]>
Commit | Line | Data |
---|---|---|
35728b82 | 1 | // SPDX-License-Identifier: GPL-2.0 |
8524070b | 2 | /* |
58c5fc2b TG |
3 | * Kernel timekeeping code and accessor functions. Based on code from |
4 | * timer.c, moved in commit 8524070b7982. | |
8524070b | 5 | */ |
d7b4202e | 6 | #include <linux/timekeeper_internal.h> |
8524070b JS |
7 | #include <linux/module.h> |
8 | #include <linux/interrupt.h> | |
9 | #include <linux/percpu.h> | |
10 | #include <linux/init.h> | |
11 | #include <linux/mm.h> | |
38b8d208 | 12 | #include <linux/nmi.h> |
d43c36dc | 13 | #include <linux/sched.h> |
4f17722c | 14 | #include <linux/sched/loadavg.h> |
3eca9937 | 15 | #include <linux/sched/clock.h> |
e1a85b2c | 16 | #include <linux/syscore_ops.h> |
8524070b JS |
17 | #include <linux/clocksource.h> |
18 | #include <linux/jiffies.h> | |
19 | #include <linux/time.h> | |
20 | #include <linux/tick.h> | |
75c5158f | 21 | #include <linux/stop_machine.h> |
e0b306fe | 22 | #include <linux/pvclock_gtod.h> |
52f5684c | 23 | #include <linux/compiler.h> |
2d87a067 | 24 | #include <linux/audit.h> |
8524070b | 25 | |
eb93e4d9 | 26 | #include "tick-internal.h" |
aa6f9c59 | 27 | #include "ntp_internal.h" |
5c83545f | 28 | #include "timekeeping_internal.h" |
155ec602 | 29 | |
04397fe9 DV |
30 | #define TK_CLEAR_NTP (1 << 0) |
31 | #define TK_MIRROR (1 << 1) | |
780427f0 | 32 | #define TK_CLOCK_WAS_SET (1 << 2) |
04397fe9 | 33 | |
b061c7a5 ML |
34 | enum timekeeping_adv_mode { |
35 | /* Update timekeeper when a tick has passed */ | |
36 | TK_ADV_TICK, | |
37 | ||
38 | /* Update timekeeper on a direct frequency change */ | |
39 | TK_ADV_FREQ | |
40 | }; | |
41 | ||
b923f124 | 42 | DEFINE_RAW_SPINLOCK(timekeeper_lock); |
025e82bc | 43 | |
3fdb14fd TG |
44 | /* |
45 | * The most important data for readout fits into a single 64 byte | |
46 | * cache line. | |
47 | */ | |
48 | static struct { | |
025e82bc | 49 | seqcount_raw_spinlock_t seq; |
3fdb14fd | 50 | struct timekeeper timekeeper; |
ce10a5b3 | 51 | } tk_core ____cacheline_aligned = { |
025e82bc | 52 | .seq = SEQCNT_RAW_SPINLOCK_ZERO(tk_core.seq, &timekeeper_lock), |
ce10a5b3 | 53 | }; |
3fdb14fd | 54 | |
48cdc135 | 55 | static struct timekeeper shadow_timekeeper; |
155ec602 | 56 | |
71419b30 TG |
57 | /* flag for if timekeeping is suspended */ |
58 | int __read_mostly timekeeping_suspended; | |
59 | ||
4396e058 TG |
60 | /** |
61 | * struct tk_fast - NMI safe timekeeper | |
62 | * @seq: Sequence counter for protecting updates. The lowest bit | |
63 | * is the index for the tk_read_base array | |
64 | * @base: tk_read_base array. Access is indexed by the lowest bit of | |
65 | * @seq. | |
66 | * | |
67 | * See @update_fast_timekeeper() below. | |
68 | */ | |
69 | struct tk_fast { | |
249d0538 | 70 | seqcount_latch_t seq; |
4396e058 TG |
71 | struct tk_read_base base[2]; |
72 | }; | |
73 | ||
5df32107 PB |
74 | /* Suspend-time cycles value for halted fast timekeeper. */ |
75 | static u64 cycles_at_suspend; | |
76 | ||
77 | static u64 dummy_clock_read(struct clocksource *cs) | |
78 | { | |
71419b30 TG |
79 | if (timekeeping_suspended) |
80 | return cycles_at_suspend; | |
81 | return local_clock(); | |
5df32107 PB |
82 | } |
83 | ||
84 | static struct clocksource dummy_clock = { | |
85 | .read = dummy_clock_read, | |
86 | }; | |
87 | ||
71419b30 TG |
88 | /* |
89 | * Boot time initialization which allows local_clock() to be utilized | |
90 | * during early boot when clocksources are not available. local_clock() | |
91 | * returns nanoseconds already so no conversion is required, hence mult=1 | |
92 | * and shift=0. When the first proper clocksource is installed then | |
93 | * the fast time keepers are updated with the correct values. | |
94 | */ | |
95 | #define FAST_TK_INIT \ | |
96 | { \ | |
97 | .clock = &dummy_clock, \ | |
98 | .mask = CLOCKSOURCE_MASK(64), \ | |
99 | .mult = 1, \ | |
100 | .shift = 0, \ | |
101 | } | |
102 | ||
5df32107 | 103 | static struct tk_fast tk_fast_mono ____cacheline_aligned = { |
249d0538 | 104 | .seq = SEQCNT_LATCH_ZERO(tk_fast_mono.seq), |
71419b30 TG |
105 | .base[0] = FAST_TK_INIT, |
106 | .base[1] = FAST_TK_INIT, | |
5df32107 PB |
107 | }; |
108 | ||
109 | static struct tk_fast tk_fast_raw ____cacheline_aligned = { | |
249d0538 | 110 | .seq = SEQCNT_LATCH_ZERO(tk_fast_raw.seq), |
71419b30 TG |
111 | .base[0] = FAST_TK_INIT, |
112 | .base[1] = FAST_TK_INIT, | |
5df32107 | 113 | }; |
4396e058 | 114 | |
1e75fa8b JS |
115 | static inline void tk_normalize_xtime(struct timekeeper *tk) |
116 | { | |
876e7881 PZ |
117 | while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) { |
118 | tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift; | |
1e75fa8b JS |
119 | tk->xtime_sec++; |
120 | } | |
fc6eead7 JS |
121 | while (tk->tkr_raw.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_raw.shift)) { |
122 | tk->tkr_raw.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_raw.shift; | |
123 | tk->raw_sec++; | |
124 | } | |
1e75fa8b JS |
125 | } |
126 | ||
985e6950 | 127 | static inline struct timespec64 tk_xtime(const struct timekeeper *tk) |
c905fae4 TG |
128 | { |
129 | struct timespec64 ts; | |
130 | ||
131 | ts.tv_sec = tk->xtime_sec; | |
876e7881 | 132 | ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
c905fae4 TG |
133 | return ts; |
134 | } | |
135 | ||
7d489d15 | 136 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
137 | { |
138 | tk->xtime_sec = ts->tv_sec; | |
876e7881 | 139 | tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift; |
1e75fa8b JS |
140 | } |
141 | ||
7d489d15 | 142 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts) |
1e75fa8b JS |
143 | { |
144 | tk->xtime_sec += ts->tv_sec; | |
876e7881 | 145 | tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift; |
784ffcbb | 146 | tk_normalize_xtime(tk); |
1e75fa8b | 147 | } |
8fcce546 | 148 | |
7d489d15 | 149 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm) |
6d0ef903 | 150 | { |
7d489d15 | 151 | struct timespec64 tmp; |
6d0ef903 JS |
152 | |
153 | /* | |
154 | * Verify consistency of: offset_real = -wall_to_monotonic | |
155 | * before modifying anything | |
156 | */ | |
7d489d15 | 157 | set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec, |
6d0ef903 | 158 | -tk->wall_to_monotonic.tv_nsec); |
2456e855 | 159 | WARN_ON_ONCE(tk->offs_real != timespec64_to_ktime(tmp)); |
6d0ef903 | 160 | tk->wall_to_monotonic = wtm; |
7d489d15 JS |
161 | set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec); |
162 | tk->offs_real = timespec64_to_ktime(tmp); | |
04005f60 | 163 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); |
6d0ef903 JS |
164 | } |
165 | ||
47da70d3 | 166 | static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta) |
6d0ef903 | 167 | { |
a3ed0e43 | 168 | tk->offs_boot = ktime_add(tk->offs_boot, delta); |
b99328a6 TG |
169 | /* |
170 | * Timespec representation for VDSO update to avoid 64bit division | |
171 | * on every update. | |
172 | */ | |
173 | tk->monotonic_to_boot = ktime_to_timespec64(tk->offs_boot); | |
6d0ef903 JS |
174 | } |
175 | ||
ceea5e37 JS |
176 | /* |
177 | * tk_clock_read - atomic clocksource read() helper | |
178 | * | |
179 | * This helper is necessary to use in the read paths because, while the | |
025e82bc | 180 | * seqcount ensures we don't return a bad value while structures are updated, |
ceea5e37 JS |
181 | * it doesn't protect from potential crashes. There is the possibility that |
182 | * the tkr's clocksource may change between the read reference, and the | |
183 | * clock reference passed to the read function. This can cause crashes if | |
184 | * the wrong clocksource is passed to the wrong read function. | |
185 | * This isn't necessary to use when holding the timekeeper_lock or doing | |
186 | * a read of the fast-timekeeper tkrs (which is protected by its own locking | |
187 | * and update logic). | |
188 | */ | |
985e6950 | 189 | static inline u64 tk_clock_read(const struct tk_read_base *tkr) |
ceea5e37 JS |
190 | { |
191 | struct clocksource *clock = READ_ONCE(tkr->clock); | |
192 | ||
193 | return clock->read(clock); | |
194 | } | |
195 | ||
3c17ad19 | 196 | #ifdef CONFIG_DEBUG_TIMEKEEPING |
4ca22c26 | 197 | #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */ |
4ca22c26 | 198 | |
a5a1d1c2 | 199 | static void timekeeping_check_update(struct timekeeper *tk, u64 offset) |
3c17ad19 JS |
200 | { |
201 | ||
a5a1d1c2 | 202 | u64 max_cycles = tk->tkr_mono.clock->max_cycles; |
876e7881 | 203 | const char *name = tk->tkr_mono.clock->name; |
3c17ad19 JS |
204 | |
205 | if (offset > max_cycles) { | |
a558cd02 | 206 | printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n", |
3c17ad19 | 207 | offset, name, max_cycles); |
a558cd02 | 208 | printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n"); |
3c17ad19 JS |
209 | } else { |
210 | if (offset > (max_cycles >> 1)) { | |
fc4fa6e1 | 211 | printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n", |
3c17ad19 JS |
212 | offset, name, max_cycles >> 1); |
213 | printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n"); | |
214 | } | |
215 | } | |
4ca22c26 | 216 | |
57d05a93 JS |
217 | if (tk->underflow_seen) { |
218 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
219 | printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name); |
220 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
221 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 222 | tk->last_warning = jiffies; |
4ca22c26 | 223 | } |
57d05a93 | 224 | tk->underflow_seen = 0; |
4ca22c26 JS |
225 | } |
226 | ||
57d05a93 JS |
227 | if (tk->overflow_seen) { |
228 | if (jiffies - tk->last_warning > WARNING_FREQ) { | |
4ca22c26 JS |
229 | printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name); |
230 | printk_deferred(" Please report this, consider using a different clocksource, if possible.\n"); | |
231 | printk_deferred(" Your kernel is probably still fine.\n"); | |
57d05a93 | 232 | tk->last_warning = jiffies; |
4ca22c26 | 233 | } |
57d05a93 | 234 | tk->overflow_seen = 0; |
4ca22c26 | 235 | } |
3c17ad19 | 236 | } |
a558cd02 | 237 | |
985e6950 | 238 | static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr) |
a558cd02 | 239 | { |
57d05a93 | 240 | struct timekeeper *tk = &tk_core.timekeeper; |
a5a1d1c2 | 241 | u64 now, last, mask, max, delta; |
4ca22c26 | 242 | unsigned int seq; |
a558cd02 | 243 | |
4ca22c26 | 244 | /* |
025e82bc | 245 | * Since we're called holding a seqcount, the data may shift |
4ca22c26 JS |
246 | * under us while we're doing the calculation. This can cause |
247 | * false positives, since we'd note a problem but throw the | |
025e82bc | 248 | * results away. So nest another seqcount here to atomically |
4ca22c26 JS |
249 | * grab the points we are checking with. |
250 | */ | |
251 | do { | |
252 | seq = read_seqcount_begin(&tk_core.seq); | |
ceea5e37 | 253 | now = tk_clock_read(tkr); |
4ca22c26 JS |
254 | last = tkr->cycle_last; |
255 | mask = tkr->mask; | |
256 | max = tkr->clock->max_cycles; | |
257 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
a558cd02 | 258 | |
4ca22c26 | 259 | delta = clocksource_delta(now, last, mask); |
a558cd02 | 260 | |
057b87e3 JS |
261 | /* |
262 | * Try to catch underflows by checking if we are seeing small | |
263 | * mask-relative negative values. | |
264 | */ | |
4ca22c26 | 265 | if (unlikely((~delta & mask) < (mask >> 3))) { |
57d05a93 | 266 | tk->underflow_seen = 1; |
057b87e3 | 267 | delta = 0; |
4ca22c26 | 268 | } |
057b87e3 | 269 | |
a558cd02 | 270 | /* Cap delta value to the max_cycles values to avoid mult overflows */ |
4ca22c26 | 271 | if (unlikely(delta > max)) { |
57d05a93 | 272 | tk->overflow_seen = 1; |
a558cd02 | 273 | delta = tkr->clock->max_cycles; |
4ca22c26 | 274 | } |
a558cd02 JS |
275 | |
276 | return delta; | |
277 | } | |
3c17ad19 | 278 | #else |
a5a1d1c2 | 279 | static inline void timekeeping_check_update(struct timekeeper *tk, u64 offset) |
3c17ad19 JS |
280 | { |
281 | } | |
985e6950 | 282 | static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr) |
a558cd02 | 283 | { |
a5a1d1c2 | 284 | u64 cycle_now, delta; |
a558cd02 JS |
285 | |
286 | /* read clocksource */ | |
ceea5e37 | 287 | cycle_now = tk_clock_read(tkr); |
a558cd02 JS |
288 | |
289 | /* calculate the delta since the last update_wall_time */ | |
290 | delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); | |
291 | ||
292 | return delta; | |
293 | } | |
3c17ad19 JS |
294 | #endif |
295 | ||
155ec602 | 296 | /** |
d26e4fe0 | 297 | * tk_setup_internals - Set up internals to use clocksource clock. |
155ec602 | 298 | * |
d26e4fe0 | 299 | * @tk: The target timekeeper to setup. |
155ec602 MS |
300 | * @clock: Pointer to clocksource. |
301 | * | |
302 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment | |
303 | * pair and interval request. | |
304 | * | |
305 | * Unless you're the timekeeping code, you should not be using this! | |
306 | */ | |
f726a697 | 307 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) |
155ec602 | 308 | { |
a5a1d1c2 | 309 | u64 interval; |
a386b5af | 310 | u64 tmp, ntpinterval; |
1e75fa8b | 311 | struct clocksource *old_clock; |
155ec602 | 312 | |
2c756feb | 313 | ++tk->cs_was_changed_seq; |
876e7881 PZ |
314 | old_clock = tk->tkr_mono.clock; |
315 | tk->tkr_mono.clock = clock; | |
876e7881 | 316 | tk->tkr_mono.mask = clock->mask; |
ceea5e37 | 317 | tk->tkr_mono.cycle_last = tk_clock_read(&tk->tkr_mono); |
155ec602 | 318 | |
4a4ad80d | 319 | tk->tkr_raw.clock = clock; |
4a4ad80d PZ |
320 | tk->tkr_raw.mask = clock->mask; |
321 | tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last; | |
322 | ||
155ec602 MS |
323 | /* Do the ns -> cycle conversion first, using original mult */ |
324 | tmp = NTP_INTERVAL_LENGTH; | |
325 | tmp <<= clock->shift; | |
a386b5af | 326 | ntpinterval = tmp; |
0a544198 MS |
327 | tmp += clock->mult/2; |
328 | do_div(tmp, clock->mult); | |
155ec602 MS |
329 | if (tmp == 0) |
330 | tmp = 1; | |
331 | ||
a5a1d1c2 | 332 | interval = (u64) tmp; |
f726a697 | 333 | tk->cycle_interval = interval; |
155ec602 MS |
334 | |
335 | /* Go back from cycles -> shifted ns */ | |
cbd99e3b | 336 | tk->xtime_interval = interval * clock->mult; |
f726a697 | 337 | tk->xtime_remainder = ntpinterval - tk->xtime_interval; |
3d88d56c | 338 | tk->raw_interval = interval * clock->mult; |
155ec602 | 339 | |
1e75fa8b JS |
340 | /* if changing clocks, convert xtime_nsec shift units */ |
341 | if (old_clock) { | |
342 | int shift_change = clock->shift - old_clock->shift; | |
fc6eead7 | 343 | if (shift_change < 0) { |
876e7881 | 344 | tk->tkr_mono.xtime_nsec >>= -shift_change; |
fc6eead7 JS |
345 | tk->tkr_raw.xtime_nsec >>= -shift_change; |
346 | } else { | |
876e7881 | 347 | tk->tkr_mono.xtime_nsec <<= shift_change; |
fc6eead7 JS |
348 | tk->tkr_raw.xtime_nsec <<= shift_change; |
349 | } | |
1e75fa8b | 350 | } |
4a4ad80d | 351 | |
876e7881 | 352 | tk->tkr_mono.shift = clock->shift; |
4a4ad80d | 353 | tk->tkr_raw.shift = clock->shift; |
155ec602 | 354 | |
f726a697 JS |
355 | tk->ntp_error = 0; |
356 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; | |
375f45b5 | 357 | tk->ntp_tick = ntpinterval << tk->ntp_error_shift; |
0a544198 MS |
358 | |
359 | /* | |
360 | * The timekeeper keeps its own mult values for the currently | |
361 | * active clocksource. These value will be adjusted via NTP | |
362 | * to counteract clock drifting. | |
363 | */ | |
876e7881 | 364 | tk->tkr_mono.mult = clock->mult; |
4a4ad80d | 365 | tk->tkr_raw.mult = clock->mult; |
dc491596 | 366 | tk->ntp_err_mult = 0; |
78b98e3c | 367 | tk->skip_second_overflow = 0; |
155ec602 | 368 | } |
8524070b | 369 | |
2ba2a305 | 370 | /* Timekeeper helper functions. */ |
7b1f6207 | 371 | |
985e6950 | 372 | static inline u64 timekeeping_delta_to_ns(const struct tk_read_base *tkr, u64 delta) |
6bd58f09 | 373 | { |
9c164572 | 374 | u64 nsec; |
6bd58f09 CH |
375 | |
376 | nsec = delta * tkr->mult + tkr->xtime_nsec; | |
377 | nsec >>= tkr->shift; | |
378 | ||
77f6c0b8 | 379 | return nsec; |
6bd58f09 CH |
380 | } |
381 | ||
985e6950 | 382 | static inline u64 timekeeping_get_ns(const struct tk_read_base *tkr) |
2ba2a305 | 383 | { |
a5a1d1c2 | 384 | u64 delta; |
2ba2a305 | 385 | |
a558cd02 | 386 | delta = timekeeping_get_delta(tkr); |
6bd58f09 CH |
387 | return timekeeping_delta_to_ns(tkr, delta); |
388 | } | |
2ba2a305 | 389 | |
985e6950 | 390 | static inline u64 timekeeping_cycles_to_ns(const struct tk_read_base *tkr, u64 cycles) |
6bd58f09 | 391 | { |
a5a1d1c2 | 392 | u64 delta; |
f2a5a085 | 393 | |
6bd58f09 CH |
394 | /* calculate the delta since the last update_wall_time */ |
395 | delta = clocksource_delta(cycles, tkr->cycle_last, tkr->mask); | |
396 | return timekeeping_delta_to_ns(tkr, delta); | |
2ba2a305 MS |
397 | } |
398 | ||
4396e058 TG |
399 | /** |
400 | * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper. | |
affe3e85 | 401 | * @tkr: Timekeeping readout base from which we take the update |
e025b031 | 402 | * @tkf: Pointer to NMI safe timekeeper |
4396e058 TG |
403 | * |
404 | * We want to use this from any context including NMI and tracing / | |
405 | * instrumenting the timekeeping code itself. | |
406 | * | |
6695b92a | 407 | * Employ the latch technique; see @raw_write_seqcount_latch. |
4396e058 TG |
408 | * |
409 | * So if a NMI hits the update of base[0] then it will use base[1] | |
410 | * which is still consistent. In the worst case this can result is a | |
411 | * slightly wrong timestamp (a few nanoseconds). See | |
412 | * @ktime_get_mono_fast_ns. | |
413 | */ | |
985e6950 OM |
414 | static void update_fast_timekeeper(const struct tk_read_base *tkr, |
415 | struct tk_fast *tkf) | |
4396e058 | 416 | { |
4498e746 | 417 | struct tk_read_base *base = tkf->base; |
4396e058 TG |
418 | |
419 | /* Force readers off to base[1] */ | |
4498e746 | 420 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
421 | |
422 | /* Update base[0] */ | |
affe3e85 | 423 | memcpy(base, tkr, sizeof(*base)); |
4396e058 TG |
424 | |
425 | /* Force readers back to base[0] */ | |
4498e746 | 426 | raw_write_seqcount_latch(&tkf->seq); |
4396e058 TG |
427 | |
428 | /* Update base[1] */ | |
429 | memcpy(base + 1, base, sizeof(*base)); | |
430 | } | |
431 | ||
c1ce406e TG |
432 | static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf) |
433 | { | |
434 | struct tk_read_base *tkr; | |
435 | unsigned int seq; | |
436 | u64 now; | |
437 | ||
438 | do { | |
439 | seq = raw_read_seqcount_latch(&tkf->seq); | |
440 | tkr = tkf->base + (seq & 0x01); | |
441 | now = ktime_to_ns(tkr->base); | |
442 | ||
443 | now += timekeeping_delta_to_ns(tkr, | |
444 | clocksource_delta( | |
445 | tk_clock_read(tkr), | |
446 | tkr->cycle_last, | |
447 | tkr->mask)); | |
448 | } while (read_seqcount_latch_retry(&tkf->seq, seq)); | |
449 | ||
450 | return now; | |
451 | } | |
452 | ||
4396e058 TG |
453 | /** |
454 | * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic | |
455 | * | |
456 | * This timestamp is not guaranteed to be monotonic across an update. | |
457 | * The timestamp is calculated by: | |
458 | * | |
459 | * now = base_mono + clock_delta * slope | |
460 | * | |
461 | * So if the update lowers the slope, readers who are forced to the | |
462 | * not yet updated second array are still using the old steeper slope. | |
463 | * | |
464 | * tmono | |
465 | * ^ | |
466 | * | o n | |
467 | * | o n | |
468 | * | u | |
469 | * | o | |
470 | * |o | |
471 | * |12345678---> reader order | |
472 | * | |
473 | * o = old slope | |
474 | * u = update | |
475 | * n = new slope | |
476 | * | |
477 | * So reader 6 will observe time going backwards versus reader 5. | |
478 | * | |
c1ce406e | 479 | * While other CPUs are likely to be able to observe that, the only way |
4396e058 TG |
480 | * for a CPU local observation is when an NMI hits in the middle of |
481 | * the update. Timestamps taken from that NMI context might be ahead | |
482 | * of the following timestamps. Callers need to be aware of that and | |
483 | * deal with it. | |
484 | */ | |
4498e746 PZ |
485 | u64 ktime_get_mono_fast_ns(void) |
486 | { | |
487 | return __ktime_get_fast_ns(&tk_fast_mono); | |
488 | } | |
4396e058 TG |
489 | EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns); |
490 | ||
c1ce406e TG |
491 | /** |
492 | * ktime_get_raw_fast_ns - Fast NMI safe access to clock monotonic raw | |
493 | * | |
494 | * Contrary to ktime_get_mono_fast_ns() this is always correct because the | |
495 | * conversion factor is not affected by NTP/PTP correction. | |
496 | */ | |
f09cb9a1 PZ |
497 | u64 ktime_get_raw_fast_ns(void) |
498 | { | |
499 | return __ktime_get_fast_ns(&tk_fast_raw); | |
500 | } | |
501 | EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); | |
502 | ||
a3ed0e43 TG |
503 | /** |
504 | * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock. | |
505 | * | |
506 | * To keep it NMI safe since we're accessing from tracing, we're not using a | |
507 | * separate timekeeper with updates to monotonic clock and boot offset | |
025e82bc | 508 | * protected with seqcounts. This has the following minor side effects: |
a3ed0e43 TG |
509 | * |
510 | * (1) Its possible that a timestamp be taken after the boot offset is updated | |
511 | * but before the timekeeper is updated. If this happens, the new boot offset | |
512 | * is added to the old timekeeping making the clock appear to update slightly | |
513 | * earlier: | |
514 | * CPU 0 CPU 1 | |
515 | * timekeeping_inject_sleeptime64() | |
516 | * __timekeeping_inject_sleeptime(tk, delta); | |
517 | * timestamp(); | |
518 | * timekeeping_update(tk, TK_CLEAR_NTP...); | |
519 | * | |
520 | * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be | |
521 | * partially updated. Since the tk->offs_boot update is a rare event, this | |
522 | * should be a rare occurrence which postprocessing should be able to handle. | |
c1ce406e TG |
523 | * |
524 | * The caveats vs. timestamp ordering as documented for ktime_get_fast_ns() | |
525 | * apply as well. | |
a3ed0e43 TG |
526 | */ |
527 | u64 notrace ktime_get_boot_fast_ns(void) | |
528 | { | |
529 | struct timekeeper *tk = &tk_core.timekeeper; | |
530 | ||
531 | return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot)); | |
532 | } | |
533 | EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns); | |
534 | ||
e2d977c9 | 535 | static __always_inline u64 __ktime_get_real_fast(struct tk_fast *tkf, u64 *mono) |
4c3711d7 TG |
536 | { |
537 | struct tk_read_base *tkr; | |
e2d977c9 | 538 | u64 basem, baser, delta; |
4c3711d7 | 539 | unsigned int seq; |
4c3711d7 TG |
540 | |
541 | do { | |
542 | seq = raw_read_seqcount_latch(&tkf->seq); | |
543 | tkr = tkf->base + (seq & 0x01); | |
e2d977c9 TG |
544 | basem = ktime_to_ns(tkr->base); |
545 | baser = ktime_to_ns(tkr->base_real); | |
4c3711d7 | 546 | |
e2d977c9 TG |
547 | delta = timekeeping_delta_to_ns(tkr, |
548 | clocksource_delta(tk_clock_read(tkr), | |
549 | tkr->cycle_last, tkr->mask)); | |
249d0538 | 550 | } while (read_seqcount_latch_retry(&tkf->seq, seq)); |
4c3711d7 | 551 | |
e2d977c9 TG |
552 | if (mono) |
553 | *mono = basem + delta; | |
554 | return baser + delta; | |
4c3711d7 TG |
555 | } |
556 | ||
557 | /** | |
558 | * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime. | |
c1ce406e TG |
559 | * |
560 | * See ktime_get_fast_ns() for documentation of the time stamp ordering. | |
4c3711d7 TG |
561 | */ |
562 | u64 ktime_get_real_fast_ns(void) | |
563 | { | |
e2d977c9 | 564 | return __ktime_get_real_fast(&tk_fast_mono, NULL); |
4c3711d7 | 565 | } |
df27067e | 566 | EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns); |
4c3711d7 | 567 | |
e2d977c9 TG |
568 | /** |
569 | * ktime_get_fast_timestamps: - NMI safe timestamps | |
570 | * @snapshot: Pointer to timestamp storage | |
571 | * | |
572 | * Stores clock monotonic, boottime and realtime timestamps. | |
573 | * | |
574 | * Boot time is a racy access on 32bit systems if the sleep time injection | |
575 | * happens late during resume and not in timekeeping_resume(). That could | |
576 | * be avoided by expanding struct tk_read_base with boot offset for 32bit | |
577 | * and adding more overhead to the update. As this is a hard to observe | |
578 | * once per resume event which can be filtered with reasonable effort using | |
579 | * the accurate mono/real timestamps, it's probably not worth the trouble. | |
580 | * | |
581 | * Aside of that it might be possible on 32 and 64 bit to observe the | |
582 | * following when the sleep time injection happens late: | |
583 | * | |
584 | * CPU 0 CPU 1 | |
585 | * timekeeping_resume() | |
586 | * ktime_get_fast_timestamps() | |
587 | * mono, real = __ktime_get_real_fast() | |
588 | * inject_sleep_time() | |
589 | * update boot offset | |
590 | * boot = mono + bootoffset; | |
591 | * | |
592 | * That means that boot time already has the sleep time adjustment, but | |
593 | * real time does not. On the next readout both are in sync again. | |
594 | * | |
595 | * Preventing this for 64bit is not really feasible without destroying the | |
596 | * careful cache layout of the timekeeper because the sequence count and | |
597 | * struct tk_read_base would then need two cache lines instead of one. | |
598 | * | |
599 | * Access to the time keeper clock source is disabled accross the innermost | |
600 | * steps of suspend/resume. The accessors still work, but the timestamps | |
601 | * are frozen until time keeping is resumed which happens very early. | |
602 | * | |
603 | * For regular suspend/resume there is no observable difference vs. sched | |
604 | * clock, but it might affect some of the nasty low level debug printks. | |
605 | * | |
606 | * OTOH, access to sched clock is not guaranteed accross suspend/resume on | |
607 | * all systems either so it depends on the hardware in use. | |
608 | * | |
609 | * If that turns out to be a real problem then this could be mitigated by | |
610 | * using sched clock in a similar way as during early boot. But it's not as | |
611 | * trivial as on early boot because it needs some careful protection | |
612 | * against the clock monotonic timestamp jumping backwards on resume. | |
613 | */ | |
614 | void ktime_get_fast_timestamps(struct ktime_timestamps *snapshot) | |
615 | { | |
616 | struct timekeeper *tk = &tk_core.timekeeper; | |
617 | ||
618 | snapshot->real = __ktime_get_real_fast(&tk_fast_mono, &snapshot->mono); | |
619 | snapshot->boot = snapshot->mono + ktime_to_ns(data_race(tk->offs_boot)); | |
620 | } | |
621 | ||
060407ae RW |
622 | /** |
623 | * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource. | |
624 | * @tk: Timekeeper to snapshot. | |
625 | * | |
626 | * It generally is unsafe to access the clocksource after timekeeping has been | |
627 | * suspended, so take a snapshot of the readout base of @tk and use it as the | |
628 | * fast timekeeper's readout base while suspended. It will return the same | |
629 | * number of cycles every time until timekeeping is resumed at which time the | |
630 | * proper readout base for the fast timekeeper will be restored automatically. | |
631 | */ | |
985e6950 | 632 | static void halt_fast_timekeeper(const struct timekeeper *tk) |
060407ae RW |
633 | { |
634 | static struct tk_read_base tkr_dummy; | |
985e6950 | 635 | const struct tk_read_base *tkr = &tk->tkr_mono; |
060407ae RW |
636 | |
637 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
ceea5e37 JS |
638 | cycles_at_suspend = tk_clock_read(tkr); |
639 | tkr_dummy.clock = &dummy_clock; | |
4c3711d7 | 640 | tkr_dummy.base_real = tkr->base + tk->offs_real; |
4498e746 | 641 | update_fast_timekeeper(&tkr_dummy, &tk_fast_mono); |
f09cb9a1 PZ |
642 | |
643 | tkr = &tk->tkr_raw; | |
644 | memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy)); | |
ceea5e37 | 645 | tkr_dummy.clock = &dummy_clock; |
f09cb9a1 | 646 | update_fast_timekeeper(&tkr_dummy, &tk_fast_raw); |
060407ae RW |
647 | } |
648 | ||
e0b306fe MT |
649 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); |
650 | ||
780427f0 | 651 | static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) |
e0b306fe | 652 | { |
780427f0 | 653 | raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); |
e0b306fe MT |
654 | } |
655 | ||
656 | /** | |
657 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener | |
f27f7c3f | 658 | * @nb: Pointer to the notifier block to register |
e0b306fe MT |
659 | */ |
660 | int pvclock_gtod_register_notifier(struct notifier_block *nb) | |
661 | { | |
3fdb14fd | 662 | struct timekeeper *tk = &tk_core.timekeeper; |
e0b306fe MT |
663 | unsigned long flags; |
664 | int ret; | |
665 | ||
9a7a71b1 | 666 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 667 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); |
780427f0 | 668 | update_pvclock_gtod(tk, true); |
9a7a71b1 | 669 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
670 | |
671 | return ret; | |
672 | } | |
673 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); | |
674 | ||
675 | /** | |
676 | * pvclock_gtod_unregister_notifier - unregister a pvclock | |
677 | * timedata update listener | |
f27f7c3f | 678 | * @nb: Pointer to the notifier block to unregister |
e0b306fe MT |
679 | */ |
680 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb) | |
681 | { | |
e0b306fe MT |
682 | unsigned long flags; |
683 | int ret; | |
684 | ||
9a7a71b1 | 685 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
e0b306fe | 686 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); |
9a7a71b1 | 687 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
e0b306fe MT |
688 | |
689 | return ret; | |
690 | } | |
691 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); | |
692 | ||
833f32d7 JS |
693 | /* |
694 | * tk_update_leap_state - helper to update the next_leap_ktime | |
695 | */ | |
696 | static inline void tk_update_leap_state(struct timekeeper *tk) | |
697 | { | |
698 | tk->next_leap_ktime = ntp_get_next_leap(); | |
2456e855 | 699 | if (tk->next_leap_ktime != KTIME_MAX) |
833f32d7 JS |
700 | /* Convert to monotonic time */ |
701 | tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real); | |
702 | } | |
703 | ||
7c032df5 TG |
704 | /* |
705 | * Update the ktime_t based scalar nsec members of the timekeeper | |
706 | */ | |
707 | static inline void tk_update_ktime_data(struct timekeeper *tk) | |
708 | { | |
9e3680b1 HS |
709 | u64 seconds; |
710 | u32 nsec; | |
7c032df5 TG |
711 | |
712 | /* | |
713 | * The xtime based monotonic readout is: | |
714 | * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now(); | |
715 | * The ktime based monotonic readout is: | |
716 | * nsec = base_mono + now(); | |
717 | * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec | |
718 | */ | |
9e3680b1 HS |
719 | seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec); |
720 | nsec = (u32) tk->wall_to_monotonic.tv_nsec; | |
876e7881 | 721 | tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec); |
f519b1a2 | 722 | |
9e3680b1 HS |
723 | /* |
724 | * The sum of the nanoseconds portions of xtime and | |
725 | * wall_to_monotonic can be greater/equal one second. Take | |
726 | * this into account before updating tk->ktime_sec. | |
727 | */ | |
876e7881 | 728 | nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); |
9e3680b1 HS |
729 | if (nsec >= NSEC_PER_SEC) |
730 | seconds++; | |
731 | tk->ktime_sec = seconds; | |
fc6eead7 JS |
732 | |
733 | /* Update the monotonic raw base */ | |
0bcdc098 | 734 | tk->tkr_raw.base = ns_to_ktime(tk->raw_sec * NSEC_PER_SEC); |
7c032df5 TG |
735 | } |
736 | ||
9a7a71b1 | 737 | /* must hold timekeeper_lock */ |
04397fe9 | 738 | static void timekeeping_update(struct timekeeper *tk, unsigned int action) |
cc06268c | 739 | { |
04397fe9 | 740 | if (action & TK_CLEAR_NTP) { |
f726a697 | 741 | tk->ntp_error = 0; |
cc06268c TG |
742 | ntp_clear(); |
743 | } | |
48cdc135 | 744 | |
833f32d7 | 745 | tk_update_leap_state(tk); |
7c032df5 TG |
746 | tk_update_ktime_data(tk); |
747 | ||
9bf2419f TG |
748 | update_vsyscall(tk); |
749 | update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); | |
750 | ||
4c3711d7 | 751 | tk->tkr_mono.base_real = tk->tkr_mono.base + tk->offs_real; |
4498e746 | 752 | update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono); |
f09cb9a1 | 753 | update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw); |
868a3e91 TG |
754 | |
755 | if (action & TK_CLOCK_WAS_SET) | |
756 | tk->clock_was_set_seq++; | |
d1518326 JS |
757 | /* |
758 | * The mirroring of the data to the shadow-timekeeper needs | |
759 | * to happen last here to ensure we don't over-write the | |
760 | * timekeeper structure on the next update with stale data | |
761 | */ | |
762 | if (action & TK_MIRROR) | |
763 | memcpy(&shadow_timekeeper, &tk_core.timekeeper, | |
764 | sizeof(tk_core.timekeeper)); | |
cc06268c TG |
765 | } |
766 | ||
8524070b | 767 | /** |
155ec602 | 768 | * timekeeping_forward_now - update clock to the current time |
6e5a9190 | 769 | * @tk: Pointer to the timekeeper to update |
8524070b | 770 | * |
9a055117 RZ |
771 | * Forward the current clock to update its state since the last call to |
772 | * update_wall_time(). This is useful before significant clock changes, | |
773 | * as it avoids having to deal with this time offset explicitly. | |
8524070b | 774 | */ |
f726a697 | 775 | static void timekeeping_forward_now(struct timekeeper *tk) |
8524070b | 776 | { |
a5a1d1c2 | 777 | u64 cycle_now, delta; |
8524070b | 778 | |
ceea5e37 | 779 | cycle_now = tk_clock_read(&tk->tkr_mono); |
876e7881 PZ |
780 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); |
781 | tk->tkr_mono.cycle_last = cycle_now; | |
4a4ad80d | 782 | tk->tkr_raw.cycle_last = cycle_now; |
8524070b | 783 | |
876e7881 | 784 | tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult; |
fc6eead7 JS |
785 | tk->tkr_raw.xtime_nsec += delta * tk->tkr_raw.mult; |
786 | ||
fc6eead7 | 787 | tk_normalize_xtime(tk); |
8524070b JS |
788 | } |
789 | ||
790 | /** | |
edca71fe | 791 | * ktime_get_real_ts64 - Returns the time of day in a timespec64. |
8524070b JS |
792 | * @ts: pointer to the timespec to be set |
793 | * | |
edca71fe | 794 | * Returns the time of day in a timespec64 (WARN if suspended). |
8524070b | 795 | */ |
edca71fe | 796 | void ktime_get_real_ts64(struct timespec64 *ts) |
8524070b | 797 | { |
3fdb14fd | 798 | struct timekeeper *tk = &tk_core.timekeeper; |
e1e41b6c | 799 | unsigned int seq; |
acc89612 | 800 | u64 nsecs; |
8524070b | 801 | |
edca71fe AB |
802 | WARN_ON(timekeeping_suspended); |
803 | ||
8524070b | 804 | do { |
3fdb14fd | 805 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 806 | |
4e250fdd | 807 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 808 | nsecs = timekeeping_get_ns(&tk->tkr_mono); |
8524070b | 809 | |
3fdb14fd | 810 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b | 811 | |
ec145bab | 812 | ts->tv_nsec = 0; |
d6d29896 | 813 | timespec64_add_ns(ts, nsecs); |
8524070b | 814 | } |
edca71fe | 815 | EXPORT_SYMBOL(ktime_get_real_ts64); |
8524070b | 816 | |
951ed4d3 MS |
817 | ktime_t ktime_get(void) |
818 | { | |
3fdb14fd | 819 | struct timekeeper *tk = &tk_core.timekeeper; |
951ed4d3 | 820 | unsigned int seq; |
a016a5bd | 821 | ktime_t base; |
acc89612 | 822 | u64 nsecs; |
951ed4d3 MS |
823 | |
824 | WARN_ON(timekeeping_suspended); | |
825 | ||
826 | do { | |
3fdb14fd | 827 | seq = read_seqcount_begin(&tk_core.seq); |
876e7881 PZ |
828 | base = tk->tkr_mono.base; |
829 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
951ed4d3 | 830 | |
3fdb14fd | 831 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
24e4a8c3 | 832 | |
a016a5bd | 833 | return ktime_add_ns(base, nsecs); |
951ed4d3 MS |
834 | } |
835 | EXPORT_SYMBOL_GPL(ktime_get); | |
836 | ||
6374f912 HG |
837 | u32 ktime_get_resolution_ns(void) |
838 | { | |
839 | struct timekeeper *tk = &tk_core.timekeeper; | |
840 | unsigned int seq; | |
841 | u32 nsecs; | |
842 | ||
843 | WARN_ON(timekeeping_suspended); | |
844 | ||
845 | do { | |
846 | seq = read_seqcount_begin(&tk_core.seq); | |
847 | nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift; | |
848 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
849 | ||
850 | return nsecs; | |
851 | } | |
852 | EXPORT_SYMBOL_GPL(ktime_get_resolution_ns); | |
853 | ||
0077dc60 TG |
854 | static ktime_t *offsets[TK_OFFS_MAX] = { |
855 | [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real, | |
a3ed0e43 | 856 | [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot, |
0077dc60 TG |
857 | [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai, |
858 | }; | |
859 | ||
860 | ktime_t ktime_get_with_offset(enum tk_offsets offs) | |
861 | { | |
862 | struct timekeeper *tk = &tk_core.timekeeper; | |
863 | unsigned int seq; | |
864 | ktime_t base, *offset = offsets[offs]; | |
acc89612 | 865 | u64 nsecs; |
0077dc60 TG |
866 | |
867 | WARN_ON(timekeeping_suspended); | |
868 | ||
869 | do { | |
870 | seq = read_seqcount_begin(&tk_core.seq); | |
876e7881 PZ |
871 | base = ktime_add(tk->tkr_mono.base, *offset); |
872 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
0077dc60 TG |
873 | |
874 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
875 | ||
876 | return ktime_add_ns(base, nsecs); | |
877 | ||
878 | } | |
879 | EXPORT_SYMBOL_GPL(ktime_get_with_offset); | |
880 | ||
b9ff604c AB |
881 | ktime_t ktime_get_coarse_with_offset(enum tk_offsets offs) |
882 | { | |
883 | struct timekeeper *tk = &tk_core.timekeeper; | |
884 | unsigned int seq; | |
885 | ktime_t base, *offset = offsets[offs]; | |
e3ff9c36 | 886 | u64 nsecs; |
b9ff604c AB |
887 | |
888 | WARN_ON(timekeeping_suspended); | |
889 | ||
890 | do { | |
891 | seq = read_seqcount_begin(&tk_core.seq); | |
892 | base = ktime_add(tk->tkr_mono.base, *offset); | |
e3ff9c36 | 893 | nsecs = tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift; |
b9ff604c AB |
894 | |
895 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
896 | ||
0354c1a3 | 897 | return ktime_add_ns(base, nsecs); |
b9ff604c AB |
898 | } |
899 | EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset); | |
900 | ||
9a6b5197 TG |
901 | /** |
902 | * ktime_mono_to_any() - convert mononotic time to any other time | |
903 | * @tmono: time to convert. | |
904 | * @offs: which offset to use | |
905 | */ | |
906 | ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs) | |
907 | { | |
908 | ktime_t *offset = offsets[offs]; | |
e1e41b6c | 909 | unsigned int seq; |
9a6b5197 TG |
910 | ktime_t tconv; |
911 | ||
912 | do { | |
913 | seq = read_seqcount_begin(&tk_core.seq); | |
914 | tconv = ktime_add(tmono, *offset); | |
915 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
916 | ||
917 | return tconv; | |
918 | } | |
919 | EXPORT_SYMBOL_GPL(ktime_mono_to_any); | |
920 | ||
f519b1a2 TG |
921 | /** |
922 | * ktime_get_raw - Returns the raw monotonic time in ktime_t format | |
923 | */ | |
924 | ktime_t ktime_get_raw(void) | |
925 | { | |
926 | struct timekeeper *tk = &tk_core.timekeeper; | |
927 | unsigned int seq; | |
928 | ktime_t base; | |
acc89612 | 929 | u64 nsecs; |
f519b1a2 TG |
930 | |
931 | do { | |
932 | seq = read_seqcount_begin(&tk_core.seq); | |
4a4ad80d PZ |
933 | base = tk->tkr_raw.base; |
934 | nsecs = timekeeping_get_ns(&tk->tkr_raw); | |
f519b1a2 TG |
935 | |
936 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
937 | ||
938 | return ktime_add_ns(base, nsecs); | |
939 | } | |
940 | EXPORT_SYMBOL_GPL(ktime_get_raw); | |
941 | ||
951ed4d3 | 942 | /** |
d6d29896 | 943 | * ktime_get_ts64 - get the monotonic clock in timespec64 format |
951ed4d3 MS |
944 | * @ts: pointer to timespec variable |
945 | * | |
946 | * The function calculates the monotonic clock from the realtime | |
947 | * clock and the wall_to_monotonic offset and stores the result | |
5322e4c2 | 948 | * in normalized timespec64 format in the variable pointed to by @ts. |
951ed4d3 | 949 | */ |
d6d29896 | 950 | void ktime_get_ts64(struct timespec64 *ts) |
951ed4d3 | 951 | { |
3fdb14fd | 952 | struct timekeeper *tk = &tk_core.timekeeper; |
d6d29896 | 953 | struct timespec64 tomono; |
951ed4d3 | 954 | unsigned int seq; |
acc89612 | 955 | u64 nsec; |
951ed4d3 MS |
956 | |
957 | WARN_ON(timekeeping_suspended); | |
958 | ||
959 | do { | |
3fdb14fd | 960 | seq = read_seqcount_begin(&tk_core.seq); |
d6d29896 | 961 | ts->tv_sec = tk->xtime_sec; |
876e7881 | 962 | nsec = timekeeping_get_ns(&tk->tkr_mono); |
4e250fdd | 963 | tomono = tk->wall_to_monotonic; |
951ed4d3 | 964 | |
3fdb14fd | 965 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
951ed4d3 | 966 | |
d6d29896 TG |
967 | ts->tv_sec += tomono.tv_sec; |
968 | ts->tv_nsec = 0; | |
969 | timespec64_add_ns(ts, nsec + tomono.tv_nsec); | |
951ed4d3 | 970 | } |
d6d29896 | 971 | EXPORT_SYMBOL_GPL(ktime_get_ts64); |
951ed4d3 | 972 | |
9e3680b1 HS |
973 | /** |
974 | * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC | |
975 | * | |
976 | * Returns the seconds portion of CLOCK_MONOTONIC with a single non | |
977 | * serialized read. tk->ktime_sec is of type 'unsigned long' so this | |
978 | * works on both 32 and 64 bit systems. On 32 bit systems the readout | |
979 | * covers ~136 years of uptime which should be enough to prevent | |
980 | * premature wrap arounds. | |
981 | */ | |
982 | time64_t ktime_get_seconds(void) | |
983 | { | |
984 | struct timekeeper *tk = &tk_core.timekeeper; | |
985 | ||
986 | WARN_ON(timekeeping_suspended); | |
987 | return tk->ktime_sec; | |
988 | } | |
989 | EXPORT_SYMBOL_GPL(ktime_get_seconds); | |
990 | ||
dbe7aa62 HS |
991 | /** |
992 | * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME | |
993 | * | |
994 | * Returns the wall clock seconds since 1970. This replaces the | |
995 | * get_seconds() interface which is not y2038 safe on 32bit systems. | |
996 | * | |
997 | * For 64bit systems the fast access to tk->xtime_sec is preserved. On | |
998 | * 32bit systems the access must be protected with the sequence | |
999 | * counter to provide "atomic" access to the 64bit tk->xtime_sec | |
1000 | * value. | |
1001 | */ | |
1002 | time64_t ktime_get_real_seconds(void) | |
1003 | { | |
1004 | struct timekeeper *tk = &tk_core.timekeeper; | |
1005 | time64_t seconds; | |
1006 | unsigned int seq; | |
1007 | ||
1008 | if (IS_ENABLED(CONFIG_64BIT)) | |
1009 | return tk->xtime_sec; | |
1010 | ||
1011 | do { | |
1012 | seq = read_seqcount_begin(&tk_core.seq); | |
1013 | seconds = tk->xtime_sec; | |
1014 | ||
1015 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
1016 | ||
1017 | return seconds; | |
1018 | } | |
1019 | EXPORT_SYMBOL_GPL(ktime_get_real_seconds); | |
1020 | ||
dee36654 D |
1021 | /** |
1022 | * __ktime_get_real_seconds - The same as ktime_get_real_seconds | |
1023 | * but without the sequence counter protect. This internal function | |
1024 | * is called just when timekeeping lock is already held. | |
1025 | */ | |
865d3a9a | 1026 | noinstr time64_t __ktime_get_real_seconds(void) |
dee36654 D |
1027 | { |
1028 | struct timekeeper *tk = &tk_core.timekeeper; | |
1029 | ||
1030 | return tk->xtime_sec; | |
1031 | } | |
1032 | ||
9da0f49c CH |
1033 | /** |
1034 | * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter | |
1035 | * @systime_snapshot: pointer to struct receiving the system time snapshot | |
1036 | */ | |
1037 | void ktime_get_snapshot(struct system_time_snapshot *systime_snapshot) | |
1038 | { | |
1039 | struct timekeeper *tk = &tk_core.timekeeper; | |
e1e41b6c | 1040 | unsigned int seq; |
9da0f49c CH |
1041 | ktime_t base_raw; |
1042 | ktime_t base_real; | |
acc89612 TG |
1043 | u64 nsec_raw; |
1044 | u64 nsec_real; | |
a5a1d1c2 | 1045 | u64 now; |
9da0f49c | 1046 | |
ba26621e CH |
1047 | WARN_ON_ONCE(timekeeping_suspended); |
1048 | ||
9da0f49c CH |
1049 | do { |
1050 | seq = read_seqcount_begin(&tk_core.seq); | |
ceea5e37 | 1051 | now = tk_clock_read(&tk->tkr_mono); |
2c756feb CH |
1052 | systime_snapshot->cs_was_changed_seq = tk->cs_was_changed_seq; |
1053 | systime_snapshot->clock_was_set_seq = tk->clock_was_set_seq; | |
9da0f49c CH |
1054 | base_real = ktime_add(tk->tkr_mono.base, |
1055 | tk_core.timekeeper.offs_real); | |
1056 | base_raw = tk->tkr_raw.base; | |
1057 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, now); | |
1058 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, now); | |
1059 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
1060 | ||
1061 | systime_snapshot->cycles = now; | |
1062 | systime_snapshot->real = ktime_add_ns(base_real, nsec_real); | |
1063 | systime_snapshot->raw = ktime_add_ns(base_raw, nsec_raw); | |
1064 | } | |
1065 | EXPORT_SYMBOL_GPL(ktime_get_snapshot); | |
dee36654 | 1066 | |
2c756feb CH |
1067 | /* Scale base by mult/div checking for overflow */ |
1068 | static int scale64_check_overflow(u64 mult, u64 div, u64 *base) | |
1069 | { | |
1070 | u64 tmp, rem; | |
1071 | ||
1072 | tmp = div64_u64_rem(*base, div, &rem); | |
1073 | ||
1074 | if (((int)sizeof(u64)*8 - fls64(mult) < fls64(tmp)) || | |
1075 | ((int)sizeof(u64)*8 - fls64(mult) < fls64(rem))) | |
1076 | return -EOVERFLOW; | |
1077 | tmp *= mult; | |
2c756feb | 1078 | |
4cbbc3a0 | 1079 | rem = div64_u64(rem * mult, div); |
2c756feb CH |
1080 | *base = tmp + rem; |
1081 | return 0; | |
1082 | } | |
1083 | ||
1084 | /** | |
1085 | * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval | |
1086 | * @history: Snapshot representing start of history | |
1087 | * @partial_history_cycles: Cycle offset into history (fractional part) | |
1088 | * @total_history_cycles: Total history length in cycles | |
1089 | * @discontinuity: True indicates clock was set on history period | |
1090 | * @ts: Cross timestamp that should be adjusted using | |
1091 | * partial/total ratio | |
1092 | * | |
1093 | * Helper function used by get_device_system_crosststamp() to correct the | |
1094 | * crosstimestamp corresponding to the start of the current interval to the | |
1095 | * system counter value (timestamp point) provided by the driver. The | |
1096 | * total_history_* quantities are the total history starting at the provided | |
1097 | * reference point and ending at the start of the current interval. The cycle | |
1098 | * count between the driver timestamp point and the start of the current | |
1099 | * interval is partial_history_cycles. | |
1100 | */ | |
1101 | static int adjust_historical_crosststamp(struct system_time_snapshot *history, | |
a5a1d1c2 TG |
1102 | u64 partial_history_cycles, |
1103 | u64 total_history_cycles, | |
2c756feb CH |
1104 | bool discontinuity, |
1105 | struct system_device_crosststamp *ts) | |
1106 | { | |
1107 | struct timekeeper *tk = &tk_core.timekeeper; | |
1108 | u64 corr_raw, corr_real; | |
1109 | bool interp_forward; | |
1110 | int ret; | |
1111 | ||
1112 | if (total_history_cycles == 0 || partial_history_cycles == 0) | |
1113 | return 0; | |
1114 | ||
1115 | /* Interpolate shortest distance from beginning or end of history */ | |
5fc63f95 | 1116 | interp_forward = partial_history_cycles > total_history_cycles / 2; |
2c756feb CH |
1117 | partial_history_cycles = interp_forward ? |
1118 | total_history_cycles - partial_history_cycles : | |
1119 | partial_history_cycles; | |
1120 | ||
1121 | /* | |
1122 | * Scale the monotonic raw time delta by: | |
1123 | * partial_history_cycles / total_history_cycles | |
1124 | */ | |
1125 | corr_raw = (u64)ktime_to_ns( | |
1126 | ktime_sub(ts->sys_monoraw, history->raw)); | |
1127 | ret = scale64_check_overflow(partial_history_cycles, | |
1128 | total_history_cycles, &corr_raw); | |
1129 | if (ret) | |
1130 | return ret; | |
1131 | ||
1132 | /* | |
1133 | * If there is a discontinuity in the history, scale monotonic raw | |
1134 | * correction by: | |
1135 | * mult(real)/mult(raw) yielding the realtime correction | |
1136 | * Otherwise, calculate the realtime correction similar to monotonic | |
1137 | * raw calculation | |
1138 | */ | |
1139 | if (discontinuity) { | |
1140 | corr_real = mul_u64_u32_div | |
1141 | (corr_raw, tk->tkr_mono.mult, tk->tkr_raw.mult); | |
1142 | } else { | |
1143 | corr_real = (u64)ktime_to_ns( | |
1144 | ktime_sub(ts->sys_realtime, history->real)); | |
1145 | ret = scale64_check_overflow(partial_history_cycles, | |
1146 | total_history_cycles, &corr_real); | |
1147 | if (ret) | |
1148 | return ret; | |
1149 | } | |
1150 | ||
1151 | /* Fixup monotonic raw and real time time values */ | |
1152 | if (interp_forward) { | |
1153 | ts->sys_monoraw = ktime_add_ns(history->raw, corr_raw); | |
1154 | ts->sys_realtime = ktime_add_ns(history->real, corr_real); | |
1155 | } else { | |
1156 | ts->sys_monoraw = ktime_sub_ns(ts->sys_monoraw, corr_raw); | |
1157 | ts->sys_realtime = ktime_sub_ns(ts->sys_realtime, corr_real); | |
1158 | } | |
1159 | ||
1160 | return 0; | |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * cycle_between - true if test occurs chronologically between before and after | |
1165 | */ | |
a5a1d1c2 | 1166 | static bool cycle_between(u64 before, u64 test, u64 after) |
2c756feb CH |
1167 | { |
1168 | if (test > before && test < after) | |
1169 | return true; | |
1170 | if (test < before && before > after) | |
1171 | return true; | |
1172 | return false; | |
1173 | } | |
1174 | ||
8006c245 CH |
1175 | /** |
1176 | * get_device_system_crosststamp - Synchronously capture system/device timestamp | |
2c756feb | 1177 | * @get_time_fn: Callback to get simultaneous device time and |
8006c245 | 1178 | * system counter from the device driver |
2c756feb CH |
1179 | * @ctx: Context passed to get_time_fn() |
1180 | * @history_begin: Historical reference point used to interpolate system | |
1181 | * time when counter provided by the driver is before the current interval | |
8006c245 CH |
1182 | * @xtstamp: Receives simultaneously captured system and device time |
1183 | * | |
1184 | * Reads a timestamp from a device and correlates it to system time | |
1185 | */ | |
1186 | int get_device_system_crosststamp(int (*get_time_fn) | |
1187 | (ktime_t *device_time, | |
1188 | struct system_counterval_t *sys_counterval, | |
1189 | void *ctx), | |
1190 | void *ctx, | |
2c756feb | 1191 | struct system_time_snapshot *history_begin, |
8006c245 CH |
1192 | struct system_device_crosststamp *xtstamp) |
1193 | { | |
1194 | struct system_counterval_t system_counterval; | |
1195 | struct timekeeper *tk = &tk_core.timekeeper; | |
a5a1d1c2 | 1196 | u64 cycles, now, interval_start; |
6436257b | 1197 | unsigned int clock_was_set_seq = 0; |
8006c245 | 1198 | ktime_t base_real, base_raw; |
acc89612 | 1199 | u64 nsec_real, nsec_raw; |
2c756feb | 1200 | u8 cs_was_changed_seq; |
e1e41b6c | 1201 | unsigned int seq; |
2c756feb | 1202 | bool do_interp; |
8006c245 CH |
1203 | int ret; |
1204 | ||
1205 | do { | |
1206 | seq = read_seqcount_begin(&tk_core.seq); | |
1207 | /* | |
1208 | * Try to synchronously capture device time and a system | |
1209 | * counter value calling back into the device driver | |
1210 | */ | |
1211 | ret = get_time_fn(&xtstamp->device, &system_counterval, ctx); | |
1212 | if (ret) | |
1213 | return ret; | |
1214 | ||
1215 | /* | |
1216 | * Verify that the clocksource associated with the captured | |
1217 | * system counter value is the same as the currently installed | |
1218 | * timekeeper clocksource | |
1219 | */ | |
1220 | if (tk->tkr_mono.clock != system_counterval.cs) | |
1221 | return -ENODEV; | |
2c756feb CH |
1222 | cycles = system_counterval.cycles; |
1223 | ||
1224 | /* | |
1225 | * Check whether the system counter value provided by the | |
1226 | * device driver is on the current timekeeping interval. | |
1227 | */ | |
ceea5e37 | 1228 | now = tk_clock_read(&tk->tkr_mono); |
2c756feb CH |
1229 | interval_start = tk->tkr_mono.cycle_last; |
1230 | if (!cycle_between(interval_start, cycles, now)) { | |
1231 | clock_was_set_seq = tk->clock_was_set_seq; | |
1232 | cs_was_changed_seq = tk->cs_was_changed_seq; | |
1233 | cycles = interval_start; | |
1234 | do_interp = true; | |
1235 | } else { | |
1236 | do_interp = false; | |
1237 | } | |
8006c245 CH |
1238 | |
1239 | base_real = ktime_add(tk->tkr_mono.base, | |
1240 | tk_core.timekeeper.offs_real); | |
1241 | base_raw = tk->tkr_raw.base; | |
1242 | ||
1243 | nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, | |
1244 | system_counterval.cycles); | |
1245 | nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, | |
1246 | system_counterval.cycles); | |
1247 | } while (read_seqcount_retry(&tk_core.seq, seq)); | |
1248 | ||
1249 | xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real); | |
1250 | xtstamp->sys_monoraw = ktime_add_ns(base_raw, nsec_raw); | |
2c756feb CH |
1251 | |
1252 | /* | |
1253 | * Interpolate if necessary, adjusting back from the start of the | |
1254 | * current interval | |
1255 | */ | |
1256 | if (do_interp) { | |
a5a1d1c2 | 1257 | u64 partial_history_cycles, total_history_cycles; |
2c756feb CH |
1258 | bool discontinuity; |
1259 | ||
1260 | /* | |
1261 | * Check that the counter value occurs after the provided | |
1262 | * history reference and that the history doesn't cross a | |
1263 | * clocksource change | |
1264 | */ | |
1265 | if (!history_begin || | |
1266 | !cycle_between(history_begin->cycles, | |
1267 | system_counterval.cycles, cycles) || | |
1268 | history_begin->cs_was_changed_seq != cs_was_changed_seq) | |
1269 | return -EINVAL; | |
1270 | partial_history_cycles = cycles - system_counterval.cycles; | |
1271 | total_history_cycles = cycles - history_begin->cycles; | |
1272 | discontinuity = | |
1273 | history_begin->clock_was_set_seq != clock_was_set_seq; | |
1274 | ||
1275 | ret = adjust_historical_crosststamp(history_begin, | |
1276 | partial_history_cycles, | |
1277 | total_history_cycles, | |
1278 | discontinuity, xtstamp); | |
1279 | if (ret) | |
1280 | return ret; | |
1281 | } | |
1282 | ||
8006c245 CH |
1283 | return 0; |
1284 | } | |
1285 | EXPORT_SYMBOL_GPL(get_device_system_crosststamp); | |
1286 | ||
8524070b | 1287 | /** |
21f7eca5 | 1288 | * do_settimeofday64 - Sets the time of day. |
1289 | * @ts: pointer to the timespec64 variable containing the new time | |
8524070b JS |
1290 | * |
1291 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
1292 | */ | |
21f7eca5 | 1293 | int do_settimeofday64(const struct timespec64 *ts) |
8524070b | 1294 | { |
3fdb14fd | 1295 | struct timekeeper *tk = &tk_core.timekeeper; |
21f7eca5 | 1296 | struct timespec64 ts_delta, xt; |
92c1d3ed | 1297 | unsigned long flags; |
e1d7ba87 | 1298 | int ret = 0; |
8524070b | 1299 | |
7a8e61f8 | 1300 | if (!timespec64_valid_settod(ts)) |
8524070b JS |
1301 | return -EINVAL; |
1302 | ||
9a7a71b1 | 1303 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1304 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1305 | |
4e250fdd | 1306 | timekeeping_forward_now(tk); |
9a055117 | 1307 | |
4e250fdd | 1308 | xt = tk_xtime(tk); |
21f7eca5 | 1309 | ts_delta.tv_sec = ts->tv_sec - xt.tv_sec; |
1310 | ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec; | |
1e75fa8b | 1311 | |
e1d7ba87 WY |
1312 | if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) { |
1313 | ret = -EINVAL; | |
1314 | goto out; | |
1315 | } | |
1316 | ||
7d489d15 | 1317 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta)); |
8524070b | 1318 | |
21f7eca5 | 1319 | tk_set_xtime(tk, ts); |
e1d7ba87 | 1320 | out: |
780427f0 | 1321 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
8524070b | 1322 | |
3fdb14fd | 1323 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1324 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1325 | |
1326 | /* signal hrtimers about time change */ | |
1327 | clock_was_set(); | |
1328 | ||
2d87a067 OM |
1329 | if (!ret) |
1330 | audit_tk_injoffset(ts_delta); | |
1331 | ||
e1d7ba87 | 1332 | return ret; |
8524070b | 1333 | } |
21f7eca5 | 1334 | EXPORT_SYMBOL(do_settimeofday64); |
8524070b | 1335 | |
c528f7c6 JS |
1336 | /** |
1337 | * timekeeping_inject_offset - Adds or subtracts from the current time. | |
6e5a9190 | 1338 | * @ts: Pointer to the timespec variable containing the offset |
c528f7c6 JS |
1339 | * |
1340 | * Adds or subtracts an offset value from the current time. | |
1341 | */ | |
985e6950 | 1342 | static int timekeeping_inject_offset(const struct timespec64 *ts) |
c528f7c6 | 1343 | { |
3fdb14fd | 1344 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1345 | unsigned long flags; |
1572fa03 | 1346 | struct timespec64 tmp; |
4e8b1452 | 1347 | int ret = 0; |
c528f7c6 | 1348 | |
1572fa03 | 1349 | if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC) |
c528f7c6 JS |
1350 | return -EINVAL; |
1351 | ||
9a7a71b1 | 1352 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1353 | write_seqcount_begin(&tk_core.seq); |
c528f7c6 | 1354 | |
4e250fdd | 1355 | timekeeping_forward_now(tk); |
c528f7c6 | 1356 | |
4e8b1452 | 1357 | /* Make sure the proposed value is valid */ |
1572fa03 AB |
1358 | tmp = timespec64_add(tk_xtime(tk), *ts); |
1359 | if (timespec64_compare(&tk->wall_to_monotonic, ts) > 0 || | |
7a8e61f8 | 1360 | !timespec64_valid_settod(&tmp)) { |
4e8b1452 JS |
1361 | ret = -EINVAL; |
1362 | goto error; | |
1363 | } | |
1e75fa8b | 1364 | |
1572fa03 AB |
1365 | tk_xtime_add(tk, ts); |
1366 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *ts)); | |
c528f7c6 | 1367 | |
4e8b1452 | 1368 | error: /* even if we error out, we forwarded the time, so call update */ |
780427f0 | 1369 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
c528f7c6 | 1370 | |
3fdb14fd | 1371 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1372 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
c528f7c6 JS |
1373 | |
1374 | /* signal hrtimers about time change */ | |
1375 | clock_was_set(); | |
1376 | ||
4e8b1452 | 1377 | return ret; |
c528f7c6 | 1378 | } |
e0956dcc AB |
1379 | |
1380 | /* | |
1381 | * Indicates if there is an offset between the system clock and the hardware | |
1382 | * clock/persistent clock/rtc. | |
1383 | */ | |
1384 | int persistent_clock_is_local; | |
1385 | ||
1386 | /* | |
1387 | * Adjust the time obtained from the CMOS to be UTC time instead of | |
1388 | * local time. | |
1389 | * | |
1390 | * This is ugly, but preferable to the alternatives. Otherwise we | |
1391 | * would either need to write a program to do it in /etc/rc (and risk | |
1392 | * confusion if the program gets run more than once; it would also be | |
1393 | * hard to make the program warp the clock precisely n hours) or | |
1394 | * compile in the timezone information into the kernel. Bad, bad.... | |
1395 | * | |
1396 | * - TYT, 1992-01-01 | |
1397 | * | |
1398 | * The best thing to do is to keep the CMOS clock in universal time (UTC) | |
1399 | * as real UNIX machines always do it. This avoids all headaches about | |
1400 | * daylight saving times and warping kernel clocks. | |
1401 | */ | |
1402 | void timekeeping_warp_clock(void) | |
1403 | { | |
1404 | if (sys_tz.tz_minuteswest != 0) { | |
1572fa03 | 1405 | struct timespec64 adjust; |
e0956dcc AB |
1406 | |
1407 | persistent_clock_is_local = 1; | |
1408 | adjust.tv_sec = sys_tz.tz_minuteswest * 60; | |
1409 | adjust.tv_nsec = 0; | |
1410 | timekeeping_inject_offset(&adjust); | |
1411 | } | |
1412 | } | |
c528f7c6 | 1413 | |
199d280c | 1414 | /* |
40d9f827 | 1415 | * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic |
cc244dda | 1416 | */ |
dd5d70e8 | 1417 | static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) |
cc244dda JS |
1418 | { |
1419 | tk->tai_offset = tai_offset; | |
04005f60 | 1420 | tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); |
cc244dda JS |
1421 | } |
1422 | ||
199d280c | 1423 | /* |
8524070b JS |
1424 | * change_clocksource - Swaps clocksources if a new one is available |
1425 | * | |
1426 | * Accumulates current time interval and initializes new clocksource | |
1427 | */ | |
75c5158f | 1428 | static int change_clocksource(void *data) |
8524070b | 1429 | { |
3fdb14fd | 1430 | struct timekeeper *tk = &tk_core.timekeeper; |
4614e6ad | 1431 | struct clocksource *new, *old; |
f695cf94 | 1432 | unsigned long flags; |
8524070b | 1433 | |
75c5158f | 1434 | new = (struct clocksource *) data; |
8524070b | 1435 | |
9a7a71b1 | 1436 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1437 | write_seqcount_begin(&tk_core.seq); |
f695cf94 | 1438 | |
4e250fdd | 1439 | timekeeping_forward_now(tk); |
09ac369c TG |
1440 | /* |
1441 | * If the cs is in module, get a module reference. Succeeds | |
1442 | * for built-in code (owner == NULL) as well. | |
1443 | */ | |
1444 | if (try_module_get(new->owner)) { | |
1445 | if (!new->enable || new->enable(new) == 0) { | |
876e7881 | 1446 | old = tk->tkr_mono.clock; |
09ac369c TG |
1447 | tk_setup_internals(tk, new); |
1448 | if (old->disable) | |
1449 | old->disable(old); | |
1450 | module_put(old->owner); | |
1451 | } else { | |
1452 | module_put(new->owner); | |
1453 | } | |
75c5158f | 1454 | } |
780427f0 | 1455 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
f695cf94 | 1456 | |
3fdb14fd | 1457 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1458 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
f695cf94 | 1459 | |
75c5158f MS |
1460 | return 0; |
1461 | } | |
8524070b | 1462 | |
75c5158f MS |
1463 | /** |
1464 | * timekeeping_notify - Install a new clock source | |
1465 | * @clock: pointer to the clock source | |
1466 | * | |
1467 | * This function is called from clocksource.c after a new, better clock | |
1468 | * source has been registered. The caller holds the clocksource_mutex. | |
1469 | */ | |
ba919d1c | 1470 | int timekeeping_notify(struct clocksource *clock) |
75c5158f | 1471 | { |
3fdb14fd | 1472 | struct timekeeper *tk = &tk_core.timekeeper; |
4e250fdd | 1473 | |
876e7881 | 1474 | if (tk->tkr_mono.clock == clock) |
ba919d1c | 1475 | return 0; |
75c5158f | 1476 | stop_machine(change_clocksource, clock, NULL); |
8524070b | 1477 | tick_clock_notify(); |
876e7881 | 1478 | return tk->tkr_mono.clock == clock ? 0 : -1; |
8524070b | 1479 | } |
75c5158f | 1480 | |
2d42244a | 1481 | /** |
fb7fcc96 | 1482 | * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec |
cdba2ec5 | 1483 | * @ts: pointer to the timespec64 to be set |
2d42244a JS |
1484 | * |
1485 | * Returns the raw monotonic time (completely un-modified by ntp) | |
1486 | */ | |
fb7fcc96 | 1487 | void ktime_get_raw_ts64(struct timespec64 *ts) |
2d42244a | 1488 | { |
3fdb14fd | 1489 | struct timekeeper *tk = &tk_core.timekeeper; |
e1e41b6c | 1490 | unsigned int seq; |
acc89612 | 1491 | u64 nsecs; |
2d42244a JS |
1492 | |
1493 | do { | |
3fdb14fd | 1494 | seq = read_seqcount_begin(&tk_core.seq); |
fc6eead7 | 1495 | ts->tv_sec = tk->raw_sec; |
4a4ad80d | 1496 | nsecs = timekeeping_get_ns(&tk->tkr_raw); |
2d42244a | 1497 | |
3fdb14fd | 1498 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2d42244a | 1499 | |
fc6eead7 JS |
1500 | ts->tv_nsec = 0; |
1501 | timespec64_add_ns(ts, nsecs); | |
2d42244a | 1502 | } |
fb7fcc96 | 1503 | EXPORT_SYMBOL(ktime_get_raw_ts64); |
cdba2ec5 | 1504 | |
2d42244a | 1505 | |
8524070b | 1506 | /** |
cf4fc6cb | 1507 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres |
8524070b | 1508 | */ |
cf4fc6cb | 1509 | int timekeeping_valid_for_hres(void) |
8524070b | 1510 | { |
3fdb14fd | 1511 | struct timekeeper *tk = &tk_core.timekeeper; |
e1e41b6c | 1512 | unsigned int seq; |
8524070b JS |
1513 | int ret; |
1514 | ||
1515 | do { | |
3fdb14fd | 1516 | seq = read_seqcount_begin(&tk_core.seq); |
8524070b | 1517 | |
876e7881 | 1518 | ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; |
8524070b | 1519 | |
3fdb14fd | 1520 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
8524070b JS |
1521 | |
1522 | return ret; | |
1523 | } | |
1524 | ||
98962465 JH |
1525 | /** |
1526 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred | |
98962465 JH |
1527 | */ |
1528 | u64 timekeeping_max_deferment(void) | |
1529 | { | |
3fdb14fd | 1530 | struct timekeeper *tk = &tk_core.timekeeper; |
e1e41b6c | 1531 | unsigned int seq; |
70471f2f | 1532 | u64 ret; |
42e71e81 | 1533 | |
70471f2f | 1534 | do { |
3fdb14fd | 1535 | seq = read_seqcount_begin(&tk_core.seq); |
70471f2f | 1536 | |
876e7881 | 1537 | ret = tk->tkr_mono.clock->max_idle_ns; |
70471f2f | 1538 | |
3fdb14fd | 1539 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
70471f2f JS |
1540 | |
1541 | return ret; | |
98962465 JH |
1542 | } |
1543 | ||
8524070b | 1544 | /** |
92661788 | 1545 | * read_persistent_clock64 - Return time from the persistent clock. |
6e5a9190 | 1546 | * @ts: Pointer to the storage for the readout value |
8524070b JS |
1547 | * |
1548 | * Weak dummy function for arches that do not yet support it. | |
d4f587c6 MS |
1549 | * Reads the time from the battery backed persistent clock. |
1550 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. | |
8524070b JS |
1551 | * |
1552 | * XXX - Do be sure to remove it once all arches implement it. | |
1553 | */ | |
92661788 | 1554 | void __weak read_persistent_clock64(struct timespec64 *ts) |
8524070b | 1555 | { |
d4f587c6 MS |
1556 | ts->tv_sec = 0; |
1557 | ts->tv_nsec = 0; | |
8524070b JS |
1558 | } |
1559 | ||
23970e38 | 1560 | /** |
3eca9937 PT |
1561 | * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset |
1562 | * from the boot. | |
23970e38 MS |
1563 | * |
1564 | * Weak dummy function for arches that do not yet support it. | |
29efc461 AS |
1565 | * @wall_time: - current time as returned by persistent clock |
1566 | * @boot_offset: - offset that is defined as wall_time - boot_time | |
1567 | * | |
4b1b7f80 PT |
1568 | * The default function calculates offset based on the current value of |
1569 | * local_clock(). This way architectures that support sched_clock() but don't | |
1570 | * support dedicated boot time clock will provide the best estimate of the | |
1571 | * boot time. | |
23970e38 | 1572 | */ |
3eca9937 PT |
1573 | void __weak __init |
1574 | read_persistent_wall_and_boot_offset(struct timespec64 *wall_time, | |
1575 | struct timespec64 *boot_offset) | |
23970e38 | 1576 | { |
3eca9937 | 1577 | read_persistent_clock64(wall_time); |
4b1b7f80 | 1578 | *boot_offset = ns_to_timespec64(local_clock()); |
23970e38 MS |
1579 | } |
1580 | ||
f473e5f4 MO |
1581 | /* |
1582 | * Flag reflecting whether timekeeping_resume() has injected sleeptime. | |
1583 | * | |
1584 | * The flag starts of false and is only set when a suspend reaches | |
1585 | * timekeeping_suspend(), timekeeping_resume() sets it to false when the | |
1586 | * timekeeper clocksource is not stopping across suspend and has been | |
1587 | * used to update sleep time. If the timekeeper clocksource has stopped | |
1588 | * then the flag stays true and is used by the RTC resume code to decide | |
1589 | * whether sleeptime must be injected and if so the flag gets false then. | |
1590 | * | |
1591 | * If a suspend fails before reaching timekeeping_resume() then the flag | |
1592 | * stays false and prevents erroneous sleeptime injection. | |
1593 | */ | |
1594 | static bool suspend_timing_needed; | |
0fa88cb4 XP |
1595 | |
1596 | /* Flag for if there is a persistent clock on this platform */ | |
1597 | static bool persistent_clock_exists; | |
1598 | ||
8524070b JS |
1599 | /* |
1600 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
1601 | */ | |
1602 | void __init timekeeping_init(void) | |
1603 | { | |
3eca9937 | 1604 | struct timespec64 wall_time, boot_offset, wall_to_mono; |
3fdb14fd | 1605 | struct timekeeper *tk = &tk_core.timekeeper; |
155ec602 | 1606 | struct clocksource *clock; |
8524070b | 1607 | unsigned long flags; |
4e8b1452 | 1608 | |
3eca9937 | 1609 | read_persistent_wall_and_boot_offset(&wall_time, &boot_offset); |
7a8e61f8 | 1610 | if (timespec64_valid_settod(&wall_time) && |
3eca9937 PT |
1611 | timespec64_to_ns(&wall_time) > 0) { |
1612 | persistent_clock_exists = true; | |
684ad537 | 1613 | } else if (timespec64_to_ns(&wall_time) != 0) { |
3eca9937 PT |
1614 | pr_warn("Persistent clock returned invalid value"); |
1615 | wall_time = (struct timespec64){0}; | |
4e8b1452 | 1616 | } |
8524070b | 1617 | |
3eca9937 PT |
1618 | if (timespec64_compare(&wall_time, &boot_offset) < 0) |
1619 | boot_offset = (struct timespec64){0}; | |
1620 | ||
1621 | /* | |
1622 | * We want set wall_to_mono, so the following is true: | |
1623 | * wall time + wall_to_mono = boot time | |
1624 | */ | |
1625 | wall_to_mono = timespec64_sub(boot_offset, wall_time); | |
1626 | ||
9a7a71b1 | 1627 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1628 | write_seqcount_begin(&tk_core.seq); |
06c017fd JS |
1629 | ntp_init(); |
1630 | ||
f1b82746 | 1631 | clock = clocksource_default_clock(); |
a0f7d48b MS |
1632 | if (clock->enable) |
1633 | clock->enable(clock); | |
4e250fdd | 1634 | tk_setup_internals(tk, clock); |
8524070b | 1635 | |
3eca9937 | 1636 | tk_set_xtime(tk, &wall_time); |
fc6eead7 | 1637 | tk->raw_sec = 0; |
1e75fa8b | 1638 | |
3eca9937 | 1639 | tk_set_wall_to_mono(tk, wall_to_mono); |
6d0ef903 | 1640 | |
56fd16ca | 1641 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
48cdc135 | 1642 | |
3fdb14fd | 1643 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1644 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1645 | } |
1646 | ||
264bb3f7 | 1647 | /* time in seconds when suspend began for persistent clock */ |
7d489d15 | 1648 | static struct timespec64 timekeeping_suspend_time; |
8524070b | 1649 | |
304529b1 JS |
1650 | /** |
1651 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval | |
6e5a9190 AS |
1652 | * @tk: Pointer to the timekeeper to be updated |
1653 | * @delta: Pointer to the delta value in timespec64 format | |
304529b1 JS |
1654 | * |
1655 | * Takes a timespec offset measuring a suspend interval and properly | |
1656 | * adds the sleep offset to the timekeeping variables. | |
1657 | */ | |
f726a697 | 1658 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk, |
985e6950 | 1659 | const struct timespec64 *delta) |
304529b1 | 1660 | { |
7d489d15 | 1661 | if (!timespec64_valid_strict(delta)) { |
6d9bcb62 JS |
1662 | printk_deferred(KERN_WARNING |
1663 | "__timekeeping_inject_sleeptime: Invalid " | |
1664 | "sleep delta value!\n"); | |
cb5de2f8 JS |
1665 | return; |
1666 | } | |
f726a697 | 1667 | tk_xtime_add(tk, delta); |
a3ed0e43 | 1668 | tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta)); |
47da70d3 | 1669 | tk_update_sleep_time(tk, timespec64_to_ktime(*delta)); |
5c83545f | 1670 | tk_debug_account_sleep_time(delta); |
304529b1 JS |
1671 | } |
1672 | ||
7f298139 | 1673 | #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE) |
0fa88cb4 XP |
1674 | /** |
1675 | * We have three kinds of time sources to use for sleep time | |
1676 | * injection, the preference order is: | |
1677 | * 1) non-stop clocksource | |
1678 | * 2) persistent clock (ie: RTC accessible when irqs are off) | |
1679 | * 3) RTC | |
1680 | * | |
1681 | * 1) and 2) are used by timekeeping, 3) by RTC subsystem. | |
1682 | * If system has neither 1) nor 2), 3) will be used finally. | |
1683 | * | |
1684 | * | |
1685 | * If timekeeping has injected sleeptime via either 1) or 2), | |
1686 | * 3) becomes needless, so in this case we don't need to call | |
1687 | * rtc_resume(), and this is what timekeeping_rtc_skipresume() | |
1688 | * means. | |
1689 | */ | |
1690 | bool timekeeping_rtc_skipresume(void) | |
1691 | { | |
f473e5f4 | 1692 | return !suspend_timing_needed; |
0fa88cb4 XP |
1693 | } |
1694 | ||
1695 | /** | |
1696 | * 1) can be determined whether to use or not only when doing | |
1697 | * timekeeping_resume() which is invoked after rtc_suspend(), | |
1698 | * so we can't skip rtc_suspend() surely if system has 1). | |
1699 | * | |
1700 | * But if system has 2), 2) will definitely be used, so in this | |
1701 | * case we don't need to call rtc_suspend(), and this is what | |
1702 | * timekeeping_rtc_skipsuspend() means. | |
1703 | */ | |
1704 | bool timekeeping_rtc_skipsuspend(void) | |
1705 | { | |
1706 | return persistent_clock_exists; | |
1707 | } | |
1708 | ||
304529b1 | 1709 | /** |
04d90890 | 1710 | * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values |
1711 | * @delta: pointer to a timespec64 delta value | |
304529b1 | 1712 | * |
2ee96632 | 1713 | * This hook is for architectures that cannot support read_persistent_clock64 |
304529b1 | 1714 | * because their RTC/persistent clock is only accessible when irqs are enabled. |
0fa88cb4 | 1715 | * and also don't have an effective nonstop clocksource. |
304529b1 JS |
1716 | * |
1717 | * This function should only be called by rtc_resume(), and allows | |
1718 | * a suspend offset to be injected into the timekeeping values. | |
1719 | */ | |
985e6950 | 1720 | void timekeeping_inject_sleeptime64(const struct timespec64 *delta) |
304529b1 | 1721 | { |
3fdb14fd | 1722 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1723 | unsigned long flags; |
304529b1 | 1724 | |
9a7a71b1 | 1725 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1726 | write_seqcount_begin(&tk_core.seq); |
70471f2f | 1727 | |
f473e5f4 MO |
1728 | suspend_timing_needed = false; |
1729 | ||
4e250fdd | 1730 | timekeeping_forward_now(tk); |
304529b1 | 1731 | |
04d90890 | 1732 | __timekeeping_inject_sleeptime(tk, delta); |
304529b1 | 1733 | |
780427f0 | 1734 | timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); |
304529b1 | 1735 | |
3fdb14fd | 1736 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1737 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
304529b1 JS |
1738 | |
1739 | /* signal hrtimers about time change */ | |
1740 | clock_was_set(); | |
1741 | } | |
7f298139 | 1742 | #endif |
304529b1 | 1743 | |
8524070b JS |
1744 | /** |
1745 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
8524070b | 1746 | */ |
124cf911 | 1747 | void timekeeping_resume(void) |
8524070b | 1748 | { |
3fdb14fd | 1749 | struct timekeeper *tk = &tk_core.timekeeper; |
876e7881 | 1750 | struct clocksource *clock = tk->tkr_mono.clock; |
92c1d3ed | 1751 | unsigned long flags; |
7d489d15 | 1752 | struct timespec64 ts_new, ts_delta; |
39232ed5 | 1753 | u64 cycle_now, nsec; |
f473e5f4 | 1754 | bool inject_sleeptime = false; |
d4f587c6 | 1755 | |
2ee96632 | 1756 | read_persistent_clock64(&ts_new); |
8524070b | 1757 | |
adc78e6b | 1758 | clockevents_resume(); |
d10ff3fb TG |
1759 | clocksource_resume(); |
1760 | ||
9a7a71b1 | 1761 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1762 | write_seqcount_begin(&tk_core.seq); |
8524070b | 1763 | |
e445cf1c FT |
1764 | /* |
1765 | * After system resumes, we need to calculate the suspended time and | |
1766 | * compensate it for the OS time. There are 3 sources that could be | |
1767 | * used: Nonstop clocksource during suspend, persistent clock and rtc | |
1768 | * device. | |
1769 | * | |
1770 | * One specific platform may have 1 or 2 or all of them, and the | |
1771 | * preference will be: | |
1772 | * suspend-nonstop clocksource -> persistent clock -> rtc | |
1773 | * The less preferred source will only be tried if there is no better | |
1774 | * usable source. The rtc part is handled separately in rtc core code. | |
1775 | */ | |
ceea5e37 | 1776 | cycle_now = tk_clock_read(&tk->tkr_mono); |
39232ed5 BW |
1777 | nsec = clocksource_stop_suspend_timing(clock, cycle_now); |
1778 | if (nsec > 0) { | |
7d489d15 | 1779 | ts_delta = ns_to_timespec64(nsec); |
f473e5f4 | 1780 | inject_sleeptime = true; |
7d489d15 JS |
1781 | } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) { |
1782 | ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time); | |
f473e5f4 | 1783 | inject_sleeptime = true; |
8524070b | 1784 | } |
e445cf1c | 1785 | |
f473e5f4 MO |
1786 | if (inject_sleeptime) { |
1787 | suspend_timing_needed = false; | |
e445cf1c | 1788 | __timekeeping_inject_sleeptime(tk, &ts_delta); |
f473e5f4 | 1789 | } |
e445cf1c FT |
1790 | |
1791 | /* Re-base the last cycle value */ | |
876e7881 | 1792 | tk->tkr_mono.cycle_last = cycle_now; |
4a4ad80d PZ |
1793 | tk->tkr_raw.cycle_last = cycle_now; |
1794 | ||
4e250fdd | 1795 | tk->ntp_error = 0; |
8524070b | 1796 | timekeeping_suspended = 0; |
780427f0 | 1797 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
3fdb14fd | 1798 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1799 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b JS |
1800 | |
1801 | touch_softlockup_watchdog(); | |
1802 | ||
4ffee521 | 1803 | tick_resume(); |
b12a03ce | 1804 | hrtimers_resume(); |
8524070b JS |
1805 | } |
1806 | ||
124cf911 | 1807 | int timekeeping_suspend(void) |
8524070b | 1808 | { |
3fdb14fd | 1809 | struct timekeeper *tk = &tk_core.timekeeper; |
92c1d3ed | 1810 | unsigned long flags; |
7d489d15 JS |
1811 | struct timespec64 delta, delta_delta; |
1812 | static struct timespec64 old_delta; | |
39232ed5 BW |
1813 | struct clocksource *curr_clock; |
1814 | u64 cycle_now; | |
8524070b | 1815 | |
2ee96632 | 1816 | read_persistent_clock64(&timekeeping_suspend_time); |
3be90950 | 1817 | |
0d6bd995 ZM |
1818 | /* |
1819 | * On some systems the persistent_clock can not be detected at | |
1820 | * timekeeping_init by its return value, so if we see a valid | |
1821 | * value returned, update the persistent_clock_exists flag. | |
1822 | */ | |
1823 | if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) | |
0fa88cb4 | 1824 | persistent_clock_exists = true; |
0d6bd995 | 1825 | |
f473e5f4 MO |
1826 | suspend_timing_needed = true; |
1827 | ||
9a7a71b1 | 1828 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 1829 | write_seqcount_begin(&tk_core.seq); |
4e250fdd | 1830 | timekeeping_forward_now(tk); |
8524070b | 1831 | timekeeping_suspended = 1; |
cb33217b | 1832 | |
39232ed5 BW |
1833 | /* |
1834 | * Since we've called forward_now, cycle_last stores the value | |
1835 | * just read from the current clocksource. Save this to potentially | |
1836 | * use in suspend timing. | |
1837 | */ | |
1838 | curr_clock = tk->tkr_mono.clock; | |
1839 | cycle_now = tk->tkr_mono.cycle_last; | |
1840 | clocksource_start_suspend_timing(curr_clock, cycle_now); | |
1841 | ||
0fa88cb4 | 1842 | if (persistent_clock_exists) { |
cb33217b | 1843 | /* |
264bb3f7 XP |
1844 | * To avoid drift caused by repeated suspend/resumes, |
1845 | * which each can add ~1 second drift error, | |
1846 | * try to compensate so the difference in system time | |
1847 | * and persistent_clock time stays close to constant. | |
cb33217b | 1848 | */ |
264bb3f7 XP |
1849 | delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time); |
1850 | delta_delta = timespec64_sub(delta, old_delta); | |
1851 | if (abs(delta_delta.tv_sec) >= 2) { | |
1852 | /* | |
1853 | * if delta_delta is too large, assume time correction | |
1854 | * has occurred and set old_delta to the current delta. | |
1855 | */ | |
1856 | old_delta = delta; | |
1857 | } else { | |
1858 | /* Otherwise try to adjust old_system to compensate */ | |
1859 | timekeeping_suspend_time = | |
1860 | timespec64_add(timekeeping_suspend_time, delta_delta); | |
1861 | } | |
cb33217b | 1862 | } |
330a1617 JS |
1863 | |
1864 | timekeeping_update(tk, TK_MIRROR); | |
060407ae | 1865 | halt_fast_timekeeper(tk); |
3fdb14fd | 1866 | write_seqcount_end(&tk_core.seq); |
9a7a71b1 | 1867 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
8524070b | 1868 | |
4ffee521 | 1869 | tick_suspend(); |
c54a42b1 | 1870 | clocksource_suspend(); |
adc78e6b | 1871 | clockevents_suspend(); |
8524070b JS |
1872 | |
1873 | return 0; | |
1874 | } | |
1875 | ||
1876 | /* sysfs resume/suspend bits for timekeeping */ | |
e1a85b2c | 1877 | static struct syscore_ops timekeeping_syscore_ops = { |
8524070b JS |
1878 | .resume = timekeeping_resume, |
1879 | .suspend = timekeeping_suspend, | |
8524070b JS |
1880 | }; |
1881 | ||
e1a85b2c | 1882 | static int __init timekeeping_init_ops(void) |
8524070b | 1883 | { |
e1a85b2c RW |
1884 | register_syscore_ops(&timekeeping_syscore_ops); |
1885 | return 0; | |
8524070b | 1886 | } |
e1a85b2c | 1887 | device_initcall(timekeeping_init_ops); |
8524070b JS |
1888 | |
1889 | /* | |
dc491596 | 1890 | * Apply a multiplier adjustment to the timekeeper |
8524070b | 1891 | */ |
dc491596 JS |
1892 | static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk, |
1893 | s64 offset, | |
78b98e3c | 1894 | s32 mult_adj) |
8524070b | 1895 | { |
dc491596 | 1896 | s64 interval = tk->cycle_interval; |
8524070b | 1897 | |
78b98e3c ML |
1898 | if (mult_adj == 0) { |
1899 | return; | |
1900 | } else if (mult_adj == -1) { | |
dc491596 | 1901 | interval = -interval; |
78b98e3c ML |
1902 | offset = -offset; |
1903 | } else if (mult_adj != 1) { | |
1904 | interval *= mult_adj; | |
1905 | offset *= mult_adj; | |
1d17d174 | 1906 | } |
8524070b | 1907 | |
c2bc1111 JS |
1908 | /* |
1909 | * So the following can be confusing. | |
1910 | * | |
dc491596 | 1911 | * To keep things simple, lets assume mult_adj == 1 for now. |
c2bc1111 | 1912 | * |
dc491596 | 1913 | * When mult_adj != 1, remember that the interval and offset values |
c2bc1111 JS |
1914 | * have been appropriately scaled so the math is the same. |
1915 | * | |
1916 | * The basic idea here is that we're increasing the multiplier | |
1917 | * by one, this causes the xtime_interval to be incremented by | |
1918 | * one cycle_interval. This is because: | |
1919 | * xtime_interval = cycle_interval * mult | |
1920 | * So if mult is being incremented by one: | |
1921 | * xtime_interval = cycle_interval * (mult + 1) | |
1922 | * Its the same as: | |
1923 | * xtime_interval = (cycle_interval * mult) + cycle_interval | |
1924 | * Which can be shortened to: | |
1925 | * xtime_interval += cycle_interval | |
1926 | * | |
1927 | * So offset stores the non-accumulated cycles. Thus the current | |
1928 | * time (in shifted nanoseconds) is: | |
1929 | * now = (offset * adj) + xtime_nsec | |
1930 | * Now, even though we're adjusting the clock frequency, we have | |
1931 | * to keep time consistent. In other words, we can't jump back | |
1932 | * in time, and we also want to avoid jumping forward in time. | |
1933 | * | |
1934 | * So given the same offset value, we need the time to be the same | |
1935 | * both before and after the freq adjustment. | |
1936 | * now = (offset * adj_1) + xtime_nsec_1 | |
1937 | * now = (offset * adj_2) + xtime_nsec_2 | |
1938 | * So: | |
1939 | * (offset * adj_1) + xtime_nsec_1 = | |
1940 | * (offset * adj_2) + xtime_nsec_2 | |
1941 | * And we know: | |
1942 | * adj_2 = adj_1 + 1 | |
1943 | * So: | |
1944 | * (offset * adj_1) + xtime_nsec_1 = | |
1945 | * (offset * (adj_1+1)) + xtime_nsec_2 | |
1946 | * (offset * adj_1) + xtime_nsec_1 = | |
1947 | * (offset * adj_1) + offset + xtime_nsec_2 | |
1948 | * Canceling the sides: | |
1949 | * xtime_nsec_1 = offset + xtime_nsec_2 | |
1950 | * Which gives us: | |
1951 | * xtime_nsec_2 = xtime_nsec_1 - offset | |
1952 | * Which simplfies to: | |
1953 | * xtime_nsec -= offset | |
c2bc1111 | 1954 | */ |
876e7881 | 1955 | if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) { |
6067dc5a | 1956 | /* NTP adjustment caused clocksource mult overflow */ |
1957 | WARN_ON_ONCE(1); | |
1958 | return; | |
1959 | } | |
1960 | ||
876e7881 | 1961 | tk->tkr_mono.mult += mult_adj; |
f726a697 | 1962 | tk->xtime_interval += interval; |
876e7881 | 1963 | tk->tkr_mono.xtime_nsec -= offset; |
dc491596 JS |
1964 | } |
1965 | ||
1966 | /* | |
78b98e3c ML |
1967 | * Adjust the timekeeper's multiplier to the correct frequency |
1968 | * and also to reduce the accumulated error value. | |
dc491596 | 1969 | */ |
78b98e3c | 1970 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset) |
dc491596 | 1971 | { |
78b98e3c | 1972 | u32 mult; |
dc491596 | 1973 | |
ec02b076 | 1974 | /* |
78b98e3c ML |
1975 | * Determine the multiplier from the current NTP tick length. |
1976 | * Avoid expensive division when the tick length doesn't change. | |
ec02b076 | 1977 | */ |
78b98e3c ML |
1978 | if (likely(tk->ntp_tick == ntp_tick_length())) { |
1979 | mult = tk->tkr_mono.mult - tk->ntp_err_mult; | |
1980 | } else { | |
1981 | tk->ntp_tick = ntp_tick_length(); | |
1982 | mult = div64_u64((tk->ntp_tick >> tk->ntp_error_shift) - | |
1983 | tk->xtime_remainder, tk->cycle_interval); | |
ec02b076 | 1984 | } |
dc491596 | 1985 | |
78b98e3c ML |
1986 | /* |
1987 | * If the clock is behind the NTP time, increase the multiplier by 1 | |
1988 | * to catch up with it. If it's ahead and there was a remainder in the | |
1989 | * tick division, the clock will slow down. Otherwise it will stay | |
1990 | * ahead until the tick length changes to a non-divisible value. | |
1991 | */ | |
1992 | tk->ntp_err_mult = tk->ntp_error > 0 ? 1 : 0; | |
1993 | mult += tk->ntp_err_mult; | |
dc491596 | 1994 | |
78b98e3c | 1995 | timekeeping_apply_adjustment(tk, offset, mult - tk->tkr_mono.mult); |
dc491596 | 1996 | |
876e7881 PZ |
1997 | if (unlikely(tk->tkr_mono.clock->maxadj && |
1998 | (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult) | |
1999 | > tk->tkr_mono.clock->maxadj))) { | |
dc491596 JS |
2000 | printk_once(KERN_WARNING |
2001 | "Adjusting %s more than 11%% (%ld vs %ld)\n", | |
876e7881 PZ |
2002 | tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult, |
2003 | (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj); | |
dc491596 | 2004 | } |
2a8c0883 JS |
2005 | |
2006 | /* | |
2007 | * It may be possible that when we entered this function, xtime_nsec | |
2008 | * was very small. Further, if we're slightly speeding the clocksource | |
2009 | * in the code above, its possible the required corrective factor to | |
2010 | * xtime_nsec could cause it to underflow. | |
2011 | * | |
78b98e3c ML |
2012 | * Now, since we have already accumulated the second and the NTP |
2013 | * subsystem has been notified via second_overflow(), we need to skip | |
2014 | * the next update. | |
2a8c0883 | 2015 | */ |
876e7881 | 2016 | if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) { |
78b98e3c ML |
2017 | tk->tkr_mono.xtime_nsec += (u64)NSEC_PER_SEC << |
2018 | tk->tkr_mono.shift; | |
2019 | tk->xtime_sec--; | |
2020 | tk->skip_second_overflow = 1; | |
2a8c0883 | 2021 | } |
8524070b JS |
2022 | } |
2023 | ||
199d280c | 2024 | /* |
1f4f9487 JS |
2025 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs |
2026 | * | |
571af55a | 2027 | * Helper function that accumulates the nsecs greater than a second |
1f4f9487 JS |
2028 | * from the xtime_nsec field to the xtime_secs field. |
2029 | * It also calls into the NTP code to handle leapsecond processing. | |
1f4f9487 | 2030 | */ |
780427f0 | 2031 | static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) |
1f4f9487 | 2032 | { |
876e7881 | 2033 | u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift; |
5258d3f2 | 2034 | unsigned int clock_set = 0; |
1f4f9487 | 2035 | |
876e7881 | 2036 | while (tk->tkr_mono.xtime_nsec >= nsecps) { |
1f4f9487 JS |
2037 | int leap; |
2038 | ||
876e7881 | 2039 | tk->tkr_mono.xtime_nsec -= nsecps; |
1f4f9487 JS |
2040 | tk->xtime_sec++; |
2041 | ||
78b98e3c ML |
2042 | /* |
2043 | * Skip NTP update if this second was accumulated before, | |
2044 | * i.e. xtime_nsec underflowed in timekeeping_adjust() | |
2045 | */ | |
2046 | if (unlikely(tk->skip_second_overflow)) { | |
2047 | tk->skip_second_overflow = 0; | |
2048 | continue; | |
2049 | } | |
2050 | ||
1f4f9487 JS |
2051 | /* Figure out if its a leap sec and apply if needed */ |
2052 | leap = second_overflow(tk->xtime_sec); | |
6d0ef903 | 2053 | if (unlikely(leap)) { |
7d489d15 | 2054 | struct timespec64 ts; |
6d0ef903 JS |
2055 | |
2056 | tk->xtime_sec += leap; | |
1f4f9487 | 2057 | |
6d0ef903 JS |
2058 | ts.tv_sec = leap; |
2059 | ts.tv_nsec = 0; | |
2060 | tk_set_wall_to_mono(tk, | |
7d489d15 | 2061 | timespec64_sub(tk->wall_to_monotonic, ts)); |
6d0ef903 | 2062 | |
cc244dda JS |
2063 | __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); |
2064 | ||
5258d3f2 | 2065 | clock_set = TK_CLOCK_WAS_SET; |
6d0ef903 | 2066 | } |
1f4f9487 | 2067 | } |
5258d3f2 | 2068 | return clock_set; |
1f4f9487 JS |
2069 | } |
2070 | ||
199d280c | 2071 | /* |
a092ff0f JS |
2072 | * logarithmic_accumulation - shifted accumulation of cycles |
2073 | * | |
2074 | * This functions accumulates a shifted interval of cycles into | |
b0294f30 | 2075 | * a shifted interval nanoseconds. Allows for O(log) accumulation |
a092ff0f JS |
2076 | * loop. |
2077 | * | |
2078 | * Returns the unconsumed cycles. | |
2079 | */ | |
a5a1d1c2 TG |
2080 | static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset, |
2081 | u32 shift, unsigned int *clock_set) | |
a092ff0f | 2082 | { |
a5a1d1c2 | 2083 | u64 interval = tk->cycle_interval << shift; |
3d88d56c | 2084 | u64 snsec_per_sec; |
a092ff0f | 2085 | |
571af55a | 2086 | /* If the offset is smaller than a shifted interval, do nothing */ |
23a9537a | 2087 | if (offset < interval) |
a092ff0f JS |
2088 | return offset; |
2089 | ||
2090 | /* Accumulate one shifted interval */ | |
23a9537a | 2091 | offset -= interval; |
876e7881 | 2092 | tk->tkr_mono.cycle_last += interval; |
4a4ad80d | 2093 | tk->tkr_raw.cycle_last += interval; |
a092ff0f | 2094 | |
876e7881 | 2095 | tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift; |
5258d3f2 | 2096 | *clock_set |= accumulate_nsecs_to_secs(tk); |
a092ff0f | 2097 | |
deda2e81 | 2098 | /* Accumulate raw time */ |
3d88d56c JS |
2099 | tk->tkr_raw.xtime_nsec += tk->raw_interval << shift; |
2100 | snsec_per_sec = (u64)NSEC_PER_SEC << tk->tkr_raw.shift; | |
2101 | while (tk->tkr_raw.xtime_nsec >= snsec_per_sec) { | |
2102 | tk->tkr_raw.xtime_nsec -= snsec_per_sec; | |
fc6eead7 | 2103 | tk->raw_sec++; |
a092ff0f JS |
2104 | } |
2105 | ||
2106 | /* Accumulate error between NTP and clock interval */ | |
375f45b5 | 2107 | tk->ntp_error += tk->ntp_tick << shift; |
f726a697 JS |
2108 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << |
2109 | (tk->ntp_error_shift + shift); | |
a092ff0f JS |
2110 | |
2111 | return offset; | |
2112 | } | |
2113 | ||
b061c7a5 ML |
2114 | /* |
2115 | * timekeeping_advance - Updates the timekeeper to the current time and | |
2116 | * current NTP tick length | |
8524070b | 2117 | */ |
b061c7a5 | 2118 | static void timekeeping_advance(enum timekeeping_adv_mode mode) |
8524070b | 2119 | { |
3fdb14fd | 2120 | struct timekeeper *real_tk = &tk_core.timekeeper; |
48cdc135 | 2121 | struct timekeeper *tk = &shadow_timekeeper; |
a5a1d1c2 | 2122 | u64 offset; |
a092ff0f | 2123 | int shift = 0, maxshift; |
5258d3f2 | 2124 | unsigned int clock_set = 0; |
70471f2f JS |
2125 | unsigned long flags; |
2126 | ||
9a7a71b1 | 2127 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
8524070b JS |
2128 | |
2129 | /* Make sure we're fully resumed: */ | |
2130 | if (unlikely(timekeeping_suspended)) | |
70471f2f | 2131 | goto out; |
8524070b | 2132 | |
ceea5e37 | 2133 | offset = clocksource_delta(tk_clock_read(&tk->tkr_mono), |
876e7881 | 2134 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); |
8524070b | 2135 | |
bf2ac312 | 2136 | /* Check if there's really nothing to do */ |
b061c7a5 | 2137 | if (offset < real_tk->cycle_interval && mode == TK_ADV_TICK) |
bf2ac312 JS |
2138 | goto out; |
2139 | ||
3c17ad19 | 2140 | /* Do some additional sanity checking */ |
a529bea8 | 2141 | timekeeping_check_update(tk, offset); |
3c17ad19 | 2142 | |
a092ff0f JS |
2143 | /* |
2144 | * With NO_HZ we may have to accumulate many cycle_intervals | |
2145 | * (think "ticks") worth of time at once. To do this efficiently, | |
2146 | * we calculate the largest doubling multiple of cycle_intervals | |
88b28adf | 2147 | * that is smaller than the offset. We then accumulate that |
a092ff0f JS |
2148 | * chunk in one go, and then try to consume the next smaller |
2149 | * doubled multiple. | |
8524070b | 2150 | */ |
4e250fdd | 2151 | shift = ilog2(offset) - ilog2(tk->cycle_interval); |
a092ff0f | 2152 | shift = max(0, shift); |
88b28adf | 2153 | /* Bound shift to one less than what overflows tick_length */ |
ea7cf49a | 2154 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; |
a092ff0f | 2155 | shift = min(shift, maxshift); |
4e250fdd | 2156 | while (offset >= tk->cycle_interval) { |
5258d3f2 JS |
2157 | offset = logarithmic_accumulation(tk, offset, shift, |
2158 | &clock_set); | |
4e250fdd | 2159 | if (offset < tk->cycle_interval<<shift) |
830ec045 | 2160 | shift--; |
8524070b JS |
2161 | } |
2162 | ||
78b98e3c | 2163 | /* Adjust the multiplier to correct NTP error */ |
4e250fdd | 2164 | timekeeping_adjust(tk, offset); |
8524070b | 2165 | |
6a867a39 JS |
2166 | /* |
2167 | * Finally, make sure that after the rounding | |
1e75fa8b | 2168 | * xtime_nsec isn't larger than NSEC_PER_SEC |
6a867a39 | 2169 | */ |
5258d3f2 | 2170 | clock_set |= accumulate_nsecs_to_secs(tk); |
83f57a11 | 2171 | |
3fdb14fd | 2172 | write_seqcount_begin(&tk_core.seq); |
48cdc135 TG |
2173 | /* |
2174 | * Update the real timekeeper. | |
2175 | * | |
2176 | * We could avoid this memcpy by switching pointers, but that | |
2177 | * requires changes to all other timekeeper usage sites as | |
2178 | * well, i.e. move the timekeeper pointer getter into the | |
2179 | * spinlocked/seqcount protected sections. And we trade this | |
3fdb14fd | 2180 | * memcpy under the tk_core.seq against one before we start |
48cdc135 TG |
2181 | * updating. |
2182 | */ | |
906c5557 | 2183 | timekeeping_update(tk, clock_set); |
48cdc135 | 2184 | memcpy(real_tk, tk, sizeof(*tk)); |
906c5557 | 2185 | /* The memcpy must come last. Do not put anything here! */ |
3fdb14fd | 2186 | write_seqcount_end(&tk_core.seq); |
ca4523cd | 2187 | out: |
9a7a71b1 | 2188 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
47a1b796 | 2189 | if (clock_set) |
cab5e127 JS |
2190 | /* Have to call _delayed version, since in irq context*/ |
2191 | clock_was_set_delayed(); | |
8524070b | 2192 | } |
7c3f1a57 | 2193 | |
b061c7a5 ML |
2194 | /** |
2195 | * update_wall_time - Uses the current clocksource to increment the wall time | |
2196 | * | |
2197 | */ | |
2198 | void update_wall_time(void) | |
2199 | { | |
2200 | timekeeping_advance(TK_ADV_TICK); | |
2201 | } | |
2202 | ||
7c3f1a57 | 2203 | /** |
d08c0cdd JS |
2204 | * getboottime64 - Return the real time of system boot. |
2205 | * @ts: pointer to the timespec64 to be set | |
7c3f1a57 | 2206 | * |
d08c0cdd | 2207 | * Returns the wall-time of boot in a timespec64. |
7c3f1a57 TJ |
2208 | * |
2209 | * This is based on the wall_to_monotonic offset and the total suspend | |
2210 | * time. Calls to settimeofday will affect the value returned (which | |
2211 | * basically means that however wrong your real time clock is at boot time, | |
2212 | * you get the right time here). | |
2213 | */ | |
d08c0cdd | 2214 | void getboottime64(struct timespec64 *ts) |
7c3f1a57 | 2215 | { |
3fdb14fd | 2216 | struct timekeeper *tk = &tk_core.timekeeper; |
a3ed0e43 | 2217 | ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot); |
02cba159 | 2218 | |
d08c0cdd | 2219 | *ts = ktime_to_timespec64(t); |
7c3f1a57 | 2220 | } |
d08c0cdd | 2221 | EXPORT_SYMBOL_GPL(getboottime64); |
7c3f1a57 | 2222 | |
fb7fcc96 | 2223 | void ktime_get_coarse_real_ts64(struct timespec64 *ts) |
2c6b47de | 2224 | { |
3fdb14fd | 2225 | struct timekeeper *tk = &tk_core.timekeeper; |
e1e41b6c | 2226 | unsigned int seq; |
2c6b47de JS |
2227 | |
2228 | do { | |
3fdb14fd | 2229 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2230 | |
fb7fcc96 | 2231 | *ts = tk_xtime(tk); |
3fdb14fd | 2232 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
2c6b47de | 2233 | } |
fb7fcc96 | 2234 | EXPORT_SYMBOL(ktime_get_coarse_real_ts64); |
da15cfda | 2235 | |
fb7fcc96 | 2236 | void ktime_get_coarse_ts64(struct timespec64 *ts) |
da15cfda | 2237 | { |
3fdb14fd | 2238 | struct timekeeper *tk = &tk_core.timekeeper; |
7d489d15 | 2239 | struct timespec64 now, mono; |
e1e41b6c | 2240 | unsigned int seq; |
da15cfda JS |
2241 | |
2242 | do { | |
3fdb14fd | 2243 | seq = read_seqcount_begin(&tk_core.seq); |
83f57a11 | 2244 | |
4e250fdd JS |
2245 | now = tk_xtime(tk); |
2246 | mono = tk->wall_to_monotonic; | |
3fdb14fd | 2247 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
da15cfda | 2248 | |
fb7fcc96 | 2249 | set_normalized_timespec64(ts, now.tv_sec + mono.tv_sec, |
da15cfda | 2250 | now.tv_nsec + mono.tv_nsec); |
da15cfda | 2251 | } |
fb7fcc96 | 2252 | EXPORT_SYMBOL(ktime_get_coarse_ts64); |
871cf1e5 TH |
2253 | |
2254 | /* | |
d6ad4187 | 2255 | * Must hold jiffies_lock |
871cf1e5 TH |
2256 | */ |
2257 | void do_timer(unsigned long ticks) | |
2258 | { | |
2259 | jiffies_64 += ticks; | |
46132e3a | 2260 | calc_global_load(); |
871cf1e5 | 2261 | } |
48cf76f7 | 2262 | |
f6c06abf | 2263 | /** |
76f41088 | 2264 | * ktime_get_update_offsets_now - hrtimer helper |
868a3e91 | 2265 | * @cwsseq: pointer to check and store the clock was set sequence number |
f6c06abf | 2266 | * @offs_real: pointer to storage for monotonic -> realtime offset |
a3ed0e43 | 2267 | * @offs_boot: pointer to storage for monotonic -> boottime offset |
b7bc50e4 | 2268 | * @offs_tai: pointer to storage for monotonic -> clock tai offset |
f6c06abf | 2269 | * |
868a3e91 TG |
2270 | * Returns current monotonic time and updates the offsets if the |
2271 | * sequence number in @cwsseq and timekeeper.clock_was_set_seq are | |
2272 | * different. | |
2273 | * | |
b7bc50e4 | 2274 | * Called from hrtimer_interrupt() or retrigger_next_event() |
f6c06abf | 2275 | */ |
868a3e91 | 2276 | ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real, |
a3ed0e43 | 2277 | ktime_t *offs_boot, ktime_t *offs_tai) |
f6c06abf | 2278 | { |
3fdb14fd | 2279 | struct timekeeper *tk = &tk_core.timekeeper; |
f6c06abf | 2280 | unsigned int seq; |
a37c0aad TG |
2281 | ktime_t base; |
2282 | u64 nsecs; | |
f6c06abf TG |
2283 | |
2284 | do { | |
3fdb14fd | 2285 | seq = read_seqcount_begin(&tk_core.seq); |
f6c06abf | 2286 | |
876e7881 PZ |
2287 | base = tk->tkr_mono.base; |
2288 | nsecs = timekeeping_get_ns(&tk->tkr_mono); | |
833f32d7 JS |
2289 | base = ktime_add_ns(base, nsecs); |
2290 | ||
868a3e91 TG |
2291 | if (*cwsseq != tk->clock_was_set_seq) { |
2292 | *cwsseq = tk->clock_was_set_seq; | |
2293 | *offs_real = tk->offs_real; | |
a3ed0e43 | 2294 | *offs_boot = tk->offs_boot; |
868a3e91 TG |
2295 | *offs_tai = tk->offs_tai; |
2296 | } | |
833f32d7 JS |
2297 | |
2298 | /* Handle leapsecond insertion adjustments */ | |
2456e855 | 2299 | if (unlikely(base >= tk->next_leap_ktime)) |
833f32d7 JS |
2300 | *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0)); |
2301 | ||
3fdb14fd | 2302 | } while (read_seqcount_retry(&tk_core.seq, seq)); |
f6c06abf | 2303 | |
833f32d7 | 2304 | return base; |
f6c06abf | 2305 | } |
f6c06abf | 2306 | |
199d280c | 2307 | /* |
1572fa03 | 2308 | * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex |
e0956dcc | 2309 | */ |
ead25417 | 2310 | static int timekeeping_validate_timex(const struct __kernel_timex *txc) |
e0956dcc AB |
2311 | { |
2312 | if (txc->modes & ADJ_ADJTIME) { | |
2313 | /* singleshot must not be used with any other mode bits */ | |
2314 | if (!(txc->modes & ADJ_OFFSET_SINGLESHOT)) | |
2315 | return -EINVAL; | |
2316 | if (!(txc->modes & ADJ_OFFSET_READONLY) && | |
2317 | !capable(CAP_SYS_TIME)) | |
2318 | return -EPERM; | |
2319 | } else { | |
2320 | /* In order to modify anything, you gotta be super-user! */ | |
2321 | if (txc->modes && !capable(CAP_SYS_TIME)) | |
2322 | return -EPERM; | |
2323 | /* | |
2324 | * if the quartz is off by more than 10% then | |
2325 | * something is VERY wrong! | |
2326 | */ | |
2327 | if (txc->modes & ADJ_TICK && | |
2328 | (txc->tick < 900000/USER_HZ || | |
2329 | txc->tick > 1100000/USER_HZ)) | |
2330 | return -EINVAL; | |
2331 | } | |
2332 | ||
2333 | if (txc->modes & ADJ_SETOFFSET) { | |
2334 | /* In order to inject time, you gotta be super-user! */ | |
2335 | if (!capable(CAP_SYS_TIME)) | |
2336 | return -EPERM; | |
2337 | ||
1572fa03 AB |
2338 | /* |
2339 | * Validate if a timespec/timeval used to inject a time | |
2340 | * offset is valid. Offsets can be postive or negative, so | |
2341 | * we don't check tv_sec. The value of the timeval/timespec | |
2342 | * is the sum of its fields,but *NOTE*: | |
2343 | * The field tv_usec/tv_nsec must always be non-negative and | |
2344 | * we can't have more nanoseconds/microseconds than a second. | |
2345 | */ | |
2346 | if (txc->time.tv_usec < 0) | |
2347 | return -EINVAL; | |
e0956dcc | 2348 | |
1572fa03 AB |
2349 | if (txc->modes & ADJ_NANO) { |
2350 | if (txc->time.tv_usec >= NSEC_PER_SEC) | |
e0956dcc | 2351 | return -EINVAL; |
e0956dcc | 2352 | } else { |
1572fa03 | 2353 | if (txc->time.tv_usec >= USEC_PER_SEC) |
e0956dcc AB |
2354 | return -EINVAL; |
2355 | } | |
2356 | } | |
2357 | ||
2358 | /* | |
2359 | * Check for potential multiplication overflows that can | |
2360 | * only happen on 64-bit systems: | |
2361 | */ | |
2362 | if ((txc->modes & ADJ_FREQUENCY) && (BITS_PER_LONG == 64)) { | |
2363 | if (LLONG_MIN / PPM_SCALE > txc->freq) | |
2364 | return -EINVAL; | |
2365 | if (LLONG_MAX / PPM_SCALE < txc->freq) | |
2366 | return -EINVAL; | |
2367 | } | |
2368 | ||
2369 | return 0; | |
2370 | } | |
2371 | ||
2372 | ||
aa6f9c59 JS |
2373 | /** |
2374 | * do_adjtimex() - Accessor function to NTP __do_adjtimex function | |
2375 | */ | |
ead25417 | 2376 | int do_adjtimex(struct __kernel_timex *txc) |
aa6f9c59 | 2377 | { |
3fdb14fd | 2378 | struct timekeeper *tk = &tk_core.timekeeper; |
7e8eda73 | 2379 | struct audit_ntp_data ad; |
06c017fd | 2380 | unsigned long flags; |
7d489d15 | 2381 | struct timespec64 ts; |
4e8f8b34 | 2382 | s32 orig_tai, tai; |
e4085693 JS |
2383 | int ret; |
2384 | ||
2385 | /* Validate the data before disabling interrupts */ | |
1572fa03 | 2386 | ret = timekeeping_validate_timex(txc); |
e4085693 JS |
2387 | if (ret) |
2388 | return ret; | |
2389 | ||
cef90377 | 2390 | if (txc->modes & ADJ_SETOFFSET) { |
1572fa03 | 2391 | struct timespec64 delta; |
cef90377 JS |
2392 | delta.tv_sec = txc->time.tv_sec; |
2393 | delta.tv_nsec = txc->time.tv_usec; | |
2394 | if (!(txc->modes & ADJ_NANO)) | |
2395 | delta.tv_nsec *= 1000; | |
2396 | ret = timekeeping_inject_offset(&delta); | |
2397 | if (ret) | |
2398 | return ret; | |
2d87a067 OM |
2399 | |
2400 | audit_tk_injoffset(delta); | |
cef90377 JS |
2401 | } |
2402 | ||
7e8eda73 OM |
2403 | audit_ntp_init(&ad); |
2404 | ||
d30faff9 | 2405 | ktime_get_real_ts64(&ts); |
87ace39b | 2406 | |
06c017fd | 2407 | raw_spin_lock_irqsave(&timekeeper_lock, flags); |
3fdb14fd | 2408 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2409 | |
4e8f8b34 | 2410 | orig_tai = tai = tk->tai_offset; |
7e8eda73 | 2411 | ret = __do_adjtimex(txc, &ts, &tai, &ad); |
aa6f9c59 | 2412 | |
4e8f8b34 JS |
2413 | if (tai != orig_tai) { |
2414 | __timekeeping_set_tai_offset(tk, tai); | |
f55c0760 | 2415 | timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); |
4e8f8b34 | 2416 | } |
833f32d7 JS |
2417 | tk_update_leap_state(tk); |
2418 | ||
3fdb14fd | 2419 | write_seqcount_end(&tk_core.seq); |
06c017fd JS |
2420 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
2421 | ||
7e8eda73 OM |
2422 | audit_ntp_log(&ad); |
2423 | ||
b061c7a5 ML |
2424 | /* Update the multiplier immediately if frequency was set directly */ |
2425 | if (txc->modes & (ADJ_FREQUENCY | ADJ_TICK)) | |
2426 | timekeeping_advance(TK_ADV_FREQ); | |
2427 | ||
6fdda9a9 JS |
2428 | if (tai != orig_tai) |
2429 | clock_was_set(); | |
2430 | ||
7bd36014 JS |
2431 | ntp_notify_cmos_timer(); |
2432 | ||
87ace39b JS |
2433 | return ret; |
2434 | } | |
aa6f9c59 JS |
2435 | |
2436 | #ifdef CONFIG_NTP_PPS | |
2437 | /** | |
2438 | * hardpps() - Accessor function to NTP __hardpps function | |
2439 | */ | |
7ec88e4b | 2440 | void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts) |
aa6f9c59 | 2441 | { |
06c017fd JS |
2442 | unsigned long flags; |
2443 | ||
2444 | raw_spin_lock_irqsave(&timekeeper_lock, flags); | |
3fdb14fd | 2445 | write_seqcount_begin(&tk_core.seq); |
06c017fd | 2446 | |
aa6f9c59 | 2447 | __hardpps(phase_ts, raw_ts); |
06c017fd | 2448 | |
3fdb14fd | 2449 | write_seqcount_end(&tk_core.seq); |
06c017fd | 2450 | raw_spin_unlock_irqrestore(&timekeeper_lock, flags); |
aa6f9c59 JS |
2451 | } |
2452 | EXPORT_SYMBOL(hardpps); | |
a2d81803 | 2453 | #endif /* CONFIG_NTP_PPS */ |