]>
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> |
32d3e51a | 29 | #include <linux/if_macvlan.h> |
292f1c7f | 30 | #include <net/sch_generic.h> |
1da177e4 | 31 | #include <net/pkt_sched.h> |
7fee226a | 32 | #include <net/dst.h> |
e543002f | 33 | #include <trace/events/qdisc.h> |
1da177e4 | 34 | |
34aedd3f | 35 | /* Qdisc to use by default */ |
36 | const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops; | |
37 | EXPORT_SYMBOL(default_qdisc_ops); | |
38 | ||
1da177e4 LT |
39 | /* Main transmission queue. */ |
40 | ||
0463d4ae | 41 | /* Modifications to data participating in scheduling must be protected with |
5fb66229 | 42 | * qdisc_lock(qdisc) spinlock. |
0463d4ae PM |
43 | * |
44 | * The idea is the following: | |
c7e4f3bb DM |
45 | * - enqueue, dequeue are serialized via qdisc root lock |
46 | * - ingress filtering is also serialized via qdisc root lock | |
0463d4ae | 47 | * - updates to tree and tree walking are only done under the rtnl mutex. |
1da177e4 | 48 | */ |
1da177e4 | 49 | |
37437bb2 | 50 | static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q) |
c716a81a | 51 | { |
6252352d | 52 | q->gso_skb = skb; |
53e91503 | 53 | q->qstats.requeues++; |
a27758ff | 54 | qdisc_qstats_backlog_inc(q, skb); |
bbd8a0d3 | 55 | q->q.qlen++; /* it's still part of the queue */ |
37437bb2 | 56 | __netif_schedule(q); |
6252352d | 57 | |
c716a81a JHS |
58 | return 0; |
59 | } | |
60 | ||
55a93b3e ED |
61 | static void try_bulk_dequeue_skb(struct Qdisc *q, |
62 | struct sk_buff *skb, | |
b8358d70 JDB |
63 | const struct netdev_queue *txq, |
64 | int *packets) | |
5772e9a3 | 65 | { |
55a93b3e | 66 | int bytelimit = qdisc_avail_bulklimit(txq) - skb->len; |
5772e9a3 JDB |
67 | |
68 | while (bytelimit > 0) { | |
55a93b3e | 69 | struct sk_buff *nskb = q->dequeue(q); |
5772e9a3 | 70 | |
55a93b3e | 71 | if (!nskb) |
5772e9a3 JDB |
72 | break; |
73 | ||
55a93b3e ED |
74 | bytelimit -= nskb->len; /* covers GSO len */ |
75 | skb->next = nskb; | |
76 | skb = nskb; | |
b8358d70 | 77 | (*packets)++; /* GSO counts as one pkt */ |
5772e9a3 | 78 | } |
55a93b3e | 79 | skb->next = NULL; |
5772e9a3 JDB |
80 | } |
81 | ||
4d202a0d ED |
82 | /* This variant of try_bulk_dequeue_skb() makes sure |
83 | * all skbs in the chain are for the same txq | |
84 | */ | |
85 | static void try_bulk_dequeue_skb_slow(struct Qdisc *q, | |
86 | struct sk_buff *skb, | |
87 | int *packets) | |
88 | { | |
89 | int mapping = skb_get_queue_mapping(skb); | |
90 | struct sk_buff *nskb; | |
91 | int cnt = 0; | |
92 | ||
93 | do { | |
94 | nskb = q->dequeue(q); | |
95 | if (!nskb) | |
96 | break; | |
97 | if (unlikely(skb_get_queue_mapping(nskb) != mapping)) { | |
98 | q->skb_bad_txq = nskb; | |
99 | qdisc_qstats_backlog_inc(q, nskb); | |
100 | q->q.qlen++; | |
101 | break; | |
102 | } | |
103 | skb->next = nskb; | |
104 | skb = nskb; | |
105 | } while (++cnt < 8); | |
106 | (*packets) += cnt; | |
107 | skb->next = NULL; | |
108 | } | |
109 | ||
5772e9a3 JDB |
110 | /* Note that dequeue_skb can possibly return a SKB list (via skb->next). |
111 | * A requeued skb (via q->gso_skb) can also be a SKB list. | |
112 | */ | |
b8358d70 JDB |
113 | static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate, |
114 | int *packets) | |
c716a81a | 115 | { |
554794de | 116 | struct sk_buff *skb = q->gso_skb; |
1abbe139 | 117 | const struct netdev_queue *txq = q->dev_queue; |
554794de | 118 | |
b8358d70 | 119 | *packets = 1; |
ebf05982 | 120 | if (unlikely(skb)) { |
4d202a0d ED |
121 | /* skb in gso_skb were already validated */ |
122 | *validate = false; | |
ebf05982 | 123 | /* check the reason of requeuing without tx lock first */ |
10c51b56 | 124 | txq = skb_get_tx_queue(txq->dev, skb); |
73466498 | 125 | if (!netif_xmit_frozen_or_stopped(txq)) { |
6252352d | 126 | q->gso_skb = NULL; |
a27758ff | 127 | qdisc_qstats_backlog_dec(q, skb); |
bbd8a0d3 KK |
128 | q->q.qlen--; |
129 | } else | |
ebf05982 | 130 | skb = NULL; |
e543002f | 131 | goto trace; |
4d202a0d ED |
132 | } |
133 | *validate = true; | |
134 | skb = q->skb_bad_txq; | |
135 | if (unlikely(skb)) { | |
136 | /* check the reason of requeuing without tx lock first */ | |
137 | txq = skb_get_tx_queue(txq->dev, skb); | |
138 | if (!netif_xmit_frozen_or_stopped(txq)) { | |
139 | q->skb_bad_txq = NULL; | |
140 | qdisc_qstats_backlog_dec(q, skb); | |
141 | q->q.qlen--; | |
142 | goto bulk; | |
50cbe9ab | 143 | } |
e543002f JDB |
144 | skb = NULL; |
145 | goto trace; | |
4d202a0d ED |
146 | } |
147 | if (!(q->flags & TCQ_F_ONETXQUEUE) || | |
148 | !netif_xmit_frozen_or_stopped(txq)) | |
149 | skb = q->dequeue(q); | |
150 | if (skb) { | |
151 | bulk: | |
152 | if (qdisc_may_bulk(q)) | |
153 | try_bulk_dequeue_skb(q, skb, txq, packets); | |
154 | else | |
155 | try_bulk_dequeue_skb_slow(q, skb, packets); | |
ebf05982 | 156 | } |
e543002f JDB |
157 | trace: |
158 | trace_qdisc_dequeue(q, txq, *packets, skb); | |
c716a81a JHS |
159 | return skb; |
160 | } | |
161 | ||
10297b99 | 162 | /* |
10770bc2 | 163 | * Transmit possibly several skbs, and handle the return status as |
f9eb8aea | 164 | * required. Owning running seqcount bit guarantees that |
10770bc2 | 165 | * only one CPU can execute this function. |
6c1361a6 KK |
166 | * |
167 | * Returns to the caller: | |
168 | * 0 - queue is empty or throttled. | |
169 | * >0 - queue is not empty. | |
6c1361a6 | 170 | */ |
bbd8a0d3 KK |
171 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
172 | struct net_device *dev, struct netdev_queue *txq, | |
55a93b3e | 173 | spinlock_t *root_lock, bool validate) |
1da177e4 | 174 | { |
5f1a485d | 175 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
176 | |
177 | /* And release qdisc */ | |
178 | spin_unlock(root_lock); | |
c716a81a | 179 | |
55a93b3e ED |
180 | /* Note that we validate skb (GSO, checksum, ...) outside of locks */ |
181 | if (validate) | |
182 | skb = validate_xmit_skb_list(skb, dev); | |
572a9d7b | 183 | |
3dcd493f | 184 | if (likely(skb)) { |
55a93b3e ED |
185 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
186 | if (!netif_xmit_frozen_or_stopped(txq)) | |
187 | skb = dev_hard_start_xmit(skb, dev, txq, &ret); | |
c716a81a | 188 | |
55a93b3e | 189 | HARD_TX_UNLOCK(dev, txq); |
3dcd493f | 190 | } else { |
52fbb290 | 191 | spin_lock(root_lock); |
3dcd493f | 192 | return qdisc_qlen(q); |
55a93b3e | 193 | } |
52fbb290 | 194 | spin_lock(root_lock); |
c716a81a | 195 | |
9a1654ba JP |
196 | if (dev_xmit_complete(ret)) { |
197 | /* Driver sent out skb successfully or skb was consumed */ | |
6c1361a6 | 198 | ret = qdisc_qlen(q); |
9a1654ba | 199 | } else { |
6c1361a6 | 200 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
e87cc472 JP |
201 | if (unlikely(ret != NETDEV_TX_BUSY)) |
202 | net_warn_ratelimited("BUG %s code %d qlen %d\n", | |
203 | dev->name, ret, q->q.qlen); | |
6c1361a6 | 204 | |
37437bb2 | 205 | ret = dev_requeue_skb(skb, q); |
6c1361a6 | 206 | } |
c716a81a | 207 | |
73466498 | 208 | if (ret && netif_xmit_frozen_or_stopped(txq)) |
37437bb2 DM |
209 | ret = 0; |
210 | ||
6c1361a6 | 211 | return ret; |
1da177e4 LT |
212 | } |
213 | ||
bbd8a0d3 KK |
214 | /* |
215 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. | |
216 | * | |
f9eb8aea | 217 | * running seqcount guarantees only one CPU can process |
bbd8a0d3 KK |
218 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for |
219 | * this queue. | |
220 | * | |
221 | * netif_tx_lock serializes accesses to device driver. | |
222 | * | |
223 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, | |
224 | * if one is grabbed, another must be free. | |
225 | * | |
226 | * Note, that this procedure can be called by a watchdog timer | |
227 | * | |
228 | * Returns to the caller: | |
229 | * 0 - queue is empty or throttled. | |
230 | * >0 - queue is not empty. | |
231 | * | |
232 | */ | |
b8358d70 | 233 | static inline int qdisc_restart(struct Qdisc *q, int *packets) |
bbd8a0d3 KK |
234 | { |
235 | struct netdev_queue *txq; | |
236 | struct net_device *dev; | |
237 | spinlock_t *root_lock; | |
238 | struct sk_buff *skb; | |
55a93b3e | 239 | bool validate; |
bbd8a0d3 KK |
240 | |
241 | /* Dequeue packet */ | |
b8358d70 | 242 | skb = dequeue_skb(q, &validate, packets); |
bbd8a0d3 KK |
243 | if (unlikely(!skb)) |
244 | return 0; | |
10c51b56 | 245 | |
bbd8a0d3 KK |
246 | root_lock = qdisc_lock(q); |
247 | dev = qdisc_dev(q); | |
10c51b56 | 248 | txq = skb_get_tx_queue(dev, skb); |
bbd8a0d3 | 249 | |
55a93b3e | 250 | return sch_direct_xmit(skb, q, dev, txq, root_lock, validate); |
bbd8a0d3 KK |
251 | } |
252 | ||
37437bb2 | 253 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 254 | { |
3d48b53f | 255 | int quota = dev_tx_weight; |
b8358d70 | 256 | int packets; |
2ba2506c | 257 | |
b8358d70 | 258 | while (qdisc_restart(q, &packets)) { |
2ba2506c | 259 | /* |
d5b8aa1d | 260 | * Ordered by possible occurrence: Postpone processing if |
261 | * 1. we've exceeded packet quota | |
262 | * 2. another process needs the CPU; | |
2ba2506c | 263 | */ |
b8358d70 JDB |
264 | quota -= packets; |
265 | if (quota <= 0 || need_resched()) { | |
37437bb2 | 266 | __netif_schedule(q); |
d90df3ad | 267 | break; |
2ba2506c HX |
268 | } |
269 | } | |
48d83325 | 270 | |
bc135b23 | 271 | qdisc_run_end(q); |
48d83325 HX |
272 | } |
273 | ||
9d21493b ED |
274 | unsigned long dev_trans_start(struct net_device *dev) |
275 | { | |
07ce76aa | 276 | unsigned long val, res; |
9d21493b ED |
277 | unsigned int i; |
278 | ||
07ce76aa | 279 | if (is_vlan_dev(dev)) |
280 | dev = vlan_dev_real_dev(dev); | |
32d3e51a CD |
281 | else if (netif_is_macvlan(dev)) |
282 | dev = macvlan_dev_real_dev(dev); | |
9b36627a FW |
283 | res = netdev_get_tx_queue(dev, 0)->trans_start; |
284 | for (i = 1; i < dev->num_tx_queues; i++) { | |
9d21493b ED |
285 | val = netdev_get_tx_queue(dev, i)->trans_start; |
286 | if (val && time_after(val, res)) | |
287 | res = val; | |
288 | } | |
07ce76aa | 289 | |
9d21493b ED |
290 | return res; |
291 | } | |
292 | EXPORT_SYMBOL(dev_trans_start); | |
293 | ||
cdeabbb8 | 294 | static void dev_watchdog(struct timer_list *t) |
1da177e4 | 295 | { |
cdeabbb8 | 296 | struct net_device *dev = from_timer(dev, t, watchdog_timer); |
1da177e4 | 297 | |
932ff279 | 298 | netif_tx_lock(dev); |
e8a0464c | 299 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
300 | if (netif_device_present(dev) && |
301 | netif_running(dev) && | |
302 | netif_carrier_ok(dev)) { | |
9d21493b | 303 | int some_queue_timedout = 0; |
e8a0464c | 304 | unsigned int i; |
9d21493b | 305 | unsigned long trans_start; |
e8a0464c DM |
306 | |
307 | for (i = 0; i < dev->num_tx_queues; i++) { | |
308 | struct netdev_queue *txq; | |
309 | ||
310 | txq = netdev_get_tx_queue(dev, i); | |
9b36627a | 311 | trans_start = txq->trans_start; |
73466498 | 312 | if (netif_xmit_stopped(txq) && |
9d21493b ED |
313 | time_after(jiffies, (trans_start + |
314 | dev->watchdog_timeo))) { | |
315 | some_queue_timedout = 1; | |
ccf5ff69 | 316 | txq->trans_timeout++; |
e8a0464c DM |
317 | break; |
318 | } | |
319 | } | |
338f7566 | 320 | |
9d21493b | 321 | if (some_queue_timedout) { |
9d21493b | 322 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 323 | dev->name, netdev_drivername(dev), i); |
d314774c | 324 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 325 | } |
e8a0464c DM |
326 | if (!mod_timer(&dev->watchdog_timer, |
327 | round_jiffies(jiffies + | |
328 | dev->watchdog_timeo))) | |
1da177e4 LT |
329 | dev_hold(dev); |
330 | } | |
331 | } | |
932ff279 | 332 | netif_tx_unlock(dev); |
1da177e4 LT |
333 | |
334 | dev_put(dev); | |
335 | } | |
336 | ||
1da177e4 LT |
337 | void __netdev_watchdog_up(struct net_device *dev) |
338 | { | |
d314774c | 339 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
340 | if (dev->watchdog_timeo <= 0) |
341 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
342 | if (!mod_timer(&dev->watchdog_timer, |
343 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
344 | dev_hold(dev); |
345 | } | |
346 | } | |
347 | ||
348 | static void dev_watchdog_up(struct net_device *dev) | |
349 | { | |
1da177e4 | 350 | __netdev_watchdog_up(dev); |
1da177e4 LT |
351 | } |
352 | ||
353 | static void dev_watchdog_down(struct net_device *dev) | |
354 | { | |
932ff279 | 355 | netif_tx_lock_bh(dev); |
1da177e4 | 356 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 357 | dev_put(dev); |
932ff279 | 358 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
359 | } |
360 | ||
bea3348e SH |
361 | /** |
362 | * netif_carrier_on - set carrier | |
363 | * @dev: network device | |
364 | * | |
365 | * Device has detected that carrier. | |
366 | */ | |
0a242efc DV |
367 | void netif_carrier_on(struct net_device *dev) |
368 | { | |
bfaae0f0 | 369 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
370 | if (dev->reg_state == NETREG_UNINITIALIZED) |
371 | return; | |
5e7f6597 | 372 | atomic_inc(&dev->carrier_up_count); |
0a242efc | 373 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
374 | if (netif_running(dev)) |
375 | __netdev_watchdog_up(dev); | |
376 | } | |
0a242efc | 377 | } |
62e3ba1b | 378 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 379 | |
bea3348e SH |
380 | /** |
381 | * netif_carrier_off - clear carrier | |
382 | * @dev: network device | |
383 | * | |
384 | * Device has detected loss of carrier. | |
385 | */ | |
0a242efc DV |
386 | void netif_carrier_off(struct net_device *dev) |
387 | { | |
b4730016 DM |
388 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
389 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
390 | return; | |
5e7f6597 | 391 | atomic_inc(&dev->carrier_down_count); |
0a242efc | 392 | linkwatch_fire_event(dev); |
b4730016 | 393 | } |
0a242efc | 394 | } |
62e3ba1b | 395 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 396 | |
1da177e4 LT |
397 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
398 | under all circumstances. It is difficult to invent anything faster or | |
399 | cheaper. | |
400 | */ | |
401 | ||
520ac30f ED |
402 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc, |
403 | struct sk_buff **to_free) | |
1da177e4 | 404 | { |
520ac30f | 405 | __qdisc_drop(skb, to_free); |
1da177e4 LT |
406 | return NET_XMIT_CN; |
407 | } | |
408 | ||
82d567c2 | 409 | static struct sk_buff *noop_dequeue(struct Qdisc *qdisc) |
1da177e4 LT |
410 | { |
411 | return NULL; | |
412 | } | |
413 | ||
20fea08b | 414 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
415 | .id = "noop", |
416 | .priv_size = 0, | |
417 | .enqueue = noop_enqueue, | |
418 | .dequeue = noop_dequeue, | |
99c0db26 | 419 | .peek = noop_dequeue, |
1da177e4 LT |
420 | .owner = THIS_MODULE, |
421 | }; | |
422 | ||
7698b4fc | 423 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 424 | .qdisc = &noop_qdisc, |
9f3ffae0 | 425 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
426 | }; |
427 | ||
1da177e4 LT |
428 | struct Qdisc noop_qdisc = { |
429 | .enqueue = noop_enqueue, | |
430 | .dequeue = noop_dequeue, | |
431 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 432 | .ops = &noop_qdisc_ops, |
83874000 | 433 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 434 | .dev_queue = &noop_netdev_queue, |
f9eb8aea | 435 | .running = SEQCNT_ZERO(noop_qdisc.running), |
7b5edbc4 | 436 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
1da177e4 | 437 | }; |
62e3ba1b | 438 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 439 | |
d66d6c31 PS |
440 | static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt) |
441 | { | |
442 | /* register_qdisc() assigns a default of noop_enqueue if unset, | |
443 | * but __dev_queue_xmit() treats noqueue only as such | |
444 | * if this is NULL - so clear it here. */ | |
445 | qdisc->enqueue = NULL; | |
446 | return 0; | |
447 | } | |
448 | ||
449 | struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { | |
1da177e4 LT |
450 | .id = "noqueue", |
451 | .priv_size = 0, | |
d66d6c31 | 452 | .init = noqueue_init, |
1da177e4 LT |
453 | .enqueue = noop_enqueue, |
454 | .dequeue = noop_dequeue, | |
99c0db26 | 455 | .peek = noop_dequeue, |
1da177e4 LT |
456 | .owner = THIS_MODULE, |
457 | }; | |
458 | ||
cc7ec456 ED |
459 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
460 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
461 | }; | |
d3678b46 DM |
462 | |
463 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
464 | generic prio+fifo combination. | |
465 | */ | |
466 | ||
467 | #define PFIFO_FAST_BANDS 3 | |
468 | ||
fd3ae5e8 KK |
469 | /* |
470 | * Private data for a pfifo_fast scheduler containing: | |
471 | * - queues for the three band | |
472 | * - bitmap indicating which of the bands contain skbs | |
473 | */ | |
474 | struct pfifo_fast_priv { | |
475 | u32 bitmap; | |
48da34b7 | 476 | struct qdisc_skb_head q[PFIFO_FAST_BANDS]; |
fd3ae5e8 KK |
477 | }; |
478 | ||
479 | /* | |
480 | * Convert a bitmap to the first band number where an skb is queued, where: | |
481 | * bitmap=0 means there are no skbs on any band. | |
482 | * bitmap=1 means there is an skb on band 0. | |
483 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
484 | */ | |
485 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
486 | ||
48da34b7 | 487 | static inline struct qdisc_skb_head *band2list(struct pfifo_fast_priv *priv, |
fd3ae5e8 | 488 | int band) |
d3678b46 | 489 | { |
fd3ae5e8 | 490 | return priv->q + band; |
d3678b46 DM |
491 | } |
492 | ||
520ac30f ED |
493 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc, |
494 | struct sk_buff **to_free) | |
321090e7 | 495 | { |
97d0678f | 496 | if (qdisc->q.qlen < qdisc_dev(qdisc)->tx_queue_len) { |
a453e068 KK |
497 | int band = prio2band[skb->priority & TC_PRIO_MAX]; |
498 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
48da34b7 | 499 | struct qdisc_skb_head *list = band2list(priv, band); |
1da177e4 | 500 | |
fd3ae5e8 | 501 | priv->bitmap |= (1 << band); |
d3678b46 | 502 | qdisc->q.qlen++; |
821d24ae | 503 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 504 | } |
821d24ae | 505 | |
520ac30f | 506 | return qdisc_drop(skb, qdisc, to_free); |
1da177e4 LT |
507 | } |
508 | ||
cc7ec456 | 509 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc) |
1da177e4 | 510 | { |
fd3ae5e8 KK |
511 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
512 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 513 | |
fd3ae5e8 | 514 | if (likely(band >= 0)) { |
48da34b7 FW |
515 | struct qdisc_skb_head *qh = band2list(priv, band); |
516 | struct sk_buff *skb = __qdisc_dequeue_head(qh); | |
ec323368 FW |
517 | |
518 | if (likely(skb != NULL)) { | |
519 | qdisc_qstats_backlog_dec(qdisc, skb); | |
520 | qdisc_bstats_update(qdisc, skb); | |
521 | } | |
fd3ae5e8 KK |
522 | |
523 | qdisc->q.qlen--; | |
48da34b7 | 524 | if (qh->qlen == 0) |
fd3ae5e8 KK |
525 | priv->bitmap &= ~(1 << band); |
526 | ||
527 | return skb; | |
d3678b46 | 528 | } |
f87a9c3d | 529 | |
1da177e4 LT |
530 | return NULL; |
531 | } | |
532 | ||
cc7ec456 | 533 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 534 | { |
fd3ae5e8 KK |
535 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
536 | int band = bitmap2band[priv->bitmap]; | |
537 | ||
538 | if (band >= 0) { | |
48da34b7 | 539 | struct qdisc_skb_head *qh = band2list(priv, band); |
99c0db26 | 540 | |
48da34b7 | 541 | return qh->head; |
99c0db26 JP |
542 | } |
543 | ||
544 | return NULL; | |
545 | } | |
546 | ||
cc7ec456 | 547 | static void pfifo_fast_reset(struct Qdisc *qdisc) |
1da177e4 | 548 | { |
d3678b46 | 549 | int prio; |
fd3ae5e8 | 550 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
551 | |
552 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
1b5c5493 | 553 | __qdisc_reset_queue(band2list(priv, prio)); |
d3678b46 | 554 | |
fd3ae5e8 | 555 | priv->bitmap = 0; |
821d24ae | 556 | qdisc->qstats.backlog = 0; |
d3678b46 | 557 | qdisc->q.qlen = 0; |
1da177e4 LT |
558 | } |
559 | ||
d3678b46 DM |
560 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
561 | { | |
562 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
563 | ||
cc7ec456 | 564 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
1b34ec43 DM |
565 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
566 | goto nla_put_failure; | |
d3678b46 DM |
567 | return skb->len; |
568 | ||
569 | nla_put_failure: | |
570 | return -1; | |
571 | } | |
572 | ||
573 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
574 | { | |
575 | int prio; | |
fd3ae5e8 | 576 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
577 | |
578 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
48da34b7 | 579 | qdisc_skb_head_init(band2list(priv, prio)); |
d3678b46 | 580 | |
23624935 ED |
581 | /* Can by-pass the queue discipline */ |
582 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
583 | return 0; |
584 | } | |
585 | ||
6ec1c69a | 586 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 587 | .id = "pfifo_fast", |
fd3ae5e8 | 588 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
589 | .enqueue = pfifo_fast_enqueue, |
590 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 591 | .peek = pfifo_fast_peek, |
d3678b46 DM |
592 | .init = pfifo_fast_init, |
593 | .reset = pfifo_fast_reset, | |
594 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
595 | .owner = THIS_MODULE, |
596 | }; | |
1f27cde3 | 597 | EXPORT_SYMBOL(pfifo_fast_ops); |
1da177e4 | 598 | |
23d3b8bf | 599 | static struct lock_class_key qdisc_tx_busylock; |
f9eb8aea | 600 | static struct lock_class_key qdisc_running_key; |
23d3b8bf | 601 | |
5ce2d488 | 602 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
d2a7f269 | 603 | const struct Qdisc_ops *ops) |
1da177e4 LT |
604 | { |
605 | void *p; | |
606 | struct Qdisc *sch; | |
d276055c | 607 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 608 | int err = -ENOBUFS; |
26aa0459 JSP |
609 | struct net_device *dev; |
610 | ||
611 | if (!dev_queue) { | |
612 | err = -EINVAL; | |
613 | goto errout; | |
614 | } | |
1da177e4 | 615 | |
26aa0459 | 616 | dev = dev_queue->dev; |
f2cd2d3e ED |
617 | p = kzalloc_node(size, GFP_KERNEL, |
618 | netdev_queue_numa_node_read(dev_queue)); | |
619 | ||
1da177e4 | 620 | if (!p) |
3d54b82f | 621 | goto errout; |
3d54b82f | 622 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
623 | /* if we got non aligned memory, ask more and do alignment ourself */ |
624 | if (sch != p) { | |
625 | kfree(p); | |
626 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
627 | netdev_queue_numa_node_read(dev_queue)); | |
628 | if (!p) | |
629 | goto errout; | |
630 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
631 | sch->padded = (char *) sch - (char *) p; | |
632 | } | |
48da34b7 FW |
633 | qdisc_skb_head_init(&sch->q); |
634 | spin_lock_init(&sch->q.lock); | |
23d3b8bf | 635 | |
81d947e2 DB |
636 | if (ops->static_flags & TCQ_F_CPUSTATS) { |
637 | sch->cpu_bstats = | |
638 | netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); | |
639 | if (!sch->cpu_bstats) | |
640 | goto errout1; | |
641 | ||
642 | sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); | |
643 | if (!sch->cpu_qstats) { | |
644 | free_percpu(sch->cpu_bstats); | |
645 | goto errout1; | |
646 | } | |
647 | } | |
648 | ||
79640a4c | 649 | spin_lock_init(&sch->busylock); |
23d3b8bf ED |
650 | lockdep_set_class(&sch->busylock, |
651 | dev->qdisc_tx_busylock ?: &qdisc_tx_busylock); | |
652 | ||
f9eb8aea ED |
653 | seqcount_init(&sch->running); |
654 | lockdep_set_class(&sch->running, | |
655 | dev->qdisc_running_key ?: &qdisc_running_key); | |
656 | ||
1da177e4 | 657 | sch->ops = ops; |
81d947e2 | 658 | sch->flags = ops->static_flags; |
1da177e4 LT |
659 | sch->enqueue = ops->enqueue; |
660 | sch->dequeue = ops->dequeue; | |
bb949fbd | 661 | sch->dev_queue = dev_queue; |
23d3b8bf | 662 | dev_hold(dev); |
7b936405 | 663 | refcount_set(&sch->refcnt, 1); |
3d54b82f TG |
664 | |
665 | return sch; | |
81d947e2 DB |
666 | errout1: |
667 | kfree(p); | |
3d54b82f | 668 | errout: |
01e123d7 | 669 | return ERR_PTR(err); |
3d54b82f TG |
670 | } |
671 | ||
3511c913 | 672 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
d2a7f269 | 673 | const struct Qdisc_ops *ops, |
674 | unsigned int parentid) | |
3d54b82f TG |
675 | { |
676 | struct Qdisc *sch; | |
10297b99 | 677 | |
6da7c8fc | 678 | if (!try_module_get(ops->owner)) |
166ee5b8 | 679 | return NULL; |
6da7c8fc | 680 | |
5ce2d488 | 681 | sch = qdisc_alloc(dev_queue, ops); |
166ee5b8 ED |
682 | if (IS_ERR(sch)) { |
683 | module_put(ops->owner); | |
684 | return NULL; | |
685 | } | |
9f9afec4 | 686 | sch->parent = parentid; |
3d54b82f | 687 | |
1da177e4 LT |
688 | if (!ops->init || ops->init(sch, NULL) == 0) |
689 | return sch; | |
690 | ||
0fbbeb1b | 691 | qdisc_destroy(sch); |
1da177e4 LT |
692 | return NULL; |
693 | } | |
62e3ba1b | 694 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 695 | |
5fb66229 | 696 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
697 | |
698 | void qdisc_reset(struct Qdisc *qdisc) | |
699 | { | |
20fea08b | 700 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
701 | |
702 | if (ops->reset) | |
703 | ops->reset(qdisc); | |
67305ebc | 704 | |
4d202a0d ED |
705 | kfree_skb(qdisc->skb_bad_txq); |
706 | qdisc->skb_bad_txq = NULL; | |
707 | ||
bbd8a0d3 | 708 | if (qdisc->gso_skb) { |
3f3c7eec | 709 | kfree_skb_list(qdisc->gso_skb); |
bbd8a0d3 | 710 | qdisc->gso_skb = NULL; |
bbd8a0d3 | 711 | } |
4d202a0d | 712 | qdisc->q.qlen = 0; |
c8e18129 | 713 | qdisc->qstats.backlog = 0; |
1da177e4 | 714 | } |
62e3ba1b | 715 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 716 | |
81d947e2 | 717 | void qdisc_free(struct Qdisc *qdisc) |
5d944c64 | 718 | { |
73c20a8b | 719 | if (qdisc_is_percpu_stats(qdisc)) { |
22e0f8b9 | 720 | free_percpu(qdisc->cpu_bstats); |
73c20a8b JF |
721 | free_percpu(qdisc->cpu_qstats); |
722 | } | |
22e0f8b9 | 723 | |
5d944c64 ED |
724 | kfree((char *) qdisc - qdisc->padded); |
725 | } | |
726 | ||
1e0d5a57 | 727 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 728 | { |
eaf7e146 CW |
729 | const struct Qdisc_ops *ops; |
730 | ||
731 | if (!qdisc) | |
732 | return; | |
733 | ops = qdisc->ops; | |
8a34c5dc | 734 | |
1e0d5a57 | 735 | if (qdisc->flags & TCQ_F_BUILTIN || |
7b936405 | 736 | !refcount_dec_and_test(&qdisc->refcnt)) |
1e0d5a57 DM |
737 | return; |
738 | ||
3a682fbd | 739 | #ifdef CONFIG_NET_SCHED |
59cc1f61 | 740 | qdisc_hash_del(qdisc); |
f6e0b239 | 741 | |
a2da570d | 742 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 743 | #endif |
1c0d32fd | 744 | gen_kill_estimator(&qdisc->rate_est); |
8a34c5dc DM |
745 | if (ops->reset) |
746 | ops->reset(qdisc); | |
747 | if (ops->destroy) | |
748 | ops->destroy(qdisc); | |
749 | ||
750 | module_put(ops->owner); | |
751 | dev_put(qdisc_dev(qdisc)); | |
752 | ||
3f3c7eec | 753 | kfree_skb_list(qdisc->gso_skb); |
4d202a0d | 754 | kfree_skb(qdisc->skb_bad_txq); |
752fbcc3 | 755 | qdisc_free(qdisc); |
1da177e4 | 756 | } |
62e3ba1b | 757 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 758 | |
589983cd PM |
759 | /* Attach toplevel qdisc to device queue. */ |
760 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
761 | struct Qdisc *qdisc) | |
762 | { | |
763 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
764 | spinlock_t *root_lock; | |
765 | ||
766 | root_lock = qdisc_lock(oqdisc); | |
767 | spin_lock_bh(root_lock); | |
768 | ||
589983cd PM |
769 | /* ... and graft new one */ |
770 | if (qdisc == NULL) | |
771 | qdisc = &noop_qdisc; | |
772 | dev_queue->qdisc_sleeping = qdisc; | |
773 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
774 | ||
775 | spin_unlock_bh(root_lock); | |
776 | ||
777 | return oqdisc; | |
778 | } | |
b8970f0b | 779 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 780 | |
e8a0464c DM |
781 | static void attach_one_default_qdisc(struct net_device *dev, |
782 | struct netdev_queue *dev_queue, | |
783 | void *_unused) | |
784 | { | |
3e692f21 PS |
785 | struct Qdisc *qdisc; |
786 | const struct Qdisc_ops *ops = default_qdisc_ops; | |
e8a0464c | 787 | |
3e692f21 PS |
788 | if (dev->priv_flags & IFF_NO_QUEUE) |
789 | ops = &noqueue_qdisc_ops; | |
790 | ||
791 | qdisc = qdisc_create_dflt(dev_queue, ops, TC_H_ROOT); | |
792 | if (!qdisc) { | |
793 | netdev_info(dev, "activation failed\n"); | |
794 | return; | |
e8a0464c | 795 | } |
3e692f21 | 796 | if (!netif_is_multiqueue(dev)) |
4eaf3b84 | 797 | qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT; |
e8a0464c DM |
798 | dev_queue->qdisc_sleeping = qdisc; |
799 | } | |
800 | ||
6ec1c69a DM |
801 | static void attach_default_qdiscs(struct net_device *dev) |
802 | { | |
803 | struct netdev_queue *txq; | |
804 | struct Qdisc *qdisc; | |
805 | ||
806 | txq = netdev_get_tx_queue(dev, 0); | |
807 | ||
4b469955 | 808 | if (!netif_is_multiqueue(dev) || |
4b469955 | 809 | dev->priv_flags & IFF_NO_QUEUE) { |
6ec1c69a DM |
810 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); |
811 | dev->qdisc = txq->qdisc_sleeping; | |
551143d8 | 812 | qdisc_refcount_inc(dev->qdisc); |
6ec1c69a | 813 | } else { |
3511c913 | 814 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT); |
6ec1c69a | 815 | if (qdisc) { |
6ec1c69a | 816 | dev->qdisc = qdisc; |
e57a784d | 817 | qdisc->ops->attach(qdisc); |
6ec1c69a DM |
818 | } |
819 | } | |
59cc1f61 | 820 | #ifdef CONFIG_NET_SCHED |
92f91706 | 821 | if (dev->qdisc != &noop_qdisc) |
49b49971 | 822 | qdisc_hash_add(dev->qdisc, false); |
59cc1f61 | 823 | #endif |
6ec1c69a DM |
824 | } |
825 | ||
e8a0464c DM |
826 | static void transition_one_qdisc(struct net_device *dev, |
827 | struct netdev_queue *dev_queue, | |
828 | void *_need_watchdog) | |
829 | { | |
83874000 | 830 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
831 | int *need_watchdog_p = _need_watchdog; |
832 | ||
a9312ae8 DM |
833 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
834 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
835 | ||
83874000 | 836 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
3e692f21 | 837 | if (need_watchdog_p) { |
9d21493b | 838 | dev_queue->trans_start = 0; |
e8a0464c | 839 | *need_watchdog_p = 1; |
9d21493b | 840 | } |
e8a0464c DM |
841 | } |
842 | ||
1da177e4 LT |
843 | void dev_activate(struct net_device *dev) |
844 | { | |
e8a0464c | 845 | int need_watchdog; |
b0e1e646 | 846 | |
1da177e4 | 847 | /* No queueing discipline is attached to device; |
6da7c8fc | 848 | * create default one for devices, which need queueing |
849 | * and noqueue_qdisc for virtual interfaces | |
1da177e4 LT |
850 | */ |
851 | ||
6ec1c69a DM |
852 | if (dev->qdisc == &noop_qdisc) |
853 | attach_default_qdiscs(dev); | |
af356afa | 854 | |
cacaddf5 TC |
855 | if (!netif_carrier_ok(dev)) |
856 | /* Delay activation until next carrier-on event */ | |
857 | return; | |
858 | ||
e8a0464c DM |
859 | need_watchdog = 0; |
860 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
24824a09 ED |
861 | if (dev_ingress_queue(dev)) |
862 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
863 | |
864 | if (need_watchdog) { | |
860e9538 | 865 | netif_trans_update(dev); |
1da177e4 LT |
866 | dev_watchdog_up(dev); |
867 | } | |
b0e1e646 | 868 | } |
b8970f0b | 869 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 870 | |
e8a0464c DM |
871 | static void dev_deactivate_queue(struct net_device *dev, |
872 | struct netdev_queue *dev_queue, | |
873 | void *_qdisc_default) | |
b0e1e646 | 874 | { |
e8a0464c | 875 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 876 | struct Qdisc *qdisc; |
970565bb | 877 | |
46e5da40 | 878 | qdisc = rtnl_dereference(dev_queue->qdisc); |
b0e1e646 | 879 | if (qdisc) { |
83874000 DM |
880 | spin_lock_bh(qdisc_lock(qdisc)); |
881 | ||
a9312ae8 DM |
882 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
883 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
884 | ||
f7a54c13 | 885 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 886 | qdisc_reset(qdisc); |
d3b753db | 887 | |
83874000 | 888 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 889 | } |
1da177e4 LT |
890 | } |
891 | ||
4335cd2d | 892 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
893 | { |
894 | unsigned int i; | |
895 | ||
896 | for (i = 0; i < dev->num_tx_queues; i++) { | |
897 | struct netdev_queue *dev_queue; | |
7698b4fc | 898 | spinlock_t *root_lock; |
e2627c8c | 899 | struct Qdisc *q; |
e8a0464c DM |
900 | int val; |
901 | ||
902 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 903 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 904 | root_lock = qdisc_lock(q); |
e8a0464c | 905 | |
4335cd2d | 906 | spin_lock_bh(root_lock); |
e8a0464c | 907 | |
bc135b23 | 908 | val = (qdisc_is_running(q) || |
b9a3b110 | 909 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 910 | |
4335cd2d | 911 | spin_unlock_bh(root_lock); |
e8a0464c DM |
912 | |
913 | if (val) | |
914 | return true; | |
915 | } | |
916 | return false; | |
917 | } | |
918 | ||
da4d0145 JF |
919 | static void dev_qdisc_reset(struct net_device *dev, |
920 | struct netdev_queue *dev_queue, | |
921 | void *none) | |
922 | { | |
923 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; | |
924 | ||
925 | if (qdisc) | |
926 | qdisc_reset(qdisc); | |
927 | } | |
928 | ||
3137663d ED |
929 | /** |
930 | * dev_deactivate_many - deactivate transmissions on several devices | |
931 | * @head: list of devices to deactivate | |
932 | * | |
933 | * This function returns only when all outstanding transmissions | |
934 | * have completed, unless all devices are in dismantle phase. | |
935 | */ | |
44345724 | 936 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 937 | { |
44345724 | 938 | struct net_device *dev; |
41a23b07 | 939 | |
5cde2829 | 940 | list_for_each_entry(dev, head, close_list) { |
44345724 OP |
941 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, |
942 | &noop_qdisc); | |
943 | if (dev_ingress_queue(dev)) | |
944 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
945 | &noop_qdisc); | |
946 | ||
947 | dev_watchdog_down(dev); | |
948 | } | |
1da177e4 | 949 | |
3137663d ED |
950 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
951 | * This is avoided if all devices are in dismantle phase : | |
952 | * Caller will call synchronize_net() for us | |
953 | */ | |
da4d0145 | 954 | synchronize_net(); |
1da177e4 | 955 | |
d4828d85 | 956 | /* Wait for outstanding qdisc_run calls. */ |
da4d0145 | 957 | list_for_each_entry(dev, head, close_list) { |
44345724 OP |
958 | while (some_qdisc_is_busy(dev)) |
959 | yield(); | |
da4d0145 JF |
960 | /* The new qdisc is assigned at this point so we can safely |
961 | * unwind stale skb lists and qdisc statistics | |
962 | */ | |
963 | netdev_for_each_tx_queue(dev, dev_qdisc_reset, NULL); | |
964 | if (dev_ingress_queue(dev)) | |
965 | dev_qdisc_reset(dev, dev_ingress_queue(dev), NULL); | |
966 | } | |
44345724 OP |
967 | } |
968 | ||
969 | void dev_deactivate(struct net_device *dev) | |
970 | { | |
971 | LIST_HEAD(single); | |
972 | ||
5cde2829 | 973 | list_add(&dev->close_list, &single); |
44345724 | 974 | dev_deactivate_many(&single); |
5f04d506 | 975 | list_del(&single); |
1da177e4 | 976 | } |
b8970f0b | 977 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 978 | |
b0e1e646 DM |
979 | static void dev_init_scheduler_queue(struct net_device *dev, |
980 | struct netdev_queue *dev_queue, | |
e8a0464c | 981 | void *_qdisc) |
b0e1e646 | 982 | { |
e8a0464c DM |
983 | struct Qdisc *qdisc = _qdisc; |
984 | ||
46e5da40 | 985 | rcu_assign_pointer(dev_queue->qdisc, qdisc); |
b0e1e646 | 986 | dev_queue->qdisc_sleeping = qdisc; |
b0e1e646 DM |
987 | } |
988 | ||
1da177e4 LT |
989 | void dev_init_scheduler(struct net_device *dev) |
990 | { | |
af356afa | 991 | dev->qdisc = &noop_qdisc; |
e8a0464c | 992 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
993 | if (dev_ingress_queue(dev)) |
994 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 995 | |
cdeabbb8 | 996 | timer_setup(&dev->watchdog_timer, dev_watchdog, 0); |
1da177e4 LT |
997 | } |
998 | ||
e8a0464c DM |
999 | static void shutdown_scheduler_queue(struct net_device *dev, |
1000 | struct netdev_queue *dev_queue, | |
1001 | void *_qdisc_default) | |
1da177e4 | 1002 | { |
b0e1e646 | 1003 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 1004 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
1005 | |
1006 | if (qdisc) { | |
f7a54c13 | 1007 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 1008 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 1009 | |
1da177e4 | 1010 | qdisc_destroy(qdisc); |
10297b99 | 1011 | } |
b0e1e646 DM |
1012 | } |
1013 | ||
1014 | void dev_shutdown(struct net_device *dev) | |
1015 | { | |
e8a0464c | 1016 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
1017 | if (dev_ingress_queue(dev)) |
1018 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
af356afa PM |
1019 | qdisc_destroy(dev->qdisc); |
1020 | dev->qdisc = &noop_qdisc; | |
1021 | ||
547b792c | 1022 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 1023 | } |
292f1c7f | 1024 | |
01cb71d2 | 1025 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
3e1e3aae ED |
1026 | const struct tc_ratespec *conf, |
1027 | u64 rate64) | |
292f1c7f | 1028 | { |
01cb71d2 ED |
1029 | memset(r, 0, sizeof(*r)); |
1030 | r->overhead = conf->overhead; | |
3e1e3aae | 1031 | r->rate_bytes_ps = max_t(u64, conf->rate, rate64); |
8a8e3d84 | 1032 | r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK); |
292f1c7f JP |
1033 | r->mult = 1; |
1034 | /* | |
130d3d68 ED |
1035 | * The deal here is to replace a divide by a reciprocal one |
1036 | * in fast path (a reciprocal divide is a multiply and a shift) | |
1037 | * | |
1038 | * Normal formula would be : | |
1039 | * time_in_ns = (NSEC_PER_SEC * len) / rate_bps | |
1040 | * | |
1041 | * We compute mult/shift to use instead : | |
1042 | * time_in_ns = (len * mult) >> shift; | |
1043 | * | |
1044 | * We try to get the highest possible mult value for accuracy, | |
1045 | * but have to make sure no overflows will ever happen. | |
292f1c7f | 1046 | */ |
130d3d68 ED |
1047 | if (r->rate_bytes_ps > 0) { |
1048 | u64 factor = NSEC_PER_SEC; | |
1049 | ||
1050 | for (;;) { | |
1051 | r->mult = div64_u64(factor, r->rate_bytes_ps); | |
1052 | if (r->mult & (1U << 31) || factor & (1ULL << 63)) | |
292f1c7f | 1053 | break; |
130d3d68 ED |
1054 | factor <<= 1; |
1055 | r->shift++; | |
292f1c7f | 1056 | } |
292f1c7f JP |
1057 | } |
1058 | } | |
1059 | EXPORT_SYMBOL(psched_ratecfg_precompute); | |
46209401 JP |
1060 | |
1061 | static void mini_qdisc_rcu_func(struct rcu_head *head) | |
1062 | { | |
1063 | } | |
1064 | ||
1065 | void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, | |
1066 | struct tcf_proto *tp_head) | |
1067 | { | |
1068 | struct mini_Qdisc *miniq_old = rtnl_dereference(*miniqp->p_miniq); | |
1069 | struct mini_Qdisc *miniq; | |
1070 | ||
1071 | if (!tp_head) { | |
1072 | RCU_INIT_POINTER(*miniqp->p_miniq, NULL); | |
b2fb01f4 CW |
1073 | /* Wait for flying RCU callback before it is freed. */ |
1074 | rcu_barrier_bh(); | |
46209401 JP |
1075 | return; |
1076 | } | |
1077 | ||
1078 | miniq = !miniq_old || miniq_old == &miniqp->miniq2 ? | |
1079 | &miniqp->miniq1 : &miniqp->miniq2; | |
1080 | ||
1081 | /* We need to make sure that readers won't see the miniq | |
1082 | * we are about to modify. So wait until previous call_rcu_bh callback | |
1083 | * is done. | |
1084 | */ | |
1085 | rcu_barrier_bh(); | |
1086 | miniq->filter_list = tp_head; | |
1087 | rcu_assign_pointer(*miniqp->p_miniq, miniq); | |
1088 | ||
1089 | if (miniq_old) | |
b2fb01f4 | 1090 | /* This is counterpart of the rcu barriers above. We need to |
46209401 JP |
1091 | * block potential new user of miniq_old until all readers |
1092 | * are not seeing it. | |
1093 | */ | |
1094 | call_rcu_bh(&miniq_old->rcu, mini_qdisc_rcu_func); | |
1095 | } | |
1096 | EXPORT_SYMBOL(mini_qdisc_pair_swap); | |
1097 | ||
1098 | void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, | |
1099 | struct mini_Qdisc __rcu **p_miniq) | |
1100 | { | |
1101 | miniqp->miniq1.cpu_bstats = qdisc->cpu_bstats; | |
1102 | miniqp->miniq1.cpu_qstats = qdisc->cpu_qstats; | |
1103 | miniqp->miniq2.cpu_bstats = qdisc->cpu_bstats; | |
1104 | miniqp->miniq2.cpu_qstats = qdisc->cpu_qstats; | |
1105 | miniqp->p_miniq = p_miniq; | |
1106 | } | |
1107 | EXPORT_SYMBOL(mini_qdisc_pair_init); |