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> |
e543002f | 32 | #include <trace/events/qdisc.h> |
1da177e4 | 33 | |
34aedd3f | 34 | /* Qdisc to use by default */ |
35 | const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops; | |
36 | EXPORT_SYMBOL(default_qdisc_ops); | |
37 | ||
1da177e4 LT |
38 | /* Main transmission queue. */ |
39 | ||
0463d4ae | 40 | /* Modifications to data participating in scheduling must be protected with |
5fb66229 | 41 | * qdisc_lock(qdisc) spinlock. |
0463d4ae PM |
42 | * |
43 | * The idea is the following: | |
c7e4f3bb DM |
44 | * - enqueue, dequeue are serialized via qdisc root lock |
45 | * - ingress filtering is also serialized via qdisc root lock | |
0463d4ae | 46 | * - updates to tree and tree walking are only done under the rtnl mutex. |
1da177e4 | 47 | */ |
1da177e4 | 48 | |
37437bb2 | 49 | static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q) |
c716a81a | 50 | { |
6252352d | 51 | q->gso_skb = skb; |
53e91503 | 52 | q->qstats.requeues++; |
a27758ff | 53 | qdisc_qstats_backlog_inc(q, skb); |
bbd8a0d3 | 54 | q->q.qlen++; /* it's still part of the queue */ |
37437bb2 | 55 | __netif_schedule(q); |
6252352d | 56 | |
c716a81a JHS |
57 | return 0; |
58 | } | |
59 | ||
55a93b3e ED |
60 | static void try_bulk_dequeue_skb(struct Qdisc *q, |
61 | struct sk_buff *skb, | |
b8358d70 JDB |
62 | const struct netdev_queue *txq, |
63 | int *packets) | |
5772e9a3 | 64 | { |
55a93b3e | 65 | int bytelimit = qdisc_avail_bulklimit(txq) - skb->len; |
5772e9a3 JDB |
66 | |
67 | while (bytelimit > 0) { | |
55a93b3e | 68 | struct sk_buff *nskb = q->dequeue(q); |
5772e9a3 | 69 | |
55a93b3e | 70 | if (!nskb) |
5772e9a3 JDB |
71 | break; |
72 | ||
55a93b3e ED |
73 | bytelimit -= nskb->len; /* covers GSO len */ |
74 | skb->next = nskb; | |
75 | skb = nskb; | |
b8358d70 | 76 | (*packets)++; /* GSO counts as one pkt */ |
5772e9a3 | 77 | } |
55a93b3e | 78 | skb->next = NULL; |
5772e9a3 JDB |
79 | } |
80 | ||
4d202a0d ED |
81 | /* This variant of try_bulk_dequeue_skb() makes sure |
82 | * all skbs in the chain are for the same txq | |
83 | */ | |
84 | static void try_bulk_dequeue_skb_slow(struct Qdisc *q, | |
85 | struct sk_buff *skb, | |
86 | int *packets) | |
87 | { | |
88 | int mapping = skb_get_queue_mapping(skb); | |
89 | struct sk_buff *nskb; | |
90 | int cnt = 0; | |
91 | ||
92 | do { | |
93 | nskb = q->dequeue(q); | |
94 | if (!nskb) | |
95 | break; | |
96 | if (unlikely(skb_get_queue_mapping(nskb) != mapping)) { | |
97 | q->skb_bad_txq = nskb; | |
98 | qdisc_qstats_backlog_inc(q, nskb); | |
99 | q->q.qlen++; | |
100 | break; | |
101 | } | |
102 | skb->next = nskb; | |
103 | skb = nskb; | |
104 | } while (++cnt < 8); | |
105 | (*packets) += cnt; | |
106 | skb->next = NULL; | |
107 | } | |
108 | ||
5772e9a3 JDB |
109 | /* Note that dequeue_skb can possibly return a SKB list (via skb->next). |
110 | * A requeued skb (via q->gso_skb) can also be a SKB list. | |
111 | */ | |
b8358d70 JDB |
112 | static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate, |
113 | int *packets) | |
c716a81a | 114 | { |
554794de | 115 | struct sk_buff *skb = q->gso_skb; |
1abbe139 | 116 | const struct netdev_queue *txq = q->dev_queue; |
554794de | 117 | |
b8358d70 | 118 | *packets = 1; |
ebf05982 | 119 | if (unlikely(skb)) { |
4d202a0d ED |
120 | /* skb in gso_skb were already validated */ |
121 | *validate = false; | |
ebf05982 | 122 | /* check the reason of requeuing without tx lock first */ |
10c51b56 | 123 | txq = skb_get_tx_queue(txq->dev, skb); |
73466498 | 124 | if (!netif_xmit_frozen_or_stopped(txq)) { |
6252352d | 125 | q->gso_skb = NULL; |
a27758ff | 126 | qdisc_qstats_backlog_dec(q, skb); |
bbd8a0d3 KK |
127 | q->q.qlen--; |
128 | } else | |
ebf05982 | 129 | skb = NULL; |
e543002f | 130 | goto trace; |
4d202a0d ED |
131 | } |
132 | *validate = true; | |
133 | skb = q->skb_bad_txq; | |
134 | if (unlikely(skb)) { | |
135 | /* check the reason of requeuing without tx lock first */ | |
136 | txq = skb_get_tx_queue(txq->dev, skb); | |
137 | if (!netif_xmit_frozen_or_stopped(txq)) { | |
138 | q->skb_bad_txq = NULL; | |
139 | qdisc_qstats_backlog_dec(q, skb); | |
140 | q->q.qlen--; | |
141 | goto bulk; | |
50cbe9ab | 142 | } |
e543002f JDB |
143 | skb = NULL; |
144 | goto trace; | |
4d202a0d ED |
145 | } |
146 | if (!(q->flags & TCQ_F_ONETXQUEUE) || | |
147 | !netif_xmit_frozen_or_stopped(txq)) | |
148 | skb = q->dequeue(q); | |
149 | if (skb) { | |
150 | bulk: | |
151 | if (qdisc_may_bulk(q)) | |
152 | try_bulk_dequeue_skb(q, skb, txq, packets); | |
153 | else | |
154 | try_bulk_dequeue_skb_slow(q, skb, packets); | |
ebf05982 | 155 | } |
e543002f JDB |
156 | trace: |
157 | trace_qdisc_dequeue(q, txq, *packets, skb); | |
c716a81a JHS |
158 | return skb; |
159 | } | |
160 | ||
10297b99 | 161 | /* |
10770bc2 | 162 | * Transmit possibly several skbs, and handle the return status as |
f9eb8aea | 163 | * required. Owning running seqcount bit guarantees that |
10770bc2 | 164 | * only one CPU can execute this function. |
6c1361a6 KK |
165 | * |
166 | * Returns to the caller: | |
167 | * 0 - queue is empty or throttled. | |
168 | * >0 - queue is not empty. | |
6c1361a6 | 169 | */ |
bbd8a0d3 KK |
170 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
171 | struct net_device *dev, struct netdev_queue *txq, | |
55a93b3e | 172 | spinlock_t *root_lock, bool validate) |
1da177e4 | 173 | { |
5f1a485d | 174 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
175 | |
176 | /* And release qdisc */ | |
177 | spin_unlock(root_lock); | |
c716a81a | 178 | |
55a93b3e ED |
179 | /* Note that we validate skb (GSO, checksum, ...) outside of locks */ |
180 | if (validate) | |
181 | skb = validate_xmit_skb_list(skb, dev); | |
572a9d7b | 182 | |
3dcd493f | 183 | if (likely(skb)) { |
55a93b3e ED |
184 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
185 | if (!netif_xmit_frozen_or_stopped(txq)) | |
186 | skb = dev_hard_start_xmit(skb, dev, txq, &ret); | |
c716a81a | 187 | |
55a93b3e | 188 | HARD_TX_UNLOCK(dev, txq); |
3dcd493f | 189 | } else { |
52fbb290 | 190 | spin_lock(root_lock); |
3dcd493f | 191 | return qdisc_qlen(q); |
55a93b3e | 192 | } |
52fbb290 | 193 | spin_lock(root_lock); |
c716a81a | 194 | |
9a1654ba JP |
195 | if (dev_xmit_complete(ret)) { |
196 | /* Driver sent out skb successfully or skb was consumed */ | |
6c1361a6 | 197 | ret = qdisc_qlen(q); |
9a1654ba | 198 | } else { |
6c1361a6 | 199 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
e87cc472 JP |
200 | if (unlikely(ret != NETDEV_TX_BUSY)) |
201 | net_warn_ratelimited("BUG %s code %d qlen %d\n", | |
202 | dev->name, ret, q->q.qlen); | |
6c1361a6 | 203 | |
37437bb2 | 204 | ret = dev_requeue_skb(skb, q); |
6c1361a6 | 205 | } |
c716a81a | 206 | |
73466498 | 207 | if (ret && netif_xmit_frozen_or_stopped(txq)) |
37437bb2 DM |
208 | ret = 0; |
209 | ||
6c1361a6 | 210 | return ret; |
1da177e4 LT |
211 | } |
212 | ||
bbd8a0d3 KK |
213 | /* |
214 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. | |
215 | * | |
f9eb8aea | 216 | * running seqcount guarantees only one CPU can process |
bbd8a0d3 KK |
217 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for |
218 | * this queue. | |
219 | * | |
220 | * netif_tx_lock serializes accesses to device driver. | |
221 | * | |
222 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, | |
223 | * if one is grabbed, another must be free. | |
224 | * | |
225 | * Note, that this procedure can be called by a watchdog timer | |
226 | * | |
227 | * Returns to the caller: | |
228 | * 0 - queue is empty or throttled. | |
229 | * >0 - queue is not empty. | |
230 | * | |
231 | */ | |
b8358d70 | 232 | static inline int qdisc_restart(struct Qdisc *q, int *packets) |
bbd8a0d3 KK |
233 | { |
234 | struct netdev_queue *txq; | |
235 | struct net_device *dev; | |
236 | spinlock_t *root_lock; | |
237 | struct sk_buff *skb; | |
55a93b3e | 238 | bool validate; |
bbd8a0d3 KK |
239 | |
240 | /* Dequeue packet */ | |
b8358d70 | 241 | skb = dequeue_skb(q, &validate, packets); |
bbd8a0d3 KK |
242 | if (unlikely(!skb)) |
243 | return 0; | |
10c51b56 | 244 | |
bbd8a0d3 KK |
245 | root_lock = qdisc_lock(q); |
246 | dev = qdisc_dev(q); | |
10c51b56 | 247 | txq = skb_get_tx_queue(dev, skb); |
bbd8a0d3 | 248 | |
55a93b3e | 249 | return sch_direct_xmit(skb, q, dev, txq, root_lock, validate); |
bbd8a0d3 KK |
250 | } |
251 | ||
37437bb2 | 252 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 253 | { |
3d48b53f | 254 | int quota = dev_tx_weight; |
b8358d70 | 255 | int packets; |
2ba2506c | 256 | |
b8358d70 | 257 | while (qdisc_restart(q, &packets)) { |
2ba2506c | 258 | /* |
d5b8aa1d | 259 | * Ordered by possible occurrence: Postpone processing if |
260 | * 1. we've exceeded packet quota | |
261 | * 2. another process needs the CPU; | |
2ba2506c | 262 | */ |
b8358d70 JDB |
263 | quota -= packets; |
264 | if (quota <= 0 || need_resched()) { | |
37437bb2 | 265 | __netif_schedule(q); |
d90df3ad | 266 | break; |
2ba2506c HX |
267 | } |
268 | } | |
48d83325 | 269 | |
bc135b23 | 270 | qdisc_run_end(q); |
48d83325 HX |
271 | } |
272 | ||
9d21493b ED |
273 | unsigned long dev_trans_start(struct net_device *dev) |
274 | { | |
07ce76aa | 275 | unsigned long val, res; |
9d21493b ED |
276 | unsigned int i; |
277 | ||
07ce76aa | 278 | if (is_vlan_dev(dev)) |
279 | dev = vlan_dev_real_dev(dev); | |
9b36627a FW |
280 | res = netdev_get_tx_queue(dev, 0)->trans_start; |
281 | for (i = 1; i < dev->num_tx_queues; i++) { | |
9d21493b ED |
282 | val = netdev_get_tx_queue(dev, i)->trans_start; |
283 | if (val && time_after(val, res)) | |
284 | res = val; | |
285 | } | |
07ce76aa | 286 | |
9d21493b ED |
287 | return res; |
288 | } | |
289 | EXPORT_SYMBOL(dev_trans_start); | |
290 | ||
1da177e4 LT |
291 | static void dev_watchdog(unsigned long arg) |
292 | { | |
293 | struct net_device *dev = (struct net_device *)arg; | |
294 | ||
932ff279 | 295 | netif_tx_lock(dev); |
e8a0464c | 296 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
297 | if (netif_device_present(dev) && |
298 | netif_running(dev) && | |
299 | netif_carrier_ok(dev)) { | |
9d21493b | 300 | int some_queue_timedout = 0; |
e8a0464c | 301 | unsigned int i; |
9d21493b | 302 | unsigned long trans_start; |
e8a0464c DM |
303 | |
304 | for (i = 0; i < dev->num_tx_queues; i++) { | |
305 | struct netdev_queue *txq; | |
306 | ||
307 | txq = netdev_get_tx_queue(dev, i); | |
9b36627a | 308 | trans_start = txq->trans_start; |
73466498 | 309 | if (netif_xmit_stopped(txq) && |
9d21493b ED |
310 | time_after(jiffies, (trans_start + |
311 | dev->watchdog_timeo))) { | |
312 | some_queue_timedout = 1; | |
ccf5ff69 | 313 | txq->trans_timeout++; |
e8a0464c DM |
314 | break; |
315 | } | |
316 | } | |
338f7566 | 317 | |
9d21493b | 318 | if (some_queue_timedout) { |
9d21493b | 319 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 320 | dev->name, netdev_drivername(dev), i); |
d314774c | 321 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 322 | } |
e8a0464c DM |
323 | if (!mod_timer(&dev->watchdog_timer, |
324 | round_jiffies(jiffies + | |
325 | dev->watchdog_timeo))) | |
1da177e4 LT |
326 | dev_hold(dev); |
327 | } | |
328 | } | |
932ff279 | 329 | netif_tx_unlock(dev); |
1da177e4 LT |
330 | |
331 | dev_put(dev); | |
332 | } | |
333 | ||
1da177e4 LT |
334 | void __netdev_watchdog_up(struct net_device *dev) |
335 | { | |
d314774c | 336 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
337 | if (dev->watchdog_timeo <= 0) |
338 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
339 | if (!mod_timer(&dev->watchdog_timer, |
340 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
341 | dev_hold(dev); |
342 | } | |
343 | } | |
344 | ||
345 | static void dev_watchdog_up(struct net_device *dev) | |
346 | { | |
1da177e4 | 347 | __netdev_watchdog_up(dev); |
1da177e4 LT |
348 | } |
349 | ||
350 | static void dev_watchdog_down(struct net_device *dev) | |
351 | { | |
932ff279 | 352 | netif_tx_lock_bh(dev); |
1da177e4 | 353 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 354 | dev_put(dev); |
932ff279 | 355 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
356 | } |
357 | ||
bea3348e SH |
358 | /** |
359 | * netif_carrier_on - set carrier | |
360 | * @dev: network device | |
361 | * | |
362 | * Device has detected that carrier. | |
363 | */ | |
0a242efc DV |
364 | void netif_carrier_on(struct net_device *dev) |
365 | { | |
bfaae0f0 | 366 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
367 | if (dev->reg_state == NETREG_UNINITIALIZED) |
368 | return; | |
2d3b479d | 369 | atomic_inc(&dev->carrier_changes); |
0a242efc | 370 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
371 | if (netif_running(dev)) |
372 | __netdev_watchdog_up(dev); | |
373 | } | |
0a242efc | 374 | } |
62e3ba1b | 375 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 376 | |
bea3348e SH |
377 | /** |
378 | * netif_carrier_off - clear carrier | |
379 | * @dev: network device | |
380 | * | |
381 | * Device has detected loss of carrier. | |
382 | */ | |
0a242efc DV |
383 | void netif_carrier_off(struct net_device *dev) |
384 | { | |
b4730016 DM |
385 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
386 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
387 | return; | |
2d3b479d | 388 | atomic_inc(&dev->carrier_changes); |
0a242efc | 389 | linkwatch_fire_event(dev); |
b4730016 | 390 | } |
0a242efc | 391 | } |
62e3ba1b | 392 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 393 | |
1da177e4 LT |
394 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
395 | under all circumstances. It is difficult to invent anything faster or | |
396 | cheaper. | |
397 | */ | |
398 | ||
520ac30f ED |
399 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc, |
400 | struct sk_buff **to_free) | |
1da177e4 | 401 | { |
520ac30f | 402 | __qdisc_drop(skb, to_free); |
1da177e4 LT |
403 | return NET_XMIT_CN; |
404 | } | |
405 | ||
82d567c2 | 406 | static struct sk_buff *noop_dequeue(struct Qdisc *qdisc) |
1da177e4 LT |
407 | { |
408 | return NULL; | |
409 | } | |
410 | ||
20fea08b | 411 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
412 | .id = "noop", |
413 | .priv_size = 0, | |
414 | .enqueue = noop_enqueue, | |
415 | .dequeue = noop_dequeue, | |
99c0db26 | 416 | .peek = noop_dequeue, |
1da177e4 LT |
417 | .owner = THIS_MODULE, |
418 | }; | |
419 | ||
7698b4fc | 420 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 421 | .qdisc = &noop_qdisc, |
9f3ffae0 | 422 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
423 | }; |
424 | ||
1da177e4 LT |
425 | struct Qdisc noop_qdisc = { |
426 | .enqueue = noop_enqueue, | |
427 | .dequeue = noop_dequeue, | |
428 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 429 | .ops = &noop_qdisc_ops, |
83874000 | 430 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 431 | .dev_queue = &noop_netdev_queue, |
f9eb8aea | 432 | .running = SEQCNT_ZERO(noop_qdisc.running), |
7b5edbc4 | 433 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
1da177e4 | 434 | }; |
62e3ba1b | 435 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 436 | |
d66d6c31 PS |
437 | static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt) |
438 | { | |
439 | /* register_qdisc() assigns a default of noop_enqueue if unset, | |
440 | * but __dev_queue_xmit() treats noqueue only as such | |
441 | * if this is NULL - so clear it here. */ | |
442 | qdisc->enqueue = NULL; | |
443 | return 0; | |
444 | } | |
445 | ||
446 | struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { | |
1da177e4 LT |
447 | .id = "noqueue", |
448 | .priv_size = 0, | |
d66d6c31 | 449 | .init = noqueue_init, |
1da177e4 LT |
450 | .enqueue = noop_enqueue, |
451 | .dequeue = noop_dequeue, | |
99c0db26 | 452 | .peek = noop_dequeue, |
1da177e4 LT |
453 | .owner = THIS_MODULE, |
454 | }; | |
455 | ||
cc7ec456 ED |
456 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
457 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
458 | }; | |
d3678b46 DM |
459 | |
460 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
461 | generic prio+fifo combination. | |
462 | */ | |
463 | ||
464 | #define PFIFO_FAST_BANDS 3 | |
465 | ||
fd3ae5e8 KK |
466 | /* |
467 | * Private data for a pfifo_fast scheduler containing: | |
468 | * - queues for the three band | |
469 | * - bitmap indicating which of the bands contain skbs | |
470 | */ | |
471 | struct pfifo_fast_priv { | |
472 | u32 bitmap; | |
48da34b7 | 473 | struct qdisc_skb_head q[PFIFO_FAST_BANDS]; |
fd3ae5e8 KK |
474 | }; |
475 | ||
476 | /* | |
477 | * Convert a bitmap to the first band number where an skb is queued, where: | |
478 | * bitmap=0 means there are no skbs on any band. | |
479 | * bitmap=1 means there is an skb on band 0. | |
480 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
481 | */ | |
482 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
483 | ||
48da34b7 | 484 | static inline struct qdisc_skb_head *band2list(struct pfifo_fast_priv *priv, |
fd3ae5e8 | 485 | int band) |
d3678b46 | 486 | { |
fd3ae5e8 | 487 | return priv->q + band; |
d3678b46 DM |
488 | } |
489 | ||
520ac30f ED |
490 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc, |
491 | struct sk_buff **to_free) | |
321090e7 | 492 | { |
97d0678f | 493 | if (qdisc->q.qlen < qdisc_dev(qdisc)->tx_queue_len) { |
a453e068 KK |
494 | int band = prio2band[skb->priority & TC_PRIO_MAX]; |
495 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
48da34b7 | 496 | struct qdisc_skb_head *list = band2list(priv, band); |
1da177e4 | 497 | |
fd3ae5e8 | 498 | priv->bitmap |= (1 << band); |
d3678b46 | 499 | qdisc->q.qlen++; |
821d24ae | 500 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 501 | } |
821d24ae | 502 | |
520ac30f | 503 | return qdisc_drop(skb, qdisc, to_free); |
1da177e4 LT |
504 | } |
505 | ||
cc7ec456 | 506 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc) |
1da177e4 | 507 | { |
fd3ae5e8 KK |
508 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
509 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 510 | |
fd3ae5e8 | 511 | if (likely(band >= 0)) { |
48da34b7 FW |
512 | struct qdisc_skb_head *qh = band2list(priv, band); |
513 | struct sk_buff *skb = __qdisc_dequeue_head(qh); | |
ec323368 FW |
514 | |
515 | if (likely(skb != NULL)) { | |
516 | qdisc_qstats_backlog_dec(qdisc, skb); | |
517 | qdisc_bstats_update(qdisc, skb); | |
518 | } | |
fd3ae5e8 KK |
519 | |
520 | qdisc->q.qlen--; | |
48da34b7 | 521 | if (qh->qlen == 0) |
fd3ae5e8 KK |
522 | priv->bitmap &= ~(1 << band); |
523 | ||
524 | return skb; | |
d3678b46 | 525 | } |
f87a9c3d | 526 | |
1da177e4 LT |
527 | return NULL; |
528 | } | |
529 | ||
cc7ec456 | 530 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 531 | { |
fd3ae5e8 KK |
532 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
533 | int band = bitmap2band[priv->bitmap]; | |
534 | ||
535 | if (band >= 0) { | |
48da34b7 | 536 | struct qdisc_skb_head *qh = band2list(priv, band); |
99c0db26 | 537 | |
48da34b7 | 538 | return qh->head; |
99c0db26 JP |
539 | } |
540 | ||
541 | return NULL; | |
542 | } | |
543 | ||
cc7ec456 | 544 | static void pfifo_fast_reset(struct Qdisc *qdisc) |
1da177e4 | 545 | { |
d3678b46 | 546 | int prio; |
fd3ae5e8 | 547 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
548 | |
549 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
1b5c5493 | 550 | __qdisc_reset_queue(band2list(priv, prio)); |
d3678b46 | 551 | |
fd3ae5e8 | 552 | priv->bitmap = 0; |
821d24ae | 553 | qdisc->qstats.backlog = 0; |
d3678b46 | 554 | qdisc->q.qlen = 0; |
1da177e4 LT |
555 | } |
556 | ||
d3678b46 DM |
557 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
558 | { | |
559 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
560 | ||
cc7ec456 | 561 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
1b34ec43 DM |
562 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
563 | goto nla_put_failure; | |
d3678b46 DM |
564 | return skb->len; |
565 | ||
566 | nla_put_failure: | |
567 | return -1; | |
568 | } | |
569 | ||
570 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
571 | { | |
572 | int prio; | |
fd3ae5e8 | 573 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
574 | |
575 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
48da34b7 | 576 | qdisc_skb_head_init(band2list(priv, prio)); |
d3678b46 | 577 | |
23624935 ED |
578 | /* Can by-pass the queue discipline */ |
579 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
580 | return 0; |
581 | } | |
582 | ||
6ec1c69a | 583 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 584 | .id = "pfifo_fast", |
fd3ae5e8 | 585 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
586 | .enqueue = pfifo_fast_enqueue, |
587 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 588 | .peek = pfifo_fast_peek, |
d3678b46 DM |
589 | .init = pfifo_fast_init, |
590 | .reset = pfifo_fast_reset, | |
591 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
592 | .owner = THIS_MODULE, |
593 | }; | |
1f27cde3 | 594 | EXPORT_SYMBOL(pfifo_fast_ops); |
1da177e4 | 595 | |
23d3b8bf | 596 | static struct lock_class_key qdisc_tx_busylock; |
f9eb8aea | 597 | static struct lock_class_key qdisc_running_key; |
23d3b8bf | 598 | |
5ce2d488 | 599 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
d2a7f269 | 600 | const struct Qdisc_ops *ops) |
1da177e4 LT |
601 | { |
602 | void *p; | |
603 | struct Qdisc *sch; | |
d276055c | 604 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 605 | int err = -ENOBUFS; |
23d3b8bf | 606 | struct net_device *dev = dev_queue->dev; |
1da177e4 | 607 | |
f2cd2d3e ED |
608 | p = kzalloc_node(size, GFP_KERNEL, |
609 | netdev_queue_numa_node_read(dev_queue)); | |
610 | ||
1da177e4 | 611 | if (!p) |
3d54b82f | 612 | goto errout; |
3d54b82f | 613 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
614 | /* if we got non aligned memory, ask more and do alignment ourself */ |
615 | if (sch != p) { | |
616 | kfree(p); | |
617 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
618 | netdev_queue_numa_node_read(dev_queue)); | |
619 | if (!p) | |
620 | goto errout; | |
621 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
622 | sch->padded = (char *) sch - (char *) p; | |
623 | } | |
48da34b7 FW |
624 | qdisc_skb_head_init(&sch->q); |
625 | spin_lock_init(&sch->q.lock); | |
23d3b8bf | 626 | |
79640a4c | 627 | spin_lock_init(&sch->busylock); |
23d3b8bf ED |
628 | lockdep_set_class(&sch->busylock, |
629 | dev->qdisc_tx_busylock ?: &qdisc_tx_busylock); | |
630 | ||
f9eb8aea ED |
631 | seqcount_init(&sch->running); |
632 | lockdep_set_class(&sch->running, | |
633 | dev->qdisc_running_key ?: &qdisc_running_key); | |
634 | ||
1da177e4 LT |
635 | sch->ops = ops; |
636 | sch->enqueue = ops->enqueue; | |
637 | sch->dequeue = ops->dequeue; | |
bb949fbd | 638 | sch->dev_queue = dev_queue; |
23d3b8bf | 639 | dev_hold(dev); |
7b936405 | 640 | refcount_set(&sch->refcnt, 1); |
3d54b82f TG |
641 | |
642 | return sch; | |
643 | errout: | |
01e123d7 | 644 | return ERR_PTR(err); |
3d54b82f TG |
645 | } |
646 | ||
3511c913 | 647 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
d2a7f269 | 648 | const struct Qdisc_ops *ops, |
649 | unsigned int parentid) | |
3d54b82f TG |
650 | { |
651 | struct Qdisc *sch; | |
10297b99 | 652 | |
6da7c8fc | 653 | if (!try_module_get(ops->owner)) |
166ee5b8 | 654 | return NULL; |
6da7c8fc | 655 | |
5ce2d488 | 656 | sch = qdisc_alloc(dev_queue, ops); |
166ee5b8 ED |
657 | if (IS_ERR(sch)) { |
658 | module_put(ops->owner); | |
659 | return NULL; | |
660 | } | |
9f9afec4 | 661 | sch->parent = parentid; |
3d54b82f | 662 | |
1da177e4 LT |
663 | if (!ops->init || ops->init(sch, NULL) == 0) |
664 | return sch; | |
665 | ||
0fbbeb1b | 666 | qdisc_destroy(sch); |
1da177e4 LT |
667 | return NULL; |
668 | } | |
62e3ba1b | 669 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 670 | |
5fb66229 | 671 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
672 | |
673 | void qdisc_reset(struct Qdisc *qdisc) | |
674 | { | |
20fea08b | 675 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
676 | |
677 | if (ops->reset) | |
678 | ops->reset(qdisc); | |
67305ebc | 679 | |
4d202a0d ED |
680 | kfree_skb(qdisc->skb_bad_txq); |
681 | qdisc->skb_bad_txq = NULL; | |
682 | ||
bbd8a0d3 | 683 | if (qdisc->gso_skb) { |
3f3c7eec | 684 | kfree_skb_list(qdisc->gso_skb); |
bbd8a0d3 | 685 | qdisc->gso_skb = NULL; |
bbd8a0d3 | 686 | } |
4d202a0d | 687 | qdisc->q.qlen = 0; |
c8e18129 | 688 | qdisc->qstats.backlog = 0; |
1da177e4 | 689 | } |
62e3ba1b | 690 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 691 | |
5d944c64 ED |
692 | static void qdisc_rcu_free(struct rcu_head *head) |
693 | { | |
694 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
695 | ||
73c20a8b | 696 | if (qdisc_is_percpu_stats(qdisc)) { |
22e0f8b9 | 697 | free_percpu(qdisc->cpu_bstats); |
73c20a8b JF |
698 | free_percpu(qdisc->cpu_qstats); |
699 | } | |
22e0f8b9 | 700 | |
5d944c64 ED |
701 | kfree((char *) qdisc - qdisc->padded); |
702 | } | |
703 | ||
1e0d5a57 | 704 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 705 | { |
8a34c5dc DM |
706 | const struct Qdisc_ops *ops = qdisc->ops; |
707 | ||
1e0d5a57 | 708 | if (qdisc->flags & TCQ_F_BUILTIN || |
7b936405 | 709 | !refcount_dec_and_test(&qdisc->refcnt)) |
1e0d5a57 DM |
710 | return; |
711 | ||
3a682fbd | 712 | #ifdef CONFIG_NET_SCHED |
59cc1f61 | 713 | qdisc_hash_del(qdisc); |
f6e0b239 | 714 | |
a2da570d | 715 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 716 | #endif |
1c0d32fd | 717 | gen_kill_estimator(&qdisc->rate_est); |
8a34c5dc DM |
718 | if (ops->reset) |
719 | ops->reset(qdisc); | |
720 | if (ops->destroy) | |
721 | ops->destroy(qdisc); | |
722 | ||
723 | module_put(ops->owner); | |
724 | dev_put(qdisc_dev(qdisc)); | |
725 | ||
3f3c7eec | 726 | kfree_skb_list(qdisc->gso_skb); |
4d202a0d | 727 | kfree_skb(qdisc->skb_bad_txq); |
5d944c64 ED |
728 | /* |
729 | * gen_estimator est_timer() might access qdisc->q.lock, | |
730 | * wait a RCU grace period before freeing qdisc. | |
731 | */ | |
732 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 733 | } |
62e3ba1b | 734 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 735 | |
589983cd PM |
736 | /* Attach toplevel qdisc to device queue. */ |
737 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
738 | struct Qdisc *qdisc) | |
739 | { | |
740 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
741 | spinlock_t *root_lock; | |
742 | ||
743 | root_lock = qdisc_lock(oqdisc); | |
744 | spin_lock_bh(root_lock); | |
745 | ||
746 | /* Prune old scheduler */ | |
7b936405 | 747 | if (oqdisc && refcount_read(&oqdisc->refcnt) <= 1) |
589983cd PM |
748 | qdisc_reset(oqdisc); |
749 | ||
750 | /* ... and graft new one */ | |
751 | if (qdisc == NULL) | |
752 | qdisc = &noop_qdisc; | |
753 | dev_queue->qdisc_sleeping = qdisc; | |
754 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
755 | ||
756 | spin_unlock_bh(root_lock); | |
757 | ||
758 | return oqdisc; | |
759 | } | |
b8970f0b | 760 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 761 | |
e8a0464c DM |
762 | static void attach_one_default_qdisc(struct net_device *dev, |
763 | struct netdev_queue *dev_queue, | |
764 | void *_unused) | |
765 | { | |
3e692f21 PS |
766 | struct Qdisc *qdisc; |
767 | const struct Qdisc_ops *ops = default_qdisc_ops; | |
e8a0464c | 768 | |
3e692f21 PS |
769 | if (dev->priv_flags & IFF_NO_QUEUE) |
770 | ops = &noqueue_qdisc_ops; | |
771 | ||
772 | qdisc = qdisc_create_dflt(dev_queue, ops, TC_H_ROOT); | |
773 | if (!qdisc) { | |
774 | netdev_info(dev, "activation failed\n"); | |
775 | return; | |
e8a0464c | 776 | } |
3e692f21 | 777 | if (!netif_is_multiqueue(dev)) |
4eaf3b84 | 778 | qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT; |
e8a0464c DM |
779 | dev_queue->qdisc_sleeping = qdisc; |
780 | } | |
781 | ||
6ec1c69a DM |
782 | static void attach_default_qdiscs(struct net_device *dev) |
783 | { | |
784 | struct netdev_queue *txq; | |
785 | struct Qdisc *qdisc; | |
786 | ||
787 | txq = netdev_get_tx_queue(dev, 0); | |
788 | ||
4b469955 | 789 | if (!netif_is_multiqueue(dev) || |
4b469955 | 790 | dev->priv_flags & IFF_NO_QUEUE) { |
6ec1c69a DM |
791 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); |
792 | dev->qdisc = txq->qdisc_sleeping; | |
551143d8 | 793 | qdisc_refcount_inc(dev->qdisc); |
6ec1c69a | 794 | } else { |
3511c913 | 795 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT); |
6ec1c69a | 796 | if (qdisc) { |
6ec1c69a | 797 | dev->qdisc = qdisc; |
e57a784d | 798 | qdisc->ops->attach(qdisc); |
6ec1c69a DM |
799 | } |
800 | } | |
59cc1f61 | 801 | #ifdef CONFIG_NET_SCHED |
92f91706 | 802 | if (dev->qdisc != &noop_qdisc) |
49b49971 | 803 | qdisc_hash_add(dev->qdisc, false); |
59cc1f61 | 804 | #endif |
6ec1c69a DM |
805 | } |
806 | ||
e8a0464c DM |
807 | static void transition_one_qdisc(struct net_device *dev, |
808 | struct netdev_queue *dev_queue, | |
809 | void *_need_watchdog) | |
810 | { | |
83874000 | 811 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
812 | int *need_watchdog_p = _need_watchdog; |
813 | ||
a9312ae8 DM |
814 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
815 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
816 | ||
83874000 | 817 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
3e692f21 | 818 | if (need_watchdog_p) { |
9d21493b | 819 | dev_queue->trans_start = 0; |
e8a0464c | 820 | *need_watchdog_p = 1; |
9d21493b | 821 | } |
e8a0464c DM |
822 | } |
823 | ||
1da177e4 LT |
824 | void dev_activate(struct net_device *dev) |
825 | { | |
e8a0464c | 826 | int need_watchdog; |
b0e1e646 | 827 | |
1da177e4 | 828 | /* No queueing discipline is attached to device; |
6da7c8fc | 829 | * create default one for devices, which need queueing |
830 | * and noqueue_qdisc for virtual interfaces | |
1da177e4 LT |
831 | */ |
832 | ||
6ec1c69a DM |
833 | if (dev->qdisc == &noop_qdisc) |
834 | attach_default_qdiscs(dev); | |
af356afa | 835 | |
cacaddf5 TC |
836 | if (!netif_carrier_ok(dev)) |
837 | /* Delay activation until next carrier-on event */ | |
838 | return; | |
839 | ||
e8a0464c DM |
840 | need_watchdog = 0; |
841 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
24824a09 ED |
842 | if (dev_ingress_queue(dev)) |
843 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
844 | |
845 | if (need_watchdog) { | |
860e9538 | 846 | netif_trans_update(dev); |
1da177e4 LT |
847 | dev_watchdog_up(dev); |
848 | } | |
b0e1e646 | 849 | } |
b8970f0b | 850 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 851 | |
e8a0464c DM |
852 | static void dev_deactivate_queue(struct net_device *dev, |
853 | struct netdev_queue *dev_queue, | |
854 | void *_qdisc_default) | |
b0e1e646 | 855 | { |
e8a0464c | 856 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 857 | struct Qdisc *qdisc; |
970565bb | 858 | |
46e5da40 | 859 | qdisc = rtnl_dereference(dev_queue->qdisc); |
b0e1e646 | 860 | if (qdisc) { |
83874000 DM |
861 | spin_lock_bh(qdisc_lock(qdisc)); |
862 | ||
a9312ae8 DM |
863 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
864 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
865 | ||
f7a54c13 | 866 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 867 | qdisc_reset(qdisc); |
d3b753db | 868 | |
83874000 | 869 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 870 | } |
1da177e4 LT |
871 | } |
872 | ||
4335cd2d | 873 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
874 | { |
875 | unsigned int i; | |
876 | ||
877 | for (i = 0; i < dev->num_tx_queues; i++) { | |
878 | struct netdev_queue *dev_queue; | |
7698b4fc | 879 | spinlock_t *root_lock; |
e2627c8c | 880 | struct Qdisc *q; |
e8a0464c DM |
881 | int val; |
882 | ||
883 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 884 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 885 | root_lock = qdisc_lock(q); |
e8a0464c | 886 | |
4335cd2d | 887 | spin_lock_bh(root_lock); |
e8a0464c | 888 | |
bc135b23 | 889 | val = (qdisc_is_running(q) || |
b9a3b110 | 890 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 891 | |
4335cd2d | 892 | spin_unlock_bh(root_lock); |
e8a0464c DM |
893 | |
894 | if (val) | |
895 | return true; | |
896 | } | |
897 | return false; | |
898 | } | |
899 | ||
3137663d ED |
900 | /** |
901 | * dev_deactivate_many - deactivate transmissions on several devices | |
902 | * @head: list of devices to deactivate | |
903 | * | |
904 | * This function returns only when all outstanding transmissions | |
905 | * have completed, unless all devices are in dismantle phase. | |
906 | */ | |
44345724 | 907 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 908 | { |
44345724 | 909 | struct net_device *dev; |
3137663d | 910 | bool sync_needed = false; |
41a23b07 | 911 | |
5cde2829 | 912 | list_for_each_entry(dev, head, close_list) { |
44345724 OP |
913 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, |
914 | &noop_qdisc); | |
915 | if (dev_ingress_queue(dev)) | |
916 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
917 | &noop_qdisc); | |
918 | ||
919 | dev_watchdog_down(dev); | |
3137663d | 920 | sync_needed |= !dev->dismantle; |
44345724 | 921 | } |
1da177e4 | 922 | |
3137663d ED |
923 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
924 | * This is avoided if all devices are in dismantle phase : | |
925 | * Caller will call synchronize_net() for us | |
926 | */ | |
927 | if (sync_needed) | |
928 | synchronize_net(); | |
1da177e4 | 929 | |
d4828d85 | 930 | /* Wait for outstanding qdisc_run calls. */ |
5cde2829 | 931 | list_for_each_entry(dev, head, close_list) |
44345724 OP |
932 | while (some_qdisc_is_busy(dev)) |
933 | yield(); | |
934 | } | |
935 | ||
936 | void dev_deactivate(struct net_device *dev) | |
937 | { | |
938 | LIST_HEAD(single); | |
939 | ||
5cde2829 | 940 | list_add(&dev->close_list, &single); |
44345724 | 941 | dev_deactivate_many(&single); |
5f04d506 | 942 | list_del(&single); |
1da177e4 | 943 | } |
b8970f0b | 944 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 945 | |
b0e1e646 DM |
946 | static void dev_init_scheduler_queue(struct net_device *dev, |
947 | struct netdev_queue *dev_queue, | |
e8a0464c | 948 | void *_qdisc) |
b0e1e646 | 949 | { |
e8a0464c DM |
950 | struct Qdisc *qdisc = _qdisc; |
951 | ||
46e5da40 | 952 | rcu_assign_pointer(dev_queue->qdisc, qdisc); |
b0e1e646 | 953 | dev_queue->qdisc_sleeping = qdisc; |
b0e1e646 DM |
954 | } |
955 | ||
1da177e4 LT |
956 | void dev_init_scheduler(struct net_device *dev) |
957 | { | |
af356afa | 958 | dev->qdisc = &noop_qdisc; |
e8a0464c | 959 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
960 | if (dev_ingress_queue(dev)) |
961 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 962 | |
b24b8a24 | 963 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
964 | } |
965 | ||
e8a0464c DM |
966 | static void shutdown_scheduler_queue(struct net_device *dev, |
967 | struct netdev_queue *dev_queue, | |
968 | void *_qdisc_default) | |
1da177e4 | 969 | { |
b0e1e646 | 970 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 971 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
972 | |
973 | if (qdisc) { | |
f7a54c13 | 974 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 975 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 976 | |
1da177e4 | 977 | qdisc_destroy(qdisc); |
10297b99 | 978 | } |
b0e1e646 DM |
979 | } |
980 | ||
981 | void dev_shutdown(struct net_device *dev) | |
982 | { | |
e8a0464c | 983 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
984 | if (dev_ingress_queue(dev)) |
985 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
af356afa PM |
986 | qdisc_destroy(dev->qdisc); |
987 | dev->qdisc = &noop_qdisc; | |
988 | ||
547b792c | 989 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 990 | } |
292f1c7f | 991 | |
01cb71d2 | 992 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
3e1e3aae ED |
993 | const struct tc_ratespec *conf, |
994 | u64 rate64) | |
292f1c7f | 995 | { |
01cb71d2 ED |
996 | memset(r, 0, sizeof(*r)); |
997 | r->overhead = conf->overhead; | |
3e1e3aae | 998 | r->rate_bytes_ps = max_t(u64, conf->rate, rate64); |
8a8e3d84 | 999 | r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK); |
292f1c7f JP |
1000 | r->mult = 1; |
1001 | /* | |
130d3d68 ED |
1002 | * The deal here is to replace a divide by a reciprocal one |
1003 | * in fast path (a reciprocal divide is a multiply and a shift) | |
1004 | * | |
1005 | * Normal formula would be : | |
1006 | * time_in_ns = (NSEC_PER_SEC * len) / rate_bps | |
1007 | * | |
1008 | * We compute mult/shift to use instead : | |
1009 | * time_in_ns = (len * mult) >> shift; | |
1010 | * | |
1011 | * We try to get the highest possible mult value for accuracy, | |
1012 | * but have to make sure no overflows will ever happen. | |
292f1c7f | 1013 | */ |
130d3d68 ED |
1014 | if (r->rate_bytes_ps > 0) { |
1015 | u64 factor = NSEC_PER_SEC; | |
1016 | ||
1017 | for (;;) { | |
1018 | r->mult = div64_u64(factor, r->rate_bytes_ps); | |
1019 | if (r->mult & (1U << 31) || factor & (1ULL << 63)) | |
292f1c7f | 1020 | break; |
130d3d68 ED |
1021 | factor <<= 1; |
1022 | r->shift++; | |
292f1c7f | 1023 | } |
292f1c7f JP |
1024 | } |
1025 | } | |
1026 | EXPORT_SYMBOL(psched_ratecfg_precompute); |