]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - kernel/timer.c
[PATCH] Char: stallion, implement fail paths
[mirror_ubuntu-bionic-kernel.git] / kernel / timer.c
CommitLineData
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>
97a41e26 36#include <linux/delay.h>
1da177e4
LT
37
38#include <asm/uaccess.h>
39#include <asm/unistd.h>
40#include <asm/div64.h>
41#include <asm/timex.h>
42#include <asm/io.h>
43
ecea8d19
TG
44u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
45
46EXPORT_SYMBOL(jiffies_64);
47
1da177e4
LT
48/*
49 * per-CPU timer vector definitions:
50 */
1da177e4
LT
51#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
52#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
53#define TVN_SIZE (1 << TVN_BITS)
54#define TVR_SIZE (1 << TVR_BITS)
55#define TVN_MASK (TVN_SIZE - 1)
56#define TVR_MASK (TVR_SIZE - 1)
57
58typedef struct tvec_s {
59 struct list_head vec[TVN_SIZE];
60} tvec_t;
61
62typedef struct tvec_root_s {
63 struct list_head vec[TVR_SIZE];
64} tvec_root_t;
65
66struct tvec_t_base_s {
3691c519
ON
67 spinlock_t lock;
68 struct timer_list *running_timer;
1da177e4 69 unsigned long timer_jiffies;
1da177e4
LT
70 tvec_root_t tv1;
71 tvec_t tv2;
72 tvec_t tv3;
73 tvec_t tv4;
74 tvec_t tv5;
75} ____cacheline_aligned_in_smp;
76
77typedef struct tvec_t_base_s tvec_base_t;
ba6edfcd 78
3691c519
ON
79tvec_base_t boot_tvec_bases;
80EXPORT_SYMBOL(boot_tvec_bases);
51d8c5ed 81static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = &boot_tvec_bases;
1da177e4
LT
82
83static inline void set_running_timer(tvec_base_t *base,
84 struct timer_list *timer)
85{
86#ifdef CONFIG_SMP
3691c519 87 base->running_timer = timer;
1da177e4
LT
88#endif
89}
90
1da177e4
LT
91static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
92{
93 unsigned long expires = timer->expires;
94 unsigned long idx = expires - base->timer_jiffies;
95 struct list_head *vec;
96
97 if (idx < TVR_SIZE) {
98 int i = expires & TVR_MASK;
99 vec = base->tv1.vec + i;
100 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
101 int i = (expires >> TVR_BITS) & TVN_MASK;
102 vec = base->tv2.vec + i;
103 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
104 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
105 vec = base->tv3.vec + i;
106 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
107 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
108 vec = base->tv4.vec + i;
109 } else if ((signed long) idx < 0) {
110 /*
111 * Can happen if you add a timer with expires == jiffies,
112 * or you set a timer to go off in the past
113 */
114 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
115 } else {
116 int i;
117 /* If the timeout is larger than 0xffffffff on 64-bit
118 * architectures then we use the maximum timeout:
119 */
120 if (idx > 0xffffffffUL) {
121 idx = 0xffffffffUL;
122 expires = idx + base->timer_jiffies;
123 }
124 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
125 vec = base->tv5.vec + i;
126 }
127 /*
128 * Timers are FIFO:
129 */
130 list_add_tail(&timer->entry, vec);
131}
132
2aae4a10 133/**
55c888d6
ON
134 * init_timer - initialize a timer.
135 * @timer: the timer to be initialized
136 *
137 * init_timer() must be done to a timer prior calling *any* of the
138 * other timer functions.
139 */
140void fastcall init_timer(struct timer_list *timer)
141{
142 timer->entry.next = NULL;
bfe5d834 143 timer->base = __raw_get_cpu_var(tvec_bases);
55c888d6
ON
144}
145EXPORT_SYMBOL(init_timer);
146
147static inline void detach_timer(struct timer_list *timer,
148 int clear_pending)
149{
150 struct list_head *entry = &timer->entry;
151
152 __list_del(entry->prev, entry->next);
153 if (clear_pending)
154 entry->next = NULL;
155 entry->prev = LIST_POISON2;
156}
157
158/*
3691c519 159 * We are using hashed locking: holding per_cpu(tvec_bases).lock
55c888d6
ON
160 * means that all timers which are tied to this base via timer->base are
161 * locked, and the base itself is locked too.
162 *
163 * So __run_timers/migrate_timers can safely modify all timers which could
164 * be found on ->tvX lists.
165 *
166 * When the timer's base is locked, and the timer removed from list, it is
167 * possible to set timer->base = NULL and drop the lock: the timer remains
168 * locked.
169 */
3691c519 170static tvec_base_t *lock_timer_base(struct timer_list *timer,
55c888d6 171 unsigned long *flags)
89e7e374 172 __acquires(timer->base->lock)
55c888d6 173{
3691c519 174 tvec_base_t *base;
55c888d6
ON
175
176 for (;;) {
177 base = timer->base;
178 if (likely(base != NULL)) {
179 spin_lock_irqsave(&base->lock, *flags);
180 if (likely(base == timer->base))
181 return base;
182 /* The timer has migrated to another CPU */
183 spin_unlock_irqrestore(&base->lock, *flags);
184 }
185 cpu_relax();
186 }
187}
188
1da177e4
LT
189int __mod_timer(struct timer_list *timer, unsigned long expires)
190{
3691c519 191 tvec_base_t *base, *new_base;
1da177e4
LT
192 unsigned long flags;
193 int ret = 0;
194
195 BUG_ON(!timer->function);
1da177e4 196
55c888d6
ON
197 base = lock_timer_base(timer, &flags);
198
199 if (timer_pending(timer)) {
200 detach_timer(timer, 0);
201 ret = 1;
202 }
203
a4a6198b 204 new_base = __get_cpu_var(tvec_bases);
1da177e4 205
3691c519 206 if (base != new_base) {
1da177e4 207 /*
55c888d6
ON
208 * We are trying to schedule the timer on the local CPU.
209 * However we can't change timer's base while it is running,
210 * otherwise del_timer_sync() can't detect that the timer's
211 * handler yet has not finished. This also guarantees that
212 * the timer is serialized wrt itself.
1da177e4 213 */
a2c348fe 214 if (likely(base->running_timer != timer)) {
55c888d6
ON
215 /* See the comment in lock_timer_base() */
216 timer->base = NULL;
217 spin_unlock(&base->lock);
a2c348fe
ON
218 base = new_base;
219 spin_lock(&base->lock);
220 timer->base = base;
1da177e4
LT
221 }
222 }
223
1da177e4 224 timer->expires = expires;
a2c348fe
ON
225 internal_add_timer(base, timer);
226 spin_unlock_irqrestore(&base->lock, flags);
1da177e4
LT
227
228 return ret;
229}
230
231EXPORT_SYMBOL(__mod_timer);
232
2aae4a10 233/**
1da177e4
LT
234 * add_timer_on - start a timer on a particular CPU
235 * @timer: the timer to be added
236 * @cpu: the CPU to start it on
237 *
238 * This is not very scalable on SMP. Double adds are not possible.
239 */
240void add_timer_on(struct timer_list *timer, int cpu)
241{
a4a6198b 242 tvec_base_t *base = per_cpu(tvec_bases, cpu);
1da177e4 243 unsigned long flags;
55c888d6 244
1da177e4 245 BUG_ON(timer_pending(timer) || !timer->function);
3691c519
ON
246 spin_lock_irqsave(&base->lock, flags);
247 timer->base = base;
1da177e4 248 internal_add_timer(base, timer);
3691c519 249 spin_unlock_irqrestore(&base->lock, flags);
1da177e4
LT
250}
251
252
2aae4a10 253/**
1da177e4
LT
254 * mod_timer - modify a timer's timeout
255 * @timer: the timer to be modified
2aae4a10 256 * @expires: new timeout in jiffies
1da177e4
LT
257 *
258 * mod_timer is a more efficient way to update the expire field of an
259 * active timer (if the timer is inactive it will be activated)
260 *
261 * mod_timer(timer, expires) is equivalent to:
262 *
263 * del_timer(timer); timer->expires = expires; add_timer(timer);
264 *
265 * Note that if there are multiple unserialized concurrent users of the
266 * same timer, then mod_timer() is the only safe way to modify the timeout,
267 * since add_timer() cannot modify an already running timer.
268 *
269 * The function returns whether it has modified a pending timer or not.
270 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
271 * active timer returns 1.)
272 */
273int mod_timer(struct timer_list *timer, unsigned long expires)
274{
275 BUG_ON(!timer->function);
276
1da177e4
LT
277 /*
278 * This is a common optimization triggered by the
279 * networking code - if the timer is re-modified
280 * to be the same thing then just return:
281 */
282 if (timer->expires == expires && timer_pending(timer))
283 return 1;
284
285 return __mod_timer(timer, expires);
286}
287
288EXPORT_SYMBOL(mod_timer);
289
2aae4a10 290/**
1da177e4
LT
291 * del_timer - deactive a timer.
292 * @timer: the timer to be deactivated
293 *
294 * del_timer() deactivates a timer - this works on both active and inactive
295 * timers.
296 *
297 * The function returns whether it has deactivated a pending timer or not.
298 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
299 * active timer returns 1.)
300 */
301int del_timer(struct timer_list *timer)
302{
3691c519 303 tvec_base_t *base;
1da177e4 304 unsigned long flags;
55c888d6 305 int ret = 0;
1da177e4 306
55c888d6
ON
307 if (timer_pending(timer)) {
308 base = lock_timer_base(timer, &flags);
309 if (timer_pending(timer)) {
310 detach_timer(timer, 1);
311 ret = 1;
312 }
1da177e4 313 spin_unlock_irqrestore(&base->lock, flags);
1da177e4 314 }
1da177e4 315
55c888d6 316 return ret;
1da177e4
LT
317}
318
319EXPORT_SYMBOL(del_timer);
320
321#ifdef CONFIG_SMP
2aae4a10
REB
322/**
323 * try_to_del_timer_sync - Try to deactivate a timer
324 * @timer: timer do del
325 *
fd450b73
ON
326 * This function tries to deactivate a timer. Upon successful (ret >= 0)
327 * exit the timer is not queued and the handler is not running on any CPU.
328 *
329 * It must not be called from interrupt contexts.
330 */
331int try_to_del_timer_sync(struct timer_list *timer)
332{
3691c519 333 tvec_base_t *base;
fd450b73
ON
334 unsigned long flags;
335 int ret = -1;
336
337 base = lock_timer_base(timer, &flags);
338
339 if (base->running_timer == timer)
340 goto out;
341
342 ret = 0;
343 if (timer_pending(timer)) {
344 detach_timer(timer, 1);
345 ret = 1;
346 }
347out:
348 spin_unlock_irqrestore(&base->lock, flags);
349
350 return ret;
351}
352
2aae4a10 353/**
1da177e4
LT
354 * del_timer_sync - deactivate a timer and wait for the handler to finish.
355 * @timer: the timer to be deactivated
356 *
357 * This function only differs from del_timer() on SMP: besides deactivating
358 * the timer it also makes sure the handler has finished executing on other
359 * CPUs.
360 *
361 * Synchronization rules: callers must prevent restarting of the timer,
362 * otherwise this function is meaningless. It must not be called from
363 * interrupt contexts. The caller must not hold locks which would prevent
55c888d6
ON
364 * completion of the timer's handler. The timer's handler must not call
365 * add_timer_on(). Upon exit the timer is not queued and the handler is
366 * not running on any CPU.
1da177e4
LT
367 *
368 * The function returns whether it has deactivated a pending timer or not.
1da177e4
LT
369 */
370int del_timer_sync(struct timer_list *timer)
371{
fd450b73
ON
372 for (;;) {
373 int ret = try_to_del_timer_sync(timer);
374 if (ret >= 0)
375 return ret;
a0009652 376 cpu_relax();
fd450b73 377 }
1da177e4 378}
1da177e4 379
55c888d6 380EXPORT_SYMBOL(del_timer_sync);
1da177e4
LT
381#endif
382
383static int cascade(tvec_base_t *base, tvec_t *tv, int index)
384{
385 /* cascade all the timers from tv up one level */
3439dd86
P
386 struct timer_list *timer, *tmp;
387 struct list_head tv_list;
388
389 list_replace_init(tv->vec + index, &tv_list);
1da177e4 390
1da177e4 391 /*
3439dd86
P
392 * We are removing _all_ timers from the list, so we
393 * don't have to detach them individually.
1da177e4 394 */
3439dd86
P
395 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
396 BUG_ON(timer->base != base);
397 internal_add_timer(base, timer);
1da177e4 398 }
1da177e4
LT
399
400 return index;
401}
402
2aae4a10
REB
403#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
404
405/**
1da177e4
LT
406 * __run_timers - run all expired timers (if any) on this CPU.
407 * @base: the timer vector to be processed.
408 *
409 * This function cascades all vectors and executes all expired timer
410 * vectors.
411 */
1da177e4
LT
412static inline void __run_timers(tvec_base_t *base)
413{
414 struct timer_list *timer;
415
3691c519 416 spin_lock_irq(&base->lock);
1da177e4 417 while (time_after_eq(jiffies, base->timer_jiffies)) {
626ab0e6 418 struct list_head work_list;
1da177e4
LT
419 struct list_head *head = &work_list;
420 int index = base->timer_jiffies & TVR_MASK;
626ab0e6 421
1da177e4
LT
422 /*
423 * Cascade timers:
424 */
425 if (!index &&
426 (!cascade(base, &base->tv2, INDEX(0))) &&
427 (!cascade(base, &base->tv3, INDEX(1))) &&
428 !cascade(base, &base->tv4, INDEX(2)))
429 cascade(base, &base->tv5, INDEX(3));
626ab0e6
ON
430 ++base->timer_jiffies;
431 list_replace_init(base->tv1.vec + index, &work_list);
55c888d6 432 while (!list_empty(head)) {
1da177e4
LT
433 void (*fn)(unsigned long);
434 unsigned long data;
435
436 timer = list_entry(head->next,struct timer_list,entry);
437 fn = timer->function;
438 data = timer->data;
439
1da177e4 440 set_running_timer(base, timer);
55c888d6 441 detach_timer(timer, 1);
3691c519 442 spin_unlock_irq(&base->lock);
1da177e4 443 {
be5b4fbd 444 int preempt_count = preempt_count();
1da177e4
LT
445 fn(data);
446 if (preempt_count != preempt_count()) {
be5b4fbd
JJ
447 printk(KERN_WARNING "huh, entered %p "
448 "with preempt_count %08x, exited"
449 " with %08x?\n",
450 fn, preempt_count,
451 preempt_count());
1da177e4
LT
452 BUG();
453 }
454 }
3691c519 455 spin_lock_irq(&base->lock);
1da177e4
LT
456 }
457 }
458 set_running_timer(base, NULL);
3691c519 459 spin_unlock_irq(&base->lock);
1da177e4
LT
460}
461
462#ifdef CONFIG_NO_IDLE_HZ
463/*
464 * Find out when the next timer event is due to happen. This
465 * is used on S/390 to stop all activity when a cpus is idle.
466 * This functions needs to be called disabled.
467 */
468unsigned long next_timer_interrupt(void)
469{
470 tvec_base_t *base;
471 struct list_head *list;
472 struct timer_list *nte;
473 unsigned long expires;
69239749
TL
474 unsigned long hr_expires = MAX_JIFFY_OFFSET;
475 ktime_t hr_delta;
1da177e4
LT
476 tvec_t *varray[4];
477 int i, j;
478
69239749
TL
479 hr_delta = hrtimer_get_next_event();
480 if (hr_delta.tv64 != KTIME_MAX) {
481 struct timespec tsdelta;
482 tsdelta = ktime_to_timespec(hr_delta);
483 hr_expires = timespec_to_jiffies(&tsdelta);
484 if (hr_expires < 3)
485 return hr_expires + jiffies;
486 }
487 hr_expires += jiffies;
488
a4a6198b 489 base = __get_cpu_var(tvec_bases);
3691c519 490 spin_lock(&base->lock);
1da177e4 491 expires = base->timer_jiffies + (LONG_MAX >> 1);
53f087fe 492 list = NULL;
1da177e4
LT
493
494 /* Look for timer events in tv1. */
495 j = base->timer_jiffies & TVR_MASK;
496 do {
497 list_for_each_entry(nte, base->tv1.vec + j, entry) {
498 expires = nte->expires;
499 if (j < (base->timer_jiffies & TVR_MASK))
500 list = base->tv2.vec + (INDEX(0));
501 goto found;
502 }
503 j = (j + 1) & TVR_MASK;
504 } while (j != (base->timer_jiffies & TVR_MASK));
505
506 /* Check tv2-tv5. */
507 varray[0] = &base->tv2;
508 varray[1] = &base->tv3;
509 varray[2] = &base->tv4;
510 varray[3] = &base->tv5;
511 for (i = 0; i < 4; i++) {
512 j = INDEX(i);
513 do {
514 if (list_empty(varray[i]->vec + j)) {
515 j = (j + 1) & TVN_MASK;
516 continue;
517 }
518 list_for_each_entry(nte, varray[i]->vec + j, entry)
519 if (time_before(nte->expires, expires))
520 expires = nte->expires;
521 if (j < (INDEX(i)) && i < 3)
522 list = varray[i + 1]->vec + (INDEX(i + 1));
523 goto found;
524 } while (j != (INDEX(i)));
525 }
526found:
527 if (list) {
528 /*
529 * The search wrapped. We need to look at the next list
530 * from next tv element that would cascade into tv element
531 * where we found the timer element.
532 */
533 list_for_each_entry(nte, list, entry) {
534 if (time_before(nte->expires, expires))
535 expires = nte->expires;
536 }
537 }
3691c519 538 spin_unlock(&base->lock);
69239749 539
0662b713
ZA
540 /*
541 * It can happen that other CPUs service timer IRQs and increment
542 * jiffies, but we have not yet got a local timer tick to process
543 * the timer wheels. In that case, the expiry time can be before
544 * jiffies, but since the high-resolution timer here is relative to
545 * jiffies, the default expression when high-resolution timers are
546 * not active,
547 *
548 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
549 *
550 * would falsely evaluate to true. If that is the case, just
551 * return jiffies so that we can immediately fire the local timer
552 */
553 if (time_before(expires, jiffies))
554 return jiffies;
555
69239749
TL
556 if (time_before(hr_expires, expires))
557 return hr_expires;
558
1da177e4
LT
559 return expires;
560}
561#endif
562
563/******************************************************************/
564
1da177e4
LT
565/*
566 * The current time
567 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
568 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
569 * at zero at system boot time, so wall_to_monotonic will be negative,
570 * however, we will ALWAYS keep the tv_nsec part positive so we can use
571 * the usual normalization.
572 */
573struct timespec xtime __attribute__ ((aligned (16)));
574struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
575
576EXPORT_SYMBOL(xtime);
577
726c14bf 578
ad596171
JS
579/* XXX - all of this timekeeping code should be later moved to time.c */
580#include <linux/clocksource.h>
581static struct clocksource *clock; /* pointer to current clocksource */
cf3c769b
JS
582
583#ifdef CONFIG_GENERIC_TIME
584/**
585 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
586 *
587 * private function, must hold xtime_lock lock when being
588 * called. Returns the number of nanoseconds since the
589 * last call to update_wall_time() (adjusted by NTP scaling)
590 */
591static inline s64 __get_nsec_offset(void)
592{
593 cycle_t cycle_now, cycle_delta;
594 s64 ns_offset;
595
596 /* read clocksource: */
a2752549 597 cycle_now = clocksource_read(clock);
cf3c769b
JS
598
599 /* calculate the delta since the last update_wall_time: */
19923c19 600 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
cf3c769b
JS
601
602 /* convert to nanoseconds: */
603 ns_offset = cyc2ns(clock, cycle_delta);
604
605 return ns_offset;
606}
607
608/**
609 * __get_realtime_clock_ts - Returns the time of day in a timespec
610 * @ts: pointer to the timespec to be set
611 *
612 * Returns the time of day in a timespec. Used by
613 * do_gettimeofday() and get_realtime_clock_ts().
614 */
615static inline void __get_realtime_clock_ts(struct timespec *ts)
616{
617 unsigned long seq;
618 s64 nsecs;
619
620 do {
621 seq = read_seqbegin(&xtime_lock);
622
623 *ts = xtime;
624 nsecs = __get_nsec_offset();
625
626 } while (read_seqretry(&xtime_lock, seq));
627
628 timespec_add_ns(ts, nsecs);
629}
630
631/**
a2752549 632 * getnstimeofday - Returns the time of day in a timespec
cf3c769b
JS
633 * @ts: pointer to the timespec to be set
634 *
635 * Returns the time of day in a timespec.
636 */
637void getnstimeofday(struct timespec *ts)
638{
639 __get_realtime_clock_ts(ts);
640}
641
642EXPORT_SYMBOL(getnstimeofday);
643
644/**
645 * do_gettimeofday - Returns the time of day in a timeval
646 * @tv: pointer to the timeval to be set
647 *
648 * NOTE: Users should be converted to using get_realtime_clock_ts()
649 */
650void do_gettimeofday(struct timeval *tv)
651{
652 struct timespec now;
653
654 __get_realtime_clock_ts(&now);
655 tv->tv_sec = now.tv_sec;
656 tv->tv_usec = now.tv_nsec/1000;
657}
658
659EXPORT_SYMBOL(do_gettimeofday);
660/**
661 * do_settimeofday - Sets the time of day
662 * @tv: pointer to the timespec variable containing the new time
663 *
664 * Sets the time of day to the new time and update NTP and notify hrtimers
665 */
666int do_settimeofday(struct timespec *tv)
667{
668 unsigned long flags;
669 time_t wtm_sec, sec = tv->tv_sec;
670 long wtm_nsec, nsec = tv->tv_nsec;
671
672 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
673 return -EINVAL;
674
675 write_seqlock_irqsave(&xtime_lock, flags);
676
677 nsec -= __get_nsec_offset();
678
679 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
680 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
681
682 set_normalized_timespec(&xtime, sec, nsec);
683 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
684
e154ff3d 685 clock->error = 0;
cf3c769b
JS
686 ntp_clear();
687
688 write_sequnlock_irqrestore(&xtime_lock, flags);
689
690 /* signal hrtimers about time change */
691 clock_was_set();
692
693 return 0;
694}
695
696EXPORT_SYMBOL(do_settimeofday);
697
698/**
699 * change_clocksource - Swaps clocksources if a new one is available
700 *
701 * Accumulates current time interval and initializes new clocksource
702 */
703static int change_clocksource(void)
704{
705 struct clocksource *new;
706 cycle_t now;
707 u64 nsec;
a2752549 708 new = clocksource_get_next();
cf3c769b 709 if (clock != new) {
a2752549 710 now = clocksource_read(new);
cf3c769b
JS
711 nsec = __get_nsec_offset();
712 timespec_add_ns(&xtime, nsec);
713
714 clock = new;
19923c19 715 clock->cycle_last = now;
cf3c769b
JS
716 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
717 clock->name);
718 return 1;
719 } else if (clock->update_callback) {
720 return clock->update_callback();
721 }
722 return 0;
723}
724#else
725#define change_clocksource() (0)
726#endif
727
728/**
729 * timeofday_is_continuous - check to see if timekeeping is free running
730 */
731int timekeeping_is_continuous(void)
732{
733 unsigned long seq;
734 int ret;
735
736 do {
737 seq = read_seqbegin(&xtime_lock);
738
739 ret = clock->is_continuous;
740
741 } while (read_seqretry(&xtime_lock, seq));
742
743 return ret;
744}
745
1da177e4 746/*
ad596171 747 * timekeeping_init - Initializes the clocksource and common timekeeping values
1da177e4 748 */
ad596171 749void __init timekeeping_init(void)
1da177e4 750{
ad596171
JS
751 unsigned long flags;
752
753 write_seqlock_irqsave(&xtime_lock, flags);
b0ee7556
RZ
754
755 ntp_clear();
756
a2752549
JS
757 clock = clocksource_get_next();
758 clocksource_calculate_interval(clock, tick_nsec);
19923c19 759 clock->cycle_last = clocksource_read(clock);
b0ee7556 760
ad596171
JS
761 write_sequnlock_irqrestore(&xtime_lock, flags);
762}
763
764
3e143475 765static int timekeeping_suspended;
2aae4a10 766/**
ad596171
JS
767 * timekeeping_resume - Resumes the generic timekeeping subsystem.
768 * @dev: unused
769 *
770 * This is for the generic clocksource timekeeping.
8ef38609 771 * xtime/wall_to_monotonic/jiffies/etc are
ad596171
JS
772 * still managed by arch specific suspend/resume code.
773 */
774static int timekeeping_resume(struct sys_device *dev)
775{
776 unsigned long flags;
777
778 write_seqlock_irqsave(&xtime_lock, flags);
779 /* restart the last cycle value */
19923c19 780 clock->cycle_last = clocksource_read(clock);
3e143475
JS
781 clock->error = 0;
782 timekeeping_suspended = 0;
783 write_sequnlock_irqrestore(&xtime_lock, flags);
784 return 0;
785}
786
787static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
788{
789 unsigned long flags;
790
791 write_seqlock_irqsave(&xtime_lock, flags);
792 timekeeping_suspended = 1;
ad596171
JS
793 write_sequnlock_irqrestore(&xtime_lock, flags);
794 return 0;
795}
796
797/* sysfs resume/suspend bits for timekeeping */
798static struct sysdev_class timekeeping_sysclass = {
799 .resume = timekeeping_resume,
3e143475 800 .suspend = timekeeping_suspend,
ad596171
JS
801 set_kset_name("timekeeping"),
802};
803
804static struct sys_device device_timer = {
805 .id = 0,
806 .cls = &timekeeping_sysclass,
807};
808
809static int __init timekeeping_init_device(void)
810{
811 int error = sysdev_class_register(&timekeeping_sysclass);
812 if (!error)
813 error = sysdev_register(&device_timer);
814 return error;
815}
816
817device_initcall(timekeeping_init_device);
818
19923c19 819/*
e154ff3d 820 * If the error is already larger, we look ahead even further
19923c19
RZ
821 * to compensate for late or lost adjustments.
822 */
e154ff3d 823static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, s64 *offset)
19923c19 824{
e154ff3d
RZ
825 s64 tick_error, i;
826 u32 look_ahead, adj;
827 s32 error2, mult;
19923c19
RZ
828
829 /*
e154ff3d
RZ
830 * Use the current error value to determine how much to look ahead.
831 * The larger the error the slower we adjust for it to avoid problems
832 * with losing too many ticks, otherwise we would overadjust and
833 * produce an even larger error. The smaller the adjustment the
834 * faster we try to adjust for it, as lost ticks can do less harm
835 * here. This is tuned so that an error of about 1 msec is adusted
836 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
19923c19 837 */
e154ff3d
RZ
838 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
839 error2 = abs(error2);
840 for (look_ahead = 0; error2 > 0; look_ahead++)
841 error2 >>= 2;
19923c19
RZ
842
843 /*
e154ff3d
RZ
844 * Now calculate the error in (1 << look_ahead) ticks, but first
845 * remove the single look ahead already included in the error.
19923c19 846 */
e154ff3d
RZ
847 tick_error = current_tick_length() >> (TICK_LENGTH_SHIFT - clock->shift + 1);
848 tick_error -= clock->xtime_interval >> 1;
849 error = ((error - tick_error) >> look_ahead) + tick_error;
850
851 /* Finally calculate the adjustment shift value. */
852 i = *interval;
853 mult = 1;
854 if (error < 0) {
855 error = -error;
856 *interval = -*interval;
857 *offset = -*offset;
858 mult = -1;
19923c19 859 }
e154ff3d
RZ
860 for (adj = 0; error > i; adj++)
861 error >>= 1;
19923c19
RZ
862
863 *interval <<= adj;
864 *offset <<= adj;
e154ff3d 865 return mult << adj;
19923c19
RZ
866}
867
868/*
869 * Adjust the multiplier to reduce the error value,
870 * this is optimized for the most common adjustments of -1,0,1,
871 * for other values we can do a bit more work.
872 */
873static void clocksource_adjust(struct clocksource *clock, s64 offset)
874{
875 s64 error, interval = clock->cycle_interval;
876 int adj;
877
878 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
879 if (error > interval) {
e154ff3d
RZ
880 error >>= 2;
881 if (likely(error <= interval))
882 adj = 1;
883 else
884 adj = clocksource_bigadjust(error, &interval, &offset);
19923c19 885 } else if (error < -interval) {
e154ff3d
RZ
886 error >>= 2;
887 if (likely(error >= -interval)) {
888 adj = -1;
889 interval = -interval;
890 offset = -offset;
891 } else
892 adj = clocksource_bigadjust(error, &interval, &offset);
19923c19
RZ
893 } else
894 return;
895
896 clock->mult += adj;
897 clock->xtime_interval += interval;
898 clock->xtime_nsec -= offset;
899 clock->error -= (interval - offset) << (TICK_LENGTH_SHIFT - clock->shift);
900}
901
2aae4a10 902/**
ad596171
JS
903 * update_wall_time - Uses the current clocksource to increment the wall time
904 *
905 * Called from the timer interrupt, must hold a write on xtime_lock.
906 */
907static void update_wall_time(void)
908{
19923c19 909 cycle_t offset;
ad596171 910
3e143475
JS
911 /* Make sure we're fully resumed: */
912 if (unlikely(timekeeping_suspended))
913 return;
5eb6d205 914
19923c19
RZ
915#ifdef CONFIG_GENERIC_TIME
916 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
917#else
918 offset = clock->cycle_interval;
919#endif
3e143475 920 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
ad596171
JS
921
922 /* normally this loop will run just once, however in the
923 * case of lost or late ticks, it will accumulate correctly.
924 */
19923c19 925 while (offset >= clock->cycle_interval) {
ad596171 926 /* accumulate one interval */
19923c19
RZ
927 clock->xtime_nsec += clock->xtime_interval;
928 clock->cycle_last += clock->cycle_interval;
929 offset -= clock->cycle_interval;
930
931 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
932 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
933 xtime.tv_sec++;
934 second_overflow();
935 }
ad596171 936
5eb6d205 937 /* interpolator bits */
19923c19 938 time_interpolator_update(clock->xtime_interval
5eb6d205 939 >> clock->shift);
5eb6d205
JS
940
941 /* accumulate error between NTP and clock interval */
19923c19
RZ
942 clock->error += current_tick_length();
943 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
944 }
5eb6d205 945
19923c19
RZ
946 /* correct the clock when NTP error is too big */
947 clocksource_adjust(clock, offset);
5eb6d205 948
5eb6d205 949 /* store full nanoseconds into xtime */
e154ff3d 950 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
19923c19 951 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
cf3c769b
JS
952
953 /* check to see if there is a new clocksource to use */
954 if (change_clocksource()) {
19923c19
RZ
955 clock->error = 0;
956 clock->xtime_nsec = 0;
a2752549 957 clocksource_calculate_interval(clock, tick_nsec);
cf3c769b 958 }
1da177e4
LT
959}
960
961/*
962 * Called from the timer interrupt handler to charge one tick to the current
963 * process. user_tick is 1 if the tick is user time, 0 for system.
964 */
965void update_process_times(int user_tick)
966{
967 struct task_struct *p = current;
968 int cpu = smp_processor_id();
969
970 /* Note: this timer irq context must be accounted for as well. */
971 if (user_tick)
972 account_user_time(p, jiffies_to_cputime(1));
973 else
974 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
975 run_local_timers();
976 if (rcu_pending(cpu))
977 rcu_check_callbacks(cpu, user_tick);
978 scheduler_tick();
979 run_posix_cpu_timers(p);
980}
981
982/*
983 * Nr of active tasks - counted in fixed-point numbers
984 */
985static unsigned long count_active_tasks(void)
986{
db1b1fef 987 return nr_active() * FIXED_1;
1da177e4
LT
988}
989
990/*
991 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
992 * imply that avenrun[] is the standard name for this kind of thing.
993 * Nothing else seems to be standardized: the fractional size etc
994 * all seem to differ on different machines.
995 *
996 * Requires xtime_lock to access.
997 */
998unsigned long avenrun[3];
999
1000EXPORT_SYMBOL(avenrun);
1001
1002/*
1003 * calc_load - given tick count, update the avenrun load estimates.
1004 * This is called while holding a write_lock on xtime_lock.
1005 */
1006static inline void calc_load(unsigned long ticks)
1007{
1008 unsigned long active_tasks; /* fixed-point */
1009 static int count = LOAD_FREQ;
1010
3171a030
AN
1011 active_tasks = count_active_tasks();
1012 for (count -= ticks; count < 0; count += LOAD_FREQ) {
1da177e4
LT
1013 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1014 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1015 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1016 }
1017}
1018
1da177e4
LT
1019/*
1020 * This read-write spinlock protects us from races in SMP while
1021 * playing with xtime and avenrun.
1022 */
1023#ifndef ARCH_HAVE_XTIME_LOCK
e4d91918 1024__cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
1da177e4
LT
1025
1026EXPORT_SYMBOL(xtime_lock);
1027#endif
1028
1029/*
1030 * This function runs timers and the timer-tq in bottom half context.
1031 */
1032static void run_timer_softirq(struct softirq_action *h)
1033{
a4a6198b 1034 tvec_base_t *base = __get_cpu_var(tvec_bases);
1da177e4 1035
c0a31329 1036 hrtimer_run_queues();
1da177e4
LT
1037 if (time_after_eq(jiffies, base->timer_jiffies))
1038 __run_timers(base);
1039}
1040
1041/*
1042 * Called by the local, per-CPU timer interrupt on SMP.
1043 */
1044void run_local_timers(void)
1045{
1046 raise_softirq(TIMER_SOFTIRQ);
6687a97d 1047 softlockup_tick();
1da177e4
LT
1048}
1049
1050/*
1051 * Called by the timer interrupt. xtime_lock must already be taken
1052 * by the timer IRQ!
1053 */
3171a030 1054static inline void update_times(unsigned long ticks)
1da177e4 1055{
ad596171 1056 update_wall_time();
1da177e4
LT
1057 calc_load(ticks);
1058}
1059
1060/*
1061 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1062 * without sampling the sequence number in xtime_lock.
1063 * jiffies is defined in the linker script...
1064 */
1065
3171a030 1066void do_timer(unsigned long ticks)
1da177e4 1067{
3171a030
AN
1068 jiffies_64 += ticks;
1069 update_times(ticks);
1da177e4
LT
1070}
1071
1072#ifdef __ARCH_WANT_SYS_ALARM
1073
1074/*
1075 * For backwards compatibility? This can be done in libc so Alpha
1076 * and all newer ports shouldn't need it.
1077 */
1078asmlinkage unsigned long sys_alarm(unsigned int seconds)
1079{
c08b8a49 1080 return alarm_setitimer(seconds);
1da177e4
LT
1081}
1082
1083#endif
1084
1085#ifndef __alpha__
1086
1087/*
1088 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1089 * should be moved into arch/i386 instead?
1090 */
1091
1092/**
1093 * sys_getpid - return the thread group id of the current process
1094 *
1095 * Note, despite the name, this returns the tgid not the pid. The tgid and
1096 * the pid are identical unless CLONE_THREAD was specified on clone() in
1097 * which case the tgid is the same in all threads of the same group.
1098 *
1099 * This is SMP safe as current->tgid does not change.
1100 */
1101asmlinkage long sys_getpid(void)
1102{
1103 return current->tgid;
1104}
1105
1106/*
6997a6fa
KK
1107 * Accessing ->real_parent is not SMP-safe, it could
1108 * change from under us. However, we can use a stale
1109 * value of ->real_parent under rcu_read_lock(), see
1110 * release_task()->call_rcu(delayed_put_task_struct).
1da177e4
LT
1111 */
1112asmlinkage long sys_getppid(void)
1113{
1114 int pid;
1da177e4 1115
6997a6fa
KK
1116 rcu_read_lock();
1117 pid = rcu_dereference(current->real_parent)->tgid;
1118 rcu_read_unlock();
1da177e4 1119
1da177e4
LT
1120 return pid;
1121}
1122
1123asmlinkage long sys_getuid(void)
1124{
1125 /* Only we change this so SMP safe */
1126 return current->uid;
1127}
1128
1129asmlinkage long sys_geteuid(void)
1130{
1131 /* Only we change this so SMP safe */
1132 return current->euid;
1133}
1134
1135asmlinkage long sys_getgid(void)
1136{
1137 /* Only we change this so SMP safe */
1138 return current->gid;
1139}
1140
1141asmlinkage long sys_getegid(void)
1142{
1143 /* Only we change this so SMP safe */
1144 return current->egid;
1145}
1146
1147#endif
1148
1149static void process_timeout(unsigned long __data)
1150{
36c8b586 1151 wake_up_process((struct task_struct *)__data);
1da177e4
LT
1152}
1153
1154/**
1155 * schedule_timeout - sleep until timeout
1156 * @timeout: timeout value in jiffies
1157 *
1158 * Make the current task sleep until @timeout jiffies have
1159 * elapsed. The routine will return immediately unless
1160 * the current task state has been set (see set_current_state()).
1161 *
1162 * You can set the task state as follows -
1163 *
1164 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1165 * pass before the routine returns. The routine will return 0
1166 *
1167 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1168 * delivered to the current task. In this case the remaining time
1169 * in jiffies will be returned, or 0 if the timer expired in time
1170 *
1171 * The current task state is guaranteed to be TASK_RUNNING when this
1172 * routine returns.
1173 *
1174 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1175 * the CPU away without a bound on the timeout. In this case the return
1176 * value will be %MAX_SCHEDULE_TIMEOUT.
1177 *
1178 * In all cases the return value is guaranteed to be non-negative.
1179 */
1180fastcall signed long __sched schedule_timeout(signed long timeout)
1181{
1182 struct timer_list timer;
1183 unsigned long expire;
1184
1185 switch (timeout)
1186 {
1187 case MAX_SCHEDULE_TIMEOUT:
1188 /*
1189 * These two special cases are useful to be comfortable
1190 * in the caller. Nothing more. We could take
1191 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1192 * but I' d like to return a valid offset (>=0) to allow
1193 * the caller to do everything it want with the retval.
1194 */
1195 schedule();
1196 goto out;
1197 default:
1198 /*
1199 * Another bit of PARANOID. Note that the retval will be
1200 * 0 since no piece of kernel is supposed to do a check
1201 * for a negative retval of schedule_timeout() (since it
1202 * should never happens anyway). You just have the printk()
1203 * that will tell you if something is gone wrong and where.
1204 */
1205 if (timeout < 0)
1206 {
1207 printk(KERN_ERR "schedule_timeout: wrong timeout "
a5a0d52c
AM
1208 "value %lx from %p\n", timeout,
1209 __builtin_return_address(0));
1da177e4
LT
1210 current->state = TASK_RUNNING;
1211 goto out;
1212 }
1213 }
1214
1215 expire = timeout + jiffies;
1216
a8db2db1
ON
1217 setup_timer(&timer, process_timeout, (unsigned long)current);
1218 __mod_timer(&timer, expire);
1da177e4
LT
1219 schedule();
1220 del_singleshot_timer_sync(&timer);
1221
1222 timeout = expire - jiffies;
1223
1224 out:
1225 return timeout < 0 ? 0 : timeout;
1226}
1da177e4
LT
1227EXPORT_SYMBOL(schedule_timeout);
1228
8a1c1757
AM
1229/*
1230 * We can use __set_current_state() here because schedule_timeout() calls
1231 * schedule() unconditionally.
1232 */
64ed93a2
NA
1233signed long __sched schedule_timeout_interruptible(signed long timeout)
1234{
a5a0d52c
AM
1235 __set_current_state(TASK_INTERRUPTIBLE);
1236 return schedule_timeout(timeout);
64ed93a2
NA
1237}
1238EXPORT_SYMBOL(schedule_timeout_interruptible);
1239
1240signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1241{
a5a0d52c
AM
1242 __set_current_state(TASK_UNINTERRUPTIBLE);
1243 return schedule_timeout(timeout);
64ed93a2
NA
1244}
1245EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1246
1da177e4
LT
1247/* Thread ID - the internal kernel "pid" */
1248asmlinkage long sys_gettid(void)
1249{
1250 return current->pid;
1251}
1252
2aae4a10 1253/**
1da177e4 1254 * sys_sysinfo - fill in sysinfo struct
2aae4a10 1255 * @info: pointer to buffer to fill
1da177e4
LT
1256 */
1257asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1258{
1259 struct sysinfo val;
1260 unsigned long mem_total, sav_total;
1261 unsigned int mem_unit, bitcount;
1262 unsigned long seq;
1263
1264 memset((char *)&val, 0, sizeof(struct sysinfo));
1265
1266 do {
1267 struct timespec tp;
1268 seq = read_seqbegin(&xtime_lock);
1269
1270 /*
1271 * This is annoying. The below is the same thing
1272 * posix_get_clock_monotonic() does, but it wants to
1273 * take the lock which we want to cover the loads stuff
1274 * too.
1275 */
1276
1277 getnstimeofday(&tp);
1278 tp.tv_sec += wall_to_monotonic.tv_sec;
1279 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1280 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1281 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1282 tp.tv_sec++;
1283 }
1284 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1285
1286 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1287 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1288 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1289
1290 val.procs = nr_threads;
1291 } while (read_seqretry(&xtime_lock, seq));
1292
1293 si_meminfo(&val);
1294 si_swapinfo(&val);
1295
1296 /*
1297 * If the sum of all the available memory (i.e. ram + swap)
1298 * is less than can be stored in a 32 bit unsigned long then
1299 * we can be binary compatible with 2.2.x kernels. If not,
1300 * well, in that case 2.2.x was broken anyways...
1301 *
1302 * -Erik Andersen <andersee@debian.org>
1303 */
1304
1305 mem_total = val.totalram + val.totalswap;
1306 if (mem_total < val.totalram || mem_total < val.totalswap)
1307 goto out;
1308 bitcount = 0;
1309 mem_unit = val.mem_unit;
1310 while (mem_unit > 1) {
1311 bitcount++;
1312 mem_unit >>= 1;
1313 sav_total = mem_total;
1314 mem_total <<= 1;
1315 if (mem_total < sav_total)
1316 goto out;
1317 }
1318
1319 /*
1320 * If mem_total did not overflow, multiply all memory values by
1321 * val.mem_unit and set it to 1. This leaves things compatible
1322 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1323 * kernels...
1324 */
1325
1326 val.mem_unit = 1;
1327 val.totalram <<= bitcount;
1328 val.freeram <<= bitcount;
1329 val.sharedram <<= bitcount;
1330 val.bufferram <<= bitcount;
1331 val.totalswap <<= bitcount;
1332 val.freeswap <<= bitcount;
1333 val.totalhigh <<= bitcount;
1334 val.freehigh <<= bitcount;
1335
1336 out:
1337 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1338 return -EFAULT;
1339
1340 return 0;
1341}
1342
d730e882
IM
1343/*
1344 * lockdep: we want to track each per-CPU base as a separate lock-class,
1345 * but timer-bases are kmalloc()-ed, so we need to attach separate
1346 * keys to them:
1347 */
1348static struct lock_class_key base_lock_keys[NR_CPUS];
1349
a4a6198b 1350static int __devinit init_timers_cpu(int cpu)
1da177e4
LT
1351{
1352 int j;
1353 tvec_base_t *base;
ba6edfcd 1354 static char __devinitdata tvec_base_done[NR_CPUS];
55c888d6 1355
ba6edfcd 1356 if (!tvec_base_done[cpu]) {
a4a6198b
JB
1357 static char boot_done;
1358
a4a6198b 1359 if (boot_done) {
ba6edfcd
AM
1360 /*
1361 * The APs use this path later in boot
1362 */
a4a6198b
JB
1363 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1364 cpu_to_node(cpu));
1365 if (!base)
1366 return -ENOMEM;
1367 memset(base, 0, sizeof(*base));
ba6edfcd 1368 per_cpu(tvec_bases, cpu) = base;
a4a6198b 1369 } else {
ba6edfcd
AM
1370 /*
1371 * This is for the boot CPU - we use compile-time
1372 * static initialisation because per-cpu memory isn't
1373 * ready yet and because the memory allocators are not
1374 * initialised either.
1375 */
a4a6198b 1376 boot_done = 1;
ba6edfcd 1377 base = &boot_tvec_bases;
a4a6198b 1378 }
ba6edfcd
AM
1379 tvec_base_done[cpu] = 1;
1380 } else {
1381 base = per_cpu(tvec_bases, cpu);
a4a6198b 1382 }
ba6edfcd 1383
3691c519 1384 spin_lock_init(&base->lock);
d730e882
IM
1385 lockdep_set_class(&base->lock, base_lock_keys + cpu);
1386
1da177e4
LT
1387 for (j = 0; j < TVN_SIZE; j++) {
1388 INIT_LIST_HEAD(base->tv5.vec + j);
1389 INIT_LIST_HEAD(base->tv4.vec + j);
1390 INIT_LIST_HEAD(base->tv3.vec + j);
1391 INIT_LIST_HEAD(base->tv2.vec + j);
1392 }
1393 for (j = 0; j < TVR_SIZE; j++)
1394 INIT_LIST_HEAD(base->tv1.vec + j);
1395
1396 base->timer_jiffies = jiffies;
a4a6198b 1397 return 0;
1da177e4
LT
1398}
1399
1400#ifdef CONFIG_HOTPLUG_CPU
55c888d6 1401static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1da177e4
LT
1402{
1403 struct timer_list *timer;
1404
1405 while (!list_empty(head)) {
1406 timer = list_entry(head->next, struct timer_list, entry);
55c888d6 1407 detach_timer(timer, 0);
3691c519 1408 timer->base = new_base;
1da177e4 1409 internal_add_timer(new_base, timer);
1da177e4 1410 }
1da177e4
LT
1411}
1412
1413static void __devinit migrate_timers(int cpu)
1414{
1415 tvec_base_t *old_base;
1416 tvec_base_t *new_base;
1417 int i;
1418
1419 BUG_ON(cpu_online(cpu));
a4a6198b
JB
1420 old_base = per_cpu(tvec_bases, cpu);
1421 new_base = get_cpu_var(tvec_bases);
1da177e4
LT
1422
1423 local_irq_disable();
3691c519
ON
1424 spin_lock(&new_base->lock);
1425 spin_lock(&old_base->lock);
1426
1427 BUG_ON(old_base->running_timer);
1da177e4 1428
1da177e4 1429 for (i = 0; i < TVR_SIZE; i++)
55c888d6
ON
1430 migrate_timer_list(new_base, old_base->tv1.vec + i);
1431 for (i = 0; i < TVN_SIZE; i++) {
1432 migrate_timer_list(new_base, old_base->tv2.vec + i);
1433 migrate_timer_list(new_base, old_base->tv3.vec + i);
1434 migrate_timer_list(new_base, old_base->tv4.vec + i);
1435 migrate_timer_list(new_base, old_base->tv5.vec + i);
1436 }
1437
3691c519
ON
1438 spin_unlock(&old_base->lock);
1439 spin_unlock(&new_base->lock);
1da177e4
LT
1440 local_irq_enable();
1441 put_cpu_var(tvec_bases);
1da177e4
LT
1442}
1443#endif /* CONFIG_HOTPLUG_CPU */
1444
8c78f307 1445static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1da177e4
LT
1446 unsigned long action, void *hcpu)
1447{
1448 long cpu = (long)hcpu;
1449 switch(action) {
1450 case CPU_UP_PREPARE:
a4a6198b
JB
1451 if (init_timers_cpu(cpu) < 0)
1452 return NOTIFY_BAD;
1da177e4
LT
1453 break;
1454#ifdef CONFIG_HOTPLUG_CPU
1455 case CPU_DEAD:
1456 migrate_timers(cpu);
1457 break;
1458#endif
1459 default:
1460 break;
1461 }
1462 return NOTIFY_OK;
1463}
1464
8c78f307 1465static struct notifier_block __cpuinitdata timers_nb = {
1da177e4
LT
1466 .notifier_call = timer_cpu_notify,
1467};
1468
1469
1470void __init init_timers(void)
1471{
07dccf33 1472 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1da177e4 1473 (void *)(long)smp_processor_id());
07dccf33
AM
1474
1475 BUG_ON(err == NOTIFY_BAD);
1da177e4
LT
1476 register_cpu_notifier(&timers_nb);
1477 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1478}
1479
1480#ifdef CONFIG_TIME_INTERPOLATION
1481
67890d70
CL
1482struct time_interpolator *time_interpolator __read_mostly;
1483static struct time_interpolator *time_interpolator_list __read_mostly;
1da177e4
LT
1484static DEFINE_SPINLOCK(time_interpolator_lock);
1485
1486static inline u64 time_interpolator_get_cycles(unsigned int src)
1487{
1488 unsigned long (*x)(void);
1489
1490 switch (src)
1491 {
1492 case TIME_SOURCE_FUNCTION:
1493 x = time_interpolator->addr;
1494 return x();
1495
1496 case TIME_SOURCE_MMIO64 :
685db65e 1497 return readq_relaxed((void __iomem *)time_interpolator->addr);
1da177e4
LT
1498
1499 case TIME_SOURCE_MMIO32 :
685db65e 1500 return readl_relaxed((void __iomem *)time_interpolator->addr);
1da177e4
LT
1501
1502 default: return get_cycles();
1503 }
1504}
1505
486d46ae 1506static inline u64 time_interpolator_get_counter(int writelock)
1da177e4
LT
1507{
1508 unsigned int src = time_interpolator->source;
1509
1510 if (time_interpolator->jitter)
1511 {
1512 u64 lcycle;
1513 u64 now;
1514
1515 do {
1516 lcycle = time_interpolator->last_cycle;
1517 now = time_interpolator_get_cycles(src);
1518 if (lcycle && time_after(lcycle, now))
1519 return lcycle;
486d46ae
AW
1520
1521 /* When holding the xtime write lock, there's no need
1522 * to add the overhead of the cmpxchg. Readers are
1523 * force to retry until the write lock is released.
1524 */
1525 if (writelock) {
1526 time_interpolator->last_cycle = now;
1527 return now;
1528 }
1da177e4
LT
1529 /* Keep track of the last timer value returned. The use of cmpxchg here
1530 * will cause contention in an SMP environment.
1531 */
1532 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1533 return now;
1534 }
1535 else
1536 return time_interpolator_get_cycles(src);
1537}
1538
1539void time_interpolator_reset(void)
1540{
1541 time_interpolator->offset = 0;
486d46ae 1542 time_interpolator->last_counter = time_interpolator_get_counter(1);
1da177e4
LT
1543}
1544
1545#define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1546
1547unsigned long time_interpolator_get_offset(void)
1548{
1549 /* If we do not have a time interpolator set up then just return zero */
1550 if (!time_interpolator)
1551 return 0;
1552
1553 return time_interpolator->offset +
486d46ae 1554 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1da177e4
LT
1555}
1556
1557#define INTERPOLATOR_ADJUST 65536
1558#define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1559
4c7ee8de 1560void time_interpolator_update(long delta_nsec)
1da177e4
LT
1561{
1562 u64 counter;
1563 unsigned long offset;
1564
1565 /* If there is no time interpolator set up then do nothing */
1566 if (!time_interpolator)
1567 return;
1568
a5a0d52c
AM
1569 /*
1570 * The interpolator compensates for late ticks by accumulating the late
1571 * time in time_interpolator->offset. A tick earlier than expected will
1572 * lead to a reset of the offset and a corresponding jump of the clock
1573 * forward. Again this only works if the interpolator clock is running
1574 * slightly slower than the regular clock and the tuning logic insures
1575 * that.
1576 */
1da177e4 1577
486d46ae 1578 counter = time_interpolator_get_counter(1);
a5a0d52c
AM
1579 offset = time_interpolator->offset +
1580 GET_TI_NSECS(counter, time_interpolator);
1da177e4
LT
1581
1582 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1583 time_interpolator->offset = offset - delta_nsec;
1584 else {
1585 time_interpolator->skips++;
1586 time_interpolator->ns_skipped += delta_nsec - offset;
1587 time_interpolator->offset = 0;
1588 }
1589 time_interpolator->last_counter = counter;
1590
1591 /* Tuning logic for time interpolator invoked every minute or so.
1592 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1593 * Increase interpolator clock speed if we skip too much time.
1594 */
1595 if (jiffies % INTERPOLATOR_ADJUST == 0)
1596 {
b20367a6 1597 if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
1da177e4
LT
1598 time_interpolator->nsec_per_cyc--;
1599 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1600 time_interpolator->nsec_per_cyc++;
1601 time_interpolator->skips = 0;
1602 time_interpolator->ns_skipped = 0;
1603 }
1604}
1605
1606static inline int
1607is_better_time_interpolator(struct time_interpolator *new)
1608{
1609 if (!time_interpolator)
1610 return 1;
1611 return new->frequency > 2*time_interpolator->frequency ||
1612 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1613}
1614
1615void
1616register_time_interpolator(struct time_interpolator *ti)
1617{
1618 unsigned long flags;
1619
1620 /* Sanity check */
9f31252c 1621 BUG_ON(ti->frequency == 0 || ti->mask == 0);
1da177e4
LT
1622
1623 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1624 spin_lock(&time_interpolator_lock);
1625 write_seqlock_irqsave(&xtime_lock, flags);
1626 if (is_better_time_interpolator(ti)) {
1627 time_interpolator = ti;
1628 time_interpolator_reset();
1629 }
1630 write_sequnlock_irqrestore(&xtime_lock, flags);
1631
1632 ti->next = time_interpolator_list;
1633 time_interpolator_list = ti;
1634 spin_unlock(&time_interpolator_lock);
1635}
1636
1637void
1638unregister_time_interpolator(struct time_interpolator *ti)
1639{
1640 struct time_interpolator *curr, **prev;
1641 unsigned long flags;
1642
1643 spin_lock(&time_interpolator_lock);
1644 prev = &time_interpolator_list;
1645 for (curr = *prev; curr; curr = curr->next) {
1646 if (curr == ti) {
1647 *prev = curr->next;
1648 break;
1649 }
1650 prev = &curr->next;
1651 }
1652
1653 write_seqlock_irqsave(&xtime_lock, flags);
1654 if (ti == time_interpolator) {
1655 /* we lost the best time-interpolator: */
1656 time_interpolator = NULL;
1657 /* find the next-best interpolator */
1658 for (curr = time_interpolator_list; curr; curr = curr->next)
1659 if (is_better_time_interpolator(curr))
1660 time_interpolator = curr;
1661 time_interpolator_reset();
1662 }
1663 write_sequnlock_irqrestore(&xtime_lock, flags);
1664 spin_unlock(&time_interpolator_lock);
1665}
1666#endif /* CONFIG_TIME_INTERPOLATION */
1667
1668/**
1669 * msleep - sleep safely even with waitqueue interruptions
1670 * @msecs: Time in milliseconds to sleep for
1671 */
1672void msleep(unsigned int msecs)
1673{
1674 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1675
75bcc8c5
NA
1676 while (timeout)
1677 timeout = schedule_timeout_uninterruptible(timeout);
1da177e4
LT
1678}
1679
1680EXPORT_SYMBOL(msleep);
1681
1682/**
96ec3efd 1683 * msleep_interruptible - sleep waiting for signals
1da177e4
LT
1684 * @msecs: Time in milliseconds to sleep for
1685 */
1686unsigned long msleep_interruptible(unsigned int msecs)
1687{
1688 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1689
75bcc8c5
NA
1690 while (timeout && !signal_pending(current))
1691 timeout = schedule_timeout_interruptible(timeout);
1da177e4
LT
1692 return jiffies_to_msecs(timeout);
1693}
1694
1695EXPORT_SYMBOL(msleep_interruptible);