diff options
Diffstat (limited to 'kernel/time/timekeeping.c')
-rw-r--r-- | kernel/time/timekeeping.c | 536 |
1 files changed, 402 insertions, 134 deletions
diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c index e8c77d9c633a..c3a4e2907eaa 100644 --- a/kernel/time/timekeeping.c +++ b/kernel/time/timekeeping.c | |||
@@ -13,12 +13,123 @@ | |||
13 | #include <linux/percpu.h> | 13 | #include <linux/percpu.h> |
14 | #include <linux/init.h> | 14 | #include <linux/init.h> |
15 | #include <linux/mm.h> | 15 | #include <linux/mm.h> |
16 | #include <linux/sched.h> | ||
16 | #include <linux/sysdev.h> | 17 | #include <linux/sysdev.h> |
17 | #include <linux/clocksource.h> | 18 | #include <linux/clocksource.h> |
18 | #include <linux/jiffies.h> | 19 | #include <linux/jiffies.h> |
19 | #include <linux/time.h> | 20 | #include <linux/time.h> |
20 | #include <linux/tick.h> | 21 | #include <linux/tick.h> |
22 | #include <linux/stop_machine.h> | ||
23 | |||
24 | /* Structure holding internal timekeeping values. */ | ||
25 | struct timekeeper { | ||
26 | /* Current clocksource used for timekeeping. */ | ||
27 | struct clocksource *clock; | ||
28 | /* The shift value of the current clocksource. */ | ||
29 | int shift; | ||
30 | |||
31 | /* Number of clock cycles in one NTP interval. */ | ||
32 | cycle_t cycle_interval; | ||
33 | /* Number of clock shifted nano seconds in one NTP interval. */ | ||
34 | u64 xtime_interval; | ||
35 | /* Raw nano seconds accumulated per NTP interval. */ | ||
36 | u32 raw_interval; | ||
37 | |||
38 | /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */ | ||
39 | u64 xtime_nsec; | ||
40 | /* Difference between accumulated time and NTP time in ntp | ||
41 | * shifted nano seconds. */ | ||
42 | s64 ntp_error; | ||
43 | /* Shift conversion between clock shifted nano seconds and | ||
44 | * ntp shifted nano seconds. */ | ||
45 | int ntp_error_shift; | ||
46 | /* NTP adjusted clock multiplier */ | ||
47 | u32 mult; | ||
48 | }; | ||
49 | |||
50 | struct timekeeper timekeeper; | ||
51 | |||
52 | /** | ||
53 | * timekeeper_setup_internals - Set up internals to use clocksource clock. | ||
54 | * | ||
55 | * @clock: Pointer to clocksource. | ||
56 | * | ||
57 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | ||
58 | * pair and interval request. | ||
59 | * | ||
60 | * Unless you're the timekeeping code, you should not be using this! | ||
61 | */ | ||
62 | static void timekeeper_setup_internals(struct clocksource *clock) | ||
63 | { | ||
64 | cycle_t interval; | ||
65 | u64 tmp; | ||
66 | |||
67 | timekeeper.clock = clock; | ||
68 | clock->cycle_last = clock->read(clock); | ||
21 | 69 | ||
70 | /* Do the ns -> cycle conversion first, using original mult */ | ||
71 | tmp = NTP_INTERVAL_LENGTH; | ||
72 | tmp <<= clock->shift; | ||
73 | tmp += clock->mult/2; | ||
74 | do_div(tmp, clock->mult); | ||
75 | if (tmp == 0) | ||
76 | tmp = 1; | ||
77 | |||
78 | interval = (cycle_t) tmp; | ||
79 | timekeeper.cycle_interval = interval; | ||
80 | |||
81 | /* Go back from cycles -> shifted ns */ | ||
82 | timekeeper.xtime_interval = (u64) interval * clock->mult; | ||
83 | timekeeper.raw_interval = | ||
84 | ((u64) interval * clock->mult) >> clock->shift; | ||
85 | |||
86 | timekeeper.xtime_nsec = 0; | ||
87 | timekeeper.shift = clock->shift; | ||
88 | |||
89 | timekeeper.ntp_error = 0; | ||
90 | timekeeper.ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | ||
91 | |||
92 | /* | ||
93 | * The timekeeper keeps its own mult values for the currently | ||
94 | * active clocksource. These value will be adjusted via NTP | ||
95 | * to counteract clock drifting. | ||
96 | */ | ||
97 | timekeeper.mult = clock->mult; | ||
98 | } | ||
99 | |||
100 | /* Timekeeper helper functions. */ | ||
101 | static inline s64 timekeeping_get_ns(void) | ||
102 | { | ||
103 | cycle_t cycle_now, cycle_delta; | ||
104 | struct clocksource *clock; | ||
105 | |||
106 | /* read clocksource: */ | ||
107 | clock = timekeeper.clock; | ||
108 | cycle_now = clock->read(clock); | ||
109 | |||
110 | /* calculate the delta since the last update_wall_time: */ | ||
111 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | ||
112 | |||
113 | /* return delta convert to nanoseconds using ntp adjusted mult. */ | ||
114 | return clocksource_cyc2ns(cycle_delta, timekeeper.mult, | ||
115 | timekeeper.shift); | ||
116 | } | ||
117 | |||
118 | static inline s64 timekeeping_get_ns_raw(void) | ||
119 | { | ||
120 | cycle_t cycle_now, cycle_delta; | ||
121 | struct clocksource *clock; | ||
122 | |||
123 | /* read clocksource: */ | ||
124 | clock = timekeeper.clock; | ||
125 | cycle_now = clock->read(clock); | ||
126 | |||
127 | /* calculate the delta since the last update_wall_time: */ | ||
128 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | ||
129 | |||
130 | /* return delta convert to nanoseconds using ntp adjusted mult. */ | ||
131 | return clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); | ||
132 | } | ||
22 | 133 | ||
23 | /* | 134 | /* |
24 | * This read-write spinlock protects us from races in SMP while | 135 | * This read-write spinlock protects us from races in SMP while |
@@ -44,7 +155,12 @@ __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); | |||
44 | */ | 155 | */ |
45 | struct timespec xtime __attribute__ ((aligned (16))); | 156 | struct timespec xtime __attribute__ ((aligned (16))); |
46 | struct timespec wall_to_monotonic __attribute__ ((aligned (16))); | 157 | struct timespec wall_to_monotonic __attribute__ ((aligned (16))); |
47 | static unsigned long total_sleep_time; /* seconds */ | 158 | static struct timespec total_sleep_time; |
159 | |||
160 | /* | ||
161 | * The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. | ||
162 | */ | ||
163 | struct timespec raw_time; | ||
48 | 164 | ||
49 | /* flag for if timekeeping is suspended */ | 165 | /* flag for if timekeeping is suspended */ |
50 | int __read_mostly timekeeping_suspended; | 166 | int __read_mostly timekeeping_suspended; |
@@ -56,35 +172,44 @@ void update_xtime_cache(u64 nsec) | |||
56 | timespec_add_ns(&xtime_cache, nsec); | 172 | timespec_add_ns(&xtime_cache, nsec); |
57 | } | 173 | } |
58 | 174 | ||
59 | struct clocksource *clock; | 175 | /* must hold xtime_lock */ |
60 | 176 | void timekeeping_leap_insert(int leapsecond) | |
177 | { | ||
178 | xtime.tv_sec += leapsecond; | ||
179 | wall_to_monotonic.tv_sec -= leapsecond; | ||
180 | update_vsyscall(&xtime, timekeeper.clock); | ||
181 | } | ||
61 | 182 | ||
62 | #ifdef CONFIG_GENERIC_TIME | 183 | #ifdef CONFIG_GENERIC_TIME |
184 | |||
63 | /** | 185 | /** |
64 | * clocksource_forward_now - update clock to the current time | 186 | * timekeeping_forward_now - update clock to the current time |
65 | * | 187 | * |
66 | * Forward the current clock to update its state since the last call to | 188 | * Forward the current clock to update its state since the last call to |
67 | * update_wall_time(). This is useful before significant clock changes, | 189 | * update_wall_time(). This is useful before significant clock changes, |
68 | * as it avoids having to deal with this time offset explicitly. | 190 | * as it avoids having to deal with this time offset explicitly. |
69 | */ | 191 | */ |
70 | static void clocksource_forward_now(void) | 192 | static void timekeeping_forward_now(void) |
71 | { | 193 | { |
72 | cycle_t cycle_now, cycle_delta; | 194 | cycle_t cycle_now, cycle_delta; |
195 | struct clocksource *clock; | ||
73 | s64 nsec; | 196 | s64 nsec; |
74 | 197 | ||
75 | cycle_now = clocksource_read(clock); | 198 | clock = timekeeper.clock; |
199 | cycle_now = clock->read(clock); | ||
76 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | 200 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; |
77 | clock->cycle_last = cycle_now; | 201 | clock->cycle_last = cycle_now; |
78 | 202 | ||
79 | nsec = cyc2ns(clock, cycle_delta); | 203 | nsec = clocksource_cyc2ns(cycle_delta, timekeeper.mult, |
204 | timekeeper.shift); | ||
80 | 205 | ||
81 | /* If arch requires, add in gettimeoffset() */ | 206 | /* If arch requires, add in gettimeoffset() */ |
82 | nsec += arch_gettimeoffset(); | 207 | nsec += arch_gettimeoffset(); |
83 | 208 | ||
84 | timespec_add_ns(&xtime, nsec); | 209 | timespec_add_ns(&xtime, nsec); |
85 | 210 | ||
86 | nsec = ((s64)cycle_delta * clock->mult_orig) >> clock->shift; | 211 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); |
87 | clock->raw_time.tv_nsec += nsec; | 212 | timespec_add_ns(&raw_time, nsec); |
88 | } | 213 | } |
89 | 214 | ||
90 | /** | 215 | /** |
@@ -95,7 +220,6 @@ static void clocksource_forward_now(void) | |||
95 | */ | 220 | */ |
96 | void getnstimeofday(struct timespec *ts) | 221 | void getnstimeofday(struct timespec *ts) |
97 | { | 222 | { |
98 | cycle_t cycle_now, cycle_delta; | ||
99 | unsigned long seq; | 223 | unsigned long seq; |
100 | s64 nsecs; | 224 | s64 nsecs; |
101 | 225 | ||
@@ -105,15 +229,7 @@ void getnstimeofday(struct timespec *ts) | |||
105 | seq = read_seqbegin(&xtime_lock); | 229 | seq = read_seqbegin(&xtime_lock); |
106 | 230 | ||
107 | *ts = xtime; | 231 | *ts = xtime; |
108 | 232 | nsecs = timekeeping_get_ns(); | |
109 | /* read clocksource: */ | ||
110 | cycle_now = clocksource_read(clock); | ||
111 | |||
112 | /* calculate the delta since the last update_wall_time: */ | ||
113 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | ||
114 | |||
115 | /* convert to nanoseconds: */ | ||
116 | nsecs = cyc2ns(clock, cycle_delta); | ||
117 | 233 | ||
118 | /* If arch requires, add in gettimeoffset() */ | 234 | /* If arch requires, add in gettimeoffset() */ |
119 | nsecs += arch_gettimeoffset(); | 235 | nsecs += arch_gettimeoffset(); |
@@ -125,6 +241,57 @@ void getnstimeofday(struct timespec *ts) | |||
125 | 241 | ||
126 | EXPORT_SYMBOL(getnstimeofday); | 242 | EXPORT_SYMBOL(getnstimeofday); |
127 | 243 | ||
244 | ktime_t ktime_get(void) | ||
245 | { | ||
246 | unsigned int seq; | ||
247 | s64 secs, nsecs; | ||
248 | |||
249 | WARN_ON(timekeeping_suspended); | ||
250 | |||
251 | do { | ||
252 | seq = read_seqbegin(&xtime_lock); | ||
253 | secs = xtime.tv_sec + wall_to_monotonic.tv_sec; | ||
254 | nsecs = xtime.tv_nsec + wall_to_monotonic.tv_nsec; | ||
255 | nsecs += timekeeping_get_ns(); | ||
256 | |||
257 | } while (read_seqretry(&xtime_lock, seq)); | ||
258 | /* | ||
259 | * Use ktime_set/ktime_add_ns to create a proper ktime on | ||
260 | * 32-bit architectures without CONFIG_KTIME_SCALAR. | ||
261 | */ | ||
262 | return ktime_add_ns(ktime_set(secs, 0), nsecs); | ||
263 | } | ||
264 | EXPORT_SYMBOL_GPL(ktime_get); | ||
265 | |||
266 | /** | ||
267 | * ktime_get_ts - get the monotonic clock in timespec format | ||
268 | * @ts: pointer to timespec variable | ||
269 | * | ||
270 | * The function calculates the monotonic clock from the realtime | ||
271 | * clock and the wall_to_monotonic offset and stores the result | ||
272 | * in normalized timespec format in the variable pointed to by @ts. | ||
273 | */ | ||
274 | void ktime_get_ts(struct timespec *ts) | ||
275 | { | ||
276 | struct timespec tomono; | ||
277 | unsigned int seq; | ||
278 | s64 nsecs; | ||
279 | |||
280 | WARN_ON(timekeeping_suspended); | ||
281 | |||
282 | do { | ||
283 | seq = read_seqbegin(&xtime_lock); | ||
284 | *ts = xtime; | ||
285 | tomono = wall_to_monotonic; | ||
286 | nsecs = timekeeping_get_ns(); | ||
287 | |||
288 | } while (read_seqretry(&xtime_lock, seq)); | ||
289 | |||
290 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | ||
291 | ts->tv_nsec + tomono.tv_nsec + nsecs); | ||
292 | } | ||
293 | EXPORT_SYMBOL_GPL(ktime_get_ts); | ||
294 | |||
128 | /** | 295 | /** |
129 | * do_gettimeofday - Returns the time of day in a timeval | 296 | * do_gettimeofday - Returns the time of day in a timeval |
130 | * @tv: pointer to the timeval to be set | 297 | * @tv: pointer to the timeval to be set |
@@ -157,7 +324,7 @@ int do_settimeofday(struct timespec *tv) | |||
157 | 324 | ||
158 | write_seqlock_irqsave(&xtime_lock, flags); | 325 | write_seqlock_irqsave(&xtime_lock, flags); |
159 | 326 | ||
160 | clocksource_forward_now(); | 327 | timekeeping_forward_now(); |
161 | 328 | ||
162 | ts_delta.tv_sec = tv->tv_sec - xtime.tv_sec; | 329 | ts_delta.tv_sec = tv->tv_sec - xtime.tv_sec; |
163 | ts_delta.tv_nsec = tv->tv_nsec - xtime.tv_nsec; | 330 | ts_delta.tv_nsec = tv->tv_nsec - xtime.tv_nsec; |
@@ -167,10 +334,10 @@ int do_settimeofday(struct timespec *tv) | |||
167 | 334 | ||
168 | update_xtime_cache(0); | 335 | update_xtime_cache(0); |
169 | 336 | ||
170 | clock->error = 0; | 337 | timekeeper.ntp_error = 0; |
171 | ntp_clear(); | 338 | ntp_clear(); |
172 | 339 | ||
173 | update_vsyscall(&xtime, clock); | 340 | update_vsyscall(&xtime, timekeeper.clock); |
174 | 341 | ||
175 | write_sequnlock_irqrestore(&xtime_lock, flags); | 342 | write_sequnlock_irqrestore(&xtime_lock, flags); |
176 | 343 | ||
@@ -187,44 +354,97 @@ EXPORT_SYMBOL(do_settimeofday); | |||
187 | * | 354 | * |
188 | * Accumulates current time interval and initializes new clocksource | 355 | * Accumulates current time interval and initializes new clocksource |
189 | */ | 356 | */ |
190 | static void change_clocksource(void) | 357 | static int change_clocksource(void *data) |
191 | { | 358 | { |
192 | struct clocksource *new, *old; | 359 | struct clocksource *new, *old; |
193 | 360 | ||
194 | new = clocksource_get_next(); | 361 | new = (struct clocksource *) data; |
362 | |||
363 | timekeeping_forward_now(); | ||
364 | if (!new->enable || new->enable(new) == 0) { | ||
365 | old = timekeeper.clock; | ||
366 | timekeeper_setup_internals(new); | ||
367 | if (old->disable) | ||
368 | old->disable(old); | ||
369 | } | ||
370 | return 0; | ||
371 | } | ||
195 | 372 | ||
196 | if (clock == new) | 373 | /** |
374 | * timekeeping_notify - Install a new clock source | ||
375 | * @clock: pointer to the clock source | ||
376 | * | ||
377 | * This function is called from clocksource.c after a new, better clock | ||
378 | * source has been registered. The caller holds the clocksource_mutex. | ||
379 | */ | ||
380 | void timekeeping_notify(struct clocksource *clock) | ||
381 | { | ||
382 | if (timekeeper.clock == clock) | ||
197 | return; | 383 | return; |
384 | stop_machine(change_clocksource, clock, NULL); | ||
385 | tick_clock_notify(); | ||
386 | } | ||
198 | 387 | ||
199 | clocksource_forward_now(); | 388 | #else /* GENERIC_TIME */ |
200 | 389 | ||
201 | if (clocksource_enable(new)) | 390 | static inline void timekeeping_forward_now(void) { } |
202 | return; | ||
203 | 391 | ||
204 | new->raw_time = clock->raw_time; | 392 | /** |
205 | old = clock; | 393 | * ktime_get - get the monotonic time in ktime_t format |
206 | clock = new; | 394 | * |
207 | clocksource_disable(old); | 395 | * returns the time in ktime_t format |
396 | */ | ||
397 | ktime_t ktime_get(void) | ||
398 | { | ||
399 | struct timespec now; | ||
208 | 400 | ||
209 | clock->cycle_last = 0; | 401 | ktime_get_ts(&now); |
210 | clock->cycle_last = clocksource_read(clock); | ||
211 | clock->error = 0; | ||
212 | clock->xtime_nsec = 0; | ||
213 | clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH); | ||
214 | 402 | ||
215 | tick_clock_notify(); | 403 | return timespec_to_ktime(now); |
404 | } | ||
405 | EXPORT_SYMBOL_GPL(ktime_get); | ||
216 | 406 | ||
217 | /* | 407 | /** |
218 | * We're holding xtime lock and waking up klogd would deadlock | 408 | * ktime_get_ts - get the monotonic clock in timespec format |
219 | * us on enqueue. So no printing! | 409 | * @ts: pointer to timespec variable |
220 | printk(KERN_INFO "Time: %s clocksource has been installed.\n", | 410 | * |
221 | clock->name); | 411 | * The function calculates the monotonic clock from the realtime |
222 | */ | 412 | * clock and the wall_to_monotonic offset and stores the result |
413 | * in normalized timespec format in the variable pointed to by @ts. | ||
414 | */ | ||
415 | void ktime_get_ts(struct timespec *ts) | ||
416 | { | ||
417 | struct timespec tomono; | ||
418 | unsigned long seq; | ||
419 | |||
420 | do { | ||
421 | seq = read_seqbegin(&xtime_lock); | ||
422 | getnstimeofday(ts); | ||
423 | tomono = wall_to_monotonic; | ||
424 | |||
425 | } while (read_seqretry(&xtime_lock, seq)); | ||
426 | |||
427 | set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, | ||
428 | ts->tv_nsec + tomono.tv_nsec); | ||
223 | } | 429 | } |
224 | #else | 430 | EXPORT_SYMBOL_GPL(ktime_get_ts); |
225 | static inline void clocksource_forward_now(void) { } | 431 | |
226 | static inline void change_clocksource(void) { } | 432 | #endif /* !GENERIC_TIME */ |
227 | #endif | 433 | |
434 | /** | ||
435 | * ktime_get_real - get the real (wall-) time in ktime_t format | ||
436 | * | ||
437 | * returns the time in ktime_t format | ||
438 | */ | ||
439 | ktime_t ktime_get_real(void) | ||
440 | { | ||
441 | struct timespec now; | ||
442 | |||
443 | getnstimeofday(&now); | ||
444 | |||
445 | return timespec_to_ktime(now); | ||
446 | } | ||
447 | EXPORT_SYMBOL_GPL(ktime_get_real); | ||
228 | 448 | ||
229 | /** | 449 | /** |
230 | * getrawmonotonic - Returns the raw monotonic time in a timespec | 450 | * getrawmonotonic - Returns the raw monotonic time in a timespec |
@@ -236,21 +456,11 @@ void getrawmonotonic(struct timespec *ts) | |||
236 | { | 456 | { |
237 | unsigned long seq; | 457 | unsigned long seq; |
238 | s64 nsecs; | 458 | s64 nsecs; |
239 | cycle_t cycle_now, cycle_delta; | ||
240 | 459 | ||
241 | do { | 460 | do { |
242 | seq = read_seqbegin(&xtime_lock); | 461 | seq = read_seqbegin(&xtime_lock); |
243 | 462 | nsecs = timekeeping_get_ns_raw(); | |
244 | /* read clocksource: */ | 463 | *ts = raw_time; |
245 | cycle_now = clocksource_read(clock); | ||
246 | |||
247 | /* calculate the delta since the last update_wall_time: */ | ||
248 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | ||
249 | |||
250 | /* convert to nanoseconds: */ | ||
251 | nsecs = ((s64)cycle_delta * clock->mult_orig) >> clock->shift; | ||
252 | |||
253 | *ts = clock->raw_time; | ||
254 | 464 | ||
255 | } while (read_seqretry(&xtime_lock, seq)); | 465 | } while (read_seqretry(&xtime_lock, seq)); |
256 | 466 | ||
@@ -270,7 +480,7 @@ int timekeeping_valid_for_hres(void) | |||
270 | do { | 480 | do { |
271 | seq = read_seqbegin(&xtime_lock); | 481 | seq = read_seqbegin(&xtime_lock); |
272 | 482 | ||
273 | ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; | 483 | ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
274 | 484 | ||
275 | } while (read_seqretry(&xtime_lock, seq)); | 485 | } while (read_seqretry(&xtime_lock, seq)); |
276 | 486 | ||
@@ -278,17 +488,33 @@ int timekeeping_valid_for_hres(void) | |||
278 | } | 488 | } |
279 | 489 | ||
280 | /** | 490 | /** |
281 | * read_persistent_clock - Return time in seconds from the persistent clock. | 491 | * read_persistent_clock - Return time from the persistent clock. |
282 | * | 492 | * |
283 | * Weak dummy function for arches that do not yet support it. | 493 | * Weak dummy function for arches that do not yet support it. |
284 | * Returns seconds from epoch using the battery backed persistent clock. | 494 | * Reads the time from the battery backed persistent clock. |
285 | * Returns zero if unsupported. | 495 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. |
286 | * | 496 | * |
287 | * XXX - Do be sure to remove it once all arches implement it. | 497 | * XXX - Do be sure to remove it once all arches implement it. |
288 | */ | 498 | */ |
289 | unsigned long __attribute__((weak)) read_persistent_clock(void) | 499 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts) |
290 | { | 500 | { |
291 | return 0; | 501 | ts->tv_sec = 0; |
502 | ts->tv_nsec = 0; | ||
503 | } | ||
504 | |||
505 | /** | ||
506 | * read_boot_clock - Return time of the system start. | ||
507 | * | ||
508 | * Weak dummy function for arches that do not yet support it. | ||
509 | * Function to read the exact time the system has been started. | ||
510 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | ||
511 | * | ||
512 | * XXX - Do be sure to remove it once all arches implement it. | ||
513 | */ | ||
514 | void __attribute__((weak)) read_boot_clock(struct timespec *ts) | ||
515 | { | ||
516 | ts->tv_sec = 0; | ||
517 | ts->tv_nsec = 0; | ||
292 | } | 518 | } |
293 | 519 | ||
294 | /* | 520 | /* |
@@ -296,29 +522,40 @@ unsigned long __attribute__((weak)) read_persistent_clock(void) | |||
296 | */ | 522 | */ |
297 | void __init timekeeping_init(void) | 523 | void __init timekeeping_init(void) |
298 | { | 524 | { |
525 | struct clocksource *clock; | ||
299 | unsigned long flags; | 526 | unsigned long flags; |
300 | unsigned long sec = read_persistent_clock(); | 527 | struct timespec now, boot; |
528 | |||
529 | read_persistent_clock(&now); | ||
530 | read_boot_clock(&boot); | ||
301 | 531 | ||
302 | write_seqlock_irqsave(&xtime_lock, flags); | 532 | write_seqlock_irqsave(&xtime_lock, flags); |
303 | 533 | ||
304 | ntp_init(); | 534 | ntp_init(); |
305 | 535 | ||
306 | clock = clocksource_get_next(); | 536 | clock = clocksource_default_clock(); |
307 | clocksource_enable(clock); | 537 | if (clock->enable) |
308 | clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH); | 538 | clock->enable(clock); |
309 | clock->cycle_last = clocksource_read(clock); | 539 | timekeeper_setup_internals(clock); |
310 | 540 | ||
311 | xtime.tv_sec = sec; | 541 | xtime.tv_sec = now.tv_sec; |
312 | xtime.tv_nsec = 0; | 542 | xtime.tv_nsec = now.tv_nsec; |
543 | raw_time.tv_sec = 0; | ||
544 | raw_time.tv_nsec = 0; | ||
545 | if (boot.tv_sec == 0 && boot.tv_nsec == 0) { | ||
546 | boot.tv_sec = xtime.tv_sec; | ||
547 | boot.tv_nsec = xtime.tv_nsec; | ||
548 | } | ||
313 | set_normalized_timespec(&wall_to_monotonic, | 549 | set_normalized_timespec(&wall_to_monotonic, |
314 | -xtime.tv_sec, -xtime.tv_nsec); | 550 | -boot.tv_sec, -boot.tv_nsec); |
315 | update_xtime_cache(0); | 551 | update_xtime_cache(0); |
316 | total_sleep_time = 0; | 552 | total_sleep_time.tv_sec = 0; |
553 | total_sleep_time.tv_nsec = 0; | ||
317 | write_sequnlock_irqrestore(&xtime_lock, flags); | 554 | write_sequnlock_irqrestore(&xtime_lock, flags); |
318 | } | 555 | } |
319 | 556 | ||
320 | /* time in seconds when suspend began */ | 557 | /* time in seconds when suspend began */ |
321 | static unsigned long timekeeping_suspend_time; | 558 | static struct timespec timekeeping_suspend_time; |
322 | 559 | ||
323 | /** | 560 | /** |
324 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | 561 | * timekeeping_resume - Resumes the generic timekeeping subsystem. |
@@ -331,24 +568,24 @@ static unsigned long timekeeping_suspend_time; | |||
331 | static int timekeeping_resume(struct sys_device *dev) | 568 | static int timekeeping_resume(struct sys_device *dev) |
332 | { | 569 | { |
333 | unsigned long flags; | 570 | unsigned long flags; |
334 | unsigned long now = read_persistent_clock(); | 571 | struct timespec ts; |
572 | |||
573 | read_persistent_clock(&ts); | ||
335 | 574 | ||
336 | clocksource_resume(); | 575 | clocksource_resume(); |
337 | 576 | ||
338 | write_seqlock_irqsave(&xtime_lock, flags); | 577 | write_seqlock_irqsave(&xtime_lock, flags); |
339 | 578 | ||
340 | if (now && (now > timekeeping_suspend_time)) { | 579 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) { |
341 | unsigned long sleep_length = now - timekeeping_suspend_time; | 580 | ts = timespec_sub(ts, timekeeping_suspend_time); |
342 | 581 | xtime = timespec_add_safe(xtime, ts); | |
343 | xtime.tv_sec += sleep_length; | 582 | wall_to_monotonic = timespec_sub(wall_to_monotonic, ts); |
344 | wall_to_monotonic.tv_sec -= sleep_length; | 583 | total_sleep_time = timespec_add_safe(total_sleep_time, ts); |
345 | total_sleep_time += sleep_length; | ||
346 | } | 584 | } |
347 | update_xtime_cache(0); | 585 | update_xtime_cache(0); |
348 | /* re-base the last cycle value */ | 586 | /* re-base the last cycle value */ |
349 | clock->cycle_last = 0; | 587 | timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock); |
350 | clock->cycle_last = clocksource_read(clock); | 588 | timekeeper.ntp_error = 0; |
351 | clock->error = 0; | ||
352 | timekeeping_suspended = 0; | 589 | timekeeping_suspended = 0; |
353 | write_sequnlock_irqrestore(&xtime_lock, flags); | 590 | write_sequnlock_irqrestore(&xtime_lock, flags); |
354 | 591 | ||
@@ -366,10 +603,10 @@ static int timekeeping_suspend(struct sys_device *dev, pm_message_t state) | |||
366 | { | 603 | { |
367 | unsigned long flags; | 604 | unsigned long flags; |
368 | 605 | ||
369 | timekeeping_suspend_time = read_persistent_clock(); | 606 | read_persistent_clock(&timekeeping_suspend_time); |
370 | 607 | ||
371 | write_seqlock_irqsave(&xtime_lock, flags); | 608 | write_seqlock_irqsave(&xtime_lock, flags); |
372 | clocksource_forward_now(); | 609 | timekeeping_forward_now(); |
373 | timekeeping_suspended = 1; | 610 | timekeeping_suspended = 1; |
374 | write_sequnlock_irqrestore(&xtime_lock, flags); | 611 | write_sequnlock_irqrestore(&xtime_lock, flags); |
375 | 612 | ||
@@ -404,7 +641,7 @@ device_initcall(timekeeping_init_device); | |||
404 | * If the error is already larger, we look ahead even further | 641 | * If the error is already larger, we look ahead even further |
405 | * to compensate for late or lost adjustments. | 642 | * to compensate for late or lost adjustments. |
406 | */ | 643 | */ |
407 | static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, | 644 | static __always_inline int timekeeping_bigadjust(s64 error, s64 *interval, |
408 | s64 *offset) | 645 | s64 *offset) |
409 | { | 646 | { |
410 | s64 tick_error, i; | 647 | s64 tick_error, i; |
@@ -420,7 +657,7 @@ static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, | |||
420 | * here. This is tuned so that an error of about 1 msec is adjusted | 657 | * here. This is tuned so that an error of about 1 msec is adjusted |
421 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). | 658 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). |
422 | */ | 659 | */ |
423 | error2 = clock->error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); | 660 | error2 = timekeeper.ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); |
424 | error2 = abs(error2); | 661 | error2 = abs(error2); |
425 | for (look_ahead = 0; error2 > 0; look_ahead++) | 662 | for (look_ahead = 0; error2 > 0; look_ahead++) |
426 | error2 >>= 2; | 663 | error2 >>= 2; |
@@ -429,8 +666,8 @@ static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, | |||
429 | * Now calculate the error in (1 << look_ahead) ticks, but first | 666 | * Now calculate the error in (1 << look_ahead) ticks, but first |
430 | * remove the single look ahead already included in the error. | 667 | * remove the single look ahead already included in the error. |
431 | */ | 668 | */ |
432 | tick_error = tick_length >> (NTP_SCALE_SHIFT - clock->shift + 1); | 669 | tick_error = tick_length >> (timekeeper.ntp_error_shift + 1); |
433 | tick_error -= clock->xtime_interval >> 1; | 670 | tick_error -= timekeeper.xtime_interval >> 1; |
434 | error = ((error - tick_error) >> look_ahead) + tick_error; | 671 | error = ((error - tick_error) >> look_ahead) + tick_error; |
435 | 672 | ||
436 | /* Finally calculate the adjustment shift value. */ | 673 | /* Finally calculate the adjustment shift value. */ |
@@ -455,18 +692,18 @@ static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, | |||
455 | * this is optimized for the most common adjustments of -1,0,1, | 692 | * this is optimized for the most common adjustments of -1,0,1, |
456 | * for other values we can do a bit more work. | 693 | * for other values we can do a bit more work. |
457 | */ | 694 | */ |
458 | static void clocksource_adjust(s64 offset) | 695 | static void timekeeping_adjust(s64 offset) |
459 | { | 696 | { |
460 | s64 error, interval = clock->cycle_interval; | 697 | s64 error, interval = timekeeper.cycle_interval; |
461 | int adj; | 698 | int adj; |
462 | 699 | ||
463 | error = clock->error >> (NTP_SCALE_SHIFT - clock->shift - 1); | 700 | error = timekeeper.ntp_error >> (timekeeper.ntp_error_shift - 1); |
464 | if (error > interval) { | 701 | if (error > interval) { |
465 | error >>= 2; | 702 | error >>= 2; |
466 | if (likely(error <= interval)) | 703 | if (likely(error <= interval)) |
467 | adj = 1; | 704 | adj = 1; |
468 | else | 705 | else |
469 | adj = clocksource_bigadjust(error, &interval, &offset); | 706 | adj = timekeeping_bigadjust(error, &interval, &offset); |
470 | } else if (error < -interval) { | 707 | } else if (error < -interval) { |
471 | error >>= 2; | 708 | error >>= 2; |
472 | if (likely(error >= -interval)) { | 709 | if (likely(error >= -interval)) { |
@@ -474,15 +711,15 @@ static void clocksource_adjust(s64 offset) | |||
474 | interval = -interval; | 711 | interval = -interval; |
475 | offset = -offset; | 712 | offset = -offset; |
476 | } else | 713 | } else |
477 | adj = clocksource_bigadjust(error, &interval, &offset); | 714 | adj = timekeeping_bigadjust(error, &interval, &offset); |
478 | } else | 715 | } else |
479 | return; | 716 | return; |
480 | 717 | ||
481 | clock->mult += adj; | 718 | timekeeper.mult += adj; |
482 | clock->xtime_interval += interval; | 719 | timekeeper.xtime_interval += interval; |
483 | clock->xtime_nsec -= offset; | 720 | timekeeper.xtime_nsec -= offset; |
484 | clock->error -= (interval - offset) << | 721 | timekeeper.ntp_error -= (interval - offset) << |
485 | (NTP_SCALE_SHIFT - clock->shift); | 722 | timekeeper.ntp_error_shift; |
486 | } | 723 | } |
487 | 724 | ||
488 | /** | 725 | /** |
@@ -492,53 +729,59 @@ static void clocksource_adjust(s64 offset) | |||
492 | */ | 729 | */ |
493 | void update_wall_time(void) | 730 | void update_wall_time(void) |
494 | { | 731 | { |
732 | struct clocksource *clock; | ||
495 | cycle_t offset; | 733 | cycle_t offset; |
734 | u64 nsecs; | ||
496 | 735 | ||
497 | /* Make sure we're fully resumed: */ | 736 | /* Make sure we're fully resumed: */ |
498 | if (unlikely(timekeeping_suspended)) | 737 | if (unlikely(timekeeping_suspended)) |
499 | return; | 738 | return; |
500 | 739 | ||
740 | clock = timekeeper.clock; | ||
501 | #ifdef CONFIG_GENERIC_TIME | 741 | #ifdef CONFIG_GENERIC_TIME |
502 | offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask; | 742 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask; |
503 | #else | 743 | #else |
504 | offset = clock->cycle_interval; | 744 | offset = timekeeper.cycle_interval; |
505 | #endif | 745 | #endif |
506 | clock->xtime_nsec = (s64)xtime.tv_nsec << clock->shift; | 746 | timekeeper.xtime_nsec = (s64)xtime.tv_nsec << timekeeper.shift; |
507 | 747 | ||
508 | /* normally this loop will run just once, however in the | 748 | /* normally this loop will run just once, however in the |
509 | * case of lost or late ticks, it will accumulate correctly. | 749 | * case of lost or late ticks, it will accumulate correctly. |
510 | */ | 750 | */ |
511 | while (offset >= clock->cycle_interval) { | 751 | while (offset >= timekeeper.cycle_interval) { |
752 | u64 nsecps = (u64)NSEC_PER_SEC << timekeeper.shift; | ||
753 | |||
512 | /* accumulate one interval */ | 754 | /* accumulate one interval */ |
513 | offset -= clock->cycle_interval; | 755 | offset -= timekeeper.cycle_interval; |
514 | clock->cycle_last += clock->cycle_interval; | 756 | clock->cycle_last += timekeeper.cycle_interval; |
515 | 757 | ||
516 | clock->xtime_nsec += clock->xtime_interval; | 758 | timekeeper.xtime_nsec += timekeeper.xtime_interval; |
517 | if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) { | 759 | if (timekeeper.xtime_nsec >= nsecps) { |
518 | clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift; | 760 | timekeeper.xtime_nsec -= nsecps; |
519 | xtime.tv_sec++; | 761 | xtime.tv_sec++; |
520 | second_overflow(); | 762 | second_overflow(); |
521 | } | 763 | } |
522 | 764 | ||
523 | clock->raw_time.tv_nsec += clock->raw_interval; | 765 | raw_time.tv_nsec += timekeeper.raw_interval; |
524 | if (clock->raw_time.tv_nsec >= NSEC_PER_SEC) { | 766 | if (raw_time.tv_nsec >= NSEC_PER_SEC) { |
525 | clock->raw_time.tv_nsec -= NSEC_PER_SEC; | 767 | raw_time.tv_nsec -= NSEC_PER_SEC; |
526 | clock->raw_time.tv_sec++; | 768 | raw_time.tv_sec++; |
527 | } | 769 | } |
528 | 770 | ||
529 | /* accumulate error between NTP and clock interval */ | 771 | /* accumulate error between NTP and clock interval */ |
530 | clock->error += tick_length; | 772 | timekeeper.ntp_error += tick_length; |
531 | clock->error -= clock->xtime_interval << (NTP_SCALE_SHIFT - clock->shift); | 773 | timekeeper.ntp_error -= timekeeper.xtime_interval << |
774 | timekeeper.ntp_error_shift; | ||
532 | } | 775 | } |
533 | 776 | ||
534 | /* correct the clock when NTP error is too big */ | 777 | /* correct the clock when NTP error is too big */ |
535 | clocksource_adjust(offset); | 778 | timekeeping_adjust(offset); |
536 | 779 | ||
537 | /* | 780 | /* |
538 | * Since in the loop above, we accumulate any amount of time | 781 | * Since in the loop above, we accumulate any amount of time |
539 | * in xtime_nsec over a second into xtime.tv_sec, its possible for | 782 | * in xtime_nsec over a second into xtime.tv_sec, its possible for |
540 | * xtime_nsec to be fairly small after the loop. Further, if we're | 783 | * xtime_nsec to be fairly small after the loop. Further, if we're |
541 | * slightly speeding the clocksource up in clocksource_adjust(), | 784 | * slightly speeding the clocksource up in timekeeping_adjust(), |
542 | * its possible the required corrective factor to xtime_nsec could | 785 | * its possible the required corrective factor to xtime_nsec could |
543 | * cause it to underflow. | 786 | * cause it to underflow. |
544 | * | 787 | * |
@@ -550,24 +793,25 @@ void update_wall_time(void) | |||
550 | * We'll correct this error next time through this function, when | 793 | * We'll correct this error next time through this function, when |
551 | * xtime_nsec is not as small. | 794 | * xtime_nsec is not as small. |
552 | */ | 795 | */ |
553 | if (unlikely((s64)clock->xtime_nsec < 0)) { | 796 | if (unlikely((s64)timekeeper.xtime_nsec < 0)) { |
554 | s64 neg = -(s64)clock->xtime_nsec; | 797 | s64 neg = -(s64)timekeeper.xtime_nsec; |
555 | clock->xtime_nsec = 0; | 798 | timekeeper.xtime_nsec = 0; |
556 | clock->error += neg << (NTP_SCALE_SHIFT - clock->shift); | 799 | timekeeper.ntp_error += neg << timekeeper.ntp_error_shift; |
557 | } | 800 | } |
558 | 801 | ||
559 | /* store full nanoseconds into xtime after rounding it up and | 802 | /* store full nanoseconds into xtime after rounding it up and |
560 | * add the remainder to the error difference. | 803 | * add the remainder to the error difference. |
561 | */ | 804 | */ |
562 | xtime.tv_nsec = ((s64)clock->xtime_nsec >> clock->shift) + 1; | 805 | xtime.tv_nsec = ((s64) timekeeper.xtime_nsec >> timekeeper.shift) + 1; |
563 | clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift; | 806 | timekeeper.xtime_nsec -= (s64) xtime.tv_nsec << timekeeper.shift; |
564 | clock->error += clock->xtime_nsec << (NTP_SCALE_SHIFT - clock->shift); | 807 | timekeeper.ntp_error += timekeeper.xtime_nsec << |
808 | timekeeper.ntp_error_shift; | ||
565 | 809 | ||
566 | update_xtime_cache(cyc2ns(clock, offset)); | 810 | nsecs = clocksource_cyc2ns(offset, timekeeper.mult, timekeeper.shift); |
811 | update_xtime_cache(nsecs); | ||
567 | 812 | ||
568 | /* check to see if there is a new clocksource to use */ | 813 | /* check to see if there is a new clocksource to use */ |
569 | change_clocksource(); | 814 | update_vsyscall(&xtime, timekeeper.clock); |
570 | update_vsyscall(&xtime, clock); | ||
571 | } | 815 | } |
572 | 816 | ||
573 | /** | 817 | /** |
@@ -583,9 +827,12 @@ void update_wall_time(void) | |||
583 | */ | 827 | */ |
584 | void getboottime(struct timespec *ts) | 828 | void getboottime(struct timespec *ts) |
585 | { | 829 | { |
586 | set_normalized_timespec(ts, | 830 | struct timespec boottime = { |
587 | - (wall_to_monotonic.tv_sec + total_sleep_time), | 831 | .tv_sec = wall_to_monotonic.tv_sec + total_sleep_time.tv_sec, |
588 | - wall_to_monotonic.tv_nsec); | 832 | .tv_nsec = wall_to_monotonic.tv_nsec + total_sleep_time.tv_nsec |
833 | }; | ||
834 | |||
835 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); | ||
589 | } | 836 | } |
590 | 837 | ||
591 | /** | 838 | /** |
@@ -594,7 +841,7 @@ void getboottime(struct timespec *ts) | |||
594 | */ | 841 | */ |
595 | void monotonic_to_bootbased(struct timespec *ts) | 842 | void monotonic_to_bootbased(struct timespec *ts) |
596 | { | 843 | { |
597 | ts->tv_sec += total_sleep_time; | 844 | *ts = timespec_add_safe(*ts, total_sleep_time); |
598 | } | 845 | } |
599 | 846 | ||
600 | unsigned long get_seconds(void) | 847 | unsigned long get_seconds(void) |
@@ -603,6 +850,10 @@ unsigned long get_seconds(void) | |||
603 | } | 850 | } |
604 | EXPORT_SYMBOL(get_seconds); | 851 | EXPORT_SYMBOL(get_seconds); |
605 | 852 | ||
853 | struct timespec __current_kernel_time(void) | ||
854 | { | ||
855 | return xtime_cache; | ||
856 | } | ||
606 | 857 | ||
607 | struct timespec current_kernel_time(void) | 858 | struct timespec current_kernel_time(void) |
608 | { | 859 | { |
@@ -618,3 +869,20 @@ struct timespec current_kernel_time(void) | |||
618 | return now; | 869 | return now; |
619 | } | 870 | } |
620 | EXPORT_SYMBOL(current_kernel_time); | 871 | EXPORT_SYMBOL(current_kernel_time); |
872 | |||
873 | struct timespec get_monotonic_coarse(void) | ||
874 | { | ||
875 | struct timespec now, mono; | ||
876 | unsigned long seq; | ||
877 | |||
878 | do { | ||
879 | seq = read_seqbegin(&xtime_lock); | ||
880 | |||
881 | now = xtime_cache; | ||
882 | mono = wall_to_monotonic; | ||
883 | } while (read_seqretry(&xtime_lock, seq)); | ||
884 | |||
885 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, | ||
886 | now.tv_nsec + mono.tv_nsec); | ||
887 | return now; | ||
888 | } | ||