Commit | Line | Data |
---|---|---|
c0a31329 TG |
1 | /* |
2 | * linux/kernel/hrtimer.c | |
3 | * | |
3c8aa39d | 4 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> |
79bf2bb3 | 5 | * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar |
54cdfdb4 | 6 | * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner |
c0a31329 TG |
7 | * |
8 | * High-resolution kernel timers | |
9 | * | |
10 | * In contrast to the low-resolution timeout API implemented in | |
11 | * kernel/timer.c, hrtimers provide finer resolution and accuracy | |
12 | * depending on system configuration and capabilities. | |
13 | * | |
14 | * These timers are currently used for: | |
15 | * - itimers | |
16 | * - POSIX timers | |
17 | * - nanosleep | |
18 | * - precise in-kernel timing | |
19 | * | |
20 | * Started by: Thomas Gleixner and Ingo Molnar | |
21 | * | |
22 | * Credits: | |
23 | * based on kernel/timer.c | |
24 | * | |
66188fae TG |
25 | * Help, testing, suggestions, bugfixes, improvements were |
26 | * provided by: | |
27 | * | |
28 | * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel | |
29 | * et. al. | |
30 | * | |
c0a31329 TG |
31 | * For licencing details see kernel-base/COPYING |
32 | */ | |
33 | ||
34 | #include <linux/cpu.h> | |
9984de1a | 35 | #include <linux/export.h> |
c0a31329 TG |
36 | #include <linux/percpu.h> |
37 | #include <linux/hrtimer.h> | |
38 | #include <linux/notifier.h> | |
39 | #include <linux/syscalls.h> | |
54cdfdb4 | 40 | #include <linux/kallsyms.h> |
c0a31329 | 41 | #include <linux/interrupt.h> |
79bf2bb3 | 42 | #include <linux/tick.h> |
54cdfdb4 TG |
43 | #include <linux/seq_file.h> |
44 | #include <linux/err.h> | |
237fc6e7 | 45 | #include <linux/debugobjects.h> |
eea08f32 | 46 | #include <linux/sched.h> |
cf4aebc2 | 47 | #include <linux/sched/sysctl.h> |
8bd75c77 | 48 | #include <linux/sched/rt.h> |
eea08f32 | 49 | #include <linux/timer.h> |
6fa3eb70 | 50 | #include <linux/freezer.h> |
c0a31329 TG |
51 | |
52 | #include <asm/uaccess.h> | |
53 | ||
c6a2a177 XG |
54 | #include <trace/events/timer.h> |
55 | ||
6fa3eb70 S |
56 | #include <linux/mt_sched_mon.h> |
57 | ||
58 | //#define MTK_HRTIME_DEBUG /*MTK debug func*/ | |
c0a31329 TG |
59 | /* |
60 | * The timer bases: | |
7978672c | 61 | * |
e06383db JS |
62 | * There are more clockids then hrtimer bases. Thus, we index |
63 | * into the timer bases by the hrtimer_base_type enum. When trying | |
64 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
65 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 66 | */ |
54cdfdb4 | 67 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 68 | { |
3c8aa39d | 69 | |
84cc8fd2 | 70 | .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock), |
3c8aa39d | 71 | .clock_base = |
c0a31329 | 72 | { |
3c8aa39d | 73 | { |
ab8177bc TG |
74 | .index = HRTIMER_BASE_MONOTONIC, |
75 | .clockid = CLOCK_MONOTONIC, | |
3c8aa39d | 76 | .get_time = &ktime_get, |
54cdfdb4 | 77 | .resolution = KTIME_LOW_RES, |
3c8aa39d | 78 | }, |
68fa61c0 TG |
79 | { |
80 | .index = HRTIMER_BASE_REALTIME, | |
81 | .clockid = CLOCK_REALTIME, | |
82 | .get_time = &ktime_get_real, | |
83 | .resolution = KTIME_LOW_RES, | |
84 | }, | |
70a08cca | 85 | { |
ab8177bc TG |
86 | .index = HRTIMER_BASE_BOOTTIME, |
87 | .clockid = CLOCK_BOOTTIME, | |
70a08cca JS |
88 | .get_time = &ktime_get_boottime, |
89 | .resolution = KTIME_LOW_RES, | |
90 | }, | |
90adda98 JS |
91 | { |
92 | .index = HRTIMER_BASE_TAI, | |
93 | .clockid = CLOCK_TAI, | |
94 | .get_time = &ktime_get_clocktai, | |
95 | .resolution = KTIME_LOW_RES, | |
96 | }, | |
3c8aa39d | 97 | } |
c0a31329 TG |
98 | }; |
99 | ||
942c3c5c | 100 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
ce31332d TG |
101 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
102 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
103 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, | |
90adda98 | 104 | [CLOCK_TAI] = HRTIMER_BASE_TAI, |
ce31332d | 105 | }; |
e06383db JS |
106 | |
107 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) | |
108 | { | |
109 | return hrtimer_clock_to_base_table[clock_id]; | |
110 | } | |
111 | ||
112 | ||
92127c7a TG |
113 | /* |
114 | * Get the coarse grained time at the softirq based on xtime and | |
115 | * wall_to_monotonic. | |
116 | */ | |
3c8aa39d | 117 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a | 118 | { |
70a08cca | 119 | ktime_t xtim, mono, boot; |
314ac371 | 120 | struct timespec xts, tom, slp; |
90adda98 | 121 | s32 tai_offset; |
92127c7a | 122 | |
314ac371 | 123 | get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp); |
90adda98 | 124 | tai_offset = timekeeping_get_tai_offset(); |
92127c7a | 125 | |
f4304ab2 | 126 | xtim = timespec_to_ktime(xts); |
70a08cca JS |
127 | mono = ktime_add(xtim, timespec_to_ktime(tom)); |
128 | boot = ktime_add(mono, timespec_to_ktime(slp)); | |
e06383db | 129 | base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim; |
70a08cca JS |
130 | base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono; |
131 | base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot; | |
90adda98 JS |
132 | base->clock_base[HRTIMER_BASE_TAI].softirq_time = |
133 | ktime_add(xtim, ktime_set(tai_offset, 0)); | |
92127c7a TG |
134 | } |
135 | ||
c0a31329 TG |
136 | /* |
137 | * Functions and macros which are different for UP/SMP systems are kept in a | |
138 | * single place | |
139 | */ | |
140 | #ifdef CONFIG_SMP | |
141 | ||
c0a31329 TG |
142 | /* |
143 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
144 | * means that all timers which are tied to this base via timer->base are | |
145 | * locked, and the base itself is locked too. | |
146 | * | |
147 | * So __run_timers/migrate_timers can safely modify all timers which could | |
148 | * be found on the lists/queues. | |
149 | * | |
150 | * When the timer's base is locked, and the timer removed from list, it is | |
151 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
152 | * locked. | |
153 | */ | |
3c8aa39d TG |
154 | static |
155 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
156 | unsigned long *flags) | |
c0a31329 | 157 | { |
3c8aa39d | 158 | struct hrtimer_clock_base *base; |
c0a31329 TG |
159 | |
160 | for (;;) { | |
161 | base = timer->base; | |
162 | if (likely(base != NULL)) { | |
ecb49d1a | 163 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
164 | if (likely(base == timer->base)) |
165 | return base; | |
166 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 167 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
168 | } |
169 | cpu_relax(); | |
170 | } | |
171 | } | |
172 | ||
6ff7041d TG |
173 | |
174 | /* | |
175 | * Get the preferred target CPU for NOHZ | |
176 | */ | |
177 | static int hrtimer_get_target(int this_cpu, int pinned) | |
178 | { | |
3451d024 | 179 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
180 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) |
181 | return get_nohz_timer_target(); | |
6ff7041d TG |
182 | #endif |
183 | return this_cpu; | |
184 | } | |
185 | ||
186 | /* | |
187 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
188 | * before the next event on the target cpu because we cannot reprogram | |
189 | * the target cpu hardware and we would cause it to fire late. | |
190 | * | |
191 | * Called with cpu_base->lock of target cpu held. | |
192 | */ | |
193 | static int | |
194 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
195 | { | |
196 | #ifdef CONFIG_HIGH_RES_TIMERS | |
197 | ktime_t expires; | |
198 | ||
199 | if (!new_base->cpu_base->hres_active) | |
200 | return 0; | |
201 | ||
202 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
203 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
204 | #else | |
205 | return 0; | |
206 | #endif | |
207 | } | |
208 | ||
c0a31329 TG |
209 | /* |
210 | * Switch the timer base to the current CPU when possible. | |
211 | */ | |
3c8aa39d | 212 | static inline struct hrtimer_clock_base * |
597d0275 AB |
213 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
214 | int pinned) | |
c0a31329 | 215 | { |
3c8aa39d TG |
216 | struct hrtimer_clock_base *new_base; |
217 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d TG |
218 | int this_cpu = smp_processor_id(); |
219 | int cpu = hrtimer_get_target(this_cpu, pinned); | |
ab8177bc | 220 | int basenum = base->index; |
c0a31329 | 221 | |
eea08f32 AB |
222 | again: |
223 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
e06383db | 224 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
225 | |
226 | if (base != new_base) { | |
227 | /* | |
6ff7041d | 228 | * We are trying to move timer to new_base. |
c0a31329 TG |
229 | * However we can't change timer's base while it is running, |
230 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
231 | * the event source in the high resolution case. The softirq | |
232 | * code will take care of this when the timer function has | |
233 | * completed. There is no conflict as we hold the lock until | |
234 | * the timer is enqueued. | |
235 | */ | |
54cdfdb4 | 236 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
237 | return base; |
238 | ||
239 | /* See the comment in lock_timer_base() */ | |
240 | timer->base = NULL; | |
ecb49d1a TG |
241 | raw_spin_unlock(&base->cpu_base->lock); |
242 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 243 | |
6ff7041d TG |
244 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
245 | cpu = this_cpu; | |
ecb49d1a TG |
246 | raw_spin_unlock(&new_base->cpu_base->lock); |
247 | raw_spin_lock(&base->cpu_base->lock); | |
6ff7041d TG |
248 | timer->base = base; |
249 | goto again; | |
eea08f32 | 250 | } |
c0a31329 | 251 | timer->base = new_base; |
7f7bb020 LM |
252 | } else { |
253 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { | |
254 | cpu = this_cpu; | |
255 | goto again; | |
256 | } | |
c0a31329 TG |
257 | } |
258 | return new_base; | |
259 | } | |
260 | ||
261 | #else /* CONFIG_SMP */ | |
262 | ||
3c8aa39d | 263 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
264 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
265 | { | |
3c8aa39d | 266 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 267 | |
ecb49d1a | 268 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
269 | |
270 | return base; | |
271 | } | |
272 | ||
eea08f32 | 273 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
274 | |
275 | #endif /* !CONFIG_SMP */ | |
276 | ||
277 | /* | |
278 | * Functions for the union type storage format of ktime_t which are | |
279 | * too large for inlining: | |
280 | */ | |
281 | #if BITS_PER_LONG < 64 | |
282 | # ifndef CONFIG_KTIME_SCALAR | |
283 | /** | |
284 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
285 | * @kt: addend |
286 | * @nsec: the scalar nsec value to add | |
287 | * | |
288 | * Returns the sum of kt and nsec in ktime_t format | |
289 | */ | |
290 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
291 | { | |
292 | ktime_t tmp; | |
293 | ||
294 | if (likely(nsec < NSEC_PER_SEC)) { | |
295 | tmp.tv64 = nsec; | |
296 | } else { | |
297 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
298 | ||
51fd36f3 DE |
299 | /* Make sure nsec fits into long */ |
300 | if (unlikely(nsec > KTIME_SEC_MAX)) | |
301 | return (ktime_t){ .tv64 = KTIME_MAX }; | |
302 | ||
c0a31329 TG |
303 | tmp = ktime_set((long)nsec, rem); |
304 | } | |
305 | ||
306 | return ktime_add(kt, tmp); | |
307 | } | |
b8b8fd2d DH |
308 | |
309 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
310 | |
311 | /** | |
312 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
313 | * @kt: minuend | |
314 | * @nsec: the scalar nsec value to subtract | |
315 | * | |
316 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
317 | */ | |
318 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
319 | { | |
320 | ktime_t tmp; | |
321 | ||
322 | if (likely(nsec < NSEC_PER_SEC)) { | |
323 | tmp.tv64 = nsec; | |
324 | } else { | |
325 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
326 | ||
327 | tmp = ktime_set((long)nsec, rem); | |
328 | } | |
329 | ||
330 | return ktime_sub(kt, tmp); | |
331 | } | |
332 | ||
333 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
334 | # endif /* !CONFIG_KTIME_SCALAR */ |
335 | ||
336 | /* | |
337 | * Divide a ktime value by a nanosecond value | |
338 | */ | |
4d672e7a | 339 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 340 | { |
900cfa46 | 341 | u64 dclc; |
c0a31329 TG |
342 | int sft = 0; |
343 | ||
900cfa46 | 344 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
345 | /* Make sure the divisor is less than 2^32: */ |
346 | while (div >> 32) { | |
347 | sft++; | |
348 | div >>= 1; | |
349 | } | |
350 | dclc >>= sft; | |
351 | do_div(dclc, (unsigned long) div); | |
352 | ||
4d672e7a | 353 | return dclc; |
c0a31329 | 354 | } |
c0a31329 TG |
355 | #endif /* BITS_PER_LONG >= 64 */ |
356 | ||
5a7780e7 TG |
357 | /* |
358 | * Add two ktime values and do a safety check for overflow: | |
359 | */ | |
360 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
361 | { | |
362 | ktime_t res = ktime_add(lhs, rhs); | |
363 | ||
364 | /* | |
365 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
366 | * return to user space in a timespec: | |
367 | */ | |
368 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
369 | res = ktime_set(KTIME_SEC_MAX, 0); | |
370 | ||
371 | return res; | |
372 | } | |
373 | ||
8daa21e6 AB |
374 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
375 | ||
237fc6e7 TG |
376 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
377 | ||
378 | static struct debug_obj_descr hrtimer_debug_descr; | |
379 | ||
99777288 SG |
380 | static void *hrtimer_debug_hint(void *addr) |
381 | { | |
382 | return ((struct hrtimer *) addr)->function; | |
383 | } | |
384 | ||
237fc6e7 TG |
385 | /* |
386 | * fixup_init is called when: | |
387 | * - an active object is initialized | |
388 | */ | |
389 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
390 | { | |
391 | struct hrtimer *timer = addr; | |
392 | ||
393 | switch (state) { | |
394 | case ODEBUG_STATE_ACTIVE: | |
395 | hrtimer_cancel(timer); | |
396 | debug_object_init(timer, &hrtimer_debug_descr); | |
397 | return 1; | |
398 | default: | |
399 | return 0; | |
400 | } | |
401 | } | |
402 | ||
403 | /* | |
404 | * fixup_activate is called when: | |
405 | * - an active object is activated | |
406 | * - an unknown object is activated (might be a statically initialized object) | |
407 | */ | |
408 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
409 | { | |
410 | switch (state) { | |
411 | ||
412 | case ODEBUG_STATE_NOTAVAILABLE: | |
413 | WARN_ON_ONCE(1); | |
414 | return 0; | |
415 | ||
416 | case ODEBUG_STATE_ACTIVE: | |
417 | WARN_ON(1); | |
418 | ||
419 | default: | |
420 | return 0; | |
421 | } | |
422 | } | |
423 | ||
424 | /* | |
425 | * fixup_free is called when: | |
426 | * - an active object is freed | |
427 | */ | |
428 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
429 | { | |
430 | struct hrtimer *timer = addr; | |
431 | ||
432 | switch (state) { | |
433 | case ODEBUG_STATE_ACTIVE: | |
434 | hrtimer_cancel(timer); | |
435 | debug_object_free(timer, &hrtimer_debug_descr); | |
436 | return 1; | |
437 | default: | |
438 | return 0; | |
439 | } | |
440 | } | |
441 | ||
442 | static struct debug_obj_descr hrtimer_debug_descr = { | |
443 | .name = "hrtimer", | |
99777288 | 444 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
445 | .fixup_init = hrtimer_fixup_init, |
446 | .fixup_activate = hrtimer_fixup_activate, | |
447 | .fixup_free = hrtimer_fixup_free, | |
448 | }; | |
449 | ||
450 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
451 | { | |
452 | debug_object_init(timer, &hrtimer_debug_descr); | |
453 | } | |
454 | ||
455 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
456 | { | |
457 | debug_object_activate(timer, &hrtimer_debug_descr); | |
458 | } | |
459 | ||
460 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
461 | { | |
462 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
463 | } | |
464 | ||
465 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
466 | { | |
467 | debug_object_free(timer, &hrtimer_debug_descr); | |
468 | } | |
469 | ||
470 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
471 | enum hrtimer_mode mode); | |
472 | ||
473 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
474 | enum hrtimer_mode mode) | |
475 | { | |
476 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
477 | __hrtimer_init(timer, clock_id, mode); | |
478 | } | |
2bc481cf | 479 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
480 | |
481 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
482 | { | |
483 | debug_object_free(timer, &hrtimer_debug_descr); | |
484 | } | |
485 | ||
486 | #else | |
487 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
488 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
489 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
490 | #endif | |
491 | ||
c6a2a177 XG |
492 | static inline void |
493 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
494 | enum hrtimer_mode mode) | |
495 | { | |
496 | debug_hrtimer_init(timer); | |
497 | trace_hrtimer_init(timer, clockid, mode); | |
498 | } | |
499 | ||
500 | static inline void debug_activate(struct hrtimer *timer) | |
501 | { | |
502 | debug_hrtimer_activate(timer); | |
503 | trace_hrtimer_start(timer); | |
504 | } | |
505 | ||
506 | static inline void debug_deactivate(struct hrtimer *timer) | |
507 | { | |
508 | debug_hrtimer_deactivate(timer); | |
509 | trace_hrtimer_cancel(timer); | |
510 | } | |
511 | ||
54cdfdb4 TG |
512 | /* High resolution timer related functions */ |
513 | #ifdef CONFIG_HIGH_RES_TIMERS | |
514 | ||
515 | /* | |
516 | * High resolution timer enabled ? | |
517 | */ | |
518 | static int hrtimer_hres_enabled __read_mostly = 1; | |
519 | ||
520 | /* | |
521 | * Enable / Disable high resolution mode | |
522 | */ | |
523 | static int __init setup_hrtimer_hres(char *str) | |
524 | { | |
525 | if (!strcmp(str, "off")) | |
526 | hrtimer_hres_enabled = 0; | |
527 | else if (!strcmp(str, "on")) | |
528 | hrtimer_hres_enabled = 1; | |
529 | else | |
530 | return 0; | |
531 | return 1; | |
532 | } | |
533 | ||
534 | __setup("highres=", setup_hrtimer_hres); | |
535 | ||
536 | /* | |
537 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
538 | */ | |
539 | static inline int hrtimer_is_hres_enabled(void) | |
540 | { | |
541 | return hrtimer_hres_enabled; | |
542 | } | |
543 | ||
544 | /* | |
545 | * Is the high resolution mode active ? | |
546 | */ | |
547 | static inline int hrtimer_hres_active(void) | |
548 | { | |
909ea964 | 549 | return __this_cpu_read(hrtimer_bases.hres_active); |
54cdfdb4 TG |
550 | } |
551 | ||
552 | /* | |
553 | * Reprogram the event source with checking both queues for the | |
554 | * next event | |
555 | * Called with interrupts disabled and base->lock held | |
556 | */ | |
7403f41f AC |
557 | static void |
558 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 TG |
559 | { |
560 | int i; | |
561 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
7403f41f | 562 | ktime_t expires, expires_next; |
54cdfdb4 | 563 | |
7403f41f | 564 | expires_next.tv64 = KTIME_MAX; |
54cdfdb4 TG |
565 | |
566 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
567 | struct hrtimer *timer; | |
998adc3d | 568 | struct timerqueue_node *next; |
54cdfdb4 | 569 | |
998adc3d JS |
570 | next = timerqueue_getnext(&base->active); |
571 | if (!next) | |
54cdfdb4 | 572 | continue; |
998adc3d JS |
573 | timer = container_of(next, struct hrtimer, node); |
574 | ||
cc584b21 | 575 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
576 | /* |
577 | * clock_was_set() has changed base->offset so the | |
578 | * result might be negative. Fix it up to prevent a | |
579 | * false positive in clockevents_program_event() | |
580 | */ | |
581 | if (expires.tv64 < 0) | |
582 | expires.tv64 = 0; | |
7403f41f AC |
583 | if (expires.tv64 < expires_next.tv64) |
584 | expires_next = expires; | |
54cdfdb4 TG |
585 | } |
586 | ||
7403f41f AC |
587 | if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64) |
588 | return; | |
589 | ||
590 | cpu_base->expires_next.tv64 = expires_next.tv64; | |
591 | ||
be6e0ece SH |
592 | /* |
593 | * If a hang was detected in the last timer interrupt then we | |
594 | * leave the hang delay active in the hardware. We want the | |
595 | * system to make progress. That also prevents the following | |
596 | * scenario: | |
597 | * T1 expires 50ms from now | |
598 | * T2 expires 5s from now | |
599 | * | |
600 | * T1 is removed, so this code is called and would reprogram | |
601 | * the hardware to 5s from now. Any hrtimer_start after that | |
602 | * will not reprogram the hardware due to hang_detected being | |
603 | * set. So we'd effectivly block all timers until the T2 event | |
604 | * fires. | |
605 | */ | |
606 | if (cpu_base->hang_detected) | |
607 | return; | |
608 | ||
54cdfdb4 TG |
609 | if (cpu_base->expires_next.tv64 != KTIME_MAX) |
610 | tick_program_event(cpu_base->expires_next, 1); | |
611 | } | |
612 | ||
613 | /* | |
614 | * Shared reprogramming for clock_realtime and clock_monotonic | |
615 | * | |
616 | * When a timer is enqueued and expires earlier than the already enqueued | |
617 | * timers, we have to check, whether it expires earlier than the timer for | |
618 | * which the clock event device was armed. | |
619 | * | |
620 | * Called with interrupts disabled and base->cpu_base.lock held | |
621 | */ | |
622 | static int hrtimer_reprogram(struct hrtimer *timer, | |
623 | struct hrtimer_clock_base *base) | |
624 | { | |
41d2e494 | 625 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
cc584b21 | 626 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
627 | int res; |
628 | ||
cc584b21 | 629 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 630 | |
54cdfdb4 TG |
631 | /* |
632 | * When the callback is running, we do not reprogram the clock event | |
633 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 634 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
635 | * reprogramming is handled either by the softirq, which called the |
636 | * callback or at the end of the hrtimer_interrupt. | |
637 | */ | |
638 | if (hrtimer_callback_running(timer)) | |
639 | return 0; | |
640 | ||
63070a79 TG |
641 | /* |
642 | * CLOCK_REALTIME timer might be requested with an absolute | |
643 | * expiry time which is less than base->offset. Nothing wrong | |
644 | * about that, just avoid to call into the tick code, which | |
645 | * has now objections against negative expiry values. | |
646 | */ | |
647 | if (expires.tv64 < 0) | |
648 | return -ETIME; | |
649 | ||
41d2e494 TG |
650 | if (expires.tv64 >= cpu_base->expires_next.tv64) |
651 | return 0; | |
652 | ||
653 | /* | |
654 | * If a hang was detected in the last timer interrupt then we | |
655 | * do not schedule a timer which is earlier than the expiry | |
656 | * which we enforced in the hang detection. We want the system | |
657 | * to make progress. | |
658 | */ | |
659 | if (cpu_base->hang_detected) | |
54cdfdb4 TG |
660 | return 0; |
661 | ||
662 | /* | |
663 | * Clockevents returns -ETIME, when the event was in the past. | |
664 | */ | |
665 | res = tick_program_event(expires, 0); | |
666 | if (!IS_ERR_VALUE(res)) | |
41d2e494 | 667 | cpu_base->expires_next = expires; |
54cdfdb4 TG |
668 | return res; |
669 | } | |
670 | ||
54cdfdb4 TG |
671 | /* |
672 | * Initialize the high resolution related parts of cpu_base | |
673 | */ | |
674 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
675 | { | |
676 | base->expires_next.tv64 = KTIME_MAX; | |
677 | base->hres_active = 0; | |
54cdfdb4 TG |
678 | } |
679 | ||
54cdfdb4 TG |
680 | /* |
681 | * When High resolution timers are active, try to reprogram. Note, that in case | |
682 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
683 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
684 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
685 | */ | |
686 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
b22affe0 | 687 | struct hrtimer_clock_base *base) |
54cdfdb4 | 688 | { |
b22affe0 | 689 | return base->cpu_base->hres_active && hrtimer_reprogram(timer, base); |
54cdfdb4 TG |
690 | } |
691 | ||
5baefd6d JS |
692 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
693 | { | |
694 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
695 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; | |
90adda98 | 696 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; |
5baefd6d | 697 | |
90adda98 | 698 | return ktime_get_update_offsets(offs_real, offs_boot, offs_tai); |
5baefd6d JS |
699 | } |
700 | ||
9ec26907 TG |
701 | /* |
702 | * Retrigger next event is called after clock was set | |
703 | * | |
704 | * Called with interrupts disabled via on_each_cpu() | |
705 | */ | |
706 | static void retrigger_next_event(void *arg) | |
707 | { | |
708 | struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases); | |
9ec26907 TG |
709 | |
710 | if (!hrtimer_hres_active()) | |
711 | return; | |
712 | ||
9ec26907 | 713 | raw_spin_lock(&base->lock); |
5baefd6d | 714 | hrtimer_update_base(base); |
9ec26907 TG |
715 | hrtimer_force_reprogram(base, 0); |
716 | raw_spin_unlock(&base->lock); | |
717 | } | |
b12a03ce | 718 | |
54cdfdb4 TG |
719 | /* |
720 | * Switch to high resolution mode | |
721 | */ | |
f8953856 | 722 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 723 | { |
b12a03ce | 724 | int i, cpu = smp_processor_id(); |
820de5c3 | 725 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); |
54cdfdb4 TG |
726 | unsigned long flags; |
727 | ||
728 | if (base->hres_active) | |
f8953856 | 729 | return 1; |
54cdfdb4 TG |
730 | |
731 | local_irq_save(flags); | |
732 | ||
733 | if (tick_init_highres()) { | |
734 | local_irq_restore(flags); | |
820de5c3 IM |
735 | printk(KERN_WARNING "Could not switch to high resolution " |
736 | "mode on CPU %d\n", cpu); | |
f8953856 | 737 | return 0; |
54cdfdb4 TG |
738 | } |
739 | base->hres_active = 1; | |
b12a03ce TG |
740 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) |
741 | base->clock_base[i].resolution = KTIME_HIGH_RES; | |
54cdfdb4 TG |
742 | |
743 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
744 | /* "Retrigger" the interrupt to get things going */ |
745 | retrigger_next_event(NULL); | |
746 | local_irq_restore(flags); | |
f8953856 | 747 | return 1; |
54cdfdb4 TG |
748 | } |
749 | ||
2d06fa0f TG |
750 | static void clock_was_set_work(struct work_struct *work) |
751 | { | |
752 | clock_was_set(); | |
753 | } | |
754 | ||
755 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
756 | ||
f55a6faa | 757 | /* |
2d06fa0f TG |
758 | * Called from timekeeping and resume code to reprogramm the hrtimer |
759 | * interrupt device on all cpus. | |
f55a6faa JS |
760 | */ |
761 | void clock_was_set_delayed(void) | |
762 | { | |
2d06fa0f | 763 | schedule_work(&hrtimer_work); |
f55a6faa JS |
764 | } |
765 | ||
54cdfdb4 TG |
766 | #else |
767 | ||
768 | static inline int hrtimer_hres_active(void) { return 0; } | |
769 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 770 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
7403f41f AC |
771 | static inline void |
772 | hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { } | |
54cdfdb4 | 773 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, |
b22affe0 | 774 | struct hrtimer_clock_base *base) |
54cdfdb4 TG |
775 | { |
776 | return 0; | |
777 | } | |
54cdfdb4 | 778 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
9ec26907 | 779 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
780 | |
781 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
782 | ||
b12a03ce TG |
783 | /* |
784 | * Clock realtime was set | |
785 | * | |
786 | * Change the offset of the realtime clock vs. the monotonic | |
787 | * clock. | |
788 | * | |
789 | * We might have to reprogram the high resolution timer interrupt. On | |
790 | * SMP we call the architecture specific code to retrigger _all_ high | |
791 | * resolution timer interrupts. On UP we just disable interrupts and | |
792 | * call the high resolution interrupt code. | |
793 | */ | |
794 | void clock_was_set(void) | |
795 | { | |
90ff1f30 | 796 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
797 | /* Retrigger the CPU local events everywhere */ |
798 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
799 | #endif |
800 | timerfd_clock_was_set(); | |
b12a03ce TG |
801 | } |
802 | ||
803 | /* | |
804 | * During resume we might have to reprogram the high resolution timer | |
805 | * interrupt (on the local CPU): | |
806 | */ | |
807 | void hrtimers_resume(void) | |
808 | { | |
809 | WARN_ONCE(!irqs_disabled(), | |
810 | KERN_INFO "hrtimers_resume() called with IRQs enabled!"); | |
811 | ||
2d06fa0f | 812 | /* Retrigger on the local CPU */ |
b12a03ce | 813 | retrigger_next_event(NULL); |
2d06fa0f TG |
814 | /* And schedule a retrigger for all others */ |
815 | clock_was_set_delayed(); | |
b12a03ce TG |
816 | } |
817 | ||
5f201907 | 818 | static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer) |
82f67cd9 | 819 | { |
5f201907 | 820 | #ifdef CONFIG_TIMER_STATS |
82f67cd9 IM |
821 | if (timer->start_site) |
822 | return; | |
5f201907 | 823 | timer->start_site = __builtin_return_address(0); |
82f67cd9 IM |
824 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); |
825 | timer->start_pid = current->pid; | |
5f201907 HC |
826 | #endif |
827 | } | |
828 | ||
829 | static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer) | |
830 | { | |
831 | #ifdef CONFIG_TIMER_STATS | |
832 | timer->start_site = NULL; | |
833 | #endif | |
82f67cd9 | 834 | } |
5f201907 HC |
835 | |
836 | static inline void timer_stats_account_hrtimer(struct hrtimer *timer) | |
837 | { | |
838 | #ifdef CONFIG_TIMER_STATS | |
839 | if (likely(!timer_stats_active)) | |
840 | return; | |
841 | timer_stats_update_stats(timer, timer->start_pid, timer->start_site, | |
842 | timer->function, timer->start_comm, 0); | |
82f67cd9 | 843 | #endif |
5f201907 | 844 | } |
82f67cd9 | 845 | |
c0a31329 | 846 | /* |
6506f2aa | 847 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
848 | */ |
849 | static inline | |
850 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
851 | { | |
ecb49d1a | 852 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
853 | } |
854 | ||
855 | /** | |
856 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 857 | * @timer: hrtimer to forward |
44f21475 | 858 | * @now: forward past this time |
c0a31329 TG |
859 | * @interval: the interval to forward |
860 | * | |
861 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 862 | * Returns the number of overruns. |
c0a31329 | 863 | */ |
4d672e7a | 864 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 865 | { |
4d672e7a | 866 | u64 orun = 1; |
44f21475 | 867 | ktime_t delta; |
c0a31329 | 868 | |
cc584b21 | 869 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
870 | |
871 | if (delta.tv64 < 0) | |
872 | return 0; | |
873 | ||
c9db4fa1 TG |
874 | if (interval.tv64 < timer->base->resolution.tv64) |
875 | interval.tv64 = timer->base->resolution.tv64; | |
876 | ||
c0a31329 | 877 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 878 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
879 | |
880 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
881 | hrtimer_add_expires_ns(timer, incr * orun); |
882 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
883 | return orun; |
884 | /* | |
885 | * This (and the ktime_add() below) is the | |
886 | * correction for exact: | |
887 | */ | |
888 | orun++; | |
889 | } | |
cc584b21 | 890 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
891 | |
892 | return orun; | |
893 | } | |
6bdb6b62 | 894 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
895 | |
896 | /* | |
897 | * enqueue_hrtimer - internal function to (re)start a timer | |
898 | * | |
899 | * The timer is inserted in expiry order. Insertion into the | |
900 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
901 | * |
902 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 903 | */ |
a6037b61 PZ |
904 | static int enqueue_hrtimer(struct hrtimer *timer, |
905 | struct hrtimer_clock_base *base) | |
c0a31329 | 906 | { |
c6a2a177 | 907 | debug_activate(timer); |
237fc6e7 | 908 | |
998adc3d | 909 | timerqueue_add(&base->active, &timer->node); |
ab8177bc | 910 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 911 | |
303e967f TG |
912 | /* |
913 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
914 | * state of a possibly running callback. | |
915 | */ | |
916 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 | 917 | |
998adc3d | 918 | return (&timer->node == base->active.next); |
288867ec | 919 | } |
c0a31329 TG |
920 | |
921 | /* | |
922 | * __remove_hrtimer - internal function to remove a timer | |
923 | * | |
924 | * Caller must hold the base lock. | |
54cdfdb4 TG |
925 | * |
926 | * High resolution timer mode reprograms the clock event device when the | |
927 | * timer is the one which expires next. The caller can disable this by setting | |
928 | * reprogram to zero. This is useful, when the context does a reprogramming | |
929 | * anyway (e.g. timer interrupt) | |
c0a31329 | 930 | */ |
3c8aa39d | 931 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 932 | struct hrtimer_clock_base *base, |
54cdfdb4 | 933 | unsigned long newstate, int reprogram) |
c0a31329 | 934 | { |
27c9cd7e | 935 | struct timerqueue_node *next_timer; |
7403f41f AC |
936 | if (!(timer->state & HRTIMER_STATE_ENQUEUED)) |
937 | goto out; | |
938 | ||
27c9cd7e JO |
939 | next_timer = timerqueue_getnext(&base->active); |
940 | timerqueue_del(&base->active, &timer->node); | |
941 | if (&timer->node == next_timer) { | |
7403f41f AC |
942 | #ifdef CONFIG_HIGH_RES_TIMERS |
943 | /* Reprogram the clock event device. if enabled */ | |
944 | if (reprogram && hrtimer_hres_active()) { | |
945 | ktime_t expires; | |
946 | ||
947 | expires = ktime_sub(hrtimer_get_expires(timer), | |
948 | base->offset); | |
949 | if (base->cpu_base->expires_next.tv64 == expires.tv64) | |
950 | hrtimer_force_reprogram(base->cpu_base, 1); | |
54cdfdb4 | 951 | } |
7403f41f | 952 | #endif |
54cdfdb4 | 953 | } |
ab8177bc TG |
954 | if (!timerqueue_getnext(&base->active)) |
955 | base->cpu_base->active_bases &= ~(1 << base->index); | |
7403f41f | 956 | out: |
303e967f | 957 | timer->state = newstate; |
c0a31329 TG |
958 | } |
959 | ||
960 | /* | |
961 | * remove hrtimer, called with base lock held | |
962 | */ | |
963 | static inline int | |
3c8aa39d | 964 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 965 | { |
303e967f | 966 | if (hrtimer_is_queued(timer)) { |
f13d4f97 | 967 | unsigned long state; |
54cdfdb4 TG |
968 | int reprogram; |
969 | ||
970 | /* | |
971 | * Remove the timer and force reprogramming when high | |
972 | * resolution mode is active and the timer is on the current | |
973 | * CPU. If we remove a timer on another CPU, reprogramming is | |
974 | * skipped. The interrupt event on this CPU is fired and | |
975 | * reprogramming happens in the interrupt handler. This is a | |
976 | * rare case and less expensive than a smp call. | |
977 | */ | |
c6a2a177 | 978 | debug_deactivate(timer); |
82f67cd9 | 979 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 | 980 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
f13d4f97 SQ |
981 | /* |
982 | * We must preserve the CALLBACK state flag here, | |
983 | * otherwise we could move the timer base in | |
984 | * switch_hrtimer_base. | |
985 | */ | |
986 | state = timer->state & HRTIMER_STATE_CALLBACK; | |
987 | __remove_hrtimer(timer, base, state, reprogram); | |
c0a31329 TG |
988 | return 1; |
989 | } | |
990 | return 0; | |
991 | } | |
992 | ||
7f1e2ca9 PZ |
993 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
994 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
995 | int wakeup) | |
c0a31329 | 996 | { |
3c8aa39d | 997 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 998 | unsigned long flags; |
a6037b61 | 999 | int ret, leftmost; |
6fa3eb70 S |
1000 | /*add MTK debug log for ALPS01804694*/ |
1001 | if(timer->function == NULL) { | |
1002 | pr_alert("add hrtimer but do nothing"); | |
1003 | dump_stack(); | |
1004 | } | |
1005 | ||
c0a31329 TG |
1006 | base = lock_hrtimer_base(timer, &flags); |
1007 | ||
1008 | /* Remove an active timer from the queue: */ | |
1009 | ret = remove_hrtimer(timer, base); | |
1010 | ||
597d0275 | 1011 | if (mode & HRTIMER_MODE_REL) { |
1c0301d1 | 1012 | tim = ktime_add_safe(tim, base->get_time()); |
06027bdd IM |
1013 | /* |
1014 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
1015 | * to signal that they simply return xtime in | |
1016 | * do_gettimeoffset(). In this case we want to round up by | |
1017 | * resolution when starting a relative timer, to avoid short | |
1018 | * timeouts. This will go away with the GTOD framework. | |
1019 | */ | |
1020 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 1021 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
1022 | #endif |
1023 | } | |
237fc6e7 | 1024 | |
da8f2e17 | 1025 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 1026 | |
1c0301d1 VK |
1027 | /* Switch the timer base, if necessary: */ |
1028 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); | |
1029 | ||
82f67cd9 IM |
1030 | timer_stats_hrtimer_set_start_info(timer); |
1031 | ||
a6037b61 PZ |
1032 | leftmost = enqueue_hrtimer(timer, new_base); |
1033 | ||
935c631d IM |
1034 | /* |
1035 | * Only allow reprogramming if the new base is on this CPU. | |
1036 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
1037 | * |
1038 | * XXX send_remote_softirq() ? | |
935c631d | 1039 | */ |
b22affe0 LS |
1040 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases) |
1041 | && hrtimer_enqueue_reprogram(timer, new_base)) { | |
1042 | if (wakeup) { | |
1043 | /* | |
1044 | * We need to drop cpu_base->lock to avoid a | |
1045 | * lock ordering issue vs. rq->lock. | |
1046 | */ | |
1047 | raw_spin_unlock(&new_base->cpu_base->lock); | |
1048 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1049 | local_irq_restore(flags); | |
1050 | return ret; | |
1051 | } else { | |
1052 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1053 | } | |
1054 | } | |
c0a31329 TG |
1055 | |
1056 | unlock_hrtimer_base(timer, &flags); | |
1057 | ||
1058 | return ret; | |
1059 | } | |
7f1e2ca9 PZ |
1060 | |
1061 | /** | |
1062 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
1063 | * @timer: the timer to be added | |
1064 | * @tim: expiry time | |
1065 | * @delta_ns: "slack" range for the timer | |
8ffbc7d9 DD |
1066 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
1067 | * relative (HRTIMER_MODE_REL) | |
7f1e2ca9 PZ |
1068 | * |
1069 | * Returns: | |
1070 | * 0 on success | |
1071 | * 1 when the timer was active | |
1072 | */ | |
1073 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1074 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
1075 | { | |
1076 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
1077 | } | |
da8f2e17 AV |
1078 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1079 | ||
1080 | /** | |
e1dd7bc5 | 1081 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
1082 | * @timer: the timer to be added |
1083 | * @tim: expiry time | |
8ffbc7d9 DD |
1084 | * @mode: expiry mode: absolute (HRTIMER_MODE_ABS) or |
1085 | * relative (HRTIMER_MODE_REL) | |
da8f2e17 AV |
1086 | * |
1087 | * Returns: | |
1088 | * 0 on success | |
1089 | * 1 when the timer was active | |
1090 | */ | |
1091 | int | |
1092 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
1093 | { | |
7f1e2ca9 | 1094 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1095 | } |
8d16b764 | 1096 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1097 | |
da8f2e17 | 1098 | |
c0a31329 TG |
1099 | /** |
1100 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1101 | * @timer: hrtimer to stop |
1102 | * | |
1103 | * Returns: | |
1104 | * 0 when the timer was not active | |
1105 | * 1 when the timer was active | |
1106 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1107 | * cannot be stopped |
c0a31329 TG |
1108 | */ |
1109 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1110 | { | |
3c8aa39d | 1111 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1112 | unsigned long flags; |
1113 | int ret = -1; | |
1114 | ||
1115 | base = lock_hrtimer_base(timer, &flags); | |
1116 | ||
303e967f | 1117 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1118 | ret = remove_hrtimer(timer, base); |
1119 | ||
1120 | unlock_hrtimer_base(timer, &flags); | |
1121 | ||
1122 | return ret; | |
1123 | ||
1124 | } | |
8d16b764 | 1125 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1126 | |
1127 | /** | |
1128 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1129 | * @timer: the timer to be cancelled |
1130 | * | |
1131 | * Returns: | |
1132 | * 0 when the timer was not active | |
1133 | * 1 when the timer was active | |
1134 | */ | |
1135 | int hrtimer_cancel(struct hrtimer *timer) | |
1136 | { | |
1137 | for (;;) { | |
1138 | int ret = hrtimer_try_to_cancel(timer); | |
1139 | ||
1140 | if (ret >= 0) | |
1141 | return ret; | |
5ef37b19 | 1142 | cpu_relax(); |
c0a31329 TG |
1143 | } |
1144 | } | |
8d16b764 | 1145 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1146 | |
1147 | /** | |
1148 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1149 | * @timer: the timer to read |
1150 | */ | |
1151 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1152 | { | |
c0a31329 TG |
1153 | unsigned long flags; |
1154 | ktime_t rem; | |
1155 | ||
b3bd3de6 | 1156 | lock_hrtimer_base(timer, &flags); |
cc584b21 | 1157 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1158 | unlock_hrtimer_base(timer, &flags); |
1159 | ||
1160 | return rem; | |
1161 | } | |
8d16b764 | 1162 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1163 | |
3451d024 | 1164 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1165 | /** |
1166 | * hrtimer_get_next_event - get the time until next expiry event | |
1167 | * | |
1168 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1169 | * is pending. | |
1170 | */ | |
1171 | ktime_t hrtimer_get_next_event(void) | |
1172 | { | |
3c8aa39d TG |
1173 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1174 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1175 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1176 | unsigned long flags; | |
1177 | int i; | |
1178 | ||
ecb49d1a | 1179 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1180 | |
54cdfdb4 TG |
1181 | if (!hrtimer_hres_active()) { |
1182 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1183 | struct hrtimer *timer; | |
998adc3d | 1184 | struct timerqueue_node *next; |
69239749 | 1185 | |
998adc3d JS |
1186 | next = timerqueue_getnext(&base->active); |
1187 | if (!next) | |
54cdfdb4 | 1188 | continue; |
3c8aa39d | 1189 | |
998adc3d | 1190 | timer = container_of(next, struct hrtimer, node); |
cc584b21 | 1191 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1192 | delta = ktime_sub(delta, base->get_time()); |
1193 | if (delta.tv64 < mindelta.tv64) | |
1194 | mindelta.tv64 = delta.tv64; | |
1195 | } | |
69239749 | 1196 | } |
3c8aa39d | 1197 | |
ecb49d1a | 1198 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1199 | |
69239749 TL |
1200 | if (mindelta.tv64 < 0) |
1201 | mindelta.tv64 = 0; | |
1202 | return mindelta; | |
1203 | } | |
1204 | #endif | |
1205 | ||
237fc6e7 TG |
1206 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1207 | enum hrtimer_mode mode) | |
c0a31329 | 1208 | { |
3c8aa39d | 1209 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1210 | int base; |
c0a31329 | 1211 | |
7978672c GA |
1212 | memset(timer, 0, sizeof(struct hrtimer)); |
1213 | ||
3c8aa39d | 1214 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1215 | |
c9cb2e3d | 1216 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1217 | clock_id = CLOCK_MONOTONIC; |
1218 | ||
e06383db JS |
1219 | base = hrtimer_clockid_to_base(clock_id); |
1220 | timer->base = &cpu_base->clock_base[base]; | |
998adc3d | 1221 | timerqueue_init(&timer->node); |
82f67cd9 IM |
1222 | |
1223 | #ifdef CONFIG_TIMER_STATS | |
1224 | timer->start_site = NULL; | |
1225 | timer->start_pid = -1; | |
1226 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1227 | #endif | |
c0a31329 | 1228 | } |
237fc6e7 TG |
1229 | |
1230 | /** | |
1231 | * hrtimer_init - initialize a timer to the given clock | |
1232 | * @timer: the timer to be initialized | |
1233 | * @clock_id: the clock to be used | |
1234 | * @mode: timer mode abs/rel | |
1235 | */ | |
1236 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1237 | enum hrtimer_mode mode) | |
1238 | { | |
c6a2a177 | 1239 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1240 | __hrtimer_init(timer, clock_id, mode); |
1241 | } | |
8d16b764 | 1242 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1243 | |
1244 | /** | |
1245 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1246 | * @which_clock: which clock to query |
1247 | * @tp: pointer to timespec variable to store the resolution | |
1248 | * | |
72fd4a35 RD |
1249 | * Store the resolution of the clock selected by @which_clock in the |
1250 | * variable pointed to by @tp. | |
c0a31329 TG |
1251 | */ |
1252 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1253 | { | |
3c8aa39d | 1254 | struct hrtimer_cpu_base *cpu_base; |
e06383db | 1255 | int base = hrtimer_clockid_to_base(which_clock); |
c0a31329 | 1256 | |
3c8aa39d | 1257 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
e06383db | 1258 | *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution); |
c0a31329 TG |
1259 | |
1260 | return 0; | |
1261 | } | |
8d16b764 | 1262 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1263 | |
6fa3eb70 S |
1264 | #ifdef MTK_HRTIME_DEBUG |
1265 | static void dump_hrtimer_callinfo(struct hrtimer *timer) | |
1266 | { | |
1267 | ||
1268 | char symname[KSYM_NAME_LEN]; | |
1269 | if (lookup_symbol_name((unsigned long)(timer->function), symname) < 0) { | |
1270 | pr_err("timer info1: state/%lx, func/%pK\n", | |
1271 | timer->state, timer->function); | |
1272 | } else { | |
1273 | pr_err("timer info2: state/%lx, func/%s\n", | |
1274 | timer->state, symname); | |
1275 | } | |
1276 | ||
1277 | #ifdef CONFIG_TIMER_STATS | |
1278 | if (lookup_symbol_name((unsigned long)(timer->start_site), | |
1279 | symname) < 0) { | |
1280 | pr_err("timer stats1: pid/%d(%s), site/%pK\n", | |
1281 | timer->start_pid, timer->start_comm, timer->start_site); | |
1282 | } else { | |
1283 | pr_err("timer stats2: pid/%d(%s), site/%s\n", | |
1284 | timer->start_pid, timer->start_comm, symname); | |
1285 | } | |
1286 | #endif | |
1287 | } | |
1288 | #endif | |
c6a2a177 | 1289 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1290 | { |
1291 | struct hrtimer_clock_base *base = timer->base; | |
1292 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1293 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1294 | int restart; | |
1295 | ||
ca109491 PZ |
1296 | WARN_ON(!irqs_disabled()); |
1297 | ||
c6a2a177 | 1298 | debug_deactivate(timer); |
d3d74453 PZ |
1299 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1300 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1301 | fn = timer->function; |
ca109491 PZ |
1302 | |
1303 | /* | |
1304 | * Because we run timers from hardirq context, there is no chance | |
1305 | * they get migrated to another cpu, therefore its safe to unlock | |
1306 | * the timer base. | |
1307 | */ | |
ecb49d1a | 1308 | raw_spin_unlock(&cpu_base->lock); |
c6a2a177 | 1309 | trace_hrtimer_expire_entry(timer, now); |
6fa3eb70 S |
1310 | |
1311 | mt_trace_hrt_start(fn); | |
ca109491 | 1312 | restart = fn(timer); |
6fa3eb70 | 1313 | mt_trace_hrt_end(fn); |
c6a2a177 | 1314 | trace_hrtimer_expire_exit(timer); |
ecb49d1a | 1315 | raw_spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1316 | |
1317 | /* | |
e3f1d883 TG |
1318 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1319 | * we do not reprogramm the event hardware. Happens either in | |
1320 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1321 | */ |
1322 | if (restart != HRTIMER_NORESTART) { | |
1323 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1324 | enqueue_hrtimer(timer, base); |
d3d74453 | 1325 | } |
f13d4f97 SQ |
1326 | |
1327 | WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK)); | |
1328 | ||
d3d74453 PZ |
1329 | timer->state &= ~HRTIMER_STATE_CALLBACK; |
1330 | } | |
1331 | ||
54cdfdb4 TG |
1332 | #ifdef CONFIG_HIGH_RES_TIMERS |
1333 | ||
1334 | /* | |
1335 | * High resolution timer interrupt | |
1336 | * Called with interrupts disabled | |
1337 | */ | |
1338 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1339 | { | |
1340 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
41d2e494 TG |
1341 | ktime_t expires_next, now, entry_time, delta; |
1342 | int i, retries = 0; | |
54cdfdb4 TG |
1343 | |
1344 | BUG_ON(!cpu_base->hres_active); | |
1345 | cpu_base->nr_events++; | |
1346 | dev->next_event.tv64 = KTIME_MAX; | |
1347 | ||
196951e9 | 1348 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1349 | entry_time = now = hrtimer_update_base(cpu_base); |
41d2e494 | 1350 | retry: |
54cdfdb4 | 1351 | expires_next.tv64 = KTIME_MAX; |
6ff7041d TG |
1352 | /* |
1353 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1354 | * held to prevent that a timer is enqueued in our queue via | |
1355 | * the migration code. This does not affect enqueueing of | |
1356 | * timers which run their callback and need to be requeued on | |
1357 | * this CPU. | |
1358 | */ | |
1359 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1360 | ||
54cdfdb4 | 1361 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ab8177bc | 1362 | struct hrtimer_clock_base *base; |
998adc3d | 1363 | struct timerqueue_node *node; |
ab8177bc TG |
1364 | ktime_t basenow; |
1365 | ||
1366 | if (!(cpu_base->active_bases & (1 << i))) | |
1367 | continue; | |
54cdfdb4 | 1368 | |
ab8177bc | 1369 | base = cpu_base->clock_base + i; |
54cdfdb4 TG |
1370 | basenow = ktime_add(now, base->offset); |
1371 | ||
998adc3d | 1372 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1373 | struct hrtimer *timer; |
1374 | ||
998adc3d | 1375 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1376 | |
654c8e0b AV |
1377 | /* |
1378 | * The immediate goal for using the softexpires is | |
1379 | * minimizing wakeups, not running timers at the | |
1380 | * earliest interrupt after their soft expiration. | |
1381 | * This allows us to avoid using a Priority Search | |
1382 | * Tree, which can answer a stabbing querry for | |
1383 | * overlapping intervals and instead use the simple | |
1384 | * BST we already have. | |
1385 | * We don't add extra wakeups by delaying timers that | |
1386 | * are right-of a not yet expired timer, because that | |
1387 | * timer will have to trigger a wakeup anyway. | |
1388 | */ | |
1389 | ||
1390 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1391 | ktime_t expires; |
1392 | ||
cc584b21 | 1393 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 | 1394 | base->offset); |
8f294b5a PB |
1395 | if (expires.tv64 < 0) |
1396 | expires.tv64 = KTIME_MAX; | |
54cdfdb4 TG |
1397 | if (expires.tv64 < expires_next.tv64) |
1398 | expires_next = expires; | |
1399 | break; | |
1400 | } | |
1401 | ||
c6a2a177 | 1402 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1403 | } |
54cdfdb4 TG |
1404 | } |
1405 | ||
6ff7041d TG |
1406 | /* |
1407 | * Store the new expiry value so the migration code can verify | |
1408 | * against it. | |
1409 | */ | |
54cdfdb4 | 1410 | cpu_base->expires_next = expires_next; |
ecb49d1a | 1411 | raw_spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1412 | |
1413 | /* Reprogramming necessary ? */ | |
41d2e494 TG |
1414 | if (expires_next.tv64 == KTIME_MAX || |
1415 | !tick_program_event(expires_next, 0)) { | |
1416 | cpu_base->hang_detected = 0; | |
1417 | return; | |
54cdfdb4 | 1418 | } |
41d2e494 TG |
1419 | |
1420 | /* | |
1421 | * The next timer was already expired due to: | |
1422 | * - tracing | |
1423 | * - long lasting callbacks | |
1424 | * - being scheduled away when running in a VM | |
1425 | * | |
1426 | * We need to prevent that we loop forever in the hrtimer | |
1427 | * interrupt routine. We give it 3 attempts to avoid | |
1428 | * overreacting on some spurious event. | |
5baefd6d JS |
1429 | * |
1430 | * Acquire base lock for updating the offsets and retrieving | |
1431 | * the current time. | |
41d2e494 | 1432 | */ |
196951e9 | 1433 | raw_spin_lock(&cpu_base->lock); |
5baefd6d | 1434 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1435 | cpu_base->nr_retries++; |
1436 | if (++retries < 3) | |
1437 | goto retry; | |
1438 | /* | |
1439 | * Give the system a chance to do something else than looping | |
1440 | * here. We stored the entry time, so we know exactly how long | |
1441 | * we spent here. We schedule the next event this amount of | |
1442 | * time away. | |
1443 | */ | |
1444 | cpu_base->nr_hangs++; | |
1445 | cpu_base->hang_detected = 1; | |
196951e9 | 1446 | raw_spin_unlock(&cpu_base->lock); |
41d2e494 TG |
1447 | delta = ktime_sub(now, entry_time); |
1448 | if (delta.tv64 > cpu_base->max_hang_time.tv64) | |
1449 | cpu_base->max_hang_time = delta; | |
1450 | /* | |
1451 | * Limit it to a sensible value as we enforce a longer | |
1452 | * delay. Give the CPU at least 100ms to catch up. | |
1453 | */ | |
1454 | if (delta.tv64 > 100 * NSEC_PER_MSEC) | |
1455 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); | |
1456 | else | |
1457 | expires_next = ktime_add(now, delta); | |
1458 | tick_program_event(expires_next, 1); | |
1459 | printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n", | |
1460 | ktime_to_ns(delta)); | |
54cdfdb4 TG |
1461 | } |
1462 | ||
8bdec955 TG |
1463 | /* |
1464 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1465 | * disabled. | |
1466 | */ | |
1467 | static void __hrtimer_peek_ahead_timers(void) | |
1468 | { | |
1469 | struct tick_device *td; | |
1470 | ||
1471 | if (!hrtimer_hres_active()) | |
1472 | return; | |
1473 | ||
1474 | td = &__get_cpu_var(tick_cpu_device); | |
1475 | if (td && td->evtdev) | |
1476 | hrtimer_interrupt(td->evtdev); | |
1477 | } | |
1478 | ||
2e94d1f7 AV |
1479 | /** |
1480 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1481 | * | |
1482 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1483 | * the current cpu and check if there are any timers for which | |
1484 | * the soft expires time has passed. If any such timers exist, | |
1485 | * they are run immediately and then removed from the timer queue. | |
1486 | * | |
1487 | */ | |
1488 | void hrtimer_peek_ahead_timers(void) | |
1489 | { | |
643bdf68 | 1490 | unsigned long flags; |
dc4304f7 | 1491 | |
2e94d1f7 | 1492 | local_irq_save(flags); |
8bdec955 | 1493 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1494 | local_irq_restore(flags); |
1495 | } | |
1496 | ||
a6037b61 PZ |
1497 | static void run_hrtimer_softirq(struct softirq_action *h) |
1498 | { | |
1499 | hrtimer_peek_ahead_timers(); | |
1500 | } | |
1501 | ||
82c5b7b5 IM |
1502 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1503 | ||
1504 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1505 | ||
1506 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1507 | |
d3d74453 PZ |
1508 | /* |
1509 | * Called from timer softirq every jiffy, expire hrtimers: | |
1510 | * | |
1511 | * For HRT its the fall back code to run the softirq in the timer | |
1512 | * softirq context in case the hrtimer initialization failed or has | |
1513 | * not been done yet. | |
1514 | */ | |
1515 | void hrtimer_run_pending(void) | |
1516 | { | |
d3d74453 PZ |
1517 | if (hrtimer_hres_active()) |
1518 | return; | |
54cdfdb4 | 1519 | |
d3d74453 PZ |
1520 | /* |
1521 | * This _is_ ugly: We have to check in the softirq context, | |
1522 | * whether we can switch to highres and / or nohz mode. The | |
1523 | * clocksource switch happens in the timer interrupt with | |
1524 | * xtime_lock held. Notification from there only sets the | |
1525 | * check bit in the tick_oneshot code, otherwise we might | |
1526 | * deadlock vs. xtime_lock. | |
1527 | */ | |
1528 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1529 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1530 | } |
1531 | ||
c0a31329 | 1532 | /* |
d3d74453 | 1533 | * Called from hardirq context every jiffy |
c0a31329 | 1534 | */ |
833883d9 | 1535 | void hrtimer_run_queues(void) |
c0a31329 | 1536 | { |
998adc3d | 1537 | struct timerqueue_node *node; |
833883d9 DS |
1538 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1539 | struct hrtimer_clock_base *base; | |
1540 | int index, gettime = 1; | |
c0a31329 | 1541 | |
833883d9 | 1542 | if (hrtimer_hres_active()) |
3055adda DS |
1543 | return; |
1544 | ||
833883d9 DS |
1545 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1546 | base = &cpu_base->clock_base[index]; | |
b007c389 | 1547 | if (!timerqueue_getnext(&base->active)) |
d3d74453 | 1548 | continue; |
833883d9 | 1549 | |
d7cfb60c | 1550 | if (gettime) { |
833883d9 DS |
1551 | hrtimer_get_softirq_time(cpu_base); |
1552 | gettime = 0; | |
b75f7a51 | 1553 | } |
d3d74453 | 1554 | |
ecb49d1a | 1555 | raw_spin_lock(&cpu_base->lock); |
c0a31329 | 1556 | |
b007c389 | 1557 | while ((node = timerqueue_getnext(&base->active))) { |
833883d9 | 1558 | struct hrtimer *timer; |
54cdfdb4 | 1559 | |
998adc3d | 1560 | timer = container_of(node, struct hrtimer, node); |
cc584b21 AV |
1561 | if (base->softirq_time.tv64 <= |
1562 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1563 | break; |
1564 | ||
c6a2a177 | 1565 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 | 1566 | } |
ecb49d1a | 1567 | raw_spin_unlock(&cpu_base->lock); |
833883d9 | 1568 | } |
c0a31329 TG |
1569 | } |
1570 | ||
10c94ec1 TG |
1571 | /* |
1572 | * Sleep related functions: | |
1573 | */ | |
c9cb2e3d | 1574 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1575 | { |
1576 | struct hrtimer_sleeper *t = | |
1577 | container_of(timer, struct hrtimer_sleeper, timer); | |
1578 | struct task_struct *task = t->task; | |
1579 | ||
1580 | t->task = NULL; | |
1581 | if (task) | |
1582 | wake_up_process(task); | |
1583 | ||
1584 | return HRTIMER_NORESTART; | |
1585 | } | |
1586 | ||
36c8b586 | 1587 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1588 | { |
1589 | sl->timer.function = hrtimer_wakeup; | |
1590 | sl->task = task; | |
1591 | } | |
2bc481cf | 1592 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1593 | |
669d7868 | 1594 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1595 | { |
669d7868 | 1596 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1597 | |
432569bb RZ |
1598 | do { |
1599 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1600 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1601 | if (!hrtimer_active(&t->timer)) |
1602 | t->task = NULL; | |
432569bb | 1603 | |
54cdfdb4 | 1604 | if (likely(t->task)) |
6fa3eb70 | 1605 | freezable_schedule(); |
432569bb | 1606 | |
669d7868 | 1607 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1608 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1609 | |
1610 | } while (t->task && !signal_pending(current)); | |
432569bb | 1611 | |
3588a085 PZ |
1612 | __set_current_state(TASK_RUNNING); |
1613 | ||
669d7868 | 1614 | return t->task == NULL; |
10c94ec1 TG |
1615 | } |
1616 | ||
080344b9 ON |
1617 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1618 | { | |
1619 | struct timespec rmt; | |
1620 | ktime_t rem; | |
1621 | ||
cc584b21 | 1622 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1623 | if (rem.tv64 <= 0) |
1624 | return 0; | |
1625 | rmt = ktime_to_timespec(rem); | |
1626 | ||
1627 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1628 | return -EFAULT; | |
1629 | ||
1630 | return 1; | |
1631 | } | |
1632 | ||
1711ef38 | 1633 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1634 | { |
669d7868 | 1635 | struct hrtimer_sleeper t; |
080344b9 | 1636 | struct timespec __user *rmtp; |
237fc6e7 | 1637 | int ret = 0; |
10c94ec1 | 1638 | |
ab8177bc | 1639 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1640 | HRTIMER_MODE_ABS); |
cc584b21 | 1641 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1642 | |
c9cb2e3d | 1643 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1644 | goto out; |
10c94ec1 | 1645 | |
029a07e0 | 1646 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1647 | if (rmtp) { |
237fc6e7 | 1648 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1649 | if (ret <= 0) |
237fc6e7 | 1650 | goto out; |
432569bb | 1651 | } |
10c94ec1 | 1652 | |
10c94ec1 | 1653 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1654 | ret = -ERESTART_RESTARTBLOCK; |
1655 | out: | |
1656 | destroy_hrtimer_on_stack(&t.timer); | |
1657 | return ret; | |
10c94ec1 TG |
1658 | } |
1659 | ||
080344b9 | 1660 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1661 | const enum hrtimer_mode mode, const clockid_t clockid) |
1662 | { | |
1663 | struct restart_block *restart; | |
669d7868 | 1664 | struct hrtimer_sleeper t; |
237fc6e7 | 1665 | int ret = 0; |
3bd01206 AV |
1666 | unsigned long slack; |
1667 | ||
1668 | slack = current->timer_slack_ns; | |
1669 | if (rt_task(current)) | |
1670 | slack = 0; | |
10c94ec1 | 1671 | |
237fc6e7 | 1672 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1673 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1674 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1675 | goto out; |
10c94ec1 | 1676 | |
7978672c | 1677 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1678 | if (mode == HRTIMER_MODE_ABS) { |
1679 | ret = -ERESTARTNOHAND; | |
1680 | goto out; | |
1681 | } | |
10c94ec1 | 1682 | |
432569bb | 1683 | if (rmtp) { |
237fc6e7 | 1684 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1685 | if (ret <= 0) |
237fc6e7 | 1686 | goto out; |
432569bb | 1687 | } |
10c94ec1 TG |
1688 | |
1689 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1690 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1691 | restart->nanosleep.clockid = t.timer.base->clockid; |
029a07e0 | 1692 | restart->nanosleep.rmtp = rmtp; |
cc584b21 | 1693 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1694 | |
237fc6e7 TG |
1695 | ret = -ERESTART_RESTARTBLOCK; |
1696 | out: | |
1697 | destroy_hrtimer_on_stack(&t.timer); | |
1698 | return ret; | |
10c94ec1 TG |
1699 | } |
1700 | ||
58fd3aa2 HC |
1701 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1702 | struct timespec __user *, rmtp) | |
6ba1b912 | 1703 | { |
080344b9 | 1704 | struct timespec tu; |
6ba1b912 TG |
1705 | |
1706 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1707 | return -EFAULT; | |
1708 | ||
1709 | if (!timespec_valid(&tu)) | |
1710 | return -EINVAL; | |
1711 | ||
080344b9 | 1712 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1713 | } |
1714 | ||
c0a31329 TG |
1715 | /* |
1716 | * Functions related to boot-time initialization: | |
1717 | */ | |
0ec160dd | 1718 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1719 | { |
3c8aa39d | 1720 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1721 | int i; |
1722 | ||
998adc3d | 1723 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1724 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1725 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1726 | } | |
3c8aa39d | 1727 | |
54cdfdb4 | 1728 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1729 | } |
1730 | ||
1731 | #ifdef CONFIG_HOTPLUG_CPU | |
1732 | ||
ca109491 | 1733 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1734 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1735 | { |
1736 | struct hrtimer *timer; | |
998adc3d | 1737 | struct timerqueue_node *node; |
c0a31329 | 1738 | |
998adc3d JS |
1739 | while ((node = timerqueue_getnext(&old_base->active))) { |
1740 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1741 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1742 | debug_deactivate(timer); |
b00c1a99 TG |
1743 | |
1744 | /* | |
1745 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1746 | * timer could be seen as !active and just vanish away | |
1747 | * under us on another CPU | |
1748 | */ | |
1749 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1750 | timer->base = new_base; |
54cdfdb4 | 1751 | /* |
e3f1d883 TG |
1752 | * Enqueue the timers on the new cpu. This does not |
1753 | * reprogram the event device in case the timer | |
1754 | * expires before the earliest on this CPU, but we run | |
1755 | * hrtimer_interrupt after we migrated everything to | |
1756 | * sort out already expired timers and reprogram the | |
1757 | * event device. | |
54cdfdb4 | 1758 | */ |
a6037b61 | 1759 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1760 | |
b00c1a99 TG |
1761 | /* Clear the migration state bit */ |
1762 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1763 | } |
1764 | } | |
1765 | ||
d5fd43c4 | 1766 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1767 | { |
3c8aa39d | 1768 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1769 | int i; |
c0a31329 | 1770 | |
37810659 | 1771 | BUG_ON(cpu_online(scpu)); |
37810659 | 1772 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1773 | |
1774 | local_irq_disable(); | |
1775 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1776 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1777 | /* |
1778 | * The caller is globally serialized and nobody else | |
1779 | * takes two locks at once, deadlock is not possible. | |
1780 | */ | |
ecb49d1a TG |
1781 | raw_spin_lock(&new_base->lock); |
1782 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1783 | |
3c8aa39d | 1784 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1785 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1786 | &new_base->clock_base[i]); |
c0a31329 TG |
1787 | } |
1788 | ||
ecb49d1a TG |
1789 | raw_spin_unlock(&old_base->lock); |
1790 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1791 | |
731a55ba TG |
1792 | /* Check, if we got expired work to do */ |
1793 | __hrtimer_peek_ahead_timers(); | |
1794 | local_irq_enable(); | |
c0a31329 | 1795 | } |
37810659 | 1796 | |
c0a31329 TG |
1797 | #endif /* CONFIG_HOTPLUG_CPU */ |
1798 | ||
8c78f307 | 1799 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1800 | unsigned long action, void *hcpu) |
1801 | { | |
b2e3c0ad | 1802 | int scpu = (long)hcpu; |
c0a31329 TG |
1803 | |
1804 | switch (action) { | |
1805 | ||
1806 | case CPU_UP_PREPARE: | |
8bb78442 | 1807 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1808 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1809 | break; |
1810 | ||
1811 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1812 | case CPU_DYING: |
1813 | case CPU_DYING_FROZEN: | |
1814 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1815 | break; | |
c0a31329 | 1816 | case CPU_DEAD: |
8bb78442 | 1817 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1818 | { |
37810659 | 1819 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1820 | migrate_hrtimers(scpu); |
c0a31329 | 1821 | break; |
b2e3c0ad | 1822 | } |
c0a31329 TG |
1823 | #endif |
1824 | ||
1825 | default: | |
1826 | break; | |
1827 | } | |
1828 | ||
1829 | return NOTIFY_OK; | |
1830 | } | |
1831 | ||
8c78f307 | 1832 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1833 | .notifier_call = hrtimer_cpu_notify, |
1834 | }; | |
1835 | ||
1836 | void __init hrtimers_init(void) | |
1837 | { | |
1838 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1839 | (void *)(long)smp_processor_id()); | |
1840 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1841 | #ifdef CONFIG_HIGH_RES_TIMERS |
1842 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1843 | #endif | |
c0a31329 TG |
1844 | } |
1845 | ||
7bb67439 | 1846 | /** |
351b3f7a | 1847 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1848 | * @expires: timeout value (ktime_t) |
654c8e0b | 1849 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 | 1850 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
351b3f7a | 1851 | * @clock: timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME |
7bb67439 | 1852 | */ |
351b3f7a CE |
1853 | int __sched |
1854 | schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta, | |
1855 | const enum hrtimer_mode mode, int clock) | |
7bb67439 AV |
1856 | { |
1857 | struct hrtimer_sleeper t; | |
1858 | ||
1859 | /* | |
1860 | * Optimize when a zero timeout value is given. It does not | |
1861 | * matter whether this is an absolute or a relative time. | |
1862 | */ | |
1863 | if (expires && !expires->tv64) { | |
1864 | __set_current_state(TASK_RUNNING); | |
1865 | return 0; | |
1866 | } | |
1867 | ||
1868 | /* | |
43b21013 | 1869 | * A NULL parameter means "infinite" |
7bb67439 AV |
1870 | */ |
1871 | if (!expires) { | |
1872 | schedule(); | |
1873 | __set_current_state(TASK_RUNNING); | |
1874 | return -EINTR; | |
1875 | } | |
1876 | ||
351b3f7a | 1877 | hrtimer_init_on_stack(&t.timer, clock, mode); |
654c8e0b | 1878 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1879 | |
1880 | hrtimer_init_sleeper(&t, current); | |
1881 | ||
cc584b21 | 1882 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1883 | if (!hrtimer_active(&t.timer)) |
1884 | t.task = NULL; | |
1885 | ||
1886 | if (likely(t.task)) | |
1887 | schedule(); | |
1888 | ||
1889 | hrtimer_cancel(&t.timer); | |
1890 | destroy_hrtimer_on_stack(&t.timer); | |
1891 | ||
1892 | __set_current_state(TASK_RUNNING); | |
1893 | ||
1894 | return !t.task ? 0 : -EINTR; | |
1895 | } | |
351b3f7a CE |
1896 | |
1897 | /** | |
1898 | * schedule_hrtimeout_range - sleep until timeout | |
1899 | * @expires: timeout value (ktime_t) | |
1900 | * @delta: slack in expires timeout (ktime_t) | |
1901 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1902 | * | |
1903 | * Make the current task sleep until the given expiry time has | |
1904 | * elapsed. The routine will return immediately unless | |
1905 | * the current task state has been set (see set_current_state()). | |
1906 | * | |
1907 | * The @delta argument gives the kernel the freedom to schedule the | |
1908 | * actual wakeup to a time that is both power and performance friendly. | |
1909 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1910 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1911 | * | |
1912 | * You can set the task state as follows - | |
1913 | * | |
1914 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1915 | * pass before the routine returns. | |
1916 | * | |
1917 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1918 | * delivered to the current task. | |
1919 | * | |
1920 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1921 | * routine returns. | |
1922 | * | |
1923 | * Returns 0 when the timer has expired otherwise -EINTR | |
1924 | */ | |
1925 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, | |
1926 | const enum hrtimer_mode mode) | |
1927 | { | |
1928 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1929 | CLOCK_MONOTONIC); | |
1930 | } | |
654c8e0b AV |
1931 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1932 | ||
1933 | /** | |
1934 | * schedule_hrtimeout - sleep until timeout | |
1935 | * @expires: timeout value (ktime_t) | |
1936 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1937 | * | |
1938 | * Make the current task sleep until the given expiry time has | |
1939 | * elapsed. The routine will return immediately unless | |
1940 | * the current task state has been set (see set_current_state()). | |
1941 | * | |
1942 | * You can set the task state as follows - | |
1943 | * | |
1944 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1945 | * pass before the routine returns. | |
1946 | * | |
1947 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1948 | * delivered to the current task. | |
1949 | * | |
1950 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1951 | * routine returns. | |
1952 | * | |
1953 | * Returns 0 when the timer has expired otherwise -EINTR | |
1954 | */ | |
1955 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1956 | const enum hrtimer_mode mode) | |
1957 | { | |
1958 | return schedule_hrtimeout_range(expires, 0, mode); | |
1959 | } | |
7bb67439 | 1960 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |