdisc->skb_bad_txq); |
673 | qdisc->skb_bad_txq = NULL; | |
674 | ||
bbd8a0d3 | 675 | if (qdisc->gso_skb) { |
3f3c7eec | 676 | kfree_skb_list(qdisc->gso_skb); |
bbd8a0d3 | 677 | qdisc->gso_skb = NULL; |
bbd8a0d3 | 678 | } |
4d202a0d | 679 | qdisc->q.qlen = 0; |
1da177e4 | 680 | } |
62e3ba1b | 681 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 682 | |
5d944c64 ED |
683 | static void qdisc_rcu_free(struct rcu_head *head) |
684 | { | |
685 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
686 | ||
73c20a8b | 687 | if (qdisc_is_percpu_stats(qdisc)) { |
22e0f8b9 | 688 | free_percpu(qdisc->cpu_bstats); |
73c20a8b JF |
689 | free_percpu(qdisc->cpu_qstats); |
690 | } | |
22e0f8b9 | 691 | |
5d944c64 ED |
692 | kfree((char *) qdisc - qdisc->padded); |
693 | } | |
694 | ||
1e0d5a57 | 695 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 696 | { |
8a34c5dc DM |
697 | const struct Qdisc_ops *ops = qdisc->ops; |
698 | ||
1e0d5a57 DM |
699 | if (qdisc->flags & TCQ_F_BUILTIN || |
700 | !atomic_dec_and_test(&qdisc->refcnt)) | |
701 | return; | |
702 | ||
3a682fbd | 703 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
704 | qdisc_list_del(qdisc); |
705 | ||
a2da570d | 706 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 707 | #endif |
8a34c5dc DM |
708 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
709 | if (ops->reset) | |
710 | ops->reset(qdisc); | |
711 | if (ops->destroy) | |
712 | ops->destroy(qdisc); | |
713 | ||
714 | module_put(ops->owner); | |
715 | dev_put(qdisc_dev(qdisc)); | |
716 | ||
3f3c7eec | 717 | kfree_skb_list(qdisc->gso_skb); |
4d202a0d | 718 | kfree_skb(qdisc->skb_bad_txq); |
5d944c64 ED |
719 | /* |
720 | * gen_estimator est_timer() might access qdisc->q.lock, | |
721 | * wait a RCU grace period before freeing qdisc. | |
722 | */ | |
723 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 724 | } |
62e3ba1b | 725 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 726 | |
589983cd PM |
727 | /* Attach toplevel qdisc to device queue. */ |
728 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
729 | struct Qdisc *qdisc) | |
730 | { | |
731 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
732 | spinlock_t *root_lock; | |
733 | ||
734 | root_lock = qdisc_lock(oqdisc); | |
735 | spin_lock_bh(root_lock); | |
736 | ||
737 | /* Prune old scheduler */ | |
738 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
739 | qdisc_reset(oqdisc); | |
740 | ||
741 | /* ... and graft new one */ | |
742 | if (qdisc == NULL) | |
743 | qdisc = &noop_qdisc; | |
744 | dev_queue->qdisc_sleeping = qdisc; | |
745 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
746 | ||
747 | spin_unlock_bh(root_lock); | |
748 | ||
749 | return oqdisc; | |
750 | } | |
b8970f0b | 751 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 752 | |
e8a0464c DM |
753 | static void attach_one_default_qdisc(struct net_device *dev, |
754 | struct netdev_queue *dev_queue, | |
755 | void *_unused) | |
756 | { | |
3e692f21 PS |
757 | struct Qdisc *qdisc; |
758 | const struct Qdisc_ops *ops = default_qdisc_ops; | |
e8a0464c | 759 | |
3e692f21 PS |
760 | if (dev->priv_flags & IFF_NO_QUEUE) |
761 | ops = &noqueue_qdisc_ops; | |
762 | ||
763 | qdisc = qdisc_create_dflt(dev_queue, ops, TC_H_ROOT); | |
764 | if (!qdisc) { | |
765 | netdev_info(dev, "activation failed\n"); | |
766 | return; | |
e8a0464c | 767 | } |
3e692f21 | 768 | if (!netif_is_multiqueue(dev)) |
4eaf3b84 | 769 | qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT; |
e8a0464c DM |
770 | dev_queue->qdisc_sleeping = qdisc; |
771 | } | |
772 | ||
6ec1c69a DM |
773 | static void attach_default_qdiscs(struct net_device *dev) |
774 | { | |
775 | struct netdev_queue *txq; | |
776 | struct Qdisc *qdisc; | |
777 | ||
778 | txq = netdev_get_tx_queue(dev, 0); | |
779 | ||
4b469955 | 780 | if (!netif_is_multiqueue(dev) || |
4b469955 | 781 | dev->priv_flags & IFF_NO_QUEUE) { |
6ec1c69a DM |
782 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); |
783 | dev->qdisc = txq->qdisc_sleeping; | |
784 | atomic_inc(&dev->qdisc->refcnt); | |
785 | } else { | |
3511c913 | 786 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT); |
6ec1c69a | 787 | if (qdisc) { |
6ec1c69a | 788 | dev->qdisc = qdisc; |
e57a784d | 789 | qdisc->ops->attach(qdisc); |
6ec1c69a DM |
790 | } |
791 | } | |
792 | } | |
793 | ||
e8a0464c DM |
794 | static void transition_one_qdisc(struct net_device *dev, |
795 | struct netdev_queue *dev_queue, | |
796 | void *_need_watchdog) | |
797 | { | |
83874000 | 798 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
799 | int *need_watchdog_p = _need_watchdog; |
800 | ||
a9312ae8 DM |
801 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
802 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
803 | ||
83874000 | 804 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
3e692f21 | 805 | if (need_watchdog_p) { |
9d21493b | 806 | dev_queue->trans_start = 0; |
e8a0464c | 807 | *need_watchdog_p = 1; |
9d21493b | 808 | } |
e8a0464c DM |
809 | } |
810 | ||
1da177e4 LT |
811 | void dev_activate(struct net_device *dev) |
812 | { | |
e8a0464c | 813 | int need_watchdog; |
b0e1e646 | 814 | |
1da177e4 | 815 | /* No queueing discipline is attached to device; |
6da7c8fc | 816 | * create default one for devices, which need queueing |
817 | * and noqueue_qdisc for virtual interfaces | |
1da177e4 LT |
818 | */ |
819 | ||
6ec1c69a DM |
820 | if (dev->qdisc == &noop_qdisc) |
821 | attach_default_qdiscs(dev); | |
af356afa | 822 | |
cacaddf5 TC |
823 | if (!netif_carrier_ok(dev)) |
824 | /* Delay activation until next carrier-on event */ | |
825 | return; | |
826 | ||
e8a0464c DM |
827 | need_watchdog = 0; |
828 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
24824a09 ED |
829 | if (dev_ingress_queue(dev)) |
830 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
831 | |
832 | if (need_watchdog) { | |
860e9538 | 833 | netif_trans_update(dev); |
1da177e4 LT |
834 | dev_watchdog_up(dev); |
835 | } | |
b0e1e646 | 836 | } |
b8970f0b | 837 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 838 | |
e8a0464c DM |
839 | static void dev_deactivate_queue(struct net_device *dev, |
840 | struct netdev_queue *dev_queue, | |
841 | void *_qdisc_default) | |
b0e1e646 | 842 | { |
e8a0464c | 843 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 844 | struct Qdisc *qdisc; |
970565bb | 845 | |
46e5da40 | 846 | qdisc = rtnl_dereference(dev_queue->qdisc); |
b0e1e646 | 847 | if (qdisc) { |
83874000 DM |
848 | spin_lock_bh(qdisc_lock(qdisc)); |
849 | ||
a9312ae8 DM |
850 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
851 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
852 | ||
f7a54c13 | 853 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 854 | qdisc_reset(qdisc); |
d3b753db | 855 | |
83874000 | 856 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 857 | } |
1da177e4 LT |
858 | } |
859 | ||
4335cd2d | 860 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
861 | { |
862 | unsigned int i; | |
863 | ||
864 | for (i = 0; i < dev->num_tx_queues; i++) { | |
865 | struct netdev_queue *dev_queue; | |
7698b4fc | 866 | spinlock_t *root_lock; |
e2627c8c | 867 | struct Qdisc *q; |
e8a0464c DM |
868 | int val; |
869 | ||
870 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 871 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 872 | root_lock = qdisc_lock(q); |
e8a0464c | 873 | |
4335cd2d | 874 | spin_lock_bh(root_lock); |
e8a0464c | 875 | |
bc135b23 | 876 | val = (qdisc_is_running(q) || |
b9a3b110 | 877 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 878 | |
4335cd2d | 879 | spin_unlock_bh(root_lock); |
e8a0464c DM |
880 | |
881 | if (val) | |
882 | return true; | |
883 | } | |
884 | return false; | |
885 | } | |
886 | ||
3137663d ED |
887 | /** |
888 | * dev_deactivate_many - deactivate transmissions on several devices | |
889 | * @head: list of devices to deactivate | |
890 | * | |
891 | * This function returns only when all outstanding transmissions | |
892 | * have completed, unless all devices are in dismantle phase. | |
893 | */ | |
44345724 | 894 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 895 | { |
44345724 | 896 | struct net_device *dev; |
3137663d | 897 | bool sync_needed = false; |
41a23b07 | 898 | |
5cde2829 | 899 | list_for_each_entry(dev, head, close_list) { |
44345724 OP |
900 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, |
901 | &noop_qdisc); | |
902 | if (dev_ingress_queue(dev)) | |
903 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
904 | &noop_qdisc); | |
905 | ||
906 | dev_watchdog_down(dev); | |
3137663d | 907 | sync_needed |= !dev->dismantle; |
44345724 | 908 | } |
1da177e4 | 909 | |
3137663d ED |
910 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
911 | * This is avoided if all devices are in dismantle phase : | |
912 | * Caller will call synchronize_net() for us | |
913 | */ | |
914 | if (sync_needed) | |
915 | synchronize_net(); | |
1da177e4 | 916 | |
d4828d85 | 917 | /* Wait for outstanding qdisc_run calls. */ |
5cde2829 | 918 | list_for_each_entry(dev, head, close_list) |
44345724 OP |
919 | while (some_qdisc_is_busy(dev)) |
920 | yield(); | |
921 | } | |
922 | ||
923 | void dev_deactivate(struct net_device *dev) | |
924 | { | |
925 | LIST_HEAD(single); | |
926 | ||
5cde2829 | 927 | list_add(&dev->close_list, &single); |
44345724 | 928 | dev_deactivate_many(&single); |
5f04d506 | 929 | list_del(&single); |
1da177e4 | 930 | } |
b8970f0b | 931 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 932 | |
b0e1e646 DM |
933 | static void dev_init_scheduler_queue(struct net_device *dev, |
934 | struct netdev_queue *dev_queue, | |
e8a0464c | 935 | void *_qdisc) |
b0e1e646 | 936 | { |
e8a0464c DM |
937 | struct Qdisc *qdisc = _qdisc; |
938 | ||
46e5da40 | 939 | rcu_assign_pointer(dev_queue->qdisc, qdisc); |
b0e1e646 | 940 | dev_queue->qdisc_sleeping = qdisc; |
b0e1e646 DM |
941 | } |
942 | ||
1da177e4 LT |
943 | void dev_init_scheduler(struct net_device *dev) |
944 | { | |
af356afa | 945 | dev->qdisc = &noop_qdisc; |
e8a0464c | 946 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
947 | if (dev_ingress_queue(dev)) |
948 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 949 | |
b24b8a24 | 950 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
951 | } |
952 | ||
e8a0464c DM |
953 | static void shutdown_scheduler_queue(struct net_device *dev, |
954 | struct netdev_queue *dev_queue, | |
955 | void *_qdisc_default) | |
1da177e4 | 956 | { |
b0e1e646 | 957 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 958 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
959 | |
960 | if (qdisc) { | |
f7a54c13 | 961 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 962 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 963 | |
1da177e4 | 964 | qdisc_destroy(qdisc); |
10297b99 | 965 | } |
b0e1e646 DM |
966 | } |
967 | ||
968 | void dev_shutdown(struct net_device *dev) | |
969 | { | |
e8a0464c | 970 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
971 | if (dev_ingress_queue(dev)) |
972 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
af356afa PM |
973 | qdisc_destroy(dev->qdisc); |
974 | dev->qdisc = &noop_qdisc; | |
975 | ||
547b792c | 976 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 977 | } |
292f1c7f | 978 | |
01cb71d2 | 979 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
3e1e3aae ED |
980 | const struct tc_ratespec *conf, |
981 | u64 rate64) | |
292f1c7f | 982 | { |
01cb71d2 ED |
983 | memset(r, 0, sizeof(*r)); |
984 | r->overhead = conf->overhead; | |
3e1e3aae | 985 | r->rate_bytes_ps = max_t(u64, conf->rate, rate64); |
8a8e3d84 | 986 | r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK); |
292f1c7f JP |
987 | r->mult = 1; |
988 | /* | |
130d3d68 ED |
989 | * The deal here is to replace a divide by a reciprocal one |
990 | * in fast path (a reciprocal divide is a multiply and a shift) | |
991 | * | |
992 | * Normal formula would be : | |
993 | * time_in_ns = (NSEC_PER_SEC * len) / rate_bps | |
994 | * | |
995 | * We compute mult/shift to use instead : | |
996 | * time_in_ns = (len * mult) >> shift; | |
997 | * | |
998 | * We try to get the highest possible mult value for accuracy, | |
999 | * but have to make sure no overflows will ever happen. | |
292f1c7f | 1000 | */ |
130d3d68 ED |
1001 | if (r->rate_bytes_ps > 0) { |
1002 | u64 factor = NSEC_PER_SEC; | |
1003 | ||
1004 | for (;;) { | |
1005 | r->mult = div64_u64(factor, r->rate_bytes_ps); | |
1006 | if (r->mult & (1U << 31) || factor & (1ULL << 63)) | |
292f1c7f | 1007 | break; |
130d3d68 ED |
1008 | factor <<= 1; |
1009 | r->shift++; | |
292f1c7f | 1010 | } |
292f1c7f JP |
1011 | } |
1012 | } | |
1013 | EXPORT_SYMBOL(psched_ratecfg_precompute); |