]>
Commit | Line | Data |
---|---|---|
35728b82 | 1 | // SPDX-License-Identifier: GPL-2.0 |
c0a31329 | 2 | /* |
3c8aa39d | 3 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> |
79bf2bb3 | 4 | * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar |
54cdfdb4 | 5 | * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner |
c0a31329 TG |
6 | * |
7 | * High-resolution kernel timers | |
8 | * | |
58c5fc2b TG |
9 | * In contrast to the low-resolution timeout API, aka timer wheel, |
10 | * hrtimers provide finer resolution and accuracy depending on system | |
11 | * configuration and capabilities. | |
c0a31329 TG |
12 | * |
13 | * Started by: Thomas Gleixner and Ingo Molnar | |
14 | * | |
15 | * Credits: | |
58c5fc2b | 16 | * Based on the original timer wheel code |
c0a31329 | 17 | * |
66188fae TG |
18 | * Help, testing, suggestions, bugfixes, improvements were |
19 | * provided by: | |
20 | * | |
21 | * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel | |
22 | * et. al. | |
c0a31329 TG |
23 | */ |
24 | ||
25 | #include <linux/cpu.h> | |
9984de1a | 26 | #include <linux/export.h> |
c0a31329 TG |
27 | #include <linux/percpu.h> |
28 | #include <linux/hrtimer.h> | |
29 | #include <linux/notifier.h> | |
30 | #include <linux/syscalls.h> | |
31 | #include <linux/interrupt.h> | |
79bf2bb3 | 32 | #include <linux/tick.h> |
54cdfdb4 | 33 | #include <linux/err.h> |
237fc6e7 | 34 | #include <linux/debugobjects.h> |
174cd4b1 | 35 | #include <linux/sched/signal.h> |
cf4aebc2 | 36 | #include <linux/sched/sysctl.h> |
8bd75c77 | 37 | #include <linux/sched/rt.h> |
aab03e05 | 38 | #include <linux/sched/deadline.h> |
370c9135 | 39 | #include <linux/sched/nohz.h> |
b17b0153 | 40 | #include <linux/sched/debug.h> |
eea08f32 | 41 | #include <linux/timer.h> |
b0f8c44f | 42 | #include <linux/freezer.h> |
edbeda46 | 43 | #include <linux/compat.h> |
c0a31329 | 44 | |
7c0f6ba6 | 45 | #include <linux/uaccess.h> |
c0a31329 | 46 | |
c6a2a177 XG |
47 | #include <trace/events/timer.h> |
48 | ||
c1797baf | 49 | #include "tick-internal.h" |
8b094cd0 | 50 | |
c458b1d1 AMG |
51 | /* |
52 | * Masks for selecting the soft and hard context timers from | |
53 | * cpu_base->active | |
54 | */ | |
55 | #define MASK_SHIFT (HRTIMER_BASE_MONOTONIC_SOFT) | |
56 | #define HRTIMER_ACTIVE_HARD ((1U << MASK_SHIFT) - 1) | |
57 | #define HRTIMER_ACTIVE_SOFT (HRTIMER_ACTIVE_HARD << MASK_SHIFT) | |
58 | #define HRTIMER_ACTIVE_ALL (HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD) | |
59 | ||
c0a31329 TG |
60 | /* |
61 | * The timer bases: | |
7978672c | 62 | * |
571af55a | 63 | * There are more clockids than hrtimer bases. Thus, we index |
e06383db JS |
64 | * into the timer bases by the hrtimer_base_type enum. When trying |
65 | * to reach a base using a clockid, hrtimer_clockid_to_base() | |
66 | * is used to convert from clockid to the proper hrtimer_base_type. | |
c0a31329 | 67 | */ |
54cdfdb4 | 68 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 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, |
3c8aa39d | 77 | }, |
68fa61c0 TG |
78 | { |
79 | .index = HRTIMER_BASE_REALTIME, | |
80 | .clockid = CLOCK_REALTIME, | |
81 | .get_time = &ktime_get_real, | |
68fa61c0 | 82 | }, |
a3ed0e43 TG |
83 | { |
84 | .index = HRTIMER_BASE_BOOTTIME, | |
85 | .clockid = CLOCK_BOOTTIME, | |
86 | .get_time = &ktime_get_boottime, | |
87 | }, | |
90adda98 JS |
88 | { |
89 | .index = HRTIMER_BASE_TAI, | |
90 | .clockid = CLOCK_TAI, | |
91 | .get_time = &ktime_get_clocktai, | |
90adda98 | 92 | }, |
98ecadd4 AMG |
93 | { |
94 | .index = HRTIMER_BASE_MONOTONIC_SOFT, | |
95 | .clockid = CLOCK_MONOTONIC, | |
96 | .get_time = &ktime_get, | |
97 | }, | |
98 | { | |
99 | .index = HRTIMER_BASE_REALTIME_SOFT, | |
100 | .clockid = CLOCK_REALTIME, | |
101 | .get_time = &ktime_get_real, | |
102 | }, | |
a3ed0e43 TG |
103 | { |
104 | .index = HRTIMER_BASE_BOOTTIME_SOFT, | |
105 | .clockid = CLOCK_BOOTTIME, | |
106 | .get_time = &ktime_get_boottime, | |
107 | }, | |
98ecadd4 AMG |
108 | { |
109 | .index = HRTIMER_BASE_TAI_SOFT, | |
110 | .clockid = CLOCK_TAI, | |
111 | .get_time = &ktime_get_clocktai, | |
112 | }, | |
3c8aa39d | 113 | } |
c0a31329 TG |
114 | }; |
115 | ||
942c3c5c | 116 | static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = { |
336a9cde MZ |
117 | /* Make sure we catch unsupported clockids */ |
118 | [0 ... MAX_CLOCKS - 1] = HRTIMER_MAX_CLOCK_BASES, | |
119 | ||
ce31332d TG |
120 | [CLOCK_REALTIME] = HRTIMER_BASE_REALTIME, |
121 | [CLOCK_MONOTONIC] = HRTIMER_BASE_MONOTONIC, | |
a3ed0e43 | 122 | [CLOCK_BOOTTIME] = HRTIMER_BASE_BOOTTIME, |
90adda98 | 123 | [CLOCK_TAI] = HRTIMER_BASE_TAI, |
ce31332d | 124 | }; |
e06383db | 125 | |
c0a31329 TG |
126 | /* |
127 | * Functions and macros which are different for UP/SMP systems are kept in a | |
128 | * single place | |
129 | */ | |
130 | #ifdef CONFIG_SMP | |
131 | ||
887d9dc9 PZ |
132 | /* |
133 | * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base() | |
134 | * such that hrtimer_callback_running() can unconditionally dereference | |
135 | * timer->base->cpu_base | |
136 | */ | |
137 | static struct hrtimer_cpu_base migration_cpu_base = { | |
887d9dc9 PZ |
138 | .clock_base = { { .cpu_base = &migration_cpu_base, }, }, |
139 | }; | |
140 | ||
141 | #define migration_base migration_cpu_base.clock_base[0] | |
142 | ||
c0a31329 TG |
143 | /* |
144 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
145 | * means that all timers which are tied to this base via timer->base are | |
146 | * locked, and the base itself is locked too. | |
147 | * | |
148 | * So __run_timers/migrate_timers can safely modify all timers which could | |
149 | * be found on the lists/queues. | |
150 | * | |
151 | * When the timer's base is locked, and the timer removed from list, it is | |
887d9dc9 PZ |
152 | * possible to set timer->base = &migration_base and drop the lock: the timer |
153 | * remains locked. | |
c0a31329 | 154 | */ |
3c8aa39d TG |
155 | static |
156 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
157 | unsigned long *flags) | |
c0a31329 | 158 | { |
3c8aa39d | 159 | struct hrtimer_clock_base *base; |
c0a31329 TG |
160 | |
161 | for (;;) { | |
162 | base = timer->base; | |
887d9dc9 | 163 | if (likely(base != &migration_base)) { |
ecb49d1a | 164 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
165 | if (likely(base == timer->base)) |
166 | return base; | |
167 | /* The timer has migrated to another CPU: */ | |
ecb49d1a | 168 | raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
169 | } |
170 | cpu_relax(); | |
171 | } | |
172 | } | |
173 | ||
6ff7041d | 174 | /* |
07a9a7ea AMG |
175 | * We do not migrate the timer when it is expiring before the next |
176 | * event on the target cpu. When high resolution is enabled, we cannot | |
177 | * reprogram the target cpu hardware and we would cause it to fire | |
178 | * late. To keep it simple, we handle the high resolution enabled and | |
179 | * disabled case similar. | |
6ff7041d TG |
180 | * |
181 | * Called with cpu_base->lock of target cpu held. | |
182 | */ | |
183 | static int | |
184 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
185 | { | |
6ff7041d TG |
186 | ktime_t expires; |
187 | ||
6ff7041d | 188 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); |
2ac2dccc | 189 | return expires < new_base->cpu_base->expires_next; |
6ff7041d TG |
190 | } |
191 | ||
bc7a34b8 TG |
192 | static inline |
193 | struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base, | |
194 | int pinned) | |
195 | { | |
ae67bada TG |
196 | #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) |
197 | if (static_branch_likely(&timers_migration_enabled) && !pinned) | |
198 | return &per_cpu(hrtimer_bases, get_nohz_timer_target()); | |
199 | #endif | |
662b3e19 | 200 | return base; |
bc7a34b8 | 201 | } |
bc7a34b8 | 202 | |
c0a31329 | 203 | /* |
b48362d8 FW |
204 | * We switch the timer base to a power-optimized selected CPU target, |
205 | * if: | |
206 | * - NO_HZ_COMMON is enabled | |
207 | * - timer migration is enabled | |
208 | * - the timer callback is not running | |
209 | * - the timer is not the first expiring timer on the new target | |
210 | * | |
211 | * If one of the above requirements is not fulfilled we move the timer | |
212 | * to the current CPU or leave it on the previously assigned CPU if | |
213 | * the timer callback is currently running. | |
c0a31329 | 214 | */ |
3c8aa39d | 215 | static inline struct hrtimer_clock_base * |
597d0275 AB |
216 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
217 | int pinned) | |
c0a31329 | 218 | { |
b48362d8 | 219 | struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base; |
3c8aa39d | 220 | struct hrtimer_clock_base *new_base; |
ab8177bc | 221 | int basenum = base->index; |
c0a31329 | 222 | |
b48362d8 FW |
223 | this_cpu_base = this_cpu_ptr(&hrtimer_bases); |
224 | new_cpu_base = get_target_base(this_cpu_base, pinned); | |
eea08f32 | 225 | again: |
e06383db | 226 | new_base = &new_cpu_base->clock_base[basenum]; |
c0a31329 TG |
227 | |
228 | if (base != new_base) { | |
229 | /* | |
6ff7041d | 230 | * We are trying to move timer to new_base. |
c0a31329 TG |
231 | * However we can't change timer's base while it is running, |
232 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
233 | * the event source in the high resolution case. The softirq | |
234 | * code will take care of this when the timer function has | |
235 | * completed. There is no conflict as we hold the lock until | |
236 | * the timer is enqueued. | |
237 | */ | |
54cdfdb4 | 238 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
239 | return base; |
240 | ||
887d9dc9 PZ |
241 | /* See the comment in lock_hrtimer_base() */ |
242 | timer->base = &migration_base; | |
ecb49d1a TG |
243 | raw_spin_unlock(&base->cpu_base->lock); |
244 | raw_spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 245 | |
b48362d8 | 246 | if (new_cpu_base != this_cpu_base && |
bc7a34b8 | 247 | hrtimer_check_target(timer, new_base)) { |
ecb49d1a TG |
248 | raw_spin_unlock(&new_base->cpu_base->lock); |
249 | raw_spin_lock(&base->cpu_base->lock); | |
b48362d8 | 250 | new_cpu_base = this_cpu_base; |
6ff7041d TG |
251 | timer->base = base; |
252 | goto again; | |
eea08f32 | 253 | } |
c0a31329 | 254 | timer->base = new_base; |
012a45e3 | 255 | } else { |
b48362d8 | 256 | if (new_cpu_base != this_cpu_base && |
bc7a34b8 | 257 | hrtimer_check_target(timer, new_base)) { |
b48362d8 | 258 | new_cpu_base = this_cpu_base; |
012a45e3 LM |
259 | goto again; |
260 | } | |
c0a31329 TG |
261 | } |
262 | return new_base; | |
263 | } | |
264 | ||
265 | #else /* CONFIG_SMP */ | |
266 | ||
3c8aa39d | 267 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
268 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
269 | { | |
3c8aa39d | 270 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 271 | |
ecb49d1a | 272 | raw_spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
273 | |
274 | return base; | |
275 | } | |
276 | ||
eea08f32 | 277 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
278 | |
279 | #endif /* !CONFIG_SMP */ | |
280 | ||
281 | /* | |
282 | * Functions for the union type storage format of ktime_t which are | |
283 | * too large for inlining: | |
284 | */ | |
285 | #if BITS_PER_LONG < 64 | |
c0a31329 TG |
286 | /* |
287 | * Divide a ktime value by a nanosecond value | |
288 | */ | |
f7bcb70e | 289 | s64 __ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 290 | { |
c0a31329 | 291 | int sft = 0; |
f7bcb70e JS |
292 | s64 dclc; |
293 | u64 tmp; | |
c0a31329 | 294 | |
900cfa46 | 295 | dclc = ktime_to_ns(kt); |
f7bcb70e JS |
296 | tmp = dclc < 0 ? -dclc : dclc; |
297 | ||
c0a31329 TG |
298 | /* Make sure the divisor is less than 2^32: */ |
299 | while (div >> 32) { | |
300 | sft++; | |
301 | div >>= 1; | |
302 | } | |
f7bcb70e JS |
303 | tmp >>= sft; |
304 | do_div(tmp, (unsigned long) div); | |
305 | return dclc < 0 ? -tmp : tmp; | |
c0a31329 | 306 | } |
8b618628 | 307 | EXPORT_SYMBOL_GPL(__ktime_divns); |
c0a31329 TG |
308 | #endif /* BITS_PER_LONG >= 64 */ |
309 | ||
5a7780e7 TG |
310 | /* |
311 | * Add two ktime values and do a safety check for overflow: | |
312 | */ | |
313 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
314 | { | |
979515c5 | 315 | ktime_t res = ktime_add_unsafe(lhs, rhs); |
5a7780e7 TG |
316 | |
317 | /* | |
318 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
319 | * return to user space in a timespec: | |
320 | */ | |
2456e855 | 321 | if (res < 0 || res < lhs || res < rhs) |
5a7780e7 TG |
322 | res = ktime_set(KTIME_SEC_MAX, 0); |
323 | ||
324 | return res; | |
325 | } | |
326 | ||
8daa21e6 AB |
327 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
328 | ||
237fc6e7 TG |
329 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
330 | ||
331 | static struct debug_obj_descr hrtimer_debug_descr; | |
332 | ||
99777288 SG |
333 | static void *hrtimer_debug_hint(void *addr) |
334 | { | |
335 | return ((struct hrtimer *) addr)->function; | |
336 | } | |
337 | ||
237fc6e7 TG |
338 | /* |
339 | * fixup_init is called when: | |
340 | * - an active object is initialized | |
341 | */ | |
e3252464 | 342 | static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
343 | { |
344 | struct hrtimer *timer = addr; | |
345 | ||
346 | switch (state) { | |
347 | case ODEBUG_STATE_ACTIVE: | |
348 | hrtimer_cancel(timer); | |
349 | debug_object_init(timer, &hrtimer_debug_descr); | |
e3252464 | 350 | return true; |
237fc6e7 | 351 | default: |
e3252464 | 352 | return false; |
237fc6e7 TG |
353 | } |
354 | } | |
355 | ||
356 | /* | |
357 | * fixup_activate is called when: | |
358 | * - an active object is activated | |
b9fdac7f | 359 | * - an unknown non-static object is activated |
237fc6e7 | 360 | */ |
e3252464 | 361 | static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
362 | { |
363 | switch (state) { | |
237fc6e7 TG |
364 | case ODEBUG_STATE_ACTIVE: |
365 | WARN_ON(1); | |
75b710af | 366 | /* fall through */ |
237fc6e7 | 367 | default: |
e3252464 | 368 | return false; |
237fc6e7 TG |
369 | } |
370 | } | |
371 | ||
372 | /* | |
373 | * fixup_free is called when: | |
374 | * - an active object is freed | |
375 | */ | |
e3252464 | 376 | static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state) |
237fc6e7 TG |
377 | { |
378 | struct hrtimer *timer = addr; | |
379 | ||
380 | switch (state) { | |
381 | case ODEBUG_STATE_ACTIVE: | |
382 | hrtimer_cancel(timer); | |
383 | debug_object_free(timer, &hrtimer_debug_descr); | |
e3252464 | 384 | return true; |
237fc6e7 | 385 | default: |
e3252464 | 386 | return false; |
237fc6e7 TG |
387 | } |
388 | } | |
389 | ||
390 | static struct debug_obj_descr hrtimer_debug_descr = { | |
391 | .name = "hrtimer", | |
99777288 | 392 | .debug_hint = hrtimer_debug_hint, |
237fc6e7 TG |
393 | .fixup_init = hrtimer_fixup_init, |
394 | .fixup_activate = hrtimer_fixup_activate, | |
395 | .fixup_free = hrtimer_fixup_free, | |
396 | }; | |
397 | ||
398 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
399 | { | |
400 | debug_object_init(timer, &hrtimer_debug_descr); | |
401 | } | |
402 | ||
5da70160 AMG |
403 | static inline void debug_hrtimer_activate(struct hrtimer *timer, |
404 | enum hrtimer_mode mode) | |
237fc6e7 TG |
405 | { |
406 | debug_object_activate(timer, &hrtimer_debug_descr); | |
407 | } | |
408 | ||
409 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
410 | { | |
411 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
412 | } | |
413 | ||
414 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
415 | { | |
416 | debug_object_free(timer, &hrtimer_debug_descr); | |
417 | } | |
418 | ||
419 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
420 | enum hrtimer_mode mode); | |
421 | ||
422 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
423 | enum hrtimer_mode mode) | |
424 | { | |
425 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
426 | __hrtimer_init(timer, clock_id, mode); | |
427 | } | |
2bc481cf | 428 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
429 | |
430 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
431 | { | |
432 | debug_object_free(timer, &hrtimer_debug_descr); | |
433 | } | |
c08376ac | 434 | EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack); |
237fc6e7 TG |
435 | |
436 | #else | |
5da70160 | 437 | |
237fc6e7 | 438 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } |
5da70160 AMG |
439 | static inline void debug_hrtimer_activate(struct hrtimer *timer, |
440 | enum hrtimer_mode mode) { } | |
237fc6e7 TG |
441 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } |
442 | #endif | |
443 | ||
c6a2a177 XG |
444 | static inline void |
445 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
446 | enum hrtimer_mode mode) | |
447 | { | |
448 | debug_hrtimer_init(timer); | |
449 | trace_hrtimer_init(timer, clockid, mode); | |
450 | } | |
451 | ||
63e2ed36 AMG |
452 | static inline void debug_activate(struct hrtimer *timer, |
453 | enum hrtimer_mode mode) | |
c6a2a177 | 454 | { |
5da70160 | 455 | debug_hrtimer_activate(timer, mode); |
63e2ed36 | 456 | trace_hrtimer_start(timer, mode); |
c6a2a177 XG |
457 | } |
458 | ||
459 | static inline void debug_deactivate(struct hrtimer *timer) | |
460 | { | |
461 | debug_hrtimer_deactivate(timer); | |
462 | trace_hrtimer_cancel(timer); | |
463 | } | |
464 | ||
c272ca58 AMG |
465 | static struct hrtimer_clock_base * |
466 | __next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active) | |
467 | { | |
468 | unsigned int idx; | |
469 | ||
470 | if (!*active) | |
471 | return NULL; | |
472 | ||
473 | idx = __ffs(*active); | |
474 | *active &= ~(1U << idx); | |
475 | ||
476 | return &cpu_base->clock_base[idx]; | |
477 | } | |
478 | ||
479 | #define for_each_active_base(base, cpu_base, active) \ | |
480 | while ((base = __next_base((cpu_base), &(active)))) | |
481 | ||
ad38f596 | 482 | static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base, |
a59855cd | 483 | const struct hrtimer *exclude, |
ad38f596 AMG |
484 | unsigned int active, |
485 | ktime_t expires_next) | |
9bc74919 | 486 | { |
c272ca58 | 487 | struct hrtimer_clock_base *base; |
ad38f596 | 488 | ktime_t expires; |
9bc74919 | 489 | |
c272ca58 | 490 | for_each_active_base(base, cpu_base, active) { |
9bc74919 TG |
491 | struct timerqueue_node *next; |
492 | struct hrtimer *timer; | |
493 | ||
34aee88a | 494 | next = timerqueue_getnext(&base->active); |
9bc74919 | 495 | timer = container_of(next, struct hrtimer, node); |
a59855cd RW |
496 | if (timer == exclude) { |
497 | /* Get to the next timer in the queue. */ | |
7d2f6abb | 498 | next = timerqueue_iterate_next(next); |
a59855cd RW |
499 | if (!next) |
500 | continue; | |
501 | ||
502 | timer = container_of(next, struct hrtimer, node); | |
503 | } | |
9bc74919 | 504 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
2456e855 | 505 | if (expires < expires_next) { |
9bc74919 | 506 | expires_next = expires; |
a59855cd RW |
507 | |
508 | /* Skip cpu_base update if a timer is being excluded. */ | |
509 | if (exclude) | |
510 | continue; | |
511 | ||
5da70160 AMG |
512 | if (timer->is_soft) |
513 | cpu_base->softirq_next_timer = timer; | |
514 | else | |
515 | cpu_base->next_timer = timer; | |
895bdfa7 | 516 | } |
9bc74919 TG |
517 | } |
518 | /* | |
519 | * clock_was_set() might have changed base->offset of any of | |
520 | * the clock bases so the result might be negative. Fix it up | |
521 | * to prevent a false positive in clockevents_program_event(). | |
522 | */ | |
2456e855 TG |
523 | if (expires_next < 0) |
524 | expires_next = 0; | |
9bc74919 TG |
525 | return expires_next; |
526 | } | |
9bc74919 | 527 | |
c458b1d1 AMG |
528 | /* |
529 | * Recomputes cpu_base::*next_timer and returns the earliest expires_next but | |
530 | * does not set cpu_base::*expires_next, that is done by hrtimer_reprogram. | |
531 | * | |
5da70160 AMG |
532 | * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases, |
533 | * those timers will get run whenever the softirq gets handled, at the end of | |
534 | * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases. | |
535 | * | |
536 | * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases. | |
537 | * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual | |
538 | * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD. | |
539 | * | |
c458b1d1 | 540 | * @active_mask must be one of: |
5da70160 | 541 | * - HRTIMER_ACTIVE_ALL, |
c458b1d1 AMG |
542 | * - HRTIMER_ACTIVE_SOFT, or |
543 | * - HRTIMER_ACTIVE_HARD. | |
544 | */ | |
5da70160 AMG |
545 | static ktime_t |
546 | __hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask) | |
ad38f596 | 547 | { |
c458b1d1 | 548 | unsigned int active; |
5da70160 | 549 | struct hrtimer *next_timer = NULL; |
ad38f596 AMG |
550 | ktime_t expires_next = KTIME_MAX; |
551 | ||
5da70160 AMG |
552 | if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) { |
553 | active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT; | |
554 | cpu_base->softirq_next_timer = NULL; | |
a59855cd RW |
555 | expires_next = __hrtimer_next_event_base(cpu_base, NULL, |
556 | active, KTIME_MAX); | |
5da70160 AMG |
557 | |
558 | next_timer = cpu_base->softirq_next_timer; | |
559 | } | |
ad38f596 | 560 | |
5da70160 AMG |
561 | if (active_mask & HRTIMER_ACTIVE_HARD) { |
562 | active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD; | |
563 | cpu_base->next_timer = next_timer; | |
a59855cd RW |
564 | expires_next = __hrtimer_next_event_base(cpu_base, NULL, active, |
565 | expires_next); | |
5da70160 | 566 | } |
ad38f596 AMG |
567 | |
568 | return expires_next; | |
569 | } | |
570 | ||
21d6d52a TG |
571 | static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base) |
572 | { | |
573 | ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset; | |
a3ed0e43 | 574 | ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset; |
21d6d52a TG |
575 | ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset; |
576 | ||
5da70160 | 577 | ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq, |
a3ed0e43 | 578 | offs_real, offs_boot, offs_tai); |
5da70160 AMG |
579 | |
580 | base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real; | |
a3ed0e43 | 581 | base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot; |
5da70160 AMG |
582 | base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai; |
583 | ||
584 | return now; | |
21d6d52a TG |
585 | } |
586 | ||
28bfd18b AMG |
587 | /* |
588 | * Is the high resolution mode active ? | |
589 | */ | |
590 | static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base) | |
591 | { | |
592 | return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ? | |
593 | cpu_base->hres_active : 0; | |
594 | } | |
595 | ||
596 | static inline int hrtimer_hres_active(void) | |
597 | { | |
598 | return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases)); | |
599 | } | |
600 | ||
54cdfdb4 TG |
601 | /* |
602 | * Reprogram the event source with checking both queues for the | |
603 | * next event | |
604 | * Called with interrupts disabled and base->lock held | |
605 | */ | |
7403f41f AC |
606 | static void |
607 | hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal) | |
54cdfdb4 | 608 | { |
21d6d52a TG |
609 | ktime_t expires_next; |
610 | ||
5da70160 AMG |
611 | /* |
612 | * Find the current next expiration time. | |
613 | */ | |
614 | expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL); | |
615 | ||
616 | if (cpu_base->next_timer && cpu_base->next_timer->is_soft) { | |
617 | /* | |
618 | * When the softirq is activated, hrtimer has to be | |
619 | * programmed with the first hard hrtimer because soft | |
620 | * timer interrupt could occur too late. | |
621 | */ | |
622 | if (cpu_base->softirq_activated) | |
623 | expires_next = __hrtimer_get_next_event(cpu_base, | |
624 | HRTIMER_ACTIVE_HARD); | |
625 | else | |
626 | cpu_base->softirq_expires_next = expires_next; | |
627 | } | |
54cdfdb4 | 628 | |
2456e855 | 629 | if (skip_equal && expires_next == cpu_base->expires_next) |
7403f41f AC |
630 | return; |
631 | ||
2456e855 | 632 | cpu_base->expires_next = expires_next; |
7403f41f | 633 | |
6c6c0d5a | 634 | /* |
61bb4bcb AMG |
635 | * If hres is not active, hardware does not have to be |
636 | * reprogrammed yet. | |
637 | * | |
6c6c0d5a SH |
638 | * If a hang was detected in the last timer interrupt then we |
639 | * leave the hang delay active in the hardware. We want the | |
640 | * system to make progress. That also prevents the following | |
641 | * scenario: | |
642 | * T1 expires 50ms from now | |
643 | * T2 expires 5s from now | |
644 | * | |
645 | * T1 is removed, so this code is called and would reprogram | |
646 | * the hardware to 5s from now. Any hrtimer_start after that | |
647 | * will not reprogram the hardware due to hang_detected being | |
648 | * set. So we'd effectivly block all timers until the T2 event | |
649 | * fires. | |
650 | */ | |
61bb4bcb | 651 | if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected) |
6c6c0d5a SH |
652 | return; |
653 | ||
d2540875 | 654 | tick_program_event(cpu_base->expires_next, 1); |
54cdfdb4 TG |
655 | } |
656 | ||
ebba2c72 AMG |
657 | /* High resolution timer related functions */ |
658 | #ifdef CONFIG_HIGH_RES_TIMERS | |
659 | ||
660 | /* | |
661 | * High resolution timer enabled ? | |
662 | */ | |
663 | static bool hrtimer_hres_enabled __read_mostly = true; | |
664 | unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC; | |
665 | EXPORT_SYMBOL_GPL(hrtimer_resolution); | |
666 | ||
667 | /* | |
668 | * Enable / Disable high resolution mode | |
669 | */ | |
670 | static int __init setup_hrtimer_hres(char *str) | |
671 | { | |
672 | return (kstrtobool(str, &hrtimer_hres_enabled) == 0); | |
673 | } | |
674 | ||
675 | __setup("highres=", setup_hrtimer_hres); | |
676 | ||
677 | /* | |
678 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
679 | */ | |
680 | static inline int hrtimer_is_hres_enabled(void) | |
681 | { | |
682 | return hrtimer_hres_enabled; | |
683 | } | |
684 | ||
9ec26907 TG |
685 | /* |
686 | * Retrigger next event is called after clock was set | |
687 | * | |
688 | * Called with interrupts disabled via on_each_cpu() | |
689 | */ | |
690 | static void retrigger_next_event(void *arg) | |
691 | { | |
dc5df73b | 692 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
9ec26907 | 693 | |
851cff8c | 694 | if (!__hrtimer_hres_active(base)) |
9ec26907 TG |
695 | return; |
696 | ||
9ec26907 | 697 | raw_spin_lock(&base->lock); |
5baefd6d | 698 | hrtimer_update_base(base); |
9ec26907 TG |
699 | hrtimer_force_reprogram(base, 0); |
700 | raw_spin_unlock(&base->lock); | |
701 | } | |
b12a03ce | 702 | |
54cdfdb4 TG |
703 | /* |
704 | * Switch to high resolution mode | |
705 | */ | |
75e3b37d | 706 | static void hrtimer_switch_to_hres(void) |
54cdfdb4 | 707 | { |
c6eb3f70 | 708 | struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases); |
54cdfdb4 TG |
709 | |
710 | if (tick_init_highres()) { | |
7a6e5537 GU |
711 | pr_warn("Could not switch to high resolution mode on CPU %u\n", |
712 | base->cpu); | |
85e1cd6e | 713 | return; |
54cdfdb4 TG |
714 | } |
715 | base->hres_active = 1; | |
398ca17f | 716 | hrtimer_resolution = HIGH_RES_NSEC; |
54cdfdb4 TG |
717 | |
718 | tick_setup_sched_timer(); | |
54cdfdb4 TG |
719 | /* "Retrigger" the interrupt to get things going */ |
720 | retrigger_next_event(NULL); | |
54cdfdb4 TG |
721 | } |
722 | ||
5ec2481b TG |
723 | static void clock_was_set_work(struct work_struct *work) |
724 | { | |
725 | clock_was_set(); | |
726 | } | |
727 | ||
728 | static DECLARE_WORK(hrtimer_work, clock_was_set_work); | |
729 | ||
f55a6faa | 730 | /* |
b4d90e9f | 731 | * Called from timekeeping and resume code to reprogram the hrtimer |
5ec2481b | 732 | * interrupt device on all cpus. |
f55a6faa JS |
733 | */ |
734 | void clock_was_set_delayed(void) | |
735 | { | |
5ec2481b | 736 | schedule_work(&hrtimer_work); |
f55a6faa JS |
737 | } |
738 | ||
54cdfdb4 TG |
739 | #else |
740 | ||
54cdfdb4 | 741 | static inline int hrtimer_is_hres_enabled(void) { return 0; } |
75e3b37d | 742 | static inline void hrtimer_switch_to_hres(void) { } |
9ec26907 | 743 | static inline void retrigger_next_event(void *arg) { } |
54cdfdb4 TG |
744 | |
745 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
746 | ||
11a9fe06 AMG |
747 | /* |
748 | * When a timer is enqueued and expires earlier than the already enqueued | |
749 | * timers, we have to check, whether it expires earlier than the timer for | |
750 | * which the clock event device was armed. | |
751 | * | |
752 | * Called with interrupts disabled and base->cpu_base.lock held | |
753 | */ | |
5da70160 | 754 | static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram) |
11a9fe06 AMG |
755 | { |
756 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
3ec7a3ee | 757 | struct hrtimer_clock_base *base = timer->base; |
11a9fe06 AMG |
758 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
759 | ||
760 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); | |
761 | ||
5da70160 AMG |
762 | /* |
763 | * CLOCK_REALTIME timer might be requested with an absolute | |
764 | * expiry time which is less than base->offset. Set it to 0. | |
765 | */ | |
766 | if (expires < 0) | |
767 | expires = 0; | |
768 | ||
769 | if (timer->is_soft) { | |
770 | /* | |
771 | * soft hrtimer could be started on a remote CPU. In this | |
772 | * case softirq_expires_next needs to be updated on the | |
773 | * remote CPU. The soft hrtimer will not expire before the | |
774 | * first hard hrtimer on the remote CPU - | |
775 | * hrtimer_check_target() prevents this case. | |
776 | */ | |
777 | struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base; | |
778 | ||
779 | if (timer_cpu_base->softirq_activated) | |
780 | return; | |
781 | ||
782 | if (!ktime_before(expires, timer_cpu_base->softirq_expires_next)) | |
783 | return; | |
784 | ||
785 | timer_cpu_base->softirq_next_timer = timer; | |
786 | timer_cpu_base->softirq_expires_next = expires; | |
787 | ||
788 | if (!ktime_before(expires, timer_cpu_base->expires_next) || | |
789 | !reprogram) | |
790 | return; | |
791 | } | |
792 | ||
11a9fe06 AMG |
793 | /* |
794 | * If the timer is not on the current cpu, we cannot reprogram | |
795 | * the other cpus clock event device. | |
796 | */ | |
797 | if (base->cpu_base != cpu_base) | |
798 | return; | |
799 | ||
800 | /* | |
801 | * If the hrtimer interrupt is running, then it will | |
802 | * reevaluate the clock bases and reprogram the clock event | |
803 | * device. The callbacks are always executed in hard interrupt | |
804 | * context so we don't need an extra check for a running | |
805 | * callback. | |
806 | */ | |
807 | if (cpu_base->in_hrtirq) | |
808 | return; | |
809 | ||
11a9fe06 AMG |
810 | if (expires >= cpu_base->expires_next) |
811 | return; | |
812 | ||
813 | /* Update the pointer to the next expiring timer */ | |
814 | cpu_base->next_timer = timer; | |
14c80341 | 815 | cpu_base->expires_next = expires; |
11a9fe06 AMG |
816 | |
817 | /* | |
14c80341 AMG |
818 | * If hres is not active, hardware does not have to be |
819 | * programmed yet. | |
820 | * | |
11a9fe06 AMG |
821 | * If a hang was detected in the last timer interrupt then we |
822 | * do not schedule a timer which is earlier than the expiry | |
823 | * which we enforced in the hang detection. We want the system | |
824 | * to make progress. | |
825 | */ | |
14c80341 | 826 | if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected) |
11a9fe06 AMG |
827 | return; |
828 | ||
829 | /* | |
830 | * Program the timer hardware. We enforce the expiry for | |
831 | * events which are already in the past. | |
832 | */ | |
11a9fe06 AMG |
833 | tick_program_event(expires, 1); |
834 | } | |
835 | ||
b12a03ce TG |
836 | /* |
837 | * Clock realtime was set | |
838 | * | |
839 | * Change the offset of the realtime clock vs. the monotonic | |
840 | * clock. | |
841 | * | |
842 | * We might have to reprogram the high resolution timer interrupt. On | |
843 | * SMP we call the architecture specific code to retrigger _all_ high | |
844 | * resolution timer interrupts. On UP we just disable interrupts and | |
845 | * call the high resolution interrupt code. | |
846 | */ | |
847 | void clock_was_set(void) | |
848 | { | |
90ff1f30 | 849 | #ifdef CONFIG_HIGH_RES_TIMERS |
b12a03ce TG |
850 | /* Retrigger the CPU local events everywhere */ |
851 | on_each_cpu(retrigger_next_event, NULL, 1); | |
9ec26907 TG |
852 | #endif |
853 | timerfd_clock_was_set(); | |
b12a03ce TG |
854 | } |
855 | ||
856 | /* | |
857 | * During resume we might have to reprogram the high resolution timer | |
7c4c3a0f DV |
858 | * interrupt on all online CPUs. However, all other CPUs will be |
859 | * stopped with IRQs interrupts disabled so the clock_was_set() call | |
5ec2481b | 860 | * must be deferred. |
b12a03ce TG |
861 | */ |
862 | void hrtimers_resume(void) | |
863 | { | |
53bef3fd | 864 | lockdep_assert_irqs_disabled(); |
5ec2481b | 865 | /* Retrigger on the local CPU */ |
b12a03ce | 866 | retrigger_next_event(NULL); |
5ec2481b TG |
867 | /* And schedule a retrigger for all others */ |
868 | clock_was_set_delayed(); | |
b12a03ce TG |
869 | } |
870 | ||
c0a31329 | 871 | /* |
6506f2aa | 872 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
873 | */ |
874 | static inline | |
875 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
876 | { | |
ecb49d1a | 877 | raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
878 | } |
879 | ||
880 | /** | |
881 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 882 | * @timer: hrtimer to forward |
44f21475 | 883 | * @now: forward past this time |
c0a31329 TG |
884 | * @interval: the interval to forward |
885 | * | |
886 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 887 | * Returns the number of overruns. |
91e5a217 TG |
888 | * |
889 | * Can be safely called from the callback function of @timer. If | |
890 | * called from other contexts @timer must neither be enqueued nor | |
891 | * running the callback and the caller needs to take care of | |
892 | * serialization. | |
893 | * | |
894 | * Note: This only updates the timer expiry value and does not requeue | |
895 | * the timer. | |
c0a31329 | 896 | */ |
4d672e7a | 897 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 898 | { |
4d672e7a | 899 | u64 orun = 1; |
44f21475 | 900 | ktime_t delta; |
c0a31329 | 901 | |
cc584b21 | 902 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 | 903 | |
2456e855 | 904 | if (delta < 0) |
c0a31329 TG |
905 | return 0; |
906 | ||
5de2755c PZ |
907 | if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED)) |
908 | return 0; | |
909 | ||
2456e855 TG |
910 | if (interval < hrtimer_resolution) |
911 | interval = hrtimer_resolution; | |
c9db4fa1 | 912 | |
2456e855 | 913 | if (unlikely(delta >= interval)) { |
df869b63 | 914 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
915 | |
916 | orun = ktime_divns(delta, incr); | |
cc584b21 | 917 | hrtimer_add_expires_ns(timer, incr * orun); |
2456e855 | 918 | if (hrtimer_get_expires_tv64(timer) > now) |
c0a31329 TG |
919 | return orun; |
920 | /* | |
921 | * This (and the ktime_add() below) is the | |
922 | * correction for exact: | |
923 | */ | |
924 | orun++; | |
925 | } | |
cc584b21 | 926 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
927 | |
928 | return orun; | |
929 | } | |
6bdb6b62 | 930 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
931 | |
932 | /* | |
933 | * enqueue_hrtimer - internal function to (re)start a timer | |
934 | * | |
935 | * The timer is inserted in expiry order. Insertion into the | |
936 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
937 | * |
938 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 939 | */ |
a6037b61 | 940 | static int enqueue_hrtimer(struct hrtimer *timer, |
63e2ed36 AMG |
941 | struct hrtimer_clock_base *base, |
942 | enum hrtimer_mode mode) | |
c0a31329 | 943 | { |
63e2ed36 | 944 | debug_activate(timer, mode); |
237fc6e7 | 945 | |
ab8177bc | 946 | base->cpu_base->active_bases |= 1 << base->index; |
54cdfdb4 | 947 | |
eb7b6f32 ED |
948 | /* Pairs with the lockless read in hrtimer_is_queued() */ |
949 | WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED); | |
a6037b61 | 950 | |
b97f44c9 | 951 | return timerqueue_add(&base->active, &timer->node); |
288867ec | 952 | } |
c0a31329 TG |
953 | |
954 | /* | |
955 | * __remove_hrtimer - internal function to remove a timer | |
956 | * | |
957 | * Caller must hold the base lock. | |
54cdfdb4 TG |
958 | * |
959 | * High resolution timer mode reprograms the clock event device when the | |
960 | * timer is the one which expires next. The caller can disable this by setting | |
961 | * reprogram to zero. This is useful, when the context does a reprogramming | |
962 | * anyway (e.g. timer interrupt) | |
c0a31329 | 963 | */ |
3c8aa39d | 964 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 965 | struct hrtimer_clock_base *base, |
203cbf77 | 966 | u8 newstate, int reprogram) |
c0a31329 | 967 | { |
e19ffe8b | 968 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; |
203cbf77 | 969 | u8 state = timer->state; |
e19ffe8b | 970 | |
eb7b6f32 ED |
971 | /* Pairs with the lockless read in hrtimer_is_queued() */ |
972 | WRITE_ONCE(timer->state, newstate); | |
895bdfa7 TG |
973 | if (!(state & HRTIMER_STATE_ENQUEUED)) |
974 | return; | |
7403f41f | 975 | |
b97f44c9 | 976 | if (!timerqueue_del(&base->active, &timer->node)) |
e19ffe8b | 977 | cpu_base->active_bases &= ~(1 << base->index); |
7403f41f | 978 | |
895bdfa7 TG |
979 | /* |
980 | * Note: If reprogram is false we do not update | |
981 | * cpu_base->next_timer. This happens when we remove the first | |
982 | * timer on a remote cpu. No harm as we never dereference | |
983 | * cpu_base->next_timer. So the worst thing what can happen is | |
984 | * an superflous call to hrtimer_force_reprogram() on the | |
985 | * remote cpu later on if the same timer gets enqueued again. | |
986 | */ | |
987 | if (reprogram && timer == cpu_base->next_timer) | |
988 | hrtimer_force_reprogram(cpu_base, 1); | |
c0a31329 TG |
989 | } |
990 | ||
991 | /* | |
992 | * remove hrtimer, called with base lock held | |
993 | */ | |
994 | static inline int | |
8edfb036 | 995 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart) |
c0a31329 | 996 | { |
eb7b6f32 ED |
997 | u8 state = timer->state; |
998 | ||
999 | if (state & HRTIMER_STATE_ENQUEUED) { | |
54cdfdb4 TG |
1000 | int reprogram; |
1001 | ||
1002 | /* | |
1003 | * Remove the timer and force reprogramming when high | |
1004 | * resolution mode is active and the timer is on the current | |
1005 | * CPU. If we remove a timer on another CPU, reprogramming is | |
1006 | * skipped. The interrupt event on this CPU is fired and | |
1007 | * reprogramming happens in the interrupt handler. This is a | |
1008 | * rare case and less expensive than a smp call. | |
1009 | */ | |
c6a2a177 | 1010 | debug_deactivate(timer); |
dc5df73b | 1011 | reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases); |
8edfb036 | 1012 | |
887d9dc9 PZ |
1013 | if (!restart) |
1014 | state = HRTIMER_STATE_INACTIVE; | |
1015 | ||
f13d4f97 | 1016 | __remove_hrtimer(timer, base, state, reprogram); |
c0a31329 TG |
1017 | return 1; |
1018 | } | |
1019 | return 0; | |
1020 | } | |
1021 | ||
203cbf77 TG |
1022 | static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim, |
1023 | const enum hrtimer_mode mode) | |
1024 | { | |
1025 | #ifdef CONFIG_TIME_LOW_RES | |
1026 | /* | |
1027 | * CONFIG_TIME_LOW_RES indicates that the system has no way to return | |
1028 | * granular time values. For relative timers we add hrtimer_resolution | |
1029 | * (i.e. one jiffie) to prevent short timeouts. | |
1030 | */ | |
1031 | timer->is_rel = mode & HRTIMER_MODE_REL; | |
1032 | if (timer->is_rel) | |
8b0e1953 | 1033 | tim = ktime_add_safe(tim, hrtimer_resolution); |
203cbf77 TG |
1034 | #endif |
1035 | return tim; | |
1036 | } | |
1037 | ||
5da70160 AMG |
1038 | static void |
1039 | hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram) | |
1040 | { | |
1041 | ktime_t expires; | |
1042 | ||
1043 | /* | |
1044 | * Find the next SOFT expiration. | |
1045 | */ | |
1046 | expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT); | |
1047 | ||
1048 | /* | |
1049 | * reprogramming needs to be triggered, even if the next soft | |
1050 | * hrtimer expires at the same time than the next hard | |
1051 | * hrtimer. cpu_base->softirq_expires_next needs to be updated! | |
1052 | */ | |
1053 | if (expires == KTIME_MAX) | |
1054 | return; | |
1055 | ||
1056 | /* | |
1057 | * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event() | |
1058 | * cpu_base->*expires_next is only set by hrtimer_reprogram() | |
1059 | */ | |
1060 | hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram); | |
1061 | } | |
1062 | ||
138a6b7a AMG |
1063 | static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
1064 | u64 delta_ns, const enum hrtimer_mode mode, | |
1065 | struct hrtimer_clock_base *base) | |
c0a31329 | 1066 | { |
138a6b7a | 1067 | struct hrtimer_clock_base *new_base; |
c0a31329 TG |
1068 | |
1069 | /* Remove an active timer from the queue: */ | |
8edfb036 | 1070 | remove_hrtimer(timer, base, true); |
c0a31329 | 1071 | |
203cbf77 | 1072 | if (mode & HRTIMER_MODE_REL) |
84ea7fe3 | 1073 | tim = ktime_add_safe(tim, base->get_time()); |
203cbf77 TG |
1074 | |
1075 | tim = hrtimer_update_lowres(timer, tim, mode); | |
237fc6e7 | 1076 | |
da8f2e17 | 1077 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 1078 | |
84ea7fe3 VK |
1079 | /* Switch the timer base, if necessary: */ |
1080 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); | |
1081 | ||
138a6b7a AMG |
1082 | return enqueue_hrtimer(timer, new_base, mode); |
1083 | } | |
5da70160 | 1084 | |
138a6b7a AMG |
1085 | /** |
1086 | * hrtimer_start_range_ns - (re)start an hrtimer | |
1087 | * @timer: the timer to be added | |
1088 | * @tim: expiry time | |
1089 | * @delta_ns: "slack" range for the timer | |
1090 | * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or | |
5da70160 AMG |
1091 | * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED); |
1092 | * softirq based mode is considered for debug purpose only! | |
138a6b7a AMG |
1093 | */ |
1094 | void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
1095 | u64 delta_ns, const enum hrtimer_mode mode) | |
1096 | { | |
1097 | struct hrtimer_clock_base *base; | |
1098 | unsigned long flags; | |
1099 | ||
5da70160 AMG |
1100 | /* |
1101 | * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft | |
1102 | * match. | |
1103 | */ | |
1104 | WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft); | |
1105 | ||
138a6b7a AMG |
1106 | base = lock_hrtimer_base(timer, &flags); |
1107 | ||
1108 | if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base)) | |
5da70160 | 1109 | hrtimer_reprogram(timer, true); |
49a2a075 | 1110 | |
c0a31329 | 1111 | unlock_hrtimer_base(timer, &flags); |
7f1e2ca9 | 1112 | } |
da8f2e17 AV |
1113 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
1114 | ||
c0a31329 TG |
1115 | /** |
1116 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1117 | * @timer: hrtimer to stop |
1118 | * | |
1119 | * Returns: | |
51633704 MCC |
1120 | * |
1121 | * * 0 when the timer was not active | |
1122 | * * 1 when the timer was active | |
1123 | * * -1 when the timer is currently executing the callback function and | |
fa9799e3 | 1124 | * cannot be stopped |
c0a31329 TG |
1125 | */ |
1126 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1127 | { | |
3c8aa39d | 1128 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1129 | unsigned long flags; |
1130 | int ret = -1; | |
1131 | ||
19d9f422 TG |
1132 | /* |
1133 | * Check lockless first. If the timer is not active (neither | |
1134 | * enqueued nor running the callback, nothing to do here. The | |
1135 | * base lock does not serialize against a concurrent enqueue, | |
1136 | * so we can avoid taking it. | |
1137 | */ | |
1138 | if (!hrtimer_active(timer)) | |
1139 | return 0; | |
1140 | ||
c0a31329 TG |
1141 | base = lock_hrtimer_base(timer, &flags); |
1142 | ||
303e967f | 1143 | if (!hrtimer_callback_running(timer)) |
8edfb036 | 1144 | ret = remove_hrtimer(timer, base, false); |
c0a31329 TG |
1145 | |
1146 | unlock_hrtimer_base(timer, &flags); | |
1147 | ||
1148 | return ret; | |
1149 | ||
1150 | } | |
8d16b764 | 1151 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1152 | |
1153 | /** | |
1154 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1155 | * @timer: the timer to be cancelled |
1156 | * | |
1157 | * Returns: | |
1158 | * 0 when the timer was not active | |
1159 | * 1 when the timer was active | |
1160 | */ | |
1161 | int hrtimer_cancel(struct hrtimer *timer) | |
1162 | { | |
1163 | for (;;) { | |
1164 | int ret = hrtimer_try_to_cancel(timer); | |
1165 | ||
1166 | if (ret >= 0) | |
1167 | return ret; | |
5ef37b19 | 1168 | cpu_relax(); |
c0a31329 TG |
1169 | } |
1170 | } | |
8d16b764 | 1171 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1172 | |
1173 | /** | |
1174 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 | 1175 | * @timer: the timer to read |
203cbf77 | 1176 | * @adjust: adjust relative timers when CONFIG_TIME_LOW_RES=y |
c0a31329 | 1177 | */ |
203cbf77 | 1178 | ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust) |
c0a31329 | 1179 | { |
c0a31329 TG |
1180 | unsigned long flags; |
1181 | ktime_t rem; | |
1182 | ||
b3bd3de6 | 1183 | lock_hrtimer_base(timer, &flags); |
203cbf77 TG |
1184 | if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust) |
1185 | rem = hrtimer_expires_remaining_adjusted(timer); | |
1186 | else | |
1187 | rem = hrtimer_expires_remaining(timer); | |
c0a31329 TG |
1188 | unlock_hrtimer_base(timer, &flags); |
1189 | ||
1190 | return rem; | |
1191 | } | |
203cbf77 | 1192 | EXPORT_SYMBOL_GPL(__hrtimer_get_remaining); |
c0a31329 | 1193 | |
3451d024 | 1194 | #ifdef CONFIG_NO_HZ_COMMON |
69239749 TL |
1195 | /** |
1196 | * hrtimer_get_next_event - get the time until next expiry event | |
1197 | * | |
c1ad348b | 1198 | * Returns the next expiry time or KTIME_MAX if no timer is pending. |
69239749 | 1199 | */ |
c1ad348b | 1200 | u64 hrtimer_get_next_event(void) |
69239749 | 1201 | { |
dc5df73b | 1202 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
c1ad348b | 1203 | u64 expires = KTIME_MAX; |
69239749 | 1204 | unsigned long flags; |
69239749 | 1205 | |
ecb49d1a | 1206 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
3c8aa39d | 1207 | |
e19ffe8b | 1208 | if (!__hrtimer_hres_active(cpu_base)) |
5da70160 | 1209 | expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL); |
3c8aa39d | 1210 | |
ecb49d1a | 1211 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
3c8aa39d | 1212 | |
c1ad348b | 1213 | return expires; |
69239749 | 1214 | } |
a59855cd RW |
1215 | |
1216 | /** | |
1217 | * hrtimer_next_event_without - time until next expiry event w/o one timer | |
1218 | * @exclude: timer to exclude | |
1219 | * | |
1220 | * Returns the next expiry time over all timers except for the @exclude one or | |
1221 | * KTIME_MAX if none of them is pending. | |
1222 | */ | |
1223 | u64 hrtimer_next_event_without(const struct hrtimer *exclude) | |
1224 | { | |
1225 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1226 | u64 expires = KTIME_MAX; | |
1227 | unsigned long flags; | |
1228 | ||
1229 | raw_spin_lock_irqsave(&cpu_base->lock, flags); | |
1230 | ||
1231 | if (__hrtimer_hres_active(cpu_base)) { | |
1232 | unsigned int active; | |
1233 | ||
1234 | if (!cpu_base->softirq_activated) { | |
1235 | active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT; | |
1236 | expires = __hrtimer_next_event_base(cpu_base, exclude, | |
1237 | active, KTIME_MAX); | |
1238 | } | |
1239 | active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD; | |
1240 | expires = __hrtimer_next_event_base(cpu_base, exclude, active, | |
1241 | expires); | |
1242 | } | |
1243 | ||
1244 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1245 | ||
1246 | return expires; | |
1247 | } | |
69239749 TL |
1248 | #endif |
1249 | ||
336a9cde MZ |
1250 | static inline int hrtimer_clockid_to_base(clockid_t clock_id) |
1251 | { | |
1252 | if (likely(clock_id < MAX_CLOCKS)) { | |
1253 | int base = hrtimer_clock_to_base_table[clock_id]; | |
1254 | ||
1255 | if (likely(base != HRTIMER_MAX_CLOCK_BASES)) | |
1256 | return base; | |
1257 | } | |
1258 | WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id); | |
1259 | return HRTIMER_BASE_MONOTONIC; | |
1260 | } | |
1261 | ||
237fc6e7 TG |
1262 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1263 | enum hrtimer_mode mode) | |
c0a31329 | 1264 | { |
42f42da4 AMG |
1265 | bool softtimer = !!(mode & HRTIMER_MODE_SOFT); |
1266 | int base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0; | |
3c8aa39d | 1267 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1268 | |
7978672c GA |
1269 | memset(timer, 0, sizeof(struct hrtimer)); |
1270 | ||
22127e93 | 1271 | cpu_base = raw_cpu_ptr(&hrtimer_bases); |
c0a31329 | 1272 | |
48d0c9be AMG |
1273 | /* |
1274 | * POSIX magic: Relative CLOCK_REALTIME timers are not affected by | |
1275 | * clock modifications, so they needs to become CLOCK_MONOTONIC to | |
1276 | * ensure POSIX compliance. | |
1277 | */ | |
1278 | if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL) | |
7978672c GA |
1279 | clock_id = CLOCK_MONOTONIC; |
1280 | ||
42f42da4 AMG |
1281 | base += hrtimer_clockid_to_base(clock_id); |
1282 | timer->is_soft = softtimer; | |
e06383db | 1283 | timer->base = &cpu_base->clock_base[base]; |
998adc3d | 1284 | timerqueue_init(&timer->node); |
c0a31329 | 1285 | } |
237fc6e7 TG |
1286 | |
1287 | /** | |
1288 | * hrtimer_init - initialize a timer to the given clock | |
1289 | * @timer: the timer to be initialized | |
1290 | * @clock_id: the clock to be used | |
42f42da4 AMG |
1291 | * @mode: The modes which are relevant for intitialization: |
1292 | * HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT, | |
1293 | * HRTIMER_MODE_REL_SOFT | |
1294 | * | |
1295 | * The PINNED variants of the above can be handed in, | |
1296 | * but the PINNED bit is ignored as pinning happens | |
1297 | * when the hrtimer is started | |
237fc6e7 TG |
1298 | */ |
1299 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1300 | enum hrtimer_mode mode) | |
1301 | { | |
c6a2a177 | 1302 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1303 | __hrtimer_init(timer, clock_id, mode); |
1304 | } | |
8d16b764 | 1305 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 | 1306 | |
887d9dc9 PZ |
1307 | /* |
1308 | * A timer is active, when it is enqueued into the rbtree or the | |
1309 | * callback function is running or it's in the state of being migrated | |
1310 | * to another cpu. | |
c0a31329 | 1311 | * |
887d9dc9 | 1312 | * It is important for this function to not return a false negative. |
c0a31329 | 1313 | */ |
887d9dc9 | 1314 | bool hrtimer_active(const struct hrtimer *timer) |
c0a31329 | 1315 | { |
3f0b9e8e | 1316 | struct hrtimer_clock_base *base; |
887d9dc9 | 1317 | unsigned int seq; |
c0a31329 | 1318 | |
887d9dc9 | 1319 | do { |
3f0b9e8e AMG |
1320 | base = READ_ONCE(timer->base); |
1321 | seq = raw_read_seqcount_begin(&base->seq); | |
c0a31329 | 1322 | |
887d9dc9 | 1323 | if (timer->state != HRTIMER_STATE_INACTIVE || |
3f0b9e8e | 1324 | base->running == timer) |
887d9dc9 PZ |
1325 | return true; |
1326 | ||
3f0b9e8e AMG |
1327 | } while (read_seqcount_retry(&base->seq, seq) || |
1328 | base != READ_ONCE(timer->base)); | |
887d9dc9 PZ |
1329 | |
1330 | return false; | |
c0a31329 | 1331 | } |
887d9dc9 | 1332 | EXPORT_SYMBOL_GPL(hrtimer_active); |
c0a31329 | 1333 | |
887d9dc9 PZ |
1334 | /* |
1335 | * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3 | |
1336 | * distinct sections: | |
1337 | * | |
1338 | * - queued: the timer is queued | |
1339 | * - callback: the timer is being ran | |
1340 | * - post: the timer is inactive or (re)queued | |
1341 | * | |
1342 | * On the read side we ensure we observe timer->state and cpu_base->running | |
1343 | * from the same section, if anything changed while we looked at it, we retry. | |
1344 | * This includes timer->base changing because sequence numbers alone are | |
1345 | * insufficient for that. | |
1346 | * | |
1347 | * The sequence numbers are required because otherwise we could still observe | |
1348 | * a false negative if the read side got smeared over multiple consequtive | |
1349 | * __run_hrtimer() invocations. | |
1350 | */ | |
1351 | ||
21d6d52a TG |
1352 | static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base, |
1353 | struct hrtimer_clock_base *base, | |
dd934aa8 AMG |
1354 | struct hrtimer *timer, ktime_t *now, |
1355 | unsigned long flags) | |
d3d74453 | 1356 | { |
d3d74453 PZ |
1357 | enum hrtimer_restart (*fn)(struct hrtimer *); |
1358 | int restart; | |
1359 | ||
887d9dc9 | 1360 | lockdep_assert_held(&cpu_base->lock); |
ca109491 | 1361 | |
c6a2a177 | 1362 | debug_deactivate(timer); |
3f0b9e8e | 1363 | base->running = timer; |
887d9dc9 PZ |
1364 | |
1365 | /* | |
1366 | * Separate the ->running assignment from the ->state assignment. | |
1367 | * | |
1368 | * As with a regular write barrier, this ensures the read side in | |
3f0b9e8e | 1369 | * hrtimer_active() cannot observe base->running == NULL && |
887d9dc9 PZ |
1370 | * timer->state == INACTIVE. |
1371 | */ | |
3f0b9e8e | 1372 | raw_write_seqcount_barrier(&base->seq); |
887d9dc9 PZ |
1373 | |
1374 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0); | |
d3d74453 | 1375 | fn = timer->function; |
ca109491 | 1376 | |
203cbf77 TG |
1377 | /* |
1378 | * Clear the 'is relative' flag for the TIME_LOW_RES case. If the | |
1379 | * timer is restarted with a period then it becomes an absolute | |
1380 | * timer. If its not restarted it does not matter. | |
1381 | */ | |
1382 | if (IS_ENABLED(CONFIG_TIME_LOW_RES)) | |
1383 | timer->is_rel = false; | |
1384 | ||
ca109491 | 1385 | /* |
d05ca13b TG |
1386 | * The timer is marked as running in the CPU base, so it is |
1387 | * protected against migration to a different CPU even if the lock | |
1388 | * is dropped. | |
ca109491 | 1389 | */ |
dd934aa8 | 1390 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
c6a2a177 | 1391 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1392 | restart = fn(timer); |
c6a2a177 | 1393 | trace_hrtimer_expire_exit(timer); |
dd934aa8 | 1394 | raw_spin_lock_irq(&cpu_base->lock); |
d3d74453 PZ |
1395 | |
1396 | /* | |
887d9dc9 | 1397 | * Note: We clear the running state after enqueue_hrtimer and |
b4d90e9f | 1398 | * we do not reprogram the event hardware. Happens either in |
e3f1d883 | 1399 | * hrtimer_start_range_ns() or in hrtimer_interrupt() |
5de2755c PZ |
1400 | * |
1401 | * Note: Because we dropped the cpu_base->lock above, | |
1402 | * hrtimer_start_range_ns() can have popped in and enqueued the timer | |
1403 | * for us already. | |
d3d74453 | 1404 | */ |
5de2755c PZ |
1405 | if (restart != HRTIMER_NORESTART && |
1406 | !(timer->state & HRTIMER_STATE_ENQUEUED)) | |
63e2ed36 | 1407 | enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS); |
f13d4f97 | 1408 | |
887d9dc9 PZ |
1409 | /* |
1410 | * Separate the ->running assignment from the ->state assignment. | |
1411 | * | |
1412 | * As with a regular write barrier, this ensures the read side in | |
3f0b9e8e | 1413 | * hrtimer_active() cannot observe base->running.timer == NULL && |
887d9dc9 PZ |
1414 | * timer->state == INACTIVE. |
1415 | */ | |
3f0b9e8e | 1416 | raw_write_seqcount_barrier(&base->seq); |
f13d4f97 | 1417 | |
3f0b9e8e AMG |
1418 | WARN_ON_ONCE(base->running != timer); |
1419 | base->running = NULL; | |
d3d74453 PZ |
1420 | } |
1421 | ||
dd934aa8 | 1422 | static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now, |
c458b1d1 | 1423 | unsigned long flags, unsigned int active_mask) |
54cdfdb4 | 1424 | { |
c272ca58 | 1425 | struct hrtimer_clock_base *base; |
c458b1d1 | 1426 | unsigned int active = cpu_base->active_bases & active_mask; |
6ff7041d | 1427 | |
c272ca58 | 1428 | for_each_active_base(base, cpu_base, active) { |
998adc3d | 1429 | struct timerqueue_node *node; |
ab8177bc TG |
1430 | ktime_t basenow; |
1431 | ||
54cdfdb4 TG |
1432 | basenow = ktime_add(now, base->offset); |
1433 | ||
998adc3d | 1434 | while ((node = timerqueue_getnext(&base->active))) { |
54cdfdb4 TG |
1435 | struct hrtimer *timer; |
1436 | ||
998adc3d | 1437 | timer = container_of(node, struct hrtimer, node); |
54cdfdb4 | 1438 | |
654c8e0b AV |
1439 | /* |
1440 | * The immediate goal for using the softexpires is | |
1441 | * minimizing wakeups, not running timers at the | |
1442 | * earliest interrupt after their soft expiration. | |
1443 | * This allows us to avoid using a Priority Search | |
1444 | * Tree, which can answer a stabbing querry for | |
1445 | * overlapping intervals and instead use the simple | |
1446 | * BST we already have. | |
1447 | * We don't add extra wakeups by delaying timers that | |
1448 | * are right-of a not yet expired timer, because that | |
1449 | * timer will have to trigger a wakeup anyway. | |
1450 | */ | |
2456e855 | 1451 | if (basenow < hrtimer_get_softexpires_tv64(timer)) |
54cdfdb4 | 1452 | break; |
54cdfdb4 | 1453 | |
dd934aa8 | 1454 | __run_hrtimer(cpu_base, base, timer, &basenow, flags); |
54cdfdb4 | 1455 | } |
54cdfdb4 | 1456 | } |
21d6d52a TG |
1457 | } |
1458 | ||
5da70160 AMG |
1459 | static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h) |
1460 | { | |
1461 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1462 | unsigned long flags; | |
1463 | ktime_t now; | |
1464 | ||
1465 | raw_spin_lock_irqsave(&cpu_base->lock, flags); | |
1466 | ||
1467 | now = hrtimer_update_base(cpu_base); | |
1468 | __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT); | |
1469 | ||
1470 | cpu_base->softirq_activated = 0; | |
1471 | hrtimer_update_softirq_timer(cpu_base, true); | |
1472 | ||
1473 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1474 | } | |
1475 | ||
21d6d52a TG |
1476 | #ifdef CONFIG_HIGH_RES_TIMERS |
1477 | ||
1478 | /* | |
1479 | * High resolution timer interrupt | |
1480 | * Called with interrupts disabled | |
1481 | */ | |
1482 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1483 | { | |
1484 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); | |
1485 | ktime_t expires_next, now, entry_time, delta; | |
dd934aa8 | 1486 | unsigned long flags; |
21d6d52a TG |
1487 | int retries = 0; |
1488 | ||
1489 | BUG_ON(!cpu_base->hres_active); | |
1490 | cpu_base->nr_events++; | |
2456e855 | 1491 | dev->next_event = KTIME_MAX; |
21d6d52a | 1492 | |
dd934aa8 | 1493 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
21d6d52a TG |
1494 | entry_time = now = hrtimer_update_base(cpu_base); |
1495 | retry: | |
1496 | cpu_base->in_hrtirq = 1; | |
1497 | /* | |
1498 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1499 | * held to prevent that a timer is enqueued in our queue via | |
1500 | * the migration code. This does not affect enqueueing of | |
1501 | * timers which run their callback and need to be requeued on | |
1502 | * this CPU. | |
1503 | */ | |
2456e855 | 1504 | cpu_base->expires_next = KTIME_MAX; |
21d6d52a | 1505 | |
5da70160 AMG |
1506 | if (!ktime_before(now, cpu_base->softirq_expires_next)) { |
1507 | cpu_base->softirq_expires_next = KTIME_MAX; | |
1508 | cpu_base->softirq_activated = 1; | |
1509 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1510 | } | |
1511 | ||
c458b1d1 | 1512 | __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD); |
21d6d52a | 1513 | |
9bc74919 | 1514 | /* Reevaluate the clock bases for the next expiry */ |
5da70160 | 1515 | expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL); |
6ff7041d TG |
1516 | /* |
1517 | * Store the new expiry value so the migration code can verify | |
1518 | * against it. | |
1519 | */ | |
54cdfdb4 | 1520 | cpu_base->expires_next = expires_next; |
9bc74919 | 1521 | cpu_base->in_hrtirq = 0; |
dd934aa8 | 1522 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
54cdfdb4 TG |
1523 | |
1524 | /* Reprogramming necessary ? */ | |
d2540875 | 1525 | if (!tick_program_event(expires_next, 0)) { |
41d2e494 TG |
1526 | cpu_base->hang_detected = 0; |
1527 | return; | |
54cdfdb4 | 1528 | } |
41d2e494 TG |
1529 | |
1530 | /* | |
1531 | * The next timer was already expired due to: | |
1532 | * - tracing | |
1533 | * - long lasting callbacks | |
1534 | * - being scheduled away when running in a VM | |
1535 | * | |
1536 | * We need to prevent that we loop forever in the hrtimer | |
1537 | * interrupt routine. We give it 3 attempts to avoid | |
1538 | * overreacting on some spurious event. | |
5baefd6d JS |
1539 | * |
1540 | * Acquire base lock for updating the offsets and retrieving | |
1541 | * the current time. | |
41d2e494 | 1542 | */ |
dd934aa8 | 1543 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
5baefd6d | 1544 | now = hrtimer_update_base(cpu_base); |
41d2e494 TG |
1545 | cpu_base->nr_retries++; |
1546 | if (++retries < 3) | |
1547 | goto retry; | |
1548 | /* | |
1549 | * Give the system a chance to do something else than looping | |
1550 | * here. We stored the entry time, so we know exactly how long | |
1551 | * we spent here. We schedule the next event this amount of | |
1552 | * time away. | |
1553 | */ | |
1554 | cpu_base->nr_hangs++; | |
1555 | cpu_base->hang_detected = 1; | |
dd934aa8 AMG |
1556 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
1557 | ||
41d2e494 | 1558 | delta = ktime_sub(now, entry_time); |
2456e855 TG |
1559 | if ((unsigned int)delta > cpu_base->max_hang_time) |
1560 | cpu_base->max_hang_time = (unsigned int) delta; | |
41d2e494 TG |
1561 | /* |
1562 | * Limit it to a sensible value as we enforce a longer | |
1563 | * delay. Give the CPU at least 100ms to catch up. | |
1564 | */ | |
2456e855 | 1565 | if (delta > 100 * NSEC_PER_MSEC) |
41d2e494 TG |
1566 | expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC); |
1567 | else | |
1568 | expires_next = ktime_add(now, delta); | |
1569 | tick_program_event(expires_next, 1); | |
7a6e5537 | 1570 | pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta)); |
54cdfdb4 TG |
1571 | } |
1572 | ||
016da201 | 1573 | /* called with interrupts disabled */ |
c6eb3f70 | 1574 | static inline void __hrtimer_peek_ahead_timers(void) |
8bdec955 TG |
1575 | { |
1576 | struct tick_device *td; | |
1577 | ||
1578 | if (!hrtimer_hres_active()) | |
1579 | return; | |
1580 | ||
22127e93 | 1581 | td = this_cpu_ptr(&tick_cpu_device); |
8bdec955 TG |
1582 | if (td && td->evtdev) |
1583 | hrtimer_interrupt(td->evtdev); | |
1584 | } | |
1585 | ||
82c5b7b5 IM |
1586 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1587 | ||
1588 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1589 | ||
1590 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1591 | |
d3d74453 | 1592 | /* |
c6eb3f70 | 1593 | * Called from run_local_timers in hardirq context every jiffy |
d3d74453 | 1594 | */ |
833883d9 | 1595 | void hrtimer_run_queues(void) |
d3d74453 | 1596 | { |
dc5df73b | 1597 | struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases); |
dd934aa8 | 1598 | unsigned long flags; |
21d6d52a | 1599 | ktime_t now; |
c0a31329 | 1600 | |
e19ffe8b | 1601 | if (__hrtimer_hres_active(cpu_base)) |
d3d74453 | 1602 | return; |
54cdfdb4 | 1603 | |
d3d74453 | 1604 | /* |
c6eb3f70 TG |
1605 | * This _is_ ugly: We have to check periodically, whether we |
1606 | * can switch to highres and / or nohz mode. The clocksource | |
1607 | * switch happens with xtime_lock held. Notification from | |
1608 | * there only sets the check bit in the tick_oneshot code, | |
1609 | * otherwise we might deadlock vs. xtime_lock. | |
d3d74453 | 1610 | */ |
c6eb3f70 | 1611 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) { |
d3d74453 | 1612 | hrtimer_switch_to_hres(); |
3055adda | 1613 | return; |
833883d9 | 1614 | } |
c6eb3f70 | 1615 | |
dd934aa8 | 1616 | raw_spin_lock_irqsave(&cpu_base->lock, flags); |
21d6d52a | 1617 | now = hrtimer_update_base(cpu_base); |
5da70160 AMG |
1618 | |
1619 | if (!ktime_before(now, cpu_base->softirq_expires_next)) { | |
1620 | cpu_base->softirq_expires_next = KTIME_MAX; | |
1621 | cpu_base->softirq_activated = 1; | |
1622 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
1623 | } | |
1624 | ||
c458b1d1 | 1625 | __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD); |
dd934aa8 | 1626 | raw_spin_unlock_irqrestore(&cpu_base->lock, flags); |
c0a31329 TG |
1627 | } |
1628 | ||
10c94ec1 TG |
1629 | /* |
1630 | * Sleep related functions: | |
1631 | */ | |
c9cb2e3d | 1632 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1633 | { |
1634 | struct hrtimer_sleeper *t = | |
1635 | container_of(timer, struct hrtimer_sleeper, timer); | |
1636 | struct task_struct *task = t->task; | |
1637 | ||
1638 | t->task = NULL; | |
1639 | if (task) | |
1640 | wake_up_process(task); | |
1641 | ||
1642 | return HRTIMER_NORESTART; | |
1643 | } | |
1644 | ||
36c8b586 | 1645 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1646 | { |
1647 | sl->timer.function = hrtimer_wakeup; | |
1648 | sl->task = task; | |
1649 | } | |
2bc481cf | 1650 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1651 | |
c0edd7c9 | 1652 | int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts) |
ce41aaf4 AV |
1653 | { |
1654 | switch(restart->nanosleep.type) { | |
0fe27955 | 1655 | #ifdef CONFIG_COMPAT_32BIT_TIME |
ce41aaf4 | 1656 | case TT_COMPAT: |
9afc5eee | 1657 | if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp)) |
ce41aaf4 AV |
1658 | return -EFAULT; |
1659 | break; | |
1660 | #endif | |
1661 | case TT_NATIVE: | |
c0edd7c9 | 1662 | if (put_timespec64(ts, restart->nanosleep.rmtp)) |
ce41aaf4 AV |
1663 | return -EFAULT; |
1664 | break; | |
1665 | default: | |
1666 | BUG(); | |
1667 | } | |
1668 | return -ERESTART_RESTARTBLOCK; | |
1669 | } | |
1670 | ||
669d7868 | 1671 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1672 | { |
edbeda46 AV |
1673 | struct restart_block *restart; |
1674 | ||
669d7868 | 1675 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1676 | |
432569bb RZ |
1677 | do { |
1678 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1679 | hrtimer_start_expires(&t->timer, mode); |
432569bb | 1680 | |
54cdfdb4 | 1681 | if (likely(t->task)) |
b0f8c44f | 1682 | freezable_schedule(); |
432569bb | 1683 | |
669d7868 | 1684 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1685 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1686 | |
1687 | } while (t->task && !signal_pending(current)); | |
432569bb | 1688 | |
3588a085 PZ |
1689 | __set_current_state(TASK_RUNNING); |
1690 | ||
a7602681 | 1691 | if (!t->task) |
080344b9 | 1692 | return 0; |
080344b9 | 1693 | |
edbeda46 AV |
1694 | restart = ¤t->restart_block; |
1695 | if (restart->nanosleep.type != TT_NONE) { | |
a7602681 | 1696 | ktime_t rem = hrtimer_expires_remaining(&t->timer); |
c0edd7c9 | 1697 | struct timespec64 rmt; |
edbeda46 | 1698 | |
a7602681 AV |
1699 | if (rem <= 0) |
1700 | return 0; | |
c0edd7c9 | 1701 | rmt = ktime_to_timespec64(rem); |
a7602681 | 1702 | |
ce41aaf4 | 1703 | return nanosleep_copyout(restart, &rmt); |
a7602681 AV |
1704 | } |
1705 | return -ERESTART_RESTARTBLOCK; | |
080344b9 ON |
1706 | } |
1707 | ||
fb923c4a | 1708 | static long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1709 | { |
669d7868 | 1710 | struct hrtimer_sleeper t; |
a7602681 | 1711 | int ret; |
10c94ec1 | 1712 | |
ab8177bc | 1713 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid, |
237fc6e7 | 1714 | HRTIMER_MODE_ABS); |
cc584b21 | 1715 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1716 | |
a7602681 | 1717 | ret = do_nanosleep(&t, HRTIMER_MODE_ABS); |
237fc6e7 TG |
1718 | destroy_hrtimer_on_stack(&t.timer); |
1719 | return ret; | |
10c94ec1 TG |
1720 | } |
1721 | ||
938e7cf2 | 1722 | long hrtimer_nanosleep(const struct timespec64 *rqtp, |
10c94ec1 TG |
1723 | const enum hrtimer_mode mode, const clockid_t clockid) |
1724 | { | |
a7602681 | 1725 | struct restart_block *restart; |
669d7868 | 1726 | struct hrtimer_sleeper t; |
237fc6e7 | 1727 | int ret = 0; |
da8b44d5 | 1728 | u64 slack; |
3bd01206 AV |
1729 | |
1730 | slack = current->timer_slack_ns; | |
aab03e05 | 1731 | if (dl_task(current) || rt_task(current)) |
3bd01206 | 1732 | slack = 0; |
10c94ec1 | 1733 | |
237fc6e7 | 1734 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
ad196384 | 1735 | hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack); |
a7602681 AV |
1736 | ret = do_nanosleep(&t, mode); |
1737 | if (ret != -ERESTART_RESTARTBLOCK) | |
237fc6e7 | 1738 | goto out; |
10c94ec1 | 1739 | |
7978672c | 1740 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1741 | if (mode == HRTIMER_MODE_ABS) { |
1742 | ret = -ERESTARTNOHAND; | |
1743 | goto out; | |
1744 | } | |
10c94ec1 | 1745 | |
a7602681 | 1746 | restart = ¤t->restart_block; |
1711ef38 | 1747 | restart->fn = hrtimer_nanosleep_restart; |
ab8177bc | 1748 | restart->nanosleep.clockid = t.timer.base->clockid; |
cc584b21 | 1749 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
237fc6e7 TG |
1750 | out: |
1751 | destroy_hrtimer_on_stack(&t.timer); | |
1752 | return ret; | |
10c94ec1 TG |
1753 | } |
1754 | ||
01909974 DD |
1755 | #if !defined(CONFIG_64BIT_TIME) || defined(CONFIG_64BIT) |
1756 | ||
1757 | SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp, | |
1758 | struct __kernel_timespec __user *, rmtp) | |
6ba1b912 | 1759 | { |
c0edd7c9 | 1760 | struct timespec64 tu; |
6ba1b912 | 1761 | |
c0edd7c9 | 1762 | if (get_timespec64(&tu, rqtp)) |
6ba1b912 TG |
1763 | return -EFAULT; |
1764 | ||
c0edd7c9 | 1765 | if (!timespec64_valid(&tu)) |
6ba1b912 TG |
1766 | return -EINVAL; |
1767 | ||
edbeda46 | 1768 | current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE; |
192a82f9 | 1769 | current->restart_block.nanosleep.rmtp = rmtp; |
c0edd7c9 | 1770 | return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1771 | } |
1772 | ||
01909974 DD |
1773 | #endif |
1774 | ||
b5793b0d | 1775 | #ifdef CONFIG_COMPAT_32BIT_TIME |
edbeda46 | 1776 | |
8dabe724 | 1777 | SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp, |
9afc5eee | 1778 | struct old_timespec32 __user *, rmtp) |
edbeda46 | 1779 | { |
c0edd7c9 | 1780 | struct timespec64 tu; |
edbeda46 | 1781 | |
9afc5eee | 1782 | if (get_old_timespec32(&tu, rqtp)) |
edbeda46 AV |
1783 | return -EFAULT; |
1784 | ||
c0edd7c9 | 1785 | if (!timespec64_valid(&tu)) |
edbeda46 AV |
1786 | return -EINVAL; |
1787 | ||
1788 | current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE; | |
1789 | current->restart_block.nanosleep.compat_rmtp = rmtp; | |
c0edd7c9 | 1790 | return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
edbeda46 AV |
1791 | } |
1792 | #endif | |
1793 | ||
c0a31329 TG |
1794 | /* |
1795 | * Functions related to boot-time initialization: | |
1796 | */ | |
27590dc1 | 1797 | int hrtimers_prepare_cpu(unsigned int cpu) |
c0a31329 | 1798 | { |
3c8aa39d | 1799 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1800 | int i; |
1801 | ||
998adc3d | 1802 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
3c8aa39d | 1803 | cpu_base->clock_base[i].cpu_base = cpu_base; |
998adc3d JS |
1804 | timerqueue_init_head(&cpu_base->clock_base[i].active); |
1805 | } | |
3c8aa39d | 1806 | |
cddd0248 | 1807 | cpu_base->cpu = cpu; |
303c146d | 1808 | cpu_base->active_bases = 0; |
28bfd18b | 1809 | cpu_base->hres_active = 0; |
303c146d TG |
1810 | cpu_base->hang_detected = 0; |
1811 | cpu_base->next_timer = NULL; | |
1812 | cpu_base->softirq_next_timer = NULL; | |
07a9a7ea | 1813 | cpu_base->expires_next = KTIME_MAX; |
5da70160 | 1814 | cpu_base->softirq_expires_next = KTIME_MAX; |
27590dc1 | 1815 | return 0; |
c0a31329 TG |
1816 | } |
1817 | ||
1818 | #ifdef CONFIG_HOTPLUG_CPU | |
1819 | ||
ca109491 | 1820 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1821 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1822 | { |
1823 | struct hrtimer *timer; | |
998adc3d | 1824 | struct timerqueue_node *node; |
c0a31329 | 1825 | |
998adc3d JS |
1826 | while ((node = timerqueue_getnext(&old_base->active))) { |
1827 | timer = container_of(node, struct hrtimer, node); | |
54cdfdb4 | 1828 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1829 | debug_deactivate(timer); |
b00c1a99 TG |
1830 | |
1831 | /* | |
c04dca02 | 1832 | * Mark it as ENQUEUED not INACTIVE otherwise the |
b00c1a99 TG |
1833 | * timer could be seen as !active and just vanish away |
1834 | * under us on another CPU | |
1835 | */ | |
c04dca02 | 1836 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0); |
c0a31329 | 1837 | timer->base = new_base; |
54cdfdb4 | 1838 | /* |
e3f1d883 TG |
1839 | * Enqueue the timers on the new cpu. This does not |
1840 | * reprogram the event device in case the timer | |
1841 | * expires before the earliest on this CPU, but we run | |
1842 | * hrtimer_interrupt after we migrated everything to | |
1843 | * sort out already expired timers and reprogram the | |
1844 | * event device. | |
54cdfdb4 | 1845 | */ |
63e2ed36 | 1846 | enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS); |
c0a31329 TG |
1847 | } |
1848 | } | |
1849 | ||
27590dc1 | 1850 | int hrtimers_dead_cpu(unsigned int scpu) |
c0a31329 | 1851 | { |
3c8aa39d | 1852 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1853 | int i; |
c0a31329 | 1854 | |
37810659 | 1855 | BUG_ON(cpu_online(scpu)); |
37810659 | 1856 | tick_cancel_sched_timer(scpu); |
731a55ba | 1857 | |
5da70160 AMG |
1858 | /* |
1859 | * this BH disable ensures that raise_softirq_irqoff() does | |
1860 | * not wakeup ksoftirqd (and acquire the pi-lock) while | |
1861 | * holding the cpu_base lock | |
1862 | */ | |
1863 | local_bh_disable(); | |
731a55ba TG |
1864 | local_irq_disable(); |
1865 | old_base = &per_cpu(hrtimer_bases, scpu); | |
dc5df73b | 1866 | new_base = this_cpu_ptr(&hrtimer_bases); |
d82f0b0f ON |
1867 | /* |
1868 | * The caller is globally serialized and nobody else | |
1869 | * takes two locks at once, deadlock is not possible. | |
1870 | */ | |
ecb49d1a TG |
1871 | raw_spin_lock(&new_base->lock); |
1872 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); | |
c0a31329 | 1873 | |
3c8aa39d | 1874 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1875 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1876 | &new_base->clock_base[i]); |
c0a31329 TG |
1877 | } |
1878 | ||
5da70160 AMG |
1879 | /* |
1880 | * The migration might have changed the first expiring softirq | |
1881 | * timer on this CPU. Update it. | |
1882 | */ | |
1883 | hrtimer_update_softirq_timer(new_base, false); | |
1884 | ||
ecb49d1a TG |
1885 | raw_spin_unlock(&old_base->lock); |
1886 | raw_spin_unlock(&new_base->lock); | |
37810659 | 1887 | |
731a55ba TG |
1888 | /* Check, if we got expired work to do */ |
1889 | __hrtimer_peek_ahead_timers(); | |
1890 | local_irq_enable(); | |
5da70160 | 1891 | local_bh_enable(); |
27590dc1 | 1892 | return 0; |
c0a31329 | 1893 | } |
37810659 | 1894 | |
c0a31329 TG |
1895 | #endif /* CONFIG_HOTPLUG_CPU */ |
1896 | ||
c0a31329 TG |
1897 | void __init hrtimers_init(void) |
1898 | { | |
27590dc1 | 1899 | hrtimers_prepare_cpu(smp_processor_id()); |
5da70160 | 1900 | open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq); |
c0a31329 TG |
1901 | } |
1902 | ||
7bb67439 | 1903 | /** |
351b3f7a | 1904 | * schedule_hrtimeout_range_clock - sleep until timeout |
7bb67439 | 1905 | * @expires: timeout value (ktime_t) |
654c8e0b | 1906 | * @delta: slack in expires timeout (ktime_t) |
90777713 AMG |
1907 | * @mode: timer mode |
1908 | * @clock_id: timer clock to be used | |
7bb67439 | 1909 | */ |
351b3f7a | 1910 | int __sched |
da8b44d5 | 1911 | schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta, |
90777713 | 1912 | const enum hrtimer_mode mode, clockid_t clock_id) |
7bb67439 AV |
1913 | { |
1914 | struct hrtimer_sleeper t; | |
1915 | ||
1916 | /* | |
1917 | * Optimize when a zero timeout value is given. It does not | |
1918 | * matter whether this is an absolute or a relative time. | |
1919 | */ | |
2456e855 | 1920 | if (expires && *expires == 0) { |
7bb67439 AV |
1921 | __set_current_state(TASK_RUNNING); |
1922 | return 0; | |
1923 | } | |
1924 | ||
1925 | /* | |
43b21013 | 1926 | * A NULL parameter means "infinite" |
7bb67439 AV |
1927 | */ |
1928 | if (!expires) { | |
1929 | schedule(); | |
7bb67439 AV |
1930 | return -EINTR; |
1931 | } | |
1932 | ||
90777713 | 1933 | hrtimer_init_on_stack(&t.timer, clock_id, mode); |
654c8e0b | 1934 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1935 | |
1936 | hrtimer_init_sleeper(&t, current); | |
1937 | ||
cc584b21 | 1938 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1939 | |
1940 | if (likely(t.task)) | |
1941 | schedule(); | |
1942 | ||
1943 | hrtimer_cancel(&t.timer); | |
1944 | destroy_hrtimer_on_stack(&t.timer); | |
1945 | ||
1946 | __set_current_state(TASK_RUNNING); | |
1947 | ||
1948 | return !t.task ? 0 : -EINTR; | |
1949 | } | |
351b3f7a CE |
1950 | |
1951 | /** | |
1952 | * schedule_hrtimeout_range - sleep until timeout | |
1953 | * @expires: timeout value (ktime_t) | |
1954 | * @delta: slack in expires timeout (ktime_t) | |
90777713 | 1955 | * @mode: timer mode |
351b3f7a CE |
1956 | * |
1957 | * Make the current task sleep until the given expiry time has | |
1958 | * elapsed. The routine will return immediately unless | |
1959 | * the current task state has been set (see set_current_state()). | |
1960 | * | |
1961 | * The @delta argument gives the kernel the freedom to schedule the | |
1962 | * actual wakeup to a time that is both power and performance friendly. | |
1963 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1964 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1965 | * | |
1966 | * You can set the task state as follows - | |
1967 | * | |
1968 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
4b7e9cf9 DA |
1969 | * pass before the routine returns unless the current task is explicitly |
1970 | * woken up, (e.g. by wake_up_process()). | |
351b3f7a CE |
1971 | * |
1972 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
4b7e9cf9 DA |
1973 | * delivered to the current task or the current task is explicitly woken |
1974 | * up. | |
351b3f7a CE |
1975 | * |
1976 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1977 | * routine returns. | |
1978 | * | |
4b7e9cf9 DA |
1979 | * Returns 0 when the timer has expired. If the task was woken before the |
1980 | * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or | |
1981 | * by an explicit wakeup, it returns -EINTR. | |
351b3f7a | 1982 | */ |
da8b44d5 | 1983 | int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta, |
351b3f7a CE |
1984 | const enum hrtimer_mode mode) |
1985 | { | |
1986 | return schedule_hrtimeout_range_clock(expires, delta, mode, | |
1987 | CLOCK_MONOTONIC); | |
1988 | } | |
654c8e0b AV |
1989 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1990 | ||
1991 | /** | |
1992 | * schedule_hrtimeout - sleep until timeout | |
1993 | * @expires: timeout value (ktime_t) | |
90777713 | 1994 | * @mode: timer mode |
654c8e0b AV |
1995 | * |
1996 | * Make the current task sleep until the given expiry time has | |
1997 | * elapsed. The routine will return immediately unless | |
1998 | * the current task state has been set (see set_current_state()). | |
1999 | * | |
2000 | * You can set the task state as follows - | |
2001 | * | |
2002 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
4b7e9cf9 DA |
2003 | * pass before the routine returns unless the current task is explicitly |
2004 | * woken up, (e.g. by wake_up_process()). | |
654c8e0b AV |
2005 | * |
2006 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
4b7e9cf9 DA |
2007 | * delivered to the current task or the current task is explicitly woken |
2008 | * up. | |
654c8e0b AV |
2009 | * |
2010 | * The current task state is guaranteed to be TASK_RUNNING when this | |
2011 | * routine returns. | |
2012 | * | |
4b7e9cf9 DA |
2013 | * Returns 0 when the timer has expired. If the task was woken before the |
2014 | * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or | |
2015 | * by an explicit wakeup, it returns -EINTR. | |
654c8e0b AV |
2016 | */ |
2017 | int __sched schedule_hrtimeout(ktime_t *expires, | |
2018 | const enum hrtimer_mode mode) | |
2019 | { | |
2020 | return schedule_hrtimeout_range(expires, 0, mode); | |
2021 | } | |
7bb67439 | 2022 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |