Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/timer.c | |
3 | * | |
4 | * Kernel internal timers, kernel timekeeping, basic process system calls | |
5 | * | |
6 | * Copyright (C) 1991, 1992 Linus Torvalds | |
7 | * | |
8 | * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better. | |
9 | * | |
10 | * 1997-09-10 Updated NTP code according to technical memorandum Jan '96 | |
11 | * "A Kernel Model for Precision Timekeeping" by Dave Mills | |
12 | * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to | |
13 | * serialize accesses to xtime/lost_ticks). | |
14 | * Copyright (C) 1998 Andrea Arcangeli | |
15 | * 1999-03-10 Improved NTP compatibility by Ulrich Windl | |
16 | * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love | |
17 | * 2000-10-05 Implemented scalable SMP per-CPU timer handling. | |
18 | * Copyright (C) 2000, 2001, 2002 Ingo Molnar | |
19 | * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar | |
20 | */ | |
21 | ||
22 | #include <linux/kernel_stat.h> | |
23 | #include <linux/module.h> | |
24 | #include <linux/interrupt.h> | |
25 | #include <linux/percpu.h> | |
26 | #include <linux/init.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/swap.h> | |
29 | #include <linux/notifier.h> | |
30 | #include <linux/thread_info.h> | |
31 | #include <linux/time.h> | |
32 | #include <linux/jiffies.h> | |
33 | #include <linux/posix-timers.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/syscalls.h> | |
36 | ||
37 | #include <asm/uaccess.h> | |
38 | #include <asm/unistd.h> | |
39 | #include <asm/div64.h> | |
40 | #include <asm/timex.h> | |
41 | #include <asm/io.h> | |
42 | ||
43 | #ifdef CONFIG_TIME_INTERPOLATION | |
44 | static void time_interpolator_update(long delta_nsec); | |
45 | #else | |
46 | #define time_interpolator_update(x) | |
47 | #endif | |
48 | ||
49 | /* | |
50 | * per-CPU timer vector definitions: | |
51 | */ | |
52 | ||
53 | #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6) | |
54 | #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8) | |
55 | #define TVN_SIZE (1 << TVN_BITS) | |
56 | #define TVR_SIZE (1 << TVR_BITS) | |
57 | #define TVN_MASK (TVN_SIZE - 1) | |
58 | #define TVR_MASK (TVR_SIZE - 1) | |
59 | ||
55c888d6 ON |
60 | struct timer_base_s { |
61 | spinlock_t lock; | |
62 | struct timer_list *running_timer; | |
63 | }; | |
64 | ||
1da177e4 LT |
65 | typedef struct tvec_s { |
66 | struct list_head vec[TVN_SIZE]; | |
67 | } tvec_t; | |
68 | ||
69 | typedef struct tvec_root_s { | |
70 | struct list_head vec[TVR_SIZE]; | |
71 | } tvec_root_t; | |
72 | ||
73 | struct tvec_t_base_s { | |
55c888d6 | 74 | struct timer_base_s t_base; |
1da177e4 | 75 | unsigned long timer_jiffies; |
1da177e4 LT |
76 | tvec_root_t tv1; |
77 | tvec_t tv2; | |
78 | tvec_t tv3; | |
79 | tvec_t tv4; | |
80 | tvec_t tv5; | |
81 | } ____cacheline_aligned_in_smp; | |
82 | ||
83 | typedef struct tvec_t_base_s tvec_base_t; | |
55c888d6 | 84 | static DEFINE_PER_CPU(tvec_base_t, tvec_bases); |
1da177e4 LT |
85 | |
86 | static inline void set_running_timer(tvec_base_t *base, | |
87 | struct timer_list *timer) | |
88 | { | |
89 | #ifdef CONFIG_SMP | |
55c888d6 | 90 | base->t_base.running_timer = timer; |
1da177e4 LT |
91 | #endif |
92 | } | |
93 | ||
1da177e4 LT |
94 | static void check_timer_failed(struct timer_list *timer) |
95 | { | |
96 | static int whine_count; | |
97 | if (whine_count < 16) { | |
98 | whine_count++; | |
99 | printk("Uninitialised timer!\n"); | |
100 | printk("This is just a warning. Your computer is OK\n"); | |
101 | printk("function=0x%p, data=0x%lx\n", | |
102 | timer->function, timer->data); | |
103 | dump_stack(); | |
104 | } | |
105 | /* | |
106 | * Now fix it up | |
107 | */ | |
1da177e4 LT |
108 | timer->magic = TIMER_MAGIC; |
109 | } | |
110 | ||
111 | static inline void check_timer(struct timer_list *timer) | |
112 | { | |
113 | if (timer->magic != TIMER_MAGIC) | |
114 | check_timer_failed(timer); | |
115 | } | |
116 | ||
117 | ||
118 | static void internal_add_timer(tvec_base_t *base, struct timer_list *timer) | |
119 | { | |
120 | unsigned long expires = timer->expires; | |
121 | unsigned long idx = expires - base->timer_jiffies; | |
122 | struct list_head *vec; | |
123 | ||
124 | if (idx < TVR_SIZE) { | |
125 | int i = expires & TVR_MASK; | |
126 | vec = base->tv1.vec + i; | |
127 | } else if (idx < 1 << (TVR_BITS + TVN_BITS)) { | |
128 | int i = (expires >> TVR_BITS) & TVN_MASK; | |
129 | vec = base->tv2.vec + i; | |
130 | } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) { | |
131 | int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK; | |
132 | vec = base->tv3.vec + i; | |
133 | } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) { | |
134 | int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK; | |
135 | vec = base->tv4.vec + i; | |
136 | } else if ((signed long) idx < 0) { | |
137 | /* | |
138 | * Can happen if you add a timer with expires == jiffies, | |
139 | * or you set a timer to go off in the past | |
140 | */ | |
141 | vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK); | |
142 | } else { | |
143 | int i; | |
144 | /* If the timeout is larger than 0xffffffff on 64-bit | |
145 | * architectures then we use the maximum timeout: | |
146 | */ | |
147 | if (idx > 0xffffffffUL) { | |
148 | idx = 0xffffffffUL; | |
149 | expires = idx + base->timer_jiffies; | |
150 | } | |
151 | i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK; | |
152 | vec = base->tv5.vec + i; | |
153 | } | |
154 | /* | |
155 | * Timers are FIFO: | |
156 | */ | |
157 | list_add_tail(&timer->entry, vec); | |
158 | } | |
159 | ||
55c888d6 ON |
160 | typedef struct timer_base_s timer_base_t; |
161 | /* | |
162 | * Used by TIMER_INITIALIZER, we can't use per_cpu(tvec_bases) | |
163 | * at compile time, and we need timer->base to lock the timer. | |
164 | */ | |
165 | timer_base_t __init_timer_base | |
166 | ____cacheline_aligned_in_smp = { .lock = SPIN_LOCK_UNLOCKED }; | |
167 | EXPORT_SYMBOL(__init_timer_base); | |
168 | ||
169 | /*** | |
170 | * init_timer - initialize a timer. | |
171 | * @timer: the timer to be initialized | |
172 | * | |
173 | * init_timer() must be done to a timer prior calling *any* of the | |
174 | * other timer functions. | |
175 | */ | |
176 | void fastcall init_timer(struct timer_list *timer) | |
177 | { | |
178 | timer->entry.next = NULL; | |
179 | timer->base = &per_cpu(tvec_bases, raw_smp_processor_id()).t_base; | |
180 | timer->magic = TIMER_MAGIC; | |
181 | } | |
182 | EXPORT_SYMBOL(init_timer); | |
183 | ||
184 | static inline void detach_timer(struct timer_list *timer, | |
185 | int clear_pending) | |
186 | { | |
187 | struct list_head *entry = &timer->entry; | |
188 | ||
189 | __list_del(entry->prev, entry->next); | |
190 | if (clear_pending) | |
191 | entry->next = NULL; | |
192 | entry->prev = LIST_POISON2; | |
193 | } | |
194 | ||
195 | /* | |
196 | * We are using hashed locking: holding per_cpu(tvec_bases).t_base.lock | |
197 | * means that all timers which are tied to this base via timer->base are | |
198 | * locked, and the base itself is locked too. | |
199 | * | |
200 | * So __run_timers/migrate_timers can safely modify all timers which could | |
201 | * be found on ->tvX lists. | |
202 | * | |
203 | * When the timer's base is locked, and the timer removed from list, it is | |
204 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
205 | * locked. | |
206 | */ | |
207 | static timer_base_t *lock_timer_base(struct timer_list *timer, | |
208 | unsigned long *flags) | |
209 | { | |
210 | timer_base_t *base; | |
211 | ||
212 | for (;;) { | |
213 | base = timer->base; | |
214 | if (likely(base != NULL)) { | |
215 | spin_lock_irqsave(&base->lock, *flags); | |
216 | if (likely(base == timer->base)) | |
217 | return base; | |
218 | /* The timer has migrated to another CPU */ | |
219 | spin_unlock_irqrestore(&base->lock, *flags); | |
220 | } | |
221 | cpu_relax(); | |
222 | } | |
223 | } | |
224 | ||
1da177e4 LT |
225 | int __mod_timer(struct timer_list *timer, unsigned long expires) |
226 | { | |
55c888d6 ON |
227 | timer_base_t *base; |
228 | tvec_base_t *new_base; | |
1da177e4 LT |
229 | unsigned long flags; |
230 | int ret = 0; | |
231 | ||
232 | BUG_ON(!timer->function); | |
1da177e4 LT |
233 | check_timer(timer); |
234 | ||
55c888d6 ON |
235 | base = lock_timer_base(timer, &flags); |
236 | ||
237 | if (timer_pending(timer)) { | |
238 | detach_timer(timer, 0); | |
239 | ret = 1; | |
240 | } | |
241 | ||
1da177e4 | 242 | new_base = &__get_cpu_var(tvec_bases); |
1da177e4 | 243 | |
55c888d6 | 244 | if (base != &new_base->t_base) { |
1da177e4 | 245 | /* |
55c888d6 ON |
246 | * We are trying to schedule the timer on the local CPU. |
247 | * However we can't change timer's base while it is running, | |
248 | * otherwise del_timer_sync() can't detect that the timer's | |
249 | * handler yet has not finished. This also guarantees that | |
250 | * the timer is serialized wrt itself. | |
1da177e4 | 251 | */ |
55c888d6 ON |
252 | if (unlikely(base->running_timer == timer)) { |
253 | /* The timer remains on a former base */ | |
254 | new_base = container_of(base, tvec_base_t, t_base); | |
255 | } else { | |
256 | /* See the comment in lock_timer_base() */ | |
257 | timer->base = NULL; | |
258 | spin_unlock(&base->lock); | |
259 | spin_lock(&new_base->t_base.lock); | |
260 | timer->base = &new_base->t_base; | |
1da177e4 LT |
261 | } |
262 | } | |
263 | ||
1da177e4 LT |
264 | timer->expires = expires; |
265 | internal_add_timer(new_base, timer); | |
55c888d6 | 266 | spin_unlock_irqrestore(&new_base->t_base.lock, flags); |
1da177e4 LT |
267 | |
268 | return ret; | |
269 | } | |
270 | ||
271 | EXPORT_SYMBOL(__mod_timer); | |
272 | ||
273 | /*** | |
274 | * add_timer_on - start a timer on a particular CPU | |
275 | * @timer: the timer to be added | |
276 | * @cpu: the CPU to start it on | |
277 | * | |
278 | * This is not very scalable on SMP. Double adds are not possible. | |
279 | */ | |
280 | void add_timer_on(struct timer_list *timer, int cpu) | |
281 | { | |
282 | tvec_base_t *base = &per_cpu(tvec_bases, cpu); | |
283 | unsigned long flags; | |
55c888d6 | 284 | |
1da177e4 LT |
285 | BUG_ON(timer_pending(timer) || !timer->function); |
286 | ||
287 | check_timer(timer); | |
288 | ||
55c888d6 ON |
289 | spin_lock_irqsave(&base->t_base.lock, flags); |
290 | timer->base = &base->t_base; | |
1da177e4 | 291 | internal_add_timer(base, timer); |
55c888d6 | 292 | spin_unlock_irqrestore(&base->t_base.lock, flags); |
1da177e4 LT |
293 | } |
294 | ||
295 | ||
296 | /*** | |
297 | * mod_timer - modify a timer's timeout | |
298 | * @timer: the timer to be modified | |
299 | * | |
300 | * mod_timer is a more efficient way to update the expire field of an | |
301 | * active timer (if the timer is inactive it will be activated) | |
302 | * | |
303 | * mod_timer(timer, expires) is equivalent to: | |
304 | * | |
305 | * del_timer(timer); timer->expires = expires; add_timer(timer); | |
306 | * | |
307 | * Note that if there are multiple unserialized concurrent users of the | |
308 | * same timer, then mod_timer() is the only safe way to modify the timeout, | |
309 | * since add_timer() cannot modify an already running timer. | |
310 | * | |
311 | * The function returns whether it has modified a pending timer or not. | |
312 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an | |
313 | * active timer returns 1.) | |
314 | */ | |
315 | int mod_timer(struct timer_list *timer, unsigned long expires) | |
316 | { | |
317 | BUG_ON(!timer->function); | |
318 | ||
319 | check_timer(timer); | |
320 | ||
321 | /* | |
322 | * This is a common optimization triggered by the | |
323 | * networking code - if the timer is re-modified | |
324 | * to be the same thing then just return: | |
325 | */ | |
326 | if (timer->expires == expires && timer_pending(timer)) | |
327 | return 1; | |
328 | ||
329 | return __mod_timer(timer, expires); | |
330 | } | |
331 | ||
332 | EXPORT_SYMBOL(mod_timer); | |
333 | ||
334 | /*** | |
335 | * del_timer - deactive a timer. | |
336 | * @timer: the timer to be deactivated | |
337 | * | |
338 | * del_timer() deactivates a timer - this works on both active and inactive | |
339 | * timers. | |
340 | * | |
341 | * The function returns whether it has deactivated a pending timer or not. | |
342 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an | |
343 | * active timer returns 1.) | |
344 | */ | |
345 | int del_timer(struct timer_list *timer) | |
346 | { | |
55c888d6 | 347 | timer_base_t *base; |
1da177e4 | 348 | unsigned long flags; |
55c888d6 | 349 | int ret = 0; |
1da177e4 LT |
350 | |
351 | check_timer(timer); | |
352 | ||
55c888d6 ON |
353 | if (timer_pending(timer)) { |
354 | base = lock_timer_base(timer, &flags); | |
355 | if (timer_pending(timer)) { | |
356 | detach_timer(timer, 1); | |
357 | ret = 1; | |
358 | } | |
1da177e4 | 359 | spin_unlock_irqrestore(&base->lock, flags); |
1da177e4 | 360 | } |
1da177e4 | 361 | |
55c888d6 | 362 | return ret; |
1da177e4 LT |
363 | } |
364 | ||
365 | EXPORT_SYMBOL(del_timer); | |
366 | ||
367 | #ifdef CONFIG_SMP | |
368 | /*** | |
369 | * del_timer_sync - deactivate a timer and wait for the handler to finish. | |
370 | * @timer: the timer to be deactivated | |
371 | * | |
372 | * This function only differs from del_timer() on SMP: besides deactivating | |
373 | * the timer it also makes sure the handler has finished executing on other | |
374 | * CPUs. | |
375 | * | |
376 | * Synchronization rules: callers must prevent restarting of the timer, | |
377 | * otherwise this function is meaningless. It must not be called from | |
378 | * interrupt contexts. The caller must not hold locks which would prevent | |
55c888d6 ON |
379 | * completion of the timer's handler. The timer's handler must not call |
380 | * add_timer_on(). Upon exit the timer is not queued and the handler is | |
381 | * not running on any CPU. | |
1da177e4 LT |
382 | * |
383 | * The function returns whether it has deactivated a pending timer or not. | |
1da177e4 LT |
384 | */ |
385 | int del_timer_sync(struct timer_list *timer) | |
386 | { | |
55c888d6 ON |
387 | timer_base_t *base; |
388 | unsigned long flags; | |
389 | int ret = -1; | |
1da177e4 LT |
390 | |
391 | check_timer(timer); | |
392 | ||
55c888d6 ON |
393 | do { |
394 | base = lock_timer_base(timer, &flags); | |
1da177e4 | 395 | |
55c888d6 ON |
396 | if (base->running_timer == timer) |
397 | goto unlock; | |
398 | ||
399 | ret = 0; | |
400 | if (timer_pending(timer)) { | |
401 | detach_timer(timer, 1); | |
402 | ret = 1; | |
1da177e4 | 403 | } |
55c888d6 ON |
404 | unlock: |
405 | spin_unlock_irqrestore(&base->lock, flags); | |
406 | } while (ret < 0); | |
1da177e4 LT |
407 | |
408 | return ret; | |
409 | } | |
1da177e4 | 410 | |
55c888d6 | 411 | EXPORT_SYMBOL(del_timer_sync); |
1da177e4 LT |
412 | #endif |
413 | ||
414 | static int cascade(tvec_base_t *base, tvec_t *tv, int index) | |
415 | { | |
416 | /* cascade all the timers from tv up one level */ | |
417 | struct list_head *head, *curr; | |
418 | ||
419 | head = tv->vec + index; | |
420 | curr = head->next; | |
421 | /* | |
422 | * We are removing _all_ timers from the list, so we don't have to | |
423 | * detach them individually, just clear the list afterwards. | |
424 | */ | |
425 | while (curr != head) { | |
426 | struct timer_list *tmp; | |
427 | ||
428 | tmp = list_entry(curr, struct timer_list, entry); | |
55c888d6 | 429 | BUG_ON(tmp->base != &base->t_base); |
1da177e4 LT |
430 | curr = curr->next; |
431 | internal_add_timer(base, tmp); | |
432 | } | |
433 | INIT_LIST_HEAD(head); | |
434 | ||
435 | return index; | |
436 | } | |
437 | ||
438 | /*** | |
439 | * __run_timers - run all expired timers (if any) on this CPU. | |
440 | * @base: the timer vector to be processed. | |
441 | * | |
442 | * This function cascades all vectors and executes all expired timer | |
443 | * vectors. | |
444 | */ | |
445 | #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK | |
446 | ||
447 | static inline void __run_timers(tvec_base_t *base) | |
448 | { | |
449 | struct timer_list *timer; | |
450 | ||
55c888d6 | 451 | spin_lock_irq(&base->t_base.lock); |
1da177e4 LT |
452 | while (time_after_eq(jiffies, base->timer_jiffies)) { |
453 | struct list_head work_list = LIST_HEAD_INIT(work_list); | |
454 | struct list_head *head = &work_list; | |
455 | int index = base->timer_jiffies & TVR_MASK; | |
456 | ||
457 | /* | |
458 | * Cascade timers: | |
459 | */ | |
460 | if (!index && | |
461 | (!cascade(base, &base->tv2, INDEX(0))) && | |
462 | (!cascade(base, &base->tv3, INDEX(1))) && | |
463 | !cascade(base, &base->tv4, INDEX(2))) | |
464 | cascade(base, &base->tv5, INDEX(3)); | |
465 | ++base->timer_jiffies; | |
466 | list_splice_init(base->tv1.vec + index, &work_list); | |
55c888d6 | 467 | while (!list_empty(head)) { |
1da177e4 LT |
468 | void (*fn)(unsigned long); |
469 | unsigned long data; | |
470 | ||
471 | timer = list_entry(head->next,struct timer_list,entry); | |
472 | fn = timer->function; | |
473 | data = timer->data; | |
474 | ||
1da177e4 | 475 | set_running_timer(base, timer); |
55c888d6 ON |
476 | detach_timer(timer, 1); |
477 | spin_unlock_irq(&base->t_base.lock); | |
1da177e4 LT |
478 | { |
479 | u32 preempt_count = preempt_count(); | |
480 | fn(data); | |
481 | if (preempt_count != preempt_count()) { | |
482 | printk("huh, entered %p with %08x, exited with %08x?\n", fn, preempt_count, preempt_count()); | |
483 | BUG(); | |
484 | } | |
485 | } | |
55c888d6 | 486 | spin_lock_irq(&base->t_base.lock); |
1da177e4 LT |
487 | } |
488 | } | |
489 | set_running_timer(base, NULL); | |
55c888d6 | 490 | spin_unlock_irq(&base->t_base.lock); |
1da177e4 LT |
491 | } |
492 | ||
493 | #ifdef CONFIG_NO_IDLE_HZ | |
494 | /* | |
495 | * Find out when the next timer event is due to happen. This | |
496 | * is used on S/390 to stop all activity when a cpus is idle. | |
497 | * This functions needs to be called disabled. | |
498 | */ | |
499 | unsigned long next_timer_interrupt(void) | |
500 | { | |
501 | tvec_base_t *base; | |
502 | struct list_head *list; | |
503 | struct timer_list *nte; | |
504 | unsigned long expires; | |
505 | tvec_t *varray[4]; | |
506 | int i, j; | |
507 | ||
508 | base = &__get_cpu_var(tvec_bases); | |
55c888d6 | 509 | spin_lock(&base->t_base.lock); |
1da177e4 LT |
510 | expires = base->timer_jiffies + (LONG_MAX >> 1); |
511 | list = 0; | |
512 | ||
513 | /* Look for timer events in tv1. */ | |
514 | j = base->timer_jiffies & TVR_MASK; | |
515 | do { | |
516 | list_for_each_entry(nte, base->tv1.vec + j, entry) { | |
517 | expires = nte->expires; | |
518 | if (j < (base->timer_jiffies & TVR_MASK)) | |
519 | list = base->tv2.vec + (INDEX(0)); | |
520 | goto found; | |
521 | } | |
522 | j = (j + 1) & TVR_MASK; | |
523 | } while (j != (base->timer_jiffies & TVR_MASK)); | |
524 | ||
525 | /* Check tv2-tv5. */ | |
526 | varray[0] = &base->tv2; | |
527 | varray[1] = &base->tv3; | |
528 | varray[2] = &base->tv4; | |
529 | varray[3] = &base->tv5; | |
530 | for (i = 0; i < 4; i++) { | |
531 | j = INDEX(i); | |
532 | do { | |
533 | if (list_empty(varray[i]->vec + j)) { | |
534 | j = (j + 1) & TVN_MASK; | |
535 | continue; | |
536 | } | |
537 | list_for_each_entry(nte, varray[i]->vec + j, entry) | |
538 | if (time_before(nte->expires, expires)) | |
539 | expires = nte->expires; | |
540 | if (j < (INDEX(i)) && i < 3) | |
541 | list = varray[i + 1]->vec + (INDEX(i + 1)); | |
542 | goto found; | |
543 | } while (j != (INDEX(i))); | |
544 | } | |
545 | found: | |
546 | if (list) { | |
547 | /* | |
548 | * The search wrapped. We need to look at the next list | |
549 | * from next tv element that would cascade into tv element | |
550 | * where we found the timer element. | |
551 | */ | |
552 | list_for_each_entry(nte, list, entry) { | |
553 | if (time_before(nte->expires, expires)) | |
554 | expires = nte->expires; | |
555 | } | |
556 | } | |
55c888d6 | 557 | spin_unlock(&base->t_base.lock); |
1da177e4 LT |
558 | return expires; |
559 | } | |
560 | #endif | |
561 | ||
562 | /******************************************************************/ | |
563 | ||
564 | /* | |
565 | * Timekeeping variables | |
566 | */ | |
567 | unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */ | |
568 | unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */ | |
569 | ||
570 | /* | |
571 | * The current time | |
572 | * wall_to_monotonic is what we need to add to xtime (or xtime corrected | |
573 | * for sub jiffie times) to get to monotonic time. Monotonic is pegged | |
574 | * at zero at system boot time, so wall_to_monotonic will be negative, | |
575 | * however, we will ALWAYS keep the tv_nsec part positive so we can use | |
576 | * the usual normalization. | |
577 | */ | |
578 | struct timespec xtime __attribute__ ((aligned (16))); | |
579 | struct timespec wall_to_monotonic __attribute__ ((aligned (16))); | |
580 | ||
581 | EXPORT_SYMBOL(xtime); | |
582 | ||
583 | /* Don't completely fail for HZ > 500. */ | |
584 | int tickadj = 500/HZ ? : 1; /* microsecs */ | |
585 | ||
586 | ||
587 | /* | |
588 | * phase-lock loop variables | |
589 | */ | |
590 | /* TIME_ERROR prevents overwriting the CMOS clock */ | |
591 | int time_state = TIME_OK; /* clock synchronization status */ | |
592 | int time_status = STA_UNSYNC; /* clock status bits */ | |
593 | long time_offset; /* time adjustment (us) */ | |
594 | long time_constant = 2; /* pll time constant */ | |
595 | long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */ | |
596 | long time_precision = 1; /* clock precision (us) */ | |
597 | long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */ | |
598 | long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */ | |
599 | static long time_phase; /* phase offset (scaled us) */ | |
600 | long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC; | |
601 | /* frequency offset (scaled ppm)*/ | |
602 | static long time_adj; /* tick adjust (scaled 1 / HZ) */ | |
603 | long time_reftime; /* time at last adjustment (s) */ | |
604 | long time_adjust; | |
605 | long time_next_adjust; | |
606 | ||
607 | /* | |
608 | * this routine handles the overflow of the microsecond field | |
609 | * | |
610 | * The tricky bits of code to handle the accurate clock support | |
611 | * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame. | |
612 | * They were originally developed for SUN and DEC kernels. | |
613 | * All the kudos should go to Dave for this stuff. | |
614 | * | |
615 | */ | |
616 | static void second_overflow(void) | |
617 | { | |
618 | long ltemp; | |
619 | ||
620 | /* Bump the maxerror field */ | |
621 | time_maxerror += time_tolerance >> SHIFT_USEC; | |
622 | if ( time_maxerror > NTP_PHASE_LIMIT ) { | |
623 | time_maxerror = NTP_PHASE_LIMIT; | |
624 | time_status |= STA_UNSYNC; | |
625 | } | |
626 | ||
627 | /* | |
628 | * Leap second processing. If in leap-insert state at | |
629 | * the end of the day, the system clock is set back one | |
630 | * second; if in leap-delete state, the system clock is | |
631 | * set ahead one second. The microtime() routine or | |
632 | * external clock driver will insure that reported time | |
633 | * is always monotonic. The ugly divides should be | |
634 | * replaced. | |
635 | */ | |
636 | switch (time_state) { | |
637 | ||
638 | case TIME_OK: | |
639 | if (time_status & STA_INS) | |
640 | time_state = TIME_INS; | |
641 | else if (time_status & STA_DEL) | |
642 | time_state = TIME_DEL; | |
643 | break; | |
644 | ||
645 | case TIME_INS: | |
646 | if (xtime.tv_sec % 86400 == 0) { | |
647 | xtime.tv_sec--; | |
648 | wall_to_monotonic.tv_sec++; | |
649 | /* The timer interpolator will make time change gradually instead | |
650 | * of an immediate jump by one second. | |
651 | */ | |
652 | time_interpolator_update(-NSEC_PER_SEC); | |
653 | time_state = TIME_OOP; | |
654 | clock_was_set(); | |
655 | printk(KERN_NOTICE "Clock: inserting leap second 23:59:60 UTC\n"); | |
656 | } | |
657 | break; | |
658 | ||
659 | case TIME_DEL: | |
660 | if ((xtime.tv_sec + 1) % 86400 == 0) { | |
661 | xtime.tv_sec++; | |
662 | wall_to_monotonic.tv_sec--; | |
663 | /* Use of time interpolator for a gradual change of time */ | |
664 | time_interpolator_update(NSEC_PER_SEC); | |
665 | time_state = TIME_WAIT; | |
666 | clock_was_set(); | |
667 | printk(KERN_NOTICE "Clock: deleting leap second 23:59:59 UTC\n"); | |
668 | } | |
669 | break; | |
670 | ||
671 | case TIME_OOP: | |
672 | time_state = TIME_WAIT; | |
673 | break; | |
674 | ||
675 | case TIME_WAIT: | |
676 | if (!(time_status & (STA_INS | STA_DEL))) | |
677 | time_state = TIME_OK; | |
678 | } | |
679 | ||
680 | /* | |
681 | * Compute the phase adjustment for the next second. In | |
682 | * PLL mode, the offset is reduced by a fixed factor | |
683 | * times the time constant. In FLL mode the offset is | |
684 | * used directly. In either mode, the maximum phase | |
685 | * adjustment for each second is clamped so as to spread | |
686 | * the adjustment over not more than the number of | |
687 | * seconds between updates. | |
688 | */ | |
689 | if (time_offset < 0) { | |
690 | ltemp = -time_offset; | |
691 | if (!(time_status & STA_FLL)) | |
692 | ltemp >>= SHIFT_KG + time_constant; | |
693 | if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE) | |
694 | ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE; | |
695 | time_offset += ltemp; | |
696 | time_adj = -ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE); | |
697 | } else { | |
698 | ltemp = time_offset; | |
699 | if (!(time_status & STA_FLL)) | |
700 | ltemp >>= SHIFT_KG + time_constant; | |
701 | if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE) | |
702 | ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE; | |
703 | time_offset -= ltemp; | |
704 | time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE); | |
705 | } | |
706 | ||
707 | /* | |
708 | * Compute the frequency estimate and additional phase | |
709 | * adjustment due to frequency error for the next | |
710 | * second. When the PPS signal is engaged, gnaw on the | |
711 | * watchdog counter and update the frequency computed by | |
712 | * the pll and the PPS signal. | |
713 | */ | |
714 | pps_valid++; | |
715 | if (pps_valid == PPS_VALID) { /* PPS signal lost */ | |
716 | pps_jitter = MAXTIME; | |
717 | pps_stabil = MAXFREQ; | |
718 | time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER | | |
719 | STA_PPSWANDER | STA_PPSERROR); | |
720 | } | |
721 | ltemp = time_freq + pps_freq; | |
722 | if (ltemp < 0) | |
723 | time_adj -= -ltemp >> | |
724 | (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE); | |
725 | else | |
726 | time_adj += ltemp >> | |
727 | (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE); | |
728 | ||
729 | #if HZ == 100 | |
730 | /* Compensate for (HZ==100) != (1 << SHIFT_HZ). | |
731 | * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14) | |
732 | */ | |
733 | if (time_adj < 0) | |
734 | time_adj -= (-time_adj >> 2) + (-time_adj >> 5); | |
735 | else | |
736 | time_adj += (time_adj >> 2) + (time_adj >> 5); | |
737 | #endif | |
738 | #if HZ == 1000 | |
739 | /* Compensate for (HZ==1000) != (1 << SHIFT_HZ). | |
740 | * Add 1.5625% and 0.78125% to get 1023.4375; => only 0.05% error (p. 14) | |
741 | */ | |
742 | if (time_adj < 0) | |
743 | time_adj -= (-time_adj >> 6) + (-time_adj >> 7); | |
744 | else | |
745 | time_adj += (time_adj >> 6) + (time_adj >> 7); | |
746 | #endif | |
747 | } | |
748 | ||
749 | /* in the NTP reference this is called "hardclock()" */ | |
750 | static void update_wall_time_one_tick(void) | |
751 | { | |
752 | long time_adjust_step, delta_nsec; | |
753 | ||
754 | if ( (time_adjust_step = time_adjust) != 0 ) { | |
755 | /* We are doing an adjtime thing. | |
756 | * | |
757 | * Prepare time_adjust_step to be within bounds. | |
758 | * Note that a positive time_adjust means we want the clock | |
759 | * to run faster. | |
760 | * | |
761 | * Limit the amount of the step to be in the range | |
762 | * -tickadj .. +tickadj | |
763 | */ | |
764 | if (time_adjust > tickadj) | |
765 | time_adjust_step = tickadj; | |
766 | else if (time_adjust < -tickadj) | |
767 | time_adjust_step = -tickadj; | |
768 | ||
769 | /* Reduce by this step the amount of time left */ | |
770 | time_adjust -= time_adjust_step; | |
771 | } | |
772 | delta_nsec = tick_nsec + time_adjust_step * 1000; | |
773 | /* | |
774 | * Advance the phase, once it gets to one microsecond, then | |
775 | * advance the tick more. | |
776 | */ | |
777 | time_phase += time_adj; | |
778 | if (time_phase <= -FINENSEC) { | |
779 | long ltemp = -time_phase >> (SHIFT_SCALE - 10); | |
780 | time_phase += ltemp << (SHIFT_SCALE - 10); | |
781 | delta_nsec -= ltemp; | |
782 | } | |
783 | else if (time_phase >= FINENSEC) { | |
784 | long ltemp = time_phase >> (SHIFT_SCALE - 10); | |
785 | time_phase -= ltemp << (SHIFT_SCALE - 10); | |
786 | delta_nsec += ltemp; | |
787 | } | |
788 | xtime.tv_nsec += delta_nsec; | |
789 | time_interpolator_update(delta_nsec); | |
790 | ||
791 | /* Changes by adjtime() do not take effect till next tick. */ | |
792 | if (time_next_adjust != 0) { | |
793 | time_adjust = time_next_adjust; | |
794 | time_next_adjust = 0; | |
795 | } | |
796 | } | |
797 | ||
798 | /* | |
799 | * Using a loop looks inefficient, but "ticks" is | |
800 | * usually just one (we shouldn't be losing ticks, | |
801 | * we're doing this this way mainly for interrupt | |
802 | * latency reasons, not because we think we'll | |
803 | * have lots of lost timer ticks | |
804 | */ | |
805 | static void update_wall_time(unsigned long ticks) | |
806 | { | |
807 | do { | |
808 | ticks--; | |
809 | update_wall_time_one_tick(); | |
810 | if (xtime.tv_nsec >= 1000000000) { | |
811 | xtime.tv_nsec -= 1000000000; | |
812 | xtime.tv_sec++; | |
813 | second_overflow(); | |
814 | } | |
815 | } while (ticks); | |
816 | } | |
817 | ||
818 | /* | |
819 | * Called from the timer interrupt handler to charge one tick to the current | |
820 | * process. user_tick is 1 if the tick is user time, 0 for system. | |
821 | */ | |
822 | void update_process_times(int user_tick) | |
823 | { | |
824 | struct task_struct *p = current; | |
825 | int cpu = smp_processor_id(); | |
826 | ||
827 | /* Note: this timer irq context must be accounted for as well. */ | |
828 | if (user_tick) | |
829 | account_user_time(p, jiffies_to_cputime(1)); | |
830 | else | |
831 | account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1)); | |
832 | run_local_timers(); | |
833 | if (rcu_pending(cpu)) | |
834 | rcu_check_callbacks(cpu, user_tick); | |
835 | scheduler_tick(); | |
836 | run_posix_cpu_timers(p); | |
837 | } | |
838 | ||
839 | /* | |
840 | * Nr of active tasks - counted in fixed-point numbers | |
841 | */ | |
842 | static unsigned long count_active_tasks(void) | |
843 | { | |
844 | return (nr_running() + nr_uninterruptible()) * FIXED_1; | |
845 | } | |
846 | ||
847 | /* | |
848 | * Hmm.. Changed this, as the GNU make sources (load.c) seems to | |
849 | * imply that avenrun[] is the standard name for this kind of thing. | |
850 | * Nothing else seems to be standardized: the fractional size etc | |
851 | * all seem to differ on different machines. | |
852 | * | |
853 | * Requires xtime_lock to access. | |
854 | */ | |
855 | unsigned long avenrun[3]; | |
856 | ||
857 | EXPORT_SYMBOL(avenrun); | |
858 | ||
859 | /* | |
860 | * calc_load - given tick count, update the avenrun load estimates. | |
861 | * This is called while holding a write_lock on xtime_lock. | |
862 | */ | |
863 | static inline void calc_load(unsigned long ticks) | |
864 | { | |
865 | unsigned long active_tasks; /* fixed-point */ | |
866 | static int count = LOAD_FREQ; | |
867 | ||
868 | count -= ticks; | |
869 | if (count < 0) { | |
870 | count += LOAD_FREQ; | |
871 | active_tasks = count_active_tasks(); | |
872 | CALC_LOAD(avenrun[0], EXP_1, active_tasks); | |
873 | CALC_LOAD(avenrun[1], EXP_5, active_tasks); | |
874 | CALC_LOAD(avenrun[2], EXP_15, active_tasks); | |
875 | } | |
876 | } | |
877 | ||
878 | /* jiffies at the most recent update of wall time */ | |
879 | unsigned long wall_jiffies = INITIAL_JIFFIES; | |
880 | ||
881 | /* | |
882 | * This read-write spinlock protects us from races in SMP while | |
883 | * playing with xtime and avenrun. | |
884 | */ | |
885 | #ifndef ARCH_HAVE_XTIME_LOCK | |
886 | seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED; | |
887 | ||
888 | EXPORT_SYMBOL(xtime_lock); | |
889 | #endif | |
890 | ||
891 | /* | |
892 | * This function runs timers and the timer-tq in bottom half context. | |
893 | */ | |
894 | static void run_timer_softirq(struct softirq_action *h) | |
895 | { | |
896 | tvec_base_t *base = &__get_cpu_var(tvec_bases); | |
897 | ||
898 | if (time_after_eq(jiffies, base->timer_jiffies)) | |
899 | __run_timers(base); | |
900 | } | |
901 | ||
902 | /* | |
903 | * Called by the local, per-CPU timer interrupt on SMP. | |
904 | */ | |
905 | void run_local_timers(void) | |
906 | { | |
907 | raise_softirq(TIMER_SOFTIRQ); | |
908 | } | |
909 | ||
910 | /* | |
911 | * Called by the timer interrupt. xtime_lock must already be taken | |
912 | * by the timer IRQ! | |
913 | */ | |
914 | static inline void update_times(void) | |
915 | { | |
916 | unsigned long ticks; | |
917 | ||
918 | ticks = jiffies - wall_jiffies; | |
919 | if (ticks) { | |
920 | wall_jiffies += ticks; | |
921 | update_wall_time(ticks); | |
922 | } | |
923 | calc_load(ticks); | |
924 | } | |
925 | ||
926 | /* | |
927 | * The 64-bit jiffies value is not atomic - you MUST NOT read it | |
928 | * without sampling the sequence number in xtime_lock. | |
929 | * jiffies is defined in the linker script... | |
930 | */ | |
931 | ||
932 | void do_timer(struct pt_regs *regs) | |
933 | { | |
934 | jiffies_64++; | |
935 | update_times(); | |
936 | } | |
937 | ||
938 | #ifdef __ARCH_WANT_SYS_ALARM | |
939 | ||
940 | /* | |
941 | * For backwards compatibility? This can be done in libc so Alpha | |
942 | * and all newer ports shouldn't need it. | |
943 | */ | |
944 | asmlinkage unsigned long sys_alarm(unsigned int seconds) | |
945 | { | |
946 | struct itimerval it_new, it_old; | |
947 | unsigned int oldalarm; | |
948 | ||
949 | it_new.it_interval.tv_sec = it_new.it_interval.tv_usec = 0; | |
950 | it_new.it_value.tv_sec = seconds; | |
951 | it_new.it_value.tv_usec = 0; | |
952 | do_setitimer(ITIMER_REAL, &it_new, &it_old); | |
953 | oldalarm = it_old.it_value.tv_sec; | |
954 | /* ehhh.. We can't return 0 if we have an alarm pending.. */ | |
955 | /* And we'd better return too much than too little anyway */ | |
956 | if ((!oldalarm && it_old.it_value.tv_usec) || it_old.it_value.tv_usec >= 500000) | |
957 | oldalarm++; | |
958 | return oldalarm; | |
959 | } | |
960 | ||
961 | #endif | |
962 | ||
963 | #ifndef __alpha__ | |
964 | ||
965 | /* | |
966 | * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this | |
967 | * should be moved into arch/i386 instead? | |
968 | */ | |
969 | ||
970 | /** | |
971 | * sys_getpid - return the thread group id of the current process | |
972 | * | |
973 | * Note, despite the name, this returns the tgid not the pid. The tgid and | |
974 | * the pid are identical unless CLONE_THREAD was specified on clone() in | |
975 | * which case the tgid is the same in all threads of the same group. | |
976 | * | |
977 | * This is SMP safe as current->tgid does not change. | |
978 | */ | |
979 | asmlinkage long sys_getpid(void) | |
980 | { | |
981 | return current->tgid; | |
982 | } | |
983 | ||
984 | /* | |
985 | * Accessing ->group_leader->real_parent is not SMP-safe, it could | |
986 | * change from under us. However, rather than getting any lock | |
987 | * we can use an optimistic algorithm: get the parent | |
988 | * pid, and go back and check that the parent is still | |
989 | * the same. If it has changed (which is extremely unlikely | |
990 | * indeed), we just try again.. | |
991 | * | |
992 | * NOTE! This depends on the fact that even if we _do_ | |
993 | * get an old value of "parent", we can happily dereference | |
994 | * the pointer (it was and remains a dereferencable kernel pointer | |
995 | * no matter what): we just can't necessarily trust the result | |
996 | * until we know that the parent pointer is valid. | |
997 | * | |
998 | * NOTE2: ->group_leader never changes from under us. | |
999 | */ | |
1000 | asmlinkage long sys_getppid(void) | |
1001 | { | |
1002 | int pid; | |
1003 | struct task_struct *me = current; | |
1004 | struct task_struct *parent; | |
1005 | ||
1006 | parent = me->group_leader->real_parent; | |
1007 | for (;;) { | |
1008 | pid = parent->tgid; | |
1009 | #ifdef CONFIG_SMP | |
1010 | { | |
1011 | struct task_struct *old = parent; | |
1012 | ||
1013 | /* | |
1014 | * Make sure we read the pid before re-reading the | |
1015 | * parent pointer: | |
1016 | */ | |
d59dd462 | 1017 | smp_rmb(); |
1da177e4 LT |
1018 | parent = me->group_leader->real_parent; |
1019 | if (old != parent) | |
1020 | continue; | |
1021 | } | |
1022 | #endif | |
1023 | break; | |
1024 | } | |
1025 | return pid; | |
1026 | } | |
1027 | ||
1028 | asmlinkage long sys_getuid(void) | |
1029 | { | |
1030 | /* Only we change this so SMP safe */ | |
1031 | return current->uid; | |
1032 | } | |
1033 | ||
1034 | asmlinkage long sys_geteuid(void) | |
1035 | { | |
1036 | /* Only we change this so SMP safe */ | |
1037 | return current->euid; | |
1038 | } | |
1039 | ||
1040 | asmlinkage long sys_getgid(void) | |
1041 | { | |
1042 | /* Only we change this so SMP safe */ | |
1043 | return current->gid; | |
1044 | } | |
1045 | ||
1046 | asmlinkage long sys_getegid(void) | |
1047 | { | |
1048 | /* Only we change this so SMP safe */ | |
1049 | return current->egid; | |
1050 | } | |
1051 | ||
1052 | #endif | |
1053 | ||
1054 | static void process_timeout(unsigned long __data) | |
1055 | { | |
1056 | wake_up_process((task_t *)__data); | |
1057 | } | |
1058 | ||
1059 | /** | |
1060 | * schedule_timeout - sleep until timeout | |
1061 | * @timeout: timeout value in jiffies | |
1062 | * | |
1063 | * Make the current task sleep until @timeout jiffies have | |
1064 | * elapsed. The routine will return immediately unless | |
1065 | * the current task state has been set (see set_current_state()). | |
1066 | * | |
1067 | * You can set the task state as follows - | |
1068 | * | |
1069 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to | |
1070 | * pass before the routine returns. The routine will return 0 | |
1071 | * | |
1072 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1073 | * delivered to the current task. In this case the remaining time | |
1074 | * in jiffies will be returned, or 0 if the timer expired in time | |
1075 | * | |
1076 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1077 | * routine returns. | |
1078 | * | |
1079 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule | |
1080 | * the CPU away without a bound on the timeout. In this case the return | |
1081 | * value will be %MAX_SCHEDULE_TIMEOUT. | |
1082 | * | |
1083 | * In all cases the return value is guaranteed to be non-negative. | |
1084 | */ | |
1085 | fastcall signed long __sched schedule_timeout(signed long timeout) | |
1086 | { | |
1087 | struct timer_list timer; | |
1088 | unsigned long expire; | |
1089 | ||
1090 | switch (timeout) | |
1091 | { | |
1092 | case MAX_SCHEDULE_TIMEOUT: | |
1093 | /* | |
1094 | * These two special cases are useful to be comfortable | |
1095 | * in the caller. Nothing more. We could take | |
1096 | * MAX_SCHEDULE_TIMEOUT from one of the negative value | |
1097 | * but I' d like to return a valid offset (>=0) to allow | |
1098 | * the caller to do everything it want with the retval. | |
1099 | */ | |
1100 | schedule(); | |
1101 | goto out; | |
1102 | default: | |
1103 | /* | |
1104 | * Another bit of PARANOID. Note that the retval will be | |
1105 | * 0 since no piece of kernel is supposed to do a check | |
1106 | * for a negative retval of schedule_timeout() (since it | |
1107 | * should never happens anyway). You just have the printk() | |
1108 | * that will tell you if something is gone wrong and where. | |
1109 | */ | |
1110 | if (timeout < 0) | |
1111 | { | |
1112 | printk(KERN_ERR "schedule_timeout: wrong timeout " | |
1113 | "value %lx from %p\n", timeout, | |
1114 | __builtin_return_address(0)); | |
1115 | current->state = TASK_RUNNING; | |
1116 | goto out; | |
1117 | } | |
1118 | } | |
1119 | ||
1120 | expire = timeout + jiffies; | |
1121 | ||
1122 | init_timer(&timer); | |
1123 | timer.expires = expire; | |
1124 | timer.data = (unsigned long) current; | |
1125 | timer.function = process_timeout; | |
1126 | ||
1127 | add_timer(&timer); | |
1128 | schedule(); | |
1129 | del_singleshot_timer_sync(&timer); | |
1130 | ||
1131 | timeout = expire - jiffies; | |
1132 | ||
1133 | out: | |
1134 | return timeout < 0 ? 0 : timeout; | |
1135 | } | |
1136 | ||
1137 | EXPORT_SYMBOL(schedule_timeout); | |
1138 | ||
1139 | /* Thread ID - the internal kernel "pid" */ | |
1140 | asmlinkage long sys_gettid(void) | |
1141 | { | |
1142 | return current->pid; | |
1143 | } | |
1144 | ||
1145 | static long __sched nanosleep_restart(struct restart_block *restart) | |
1146 | { | |
1147 | unsigned long expire = restart->arg0, now = jiffies; | |
1148 | struct timespec __user *rmtp = (struct timespec __user *) restart->arg1; | |
1149 | long ret; | |
1150 | ||
1151 | /* Did it expire while we handled signals? */ | |
1152 | if (!time_after(expire, now)) | |
1153 | return 0; | |
1154 | ||
1155 | current->state = TASK_INTERRUPTIBLE; | |
1156 | expire = schedule_timeout(expire - now); | |
1157 | ||
1158 | ret = 0; | |
1159 | if (expire) { | |
1160 | struct timespec t; | |
1161 | jiffies_to_timespec(expire, &t); | |
1162 | ||
1163 | ret = -ERESTART_RESTARTBLOCK; | |
1164 | if (rmtp && copy_to_user(rmtp, &t, sizeof(t))) | |
1165 | ret = -EFAULT; | |
1166 | /* The 'restart' block is already filled in */ | |
1167 | } | |
1168 | return ret; | |
1169 | } | |
1170 | ||
1171 | asmlinkage long sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp) | |
1172 | { | |
1173 | struct timespec t; | |
1174 | unsigned long expire; | |
1175 | long ret; | |
1176 | ||
1177 | if (copy_from_user(&t, rqtp, sizeof(t))) | |
1178 | return -EFAULT; | |
1179 | ||
1180 | if ((t.tv_nsec >= 1000000000L) || (t.tv_nsec < 0) || (t.tv_sec < 0)) | |
1181 | return -EINVAL; | |
1182 | ||
1183 | expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec); | |
1184 | current->state = TASK_INTERRUPTIBLE; | |
1185 | expire = schedule_timeout(expire); | |
1186 | ||
1187 | ret = 0; | |
1188 | if (expire) { | |
1189 | struct restart_block *restart; | |
1190 | jiffies_to_timespec(expire, &t); | |
1191 | if (rmtp && copy_to_user(rmtp, &t, sizeof(t))) | |
1192 | return -EFAULT; | |
1193 | ||
1194 | restart = ¤t_thread_info()->restart_block; | |
1195 | restart->fn = nanosleep_restart; | |
1196 | restart->arg0 = jiffies + expire; | |
1197 | restart->arg1 = (unsigned long) rmtp; | |
1198 | ret = -ERESTART_RESTARTBLOCK; | |
1199 | } | |
1200 | return ret; | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * sys_sysinfo - fill in sysinfo struct | |
1205 | */ | |
1206 | asmlinkage long sys_sysinfo(struct sysinfo __user *info) | |
1207 | { | |
1208 | struct sysinfo val; | |
1209 | unsigned long mem_total, sav_total; | |
1210 | unsigned int mem_unit, bitcount; | |
1211 | unsigned long seq; | |
1212 | ||
1213 | memset((char *)&val, 0, sizeof(struct sysinfo)); | |
1214 | ||
1215 | do { | |
1216 | struct timespec tp; | |
1217 | seq = read_seqbegin(&xtime_lock); | |
1218 | ||
1219 | /* | |
1220 | * This is annoying. The below is the same thing | |
1221 | * posix_get_clock_monotonic() does, but it wants to | |
1222 | * take the lock which we want to cover the loads stuff | |
1223 | * too. | |
1224 | */ | |
1225 | ||
1226 | getnstimeofday(&tp); | |
1227 | tp.tv_sec += wall_to_monotonic.tv_sec; | |
1228 | tp.tv_nsec += wall_to_monotonic.tv_nsec; | |
1229 | if (tp.tv_nsec - NSEC_PER_SEC >= 0) { | |
1230 | tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC; | |
1231 | tp.tv_sec++; | |
1232 | } | |
1233 | val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); | |
1234 | ||
1235 | val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT); | |
1236 | val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT); | |
1237 | val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT); | |
1238 | ||
1239 | val.procs = nr_threads; | |
1240 | } while (read_seqretry(&xtime_lock, seq)); | |
1241 | ||
1242 | si_meminfo(&val); | |
1243 | si_swapinfo(&val); | |
1244 | ||
1245 | /* | |
1246 | * If the sum of all the available memory (i.e. ram + swap) | |
1247 | * is less than can be stored in a 32 bit unsigned long then | |
1248 | * we can be binary compatible with 2.2.x kernels. If not, | |
1249 | * well, in that case 2.2.x was broken anyways... | |
1250 | * | |
1251 | * -Erik Andersen <andersee@debian.org> | |
1252 | */ | |
1253 | ||
1254 | mem_total = val.totalram + val.totalswap; | |
1255 | if (mem_total < val.totalram || mem_total < val.totalswap) | |
1256 | goto out; | |
1257 | bitcount = 0; | |
1258 | mem_unit = val.mem_unit; | |
1259 | while (mem_unit > 1) { | |
1260 | bitcount++; | |
1261 | mem_unit >>= 1; | |
1262 | sav_total = mem_total; | |
1263 | mem_total <<= 1; | |
1264 | if (mem_total < sav_total) | |
1265 | goto out; | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * If mem_total did not overflow, multiply all memory values by | |
1270 | * val.mem_unit and set it to 1. This leaves things compatible | |
1271 | * with 2.2.x, and also retains compatibility with earlier 2.4.x | |
1272 | * kernels... | |
1273 | */ | |
1274 | ||
1275 | val.mem_unit = 1; | |
1276 | val.totalram <<= bitcount; | |
1277 | val.freeram <<= bitcount; | |
1278 | val.sharedram <<= bitcount; | |
1279 | val.bufferram <<= bitcount; | |
1280 | val.totalswap <<= bitcount; | |
1281 | val.freeswap <<= bitcount; | |
1282 | val.totalhigh <<= bitcount; | |
1283 | val.freehigh <<= bitcount; | |
1284 | ||
1285 | out: | |
1286 | if (copy_to_user(info, &val, sizeof(struct sysinfo))) | |
1287 | return -EFAULT; | |
1288 | ||
1289 | return 0; | |
1290 | } | |
1291 | ||
1292 | static void __devinit init_timers_cpu(int cpu) | |
1293 | { | |
1294 | int j; | |
1295 | tvec_base_t *base; | |
55c888d6 | 1296 | |
1da177e4 | 1297 | base = &per_cpu(tvec_bases, cpu); |
55c888d6 | 1298 | spin_lock_init(&base->t_base.lock); |
1da177e4 LT |
1299 | for (j = 0; j < TVN_SIZE; j++) { |
1300 | INIT_LIST_HEAD(base->tv5.vec + j); | |
1301 | INIT_LIST_HEAD(base->tv4.vec + j); | |
1302 | INIT_LIST_HEAD(base->tv3.vec + j); | |
1303 | INIT_LIST_HEAD(base->tv2.vec + j); | |
1304 | } | |
1305 | for (j = 0; j < TVR_SIZE; j++) | |
1306 | INIT_LIST_HEAD(base->tv1.vec + j); | |
1307 | ||
1308 | base->timer_jiffies = jiffies; | |
1309 | } | |
1310 | ||
1311 | #ifdef CONFIG_HOTPLUG_CPU | |
55c888d6 | 1312 | static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head) |
1da177e4 LT |
1313 | { |
1314 | struct timer_list *timer; | |
1315 | ||
1316 | while (!list_empty(head)) { | |
1317 | timer = list_entry(head->next, struct timer_list, entry); | |
55c888d6 ON |
1318 | detach_timer(timer, 0); |
1319 | timer->base = &new_base->t_base; | |
1da177e4 | 1320 | internal_add_timer(new_base, timer); |
1da177e4 | 1321 | } |
1da177e4 LT |
1322 | } |
1323 | ||
1324 | static void __devinit migrate_timers(int cpu) | |
1325 | { | |
1326 | tvec_base_t *old_base; | |
1327 | tvec_base_t *new_base; | |
1328 | int i; | |
1329 | ||
1330 | BUG_ON(cpu_online(cpu)); | |
1331 | old_base = &per_cpu(tvec_bases, cpu); | |
1332 | new_base = &get_cpu_var(tvec_bases); | |
1333 | ||
1334 | local_irq_disable(); | |
55c888d6 ON |
1335 | spin_lock(&new_base->t_base.lock); |
1336 | spin_lock(&old_base->t_base.lock); | |
1da177e4 | 1337 | |
55c888d6 | 1338 | if (old_base->t_base.running_timer) |
1da177e4 LT |
1339 | BUG(); |
1340 | for (i = 0; i < TVR_SIZE; i++) | |
55c888d6 ON |
1341 | migrate_timer_list(new_base, old_base->tv1.vec + i); |
1342 | for (i = 0; i < TVN_SIZE; i++) { | |
1343 | migrate_timer_list(new_base, old_base->tv2.vec + i); | |
1344 | migrate_timer_list(new_base, old_base->tv3.vec + i); | |
1345 | migrate_timer_list(new_base, old_base->tv4.vec + i); | |
1346 | migrate_timer_list(new_base, old_base->tv5.vec + i); | |
1347 | } | |
1348 | ||
1349 | spin_unlock(&old_base->t_base.lock); | |
1350 | spin_unlock(&new_base->t_base.lock); | |
1da177e4 LT |
1351 | local_irq_enable(); |
1352 | put_cpu_var(tvec_bases); | |
1da177e4 LT |
1353 | } |
1354 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1355 | ||
1356 | static int __devinit timer_cpu_notify(struct notifier_block *self, | |
1357 | unsigned long action, void *hcpu) | |
1358 | { | |
1359 | long cpu = (long)hcpu; | |
1360 | switch(action) { | |
1361 | case CPU_UP_PREPARE: | |
1362 | init_timers_cpu(cpu); | |
1363 | break; | |
1364 | #ifdef CONFIG_HOTPLUG_CPU | |
1365 | case CPU_DEAD: | |
1366 | migrate_timers(cpu); | |
1367 | break; | |
1368 | #endif | |
1369 | default: | |
1370 | break; | |
1371 | } | |
1372 | return NOTIFY_OK; | |
1373 | } | |
1374 | ||
1375 | static struct notifier_block __devinitdata timers_nb = { | |
1376 | .notifier_call = timer_cpu_notify, | |
1377 | }; | |
1378 | ||
1379 | ||
1380 | void __init init_timers(void) | |
1381 | { | |
1382 | timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE, | |
1383 | (void *)(long)smp_processor_id()); | |
1384 | register_cpu_notifier(&timers_nb); | |
1385 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL); | |
1386 | } | |
1387 | ||
1388 | #ifdef CONFIG_TIME_INTERPOLATION | |
1389 | ||
1390 | struct time_interpolator *time_interpolator; | |
1391 | static struct time_interpolator *time_interpolator_list; | |
1392 | static DEFINE_SPINLOCK(time_interpolator_lock); | |
1393 | ||
1394 | static inline u64 time_interpolator_get_cycles(unsigned int src) | |
1395 | { | |
1396 | unsigned long (*x)(void); | |
1397 | ||
1398 | switch (src) | |
1399 | { | |
1400 | case TIME_SOURCE_FUNCTION: | |
1401 | x = time_interpolator->addr; | |
1402 | return x(); | |
1403 | ||
1404 | case TIME_SOURCE_MMIO64 : | |
1405 | return readq((void __iomem *) time_interpolator->addr); | |
1406 | ||
1407 | case TIME_SOURCE_MMIO32 : | |
1408 | return readl((void __iomem *) time_interpolator->addr); | |
1409 | ||
1410 | default: return get_cycles(); | |
1411 | } | |
1412 | } | |
1413 | ||
1414 | static inline u64 time_interpolator_get_counter(void) | |
1415 | { | |
1416 | unsigned int src = time_interpolator->source; | |
1417 | ||
1418 | if (time_interpolator->jitter) | |
1419 | { | |
1420 | u64 lcycle; | |
1421 | u64 now; | |
1422 | ||
1423 | do { | |
1424 | lcycle = time_interpolator->last_cycle; | |
1425 | now = time_interpolator_get_cycles(src); | |
1426 | if (lcycle && time_after(lcycle, now)) | |
1427 | return lcycle; | |
1428 | /* Keep track of the last timer value returned. The use of cmpxchg here | |
1429 | * will cause contention in an SMP environment. | |
1430 | */ | |
1431 | } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle)); | |
1432 | return now; | |
1433 | } | |
1434 | else | |
1435 | return time_interpolator_get_cycles(src); | |
1436 | } | |
1437 | ||
1438 | void time_interpolator_reset(void) | |
1439 | { | |
1440 | time_interpolator->offset = 0; | |
1441 | time_interpolator->last_counter = time_interpolator_get_counter(); | |
1442 | } | |
1443 | ||
1444 | #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift) | |
1445 | ||
1446 | unsigned long time_interpolator_get_offset(void) | |
1447 | { | |
1448 | /* If we do not have a time interpolator set up then just return zero */ | |
1449 | if (!time_interpolator) | |
1450 | return 0; | |
1451 | ||
1452 | return time_interpolator->offset + | |
1453 | GET_TI_NSECS(time_interpolator_get_counter(), time_interpolator); | |
1454 | } | |
1455 | ||
1456 | #define INTERPOLATOR_ADJUST 65536 | |
1457 | #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST | |
1458 | ||
1459 | static void time_interpolator_update(long delta_nsec) | |
1460 | { | |
1461 | u64 counter; | |
1462 | unsigned long offset; | |
1463 | ||
1464 | /* If there is no time interpolator set up then do nothing */ | |
1465 | if (!time_interpolator) | |
1466 | return; | |
1467 | ||
1468 | /* The interpolator compensates for late ticks by accumulating | |
1469 | * the late time in time_interpolator->offset. A tick earlier than | |
1470 | * expected will lead to a reset of the offset and a corresponding | |
1471 | * jump of the clock forward. Again this only works if the | |
1472 | * interpolator clock is running slightly slower than the regular clock | |
1473 | * and the tuning logic insures that. | |
1474 | */ | |
1475 | ||
1476 | counter = time_interpolator_get_counter(); | |
1477 | offset = time_interpolator->offset + GET_TI_NSECS(counter, time_interpolator); | |
1478 | ||
1479 | if (delta_nsec < 0 || (unsigned long) delta_nsec < offset) | |
1480 | time_interpolator->offset = offset - delta_nsec; | |
1481 | else { | |
1482 | time_interpolator->skips++; | |
1483 | time_interpolator->ns_skipped += delta_nsec - offset; | |
1484 | time_interpolator->offset = 0; | |
1485 | } | |
1486 | time_interpolator->last_counter = counter; | |
1487 | ||
1488 | /* Tuning logic for time interpolator invoked every minute or so. | |
1489 | * Decrease interpolator clock speed if no skips occurred and an offset is carried. | |
1490 | * Increase interpolator clock speed if we skip too much time. | |
1491 | */ | |
1492 | if (jiffies % INTERPOLATOR_ADJUST == 0) | |
1493 | { | |
1494 | if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC) | |
1495 | time_interpolator->nsec_per_cyc--; | |
1496 | if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0) | |
1497 | time_interpolator->nsec_per_cyc++; | |
1498 | time_interpolator->skips = 0; | |
1499 | time_interpolator->ns_skipped = 0; | |
1500 | } | |
1501 | } | |
1502 | ||
1503 | static inline int | |
1504 | is_better_time_interpolator(struct time_interpolator *new) | |
1505 | { | |
1506 | if (!time_interpolator) | |
1507 | return 1; | |
1508 | return new->frequency > 2*time_interpolator->frequency || | |
1509 | (unsigned long)new->drift < (unsigned long)time_interpolator->drift; | |
1510 | } | |
1511 | ||
1512 | void | |
1513 | register_time_interpolator(struct time_interpolator *ti) | |
1514 | { | |
1515 | unsigned long flags; | |
1516 | ||
1517 | /* Sanity check */ | |
1518 | if (ti->frequency == 0 || ti->mask == 0) | |
1519 | BUG(); | |
1520 | ||
1521 | ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency; | |
1522 | spin_lock(&time_interpolator_lock); | |
1523 | write_seqlock_irqsave(&xtime_lock, flags); | |
1524 | if (is_better_time_interpolator(ti)) { | |
1525 | time_interpolator = ti; | |
1526 | time_interpolator_reset(); | |
1527 | } | |
1528 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
1529 | ||
1530 | ti->next = time_interpolator_list; | |
1531 | time_interpolator_list = ti; | |
1532 | spin_unlock(&time_interpolator_lock); | |
1533 | } | |
1534 | ||
1535 | void | |
1536 | unregister_time_interpolator(struct time_interpolator *ti) | |
1537 | { | |
1538 | struct time_interpolator *curr, **prev; | |
1539 | unsigned long flags; | |
1540 | ||
1541 | spin_lock(&time_interpolator_lock); | |
1542 | prev = &time_interpolator_list; | |
1543 | for (curr = *prev; curr; curr = curr->next) { | |
1544 | if (curr == ti) { | |
1545 | *prev = curr->next; | |
1546 | break; | |
1547 | } | |
1548 | prev = &curr->next; | |
1549 | } | |
1550 | ||
1551 | write_seqlock_irqsave(&xtime_lock, flags); | |
1552 | if (ti == time_interpolator) { | |
1553 | /* we lost the best time-interpolator: */ | |
1554 | time_interpolator = NULL; | |
1555 | /* find the next-best interpolator */ | |
1556 | for (curr = time_interpolator_list; curr; curr = curr->next) | |
1557 | if (is_better_time_interpolator(curr)) | |
1558 | time_interpolator = curr; | |
1559 | time_interpolator_reset(); | |
1560 | } | |
1561 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
1562 | spin_unlock(&time_interpolator_lock); | |
1563 | } | |
1564 | #endif /* CONFIG_TIME_INTERPOLATION */ | |
1565 | ||
1566 | /** | |
1567 | * msleep - sleep safely even with waitqueue interruptions | |
1568 | * @msecs: Time in milliseconds to sleep for | |
1569 | */ | |
1570 | void msleep(unsigned int msecs) | |
1571 | { | |
1572 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1573 | ||
1574 | while (timeout) { | |
1575 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1576 | timeout = schedule_timeout(timeout); | |
1577 | } | |
1578 | } | |
1579 | ||
1580 | EXPORT_SYMBOL(msleep); | |
1581 | ||
1582 | /** | |
1583 | * msleep_interruptible - sleep waiting for waitqueue interruptions | |
1584 | * @msecs: Time in milliseconds to sleep for | |
1585 | */ | |
1586 | unsigned long msleep_interruptible(unsigned int msecs) | |
1587 | { | |
1588 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; | |
1589 | ||
1590 | while (timeout && !signal_pending(current)) { | |
1591 | set_current_state(TASK_INTERRUPTIBLE); | |
1592 | timeout = schedule_timeout(timeout); | |
1593 | } | |
1594 | return jiffies_to_msecs(timeout); | |
1595 | } | |
1596 | ||
1597 | EXPORT_SYMBOL(msleep_interruptible); |