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