1 /*
2 * Common time routines among all ppc machines.
3 *
4 * Written by Cort Dougan (cort@cs.nmt.edu) to merge
5 * Paul Mackerras' version and mine for PReP and Pmac.
6 * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net).
7 * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com)
8 *
9 * First round of bugfixes by Gabriel Paubert (paubert@iram.es)
10 * to make clock more stable (2.4.0-test5). The only thing
11 * that this code assumes is that the timebases have been synchronized
12 * by firmware on SMP and are never stopped (never do sleep
13 * on SMP then, nap and doze are OK).
14 *
15 * Speeded up do_gettimeofday by getting rid of references to
16 * xtime (which required locks for consistency). (mikejc@us.ibm.com)
17 *
18 * TODO (not necessarily in this file):
19 * - improve precision and reproducibility of timebase frequency
20 * measurement at boot time.
21 * - for astronomical applications: add a new function to get
22 * non ambiguous timestamps even around leap seconds. This needs
23 * a new timestamp format and a good name.
24 *
25 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
26 * "A Kernel Model for Precision Timekeeping" by Dave Mills
27 *
28 * This program is free software; you can redistribute it and/or
29 * modify it under the terms of the GNU General Public License
30 * as published by the Free Software Foundation; either version
31 * 2 of the License, or (at your option) any later version.
32 */
33
34 #include <linux/errno.h>
35 #include <linux/export.h>
36 #include <linux/sched.h>
37 #include <linux/kernel.h>
38 #include <linux/param.h>
39 #include <linux/string.h>
40 #include <linux/mm.h>
41 #include <linux/interrupt.h>
42 #include <linux/timex.h>
43 #include <linux/kernel_stat.h>
44 #include <linux/time.h>
45 #include <linux/clockchips.h>
46 #include <linux/init.h>
47 #include <linux/profile.h>
48 #include <linux/cpu.h>
49 #include <linux/security.h>
50 #include <linux/percpu.h>
51 #include <linux/rtc.h>
52 #include <linux/jiffies.h>
53 #include <linux/posix-timers.h>
54 #include <linux/irq.h>
55 #include <linux/delay.h>
56 #include <linux/irq_work.h>
57 #include <linux/clk-provider.h>
58 #include <linux/suspend.h>
59 #include <linux/rtc.h>
60 #include <asm/trace.h>
61
62 #include <asm/io.h>
63 #include <asm/processor.h>
64 #include <asm/nvram.h>
65 #include <asm/cache.h>
66 #include <asm/machdep.h>
67 #include <asm/uaccess.h>
68 #include <asm/time.h>
69 #include <asm/prom.h>
70 #include <asm/irq.h>
71 #include <asm/div64.h>
72 #include <asm/smp.h>
73 #include <asm/vdso_datapage.h>
74 #include <asm/firmware.h>
75 #include <asm/cputime.h>
76 #include <asm/asm-prototypes.h>
77
78 /* powerpc clocksource/clockevent code */
79
80 #include <linux/clockchips.h>
81 #include <linux/timekeeper_internal.h>
82
83 static cycle_t rtc_read(struct clocksource *);
84 static struct clocksource clocksource_rtc = {
85 .name = "rtc",
86 .rating = 400,
87 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
88 .mask = CLOCKSOURCE_MASK(64),
89 .read = rtc_read,
90 };
91
92 static cycle_t timebase_read(struct clocksource *);
93 static struct clocksource clocksource_timebase = {
94 .name = "timebase",
95 .rating = 400,
96 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
97 .mask = CLOCKSOURCE_MASK(64),
98 .read = timebase_read,
99 };
100
101 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
102 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
103
104 static int decrementer_set_next_event(unsigned long evt,
105 struct clock_event_device *dev);
106 static int decrementer_shutdown(struct clock_event_device *evt);
107
108 struct clock_event_device decrementer_clockevent = {
109 .name = "decrementer",
110 .rating = 200,
111 .irq = 0,
112 .set_next_event = decrementer_set_next_event,
113 .set_state_shutdown = decrementer_shutdown,
114 .tick_resume = decrementer_shutdown,
115 .features = CLOCK_EVT_FEAT_ONESHOT |
116 CLOCK_EVT_FEAT_C3STOP,
117 };
118 EXPORT_SYMBOL(decrementer_clockevent);
119
120 DEFINE_PER_CPU(u64, decrementers_next_tb);
121 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
122
123 #define XSEC_PER_SEC (1024*1024)
124
125 #ifdef CONFIG_PPC64
126 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
127 #else
128 /* compute ((xsec << 12) * max) >> 32 */
129 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
130 #endif
131
132 unsigned long tb_ticks_per_jiffy;
133 unsigned long tb_ticks_per_usec = 100; /* sane default */
134 EXPORT_SYMBOL(tb_ticks_per_usec);
135 unsigned long tb_ticks_per_sec;
136 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
137
138 DEFINE_SPINLOCK(rtc_lock);
139 EXPORT_SYMBOL_GPL(rtc_lock);
140
141 static u64 tb_to_ns_scale __read_mostly;
142 static unsigned tb_to_ns_shift __read_mostly;
143 static u64 boot_tb __read_mostly;
144
145 extern struct timezone sys_tz;
146 static long timezone_offset;
147
148 unsigned long ppc_proc_freq;
149 EXPORT_SYMBOL_GPL(ppc_proc_freq);
150 unsigned long ppc_tb_freq;
151 EXPORT_SYMBOL_GPL(ppc_tb_freq);
152
153 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
154 /*
155 * Factors for converting from cputime_t (timebase ticks) to
156 * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
157 * These are all stored as 0.64 fixed-point binary fractions.
158 */
159 u64 __cputime_jiffies_factor;
160 EXPORT_SYMBOL(__cputime_jiffies_factor);
161 u64 __cputime_usec_factor;
162 EXPORT_SYMBOL(__cputime_usec_factor);
163 u64 __cputime_sec_factor;
164 EXPORT_SYMBOL(__cputime_sec_factor);
165 u64 __cputime_clockt_factor;
166 EXPORT_SYMBOL(__cputime_clockt_factor);
167 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
168 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
169
170 cputime_t cputime_one_jiffy;
171
172 #ifdef CONFIG_PPC_SPLPAR
173 void (*dtl_consumer)(struct dtl_entry *, u64);
174 #endif
175
176 #ifdef CONFIG_PPC64
177 #define get_accounting(tsk) (&get_paca()->accounting)
178 #else
179 #define get_accounting(tsk) (&task_thread_info(tsk)->accounting)
180 #endif
181
calc_cputime_factors(void)182 static void calc_cputime_factors(void)
183 {
184 struct div_result res;
185
186 div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
187 __cputime_jiffies_factor = res.result_low;
188 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
189 __cputime_usec_factor = res.result_low;
190 div128_by_32(1, 0, tb_ticks_per_sec, &res);
191 __cputime_sec_factor = res.result_low;
192 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
193 __cputime_clockt_factor = res.result_low;
194 }
195
196 /*
197 * Read the SPURR on systems that have it, otherwise the PURR,
198 * or if that doesn't exist return the timebase value passed in.
199 */
read_spurr(unsigned long tb)200 static unsigned long read_spurr(unsigned long tb)
201 {
202 if (cpu_has_feature(CPU_FTR_SPURR))
203 return mfspr(SPRN_SPURR);
204 if (cpu_has_feature(CPU_FTR_PURR))
205 return mfspr(SPRN_PURR);
206 return tb;
207 }
208
209 #ifdef CONFIG_PPC_SPLPAR
210
211 /*
212 * Scan the dispatch trace log and count up the stolen time.
213 * Should be called with interrupts disabled.
214 */
scan_dispatch_log(u64 stop_tb)215 static u64 scan_dispatch_log(u64 stop_tb)
216 {
217 u64 i = local_paca->dtl_ridx;
218 struct dtl_entry *dtl = local_paca->dtl_curr;
219 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
220 struct lppaca *vpa = local_paca->lppaca_ptr;
221 u64 tb_delta;
222 u64 stolen = 0;
223 u64 dtb;
224
225 if (!dtl)
226 return 0;
227
228 if (i == be64_to_cpu(vpa->dtl_idx))
229 return 0;
230 while (i < be64_to_cpu(vpa->dtl_idx)) {
231 dtb = be64_to_cpu(dtl->timebase);
232 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
233 be32_to_cpu(dtl->ready_to_enqueue_time);
234 barrier();
235 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
236 /* buffer has overflowed */
237 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
238 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
239 continue;
240 }
241 if (dtb > stop_tb)
242 break;
243 if (dtl_consumer)
244 dtl_consumer(dtl, i);
245 stolen += tb_delta;
246 ++i;
247 ++dtl;
248 if (dtl == dtl_end)
249 dtl = local_paca->dispatch_log;
250 }
251 local_paca->dtl_ridx = i;
252 local_paca->dtl_curr = dtl;
253 return stolen;
254 }
255
256 /*
257 * Accumulate stolen time by scanning the dispatch trace log.
258 * Called on entry from user mode.
259 */
accumulate_stolen_time(void)260 void accumulate_stolen_time(void)
261 {
262 u64 sst, ust;
263 u8 save_soft_enabled = local_paca->soft_enabled;
264 struct cpu_accounting_data *acct = &local_paca->accounting;
265
266 /* We are called early in the exception entry, before
267 * soft/hard_enabled are sync'ed to the expected state
268 * for the exception. We are hard disabled but the PACA
269 * needs to reflect that so various debug stuff doesn't
270 * complain
271 */
272 local_paca->soft_enabled = 0;
273
274 sst = scan_dispatch_log(acct->starttime_user);
275 ust = scan_dispatch_log(acct->starttime);
276 acct->system_time -= sst;
277 acct->user_time -= ust;
278 local_paca->stolen_time += ust + sst;
279
280 local_paca->soft_enabled = save_soft_enabled;
281 }
282
calculate_stolen_time(u64 stop_tb)283 static inline u64 calculate_stolen_time(u64 stop_tb)
284 {
285 u64 stolen = 0;
286
287 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
288 stolen = scan_dispatch_log(stop_tb);
289 get_paca()->accounting.system_time -= stolen;
290 }
291
292 stolen += get_paca()->stolen_time;
293 get_paca()->stolen_time = 0;
294 return stolen;
295 }
296
297 #else /* CONFIG_PPC_SPLPAR */
calculate_stolen_time(u64 stop_tb)298 static inline u64 calculate_stolen_time(u64 stop_tb)
299 {
300 return 0;
301 }
302
303 #endif /* CONFIG_PPC_SPLPAR */
304
305 /*
306 * Account time for a transition between system, hard irq
307 * or soft irq state.
308 */
vtime_delta(struct task_struct * tsk,unsigned long * sys_scaled,unsigned long * stolen)309 static unsigned long vtime_delta(struct task_struct *tsk,
310 unsigned long *sys_scaled,
311 unsigned long *stolen)
312 {
313 unsigned long now, nowscaled, deltascaled;
314 unsigned long udelta, delta, user_scaled;
315 struct cpu_accounting_data *acct = get_accounting(tsk);
316
317 WARN_ON_ONCE(!irqs_disabled());
318
319 now = mftb();
320 nowscaled = read_spurr(now);
321 acct->system_time += now - acct->starttime;
322 acct->starttime = now;
323 deltascaled = nowscaled - acct->startspurr;
324 acct->startspurr = nowscaled;
325
326 *stolen = calculate_stolen_time(now);
327
328 delta = acct->system_time;
329 acct->system_time = 0;
330 udelta = acct->user_time - acct->utime_sspurr;
331 acct->utime_sspurr = acct->user_time;
332
333 /*
334 * Because we don't read the SPURR on every kernel entry/exit,
335 * deltascaled includes both user and system SPURR ticks.
336 * Apportion these ticks to system SPURR ticks and user
337 * SPURR ticks in the same ratio as the system time (delta)
338 * and user time (udelta) values obtained from the timebase
339 * over the same interval. The system ticks get accounted here;
340 * the user ticks get saved up in paca->user_time_scaled to be
341 * used by account_process_tick.
342 */
343 *sys_scaled = delta;
344 user_scaled = udelta;
345 if (deltascaled != delta + udelta) {
346 if (udelta) {
347 *sys_scaled = deltascaled * delta / (delta + udelta);
348 user_scaled = deltascaled - *sys_scaled;
349 } else {
350 *sys_scaled = deltascaled;
351 }
352 }
353 acct->user_time_scaled += user_scaled;
354
355 return delta;
356 }
357
vtime_account_system(struct task_struct * tsk)358 void vtime_account_system(struct task_struct *tsk)
359 {
360 unsigned long delta, sys_scaled, stolen;
361
362 delta = vtime_delta(tsk, &sys_scaled, &stolen);
363 account_system_time(tsk, 0, delta, sys_scaled);
364 if (stolen)
365 account_steal_time(stolen);
366 }
367 EXPORT_SYMBOL_GPL(vtime_account_system);
368
vtime_account_idle(struct task_struct * tsk)369 void vtime_account_idle(struct task_struct *tsk)
370 {
371 unsigned long delta, sys_scaled, stolen;
372
373 delta = vtime_delta(tsk, &sys_scaled, &stolen);
374 account_idle_time(delta + stolen);
375 }
376
377 /*
378 * Transfer the user time accumulated in the paca
379 * by the exception entry and exit code to the generic
380 * process user time records.
381 * Must be called with interrupts disabled.
382 * Assumes that vtime_account_system/idle() has been called
383 * recently (i.e. since the last entry from usermode) so that
384 * get_paca()->user_time_scaled is up to date.
385 */
vtime_account_user(struct task_struct * tsk)386 void vtime_account_user(struct task_struct *tsk)
387 {
388 cputime_t utime, utimescaled;
389 struct cpu_accounting_data *acct = get_accounting(tsk);
390
391 utime = acct->user_time;
392 utimescaled = acct->user_time_scaled;
393 acct->user_time = 0;
394 acct->user_time_scaled = 0;
395 acct->utime_sspurr = 0;
396 account_user_time(tsk, utime, utimescaled);
397 }
398
399 #ifdef CONFIG_PPC32
400 /*
401 * Called from the context switch with interrupts disabled, to charge all
402 * accumulated times to the current process, and to prepare accounting on
403 * the next process.
404 */
arch_vtime_task_switch(struct task_struct * prev)405 void arch_vtime_task_switch(struct task_struct *prev)
406 {
407 struct cpu_accounting_data *acct = get_accounting(current);
408
409 acct->starttime = get_accounting(prev)->starttime;
410 acct->startspurr = get_accounting(prev)->startspurr;
411 acct->system_time = 0;
412 acct->user_time = 0;
413 }
414 #endif /* CONFIG_PPC32 */
415
416 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
417 #define calc_cputime_factors()
418 #endif
419
__delay(unsigned long loops)420 void __delay(unsigned long loops)
421 {
422 unsigned long start;
423 int diff;
424
425 if (__USE_RTC()) {
426 start = get_rtcl();
427 do {
428 /* the RTCL register wraps at 1000000000 */
429 diff = get_rtcl() - start;
430 if (diff < 0)
431 diff += 1000000000;
432 } while (diff < loops);
433 } else {
434 start = get_tbl();
435 while (get_tbl() - start < loops)
436 HMT_low();
437 HMT_medium();
438 }
439 }
440 EXPORT_SYMBOL(__delay);
441
udelay(unsigned long usecs)442 void udelay(unsigned long usecs)
443 {
444 __delay(tb_ticks_per_usec * usecs);
445 }
446 EXPORT_SYMBOL(udelay);
447
448 #ifdef CONFIG_SMP
profile_pc(struct pt_regs * regs)449 unsigned long profile_pc(struct pt_regs *regs)
450 {
451 unsigned long pc = instruction_pointer(regs);
452
453 if (in_lock_functions(pc))
454 return regs->link;
455
456 return pc;
457 }
458 EXPORT_SYMBOL(profile_pc);
459 #endif
460
461 #ifdef CONFIG_IRQ_WORK
462
463 /*
464 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
465 */
466 #ifdef CONFIG_PPC64
test_irq_work_pending(void)467 static inline unsigned long test_irq_work_pending(void)
468 {
469 unsigned long x;
470
471 asm volatile("lbz %0,%1(13)"
472 : "=r" (x)
473 : "i" (offsetof(struct paca_struct, irq_work_pending)));
474 return x;
475 }
476
set_irq_work_pending_flag(void)477 static inline void set_irq_work_pending_flag(void)
478 {
479 asm volatile("stb %0,%1(13)" : :
480 "r" (1),
481 "i" (offsetof(struct paca_struct, irq_work_pending)));
482 }
483
clear_irq_work_pending(void)484 static inline void clear_irq_work_pending(void)
485 {
486 asm volatile("stb %0,%1(13)" : :
487 "r" (0),
488 "i" (offsetof(struct paca_struct, irq_work_pending)));
489 }
490
491 #else /* 32-bit */
492
493 DEFINE_PER_CPU(u8, irq_work_pending);
494
495 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
496 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
497 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
498
499 #endif /* 32 vs 64 bit */
500
arch_irq_work_raise(void)501 void arch_irq_work_raise(void)
502 {
503 preempt_disable();
504 set_irq_work_pending_flag();
505 set_dec(1);
506 preempt_enable();
507 }
508
509 #else /* CONFIG_IRQ_WORK */
510
511 #define test_irq_work_pending() 0
512 #define clear_irq_work_pending()
513
514 #endif /* CONFIG_IRQ_WORK */
515
__timer_interrupt(void)516 static void __timer_interrupt(void)
517 {
518 struct pt_regs *regs = get_irq_regs();
519 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
520 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
521 u64 now;
522
523 trace_timer_interrupt_entry(regs);
524
525 if (test_irq_work_pending()) {
526 clear_irq_work_pending();
527 irq_work_run();
528 }
529
530 now = get_tb_or_rtc();
531 if (now >= *next_tb) {
532 *next_tb = ~(u64)0;
533 if (evt->event_handler)
534 evt->event_handler(evt);
535 __this_cpu_inc(irq_stat.timer_irqs_event);
536 } else {
537 now = *next_tb - now;
538 if (now <= decrementer_max)
539 set_dec(now);
540 /* We may have raced with new irq work */
541 if (test_irq_work_pending())
542 set_dec(1);
543 __this_cpu_inc(irq_stat.timer_irqs_others);
544 }
545
546 #ifdef CONFIG_PPC64
547 /* collect purr register values often, for accurate calculations */
548 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
549 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
550 cu->current_tb = mfspr(SPRN_PURR);
551 }
552 #endif
553
554 trace_timer_interrupt_exit(regs);
555 }
556
557 /*
558 * timer_interrupt - gets called when the decrementer overflows,
559 * with interrupts disabled.
560 */
timer_interrupt(struct pt_regs * regs)561 void timer_interrupt(struct pt_regs * regs)
562 {
563 struct pt_regs *old_regs;
564 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
565
566 /* Ensure a positive value is written to the decrementer, or else
567 * some CPUs will continue to take decrementer exceptions.
568 */
569 set_dec(decrementer_max);
570
571 /* Some implementations of hotplug will get timer interrupts while
572 * offline, just ignore these and we also need to set
573 * decrementers_next_tb as MAX to make sure __check_irq_replay
574 * don't replay timer interrupt when return, otherwise we'll trap
575 * here infinitely :(
576 */
577 if (!cpu_online(smp_processor_id())) {
578 *next_tb = ~(u64)0;
579 return;
580 }
581
582 /* Conditionally hard-enable interrupts now that the DEC has been
583 * bumped to its maximum value
584 */
585 may_hard_irq_enable();
586
587
588 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
589 if (atomic_read(&ppc_n_lost_interrupts) != 0)
590 do_IRQ(regs);
591 #endif
592
593 old_regs = set_irq_regs(regs);
594 irq_enter();
595
596 __timer_interrupt();
597 irq_exit();
598 set_irq_regs(old_regs);
599 }
600 EXPORT_SYMBOL(timer_interrupt);
601
602 /*
603 * Hypervisor decrementer interrupts shouldn't occur but are sometimes
604 * left pending on exit from a KVM guest. We don't need to do anything
605 * to clear them, as they are edge-triggered.
606 */
hdec_interrupt(struct pt_regs * regs)607 void hdec_interrupt(struct pt_regs *regs)
608 {
609 }
610
611 #ifdef CONFIG_SUSPEND
generic_suspend_disable_irqs(void)612 static void generic_suspend_disable_irqs(void)
613 {
614 /* Disable the decrementer, so that it doesn't interfere
615 * with suspending.
616 */
617
618 set_dec(decrementer_max);
619 local_irq_disable();
620 set_dec(decrementer_max);
621 }
622
generic_suspend_enable_irqs(void)623 static void generic_suspend_enable_irqs(void)
624 {
625 local_irq_enable();
626 }
627
628 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_disable_irqs(void)629 void arch_suspend_disable_irqs(void)
630 {
631 if (ppc_md.suspend_disable_irqs)
632 ppc_md.suspend_disable_irqs();
633 generic_suspend_disable_irqs();
634 }
635
636 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_enable_irqs(void)637 void arch_suspend_enable_irqs(void)
638 {
639 generic_suspend_enable_irqs();
640 if (ppc_md.suspend_enable_irqs)
641 ppc_md.suspend_enable_irqs();
642 }
643 #endif
644
tb_to_ns(unsigned long long ticks)645 unsigned long long tb_to_ns(unsigned long long ticks)
646 {
647 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
648 }
649 EXPORT_SYMBOL_GPL(tb_to_ns);
650
651 /*
652 * Scheduler clock - returns current time in nanosec units.
653 *
654 * Note: mulhdu(a, b) (multiply high double unsigned) returns
655 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
656 * are 64-bit unsigned numbers.
657 */
sched_clock(void)658 unsigned long long sched_clock(void)
659 {
660 if (__USE_RTC())
661 return get_rtc();
662 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
663 }
664
665
666 #ifdef CONFIG_PPC_PSERIES
667
668 /*
669 * Running clock - attempts to give a view of time passing for a virtualised
670 * kernels.
671 * Uses the VTB register if available otherwise a next best guess.
672 */
running_clock(void)673 unsigned long long running_clock(void)
674 {
675 /*
676 * Don't read the VTB as a host since KVM does not switch in host
677 * timebase into the VTB when it takes a guest off the CPU, reading the
678 * VTB would result in reading 'last switched out' guest VTB.
679 *
680 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
681 * would be unsafe to rely only on the #ifdef above.
682 */
683 if (firmware_has_feature(FW_FEATURE_LPAR) &&
684 cpu_has_feature(CPU_FTR_ARCH_207S))
685 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
686
687 /*
688 * This is a next best approximation without a VTB.
689 * On a host which is running bare metal there should never be any stolen
690 * time and on a host which doesn't do any virtualisation TB *should* equal
691 * VTB so it makes no difference anyway.
692 */
693 return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
694 }
695 #endif
696
get_freq(char * name,int cells,unsigned long * val)697 static int __init get_freq(char *name, int cells, unsigned long *val)
698 {
699 struct device_node *cpu;
700 const __be32 *fp;
701 int found = 0;
702
703 /* The cpu node should have timebase and clock frequency properties */
704 cpu = of_find_node_by_type(NULL, "cpu");
705
706 if (cpu) {
707 fp = of_get_property(cpu, name, NULL);
708 if (fp) {
709 found = 1;
710 *val = of_read_ulong(fp, cells);
711 }
712
713 of_node_put(cpu);
714 }
715
716 return found;
717 }
718
start_cpu_decrementer(void)719 static void start_cpu_decrementer(void)
720 {
721 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
722 unsigned int tcr;
723
724 /* Clear any pending timer interrupts */
725 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
726
727 tcr = mfspr(SPRN_TCR);
728 /*
729 * The watchdog may have already been enabled by u-boot. So leave
730 * TRC[WP] (Watchdog Period) alone.
731 */
732 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
733 tcr |= TCR_DIE; /* Enable decrementer */
734 mtspr(SPRN_TCR, tcr);
735 #endif
736 }
737
generic_calibrate_decr(void)738 void __init generic_calibrate_decr(void)
739 {
740 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
741
742 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
743 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
744
745 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
746 "(not found)\n");
747 }
748
749 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
750
751 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
752 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
753
754 printk(KERN_ERR "WARNING: Estimating processor frequency "
755 "(not found)\n");
756 }
757 }
758
update_persistent_clock(struct timespec now)759 int update_persistent_clock(struct timespec now)
760 {
761 struct rtc_time tm;
762
763 if (!ppc_md.set_rtc_time)
764 return -ENODEV;
765
766 to_tm(now.tv_sec + 1 + timezone_offset, &tm);
767 tm.tm_year -= 1900;
768 tm.tm_mon -= 1;
769
770 return ppc_md.set_rtc_time(&tm);
771 }
772
__read_persistent_clock(struct timespec * ts)773 static void __read_persistent_clock(struct timespec *ts)
774 {
775 struct rtc_time tm;
776 static int first = 1;
777
778 ts->tv_nsec = 0;
779 /* XXX this is a litle fragile but will work okay in the short term */
780 if (first) {
781 first = 0;
782 if (ppc_md.time_init)
783 timezone_offset = ppc_md.time_init();
784
785 /* get_boot_time() isn't guaranteed to be safe to call late */
786 if (ppc_md.get_boot_time) {
787 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
788 return;
789 }
790 }
791 if (!ppc_md.get_rtc_time) {
792 ts->tv_sec = 0;
793 return;
794 }
795 ppc_md.get_rtc_time(&tm);
796
797 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
798 tm.tm_hour, tm.tm_min, tm.tm_sec);
799 }
800
read_persistent_clock(struct timespec * ts)801 void read_persistent_clock(struct timespec *ts)
802 {
803 __read_persistent_clock(ts);
804
805 /* Sanitize it in case real time clock is set below EPOCH */
806 if (ts->tv_sec < 0) {
807 ts->tv_sec = 0;
808 ts->tv_nsec = 0;
809 }
810
811 }
812
813 /* clocksource code */
rtc_read(struct clocksource * cs)814 static cycle_t rtc_read(struct clocksource *cs)
815 {
816 return (cycle_t)get_rtc();
817 }
818
timebase_read(struct clocksource * cs)819 static cycle_t timebase_read(struct clocksource *cs)
820 {
821 return (cycle_t)get_tb();
822 }
823
update_vsyscall_old(struct timespec * wall_time,struct timespec * wtm,struct clocksource * clock,u32 mult,cycle_t cycle_last)824 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
825 struct clocksource *clock, u32 mult, cycle_t cycle_last)
826 {
827 u64 new_tb_to_xs, new_stamp_xsec;
828 u32 frac_sec;
829
830 if (clock != &clocksource_timebase)
831 return;
832
833 /* Make userspace gettimeofday spin until we're done. */
834 ++vdso_data->tb_update_count;
835 smp_mb();
836
837 /* 19342813113834067 ~= 2^(20+64) / 1e9 */
838 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
839 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
840 do_div(new_stamp_xsec, 1000000000);
841 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
842
843 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
844 /* this is tv_nsec / 1e9 as a 0.32 fraction */
845 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
846
847 /*
848 * tb_update_count is used to allow the userspace gettimeofday code
849 * to assure itself that it sees a consistent view of the tb_to_xs and
850 * stamp_xsec variables. It reads the tb_update_count, then reads
851 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
852 * the two values of tb_update_count match and are even then the
853 * tb_to_xs and stamp_xsec values are consistent. If not, then it
854 * loops back and reads them again until this criteria is met.
855 * We expect the caller to have done the first increment of
856 * vdso_data->tb_update_count already.
857 */
858 vdso_data->tb_orig_stamp = cycle_last;
859 vdso_data->stamp_xsec = new_stamp_xsec;
860 vdso_data->tb_to_xs = new_tb_to_xs;
861 vdso_data->wtom_clock_sec = wtm->tv_sec;
862 vdso_data->wtom_clock_nsec = wtm->tv_nsec;
863 vdso_data->stamp_xtime = *wall_time;
864 vdso_data->stamp_sec_fraction = frac_sec;
865 smp_wmb();
866 ++(vdso_data->tb_update_count);
867 }
868
update_vsyscall_tz(void)869 void update_vsyscall_tz(void)
870 {
871 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
872 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
873 }
874
clocksource_init(void)875 static void __init clocksource_init(void)
876 {
877 struct clocksource *clock;
878
879 if (__USE_RTC())
880 clock = &clocksource_rtc;
881 else
882 clock = &clocksource_timebase;
883
884 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
885 printk(KERN_ERR "clocksource: %s is already registered\n",
886 clock->name);
887 return;
888 }
889
890 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
891 clock->name, clock->mult, clock->shift);
892 }
893
decrementer_set_next_event(unsigned long evt,struct clock_event_device * dev)894 static int decrementer_set_next_event(unsigned long evt,
895 struct clock_event_device *dev)
896 {
897 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
898 set_dec(evt);
899
900 /* We may have raced with new irq work */
901 if (test_irq_work_pending())
902 set_dec(1);
903
904 return 0;
905 }
906
decrementer_shutdown(struct clock_event_device * dev)907 static int decrementer_shutdown(struct clock_event_device *dev)
908 {
909 decrementer_set_next_event(decrementer_max, dev);
910 return 0;
911 }
912
913 /* Interrupt handler for the timer broadcast IPI */
tick_broadcast_ipi_handler(void)914 void tick_broadcast_ipi_handler(void)
915 {
916 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
917
918 *next_tb = get_tb_or_rtc();
919 __timer_interrupt();
920 }
921
register_decrementer_clockevent(int cpu)922 static void register_decrementer_clockevent(int cpu)
923 {
924 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
925
926 *dec = decrementer_clockevent;
927 dec->cpumask = cpumask_of(cpu);
928
929 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
930 dec->name, dec->mult, dec->shift, cpu);
931
932 clockevents_register_device(dec);
933 }
934
enable_large_decrementer(void)935 static void enable_large_decrementer(void)
936 {
937 if (!cpu_has_feature(CPU_FTR_ARCH_300))
938 return;
939
940 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
941 return;
942
943 /*
944 * If we're running as the hypervisor we need to enable the LD manually
945 * otherwise firmware should have done it for us.
946 */
947 if (cpu_has_feature(CPU_FTR_HVMODE))
948 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
949 }
950
set_decrementer_max(void)951 static void __init set_decrementer_max(void)
952 {
953 struct device_node *cpu;
954 u32 bits = 32;
955
956 /* Prior to ISAv3 the decrementer is always 32 bit */
957 if (!cpu_has_feature(CPU_FTR_ARCH_300))
958 return;
959
960 cpu = of_find_node_by_type(NULL, "cpu");
961
962 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
963 if (bits > 64 || bits < 32) {
964 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
965 bits = 32;
966 }
967
968 /* calculate the signed maximum given this many bits */
969 decrementer_max = (1ul << (bits - 1)) - 1;
970 }
971
972 of_node_put(cpu);
973
974 pr_info("time_init: %u bit decrementer (max: %llx)\n",
975 bits, decrementer_max);
976 }
977
init_decrementer_clockevent(void)978 static void __init init_decrementer_clockevent(void)
979 {
980 int cpu = smp_processor_id();
981
982 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
983
984 decrementer_clockevent.max_delta_ns =
985 clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
986 decrementer_clockevent.min_delta_ns =
987 clockevent_delta2ns(2, &decrementer_clockevent);
988
989 register_decrementer_clockevent(cpu);
990 }
991
secondary_cpu_time_init(void)992 void secondary_cpu_time_init(void)
993 {
994 /* Enable and test the large decrementer for this cpu */
995 enable_large_decrementer();
996
997 /* Start the decrementer on CPUs that have manual control
998 * such as BookE
999 */
1000 start_cpu_decrementer();
1001
1002 /* FIME: Should make unrelatred change to move snapshot_timebase
1003 * call here ! */
1004 register_decrementer_clockevent(smp_processor_id());
1005 }
1006
1007 /* This function is only called on the boot processor */
time_init(void)1008 void __init time_init(void)
1009 {
1010 struct div_result res;
1011 u64 scale;
1012 unsigned shift;
1013
1014 if (__USE_RTC()) {
1015 /* 601 processor: dec counts down by 128 every 128ns */
1016 ppc_tb_freq = 1000000000;
1017 } else {
1018 /* Normal PowerPC with timebase register */
1019 ppc_md.calibrate_decr();
1020 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1021 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1022 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1023 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1024 }
1025
1026 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1027 tb_ticks_per_sec = ppc_tb_freq;
1028 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1029 calc_cputime_factors();
1030 setup_cputime_one_jiffy();
1031
1032 /*
1033 * Compute scale factor for sched_clock.
1034 * The calibrate_decr() function has set tb_ticks_per_sec,
1035 * which is the timebase frequency.
1036 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1037 * the 128-bit result as a 64.64 fixed-point number.
1038 * We then shift that number right until it is less than 1.0,
1039 * giving us the scale factor and shift count to use in
1040 * sched_clock().
1041 */
1042 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1043 scale = res.result_low;
1044 for (shift = 0; res.result_high != 0; ++shift) {
1045 scale = (scale >> 1) | (res.result_high << 63);
1046 res.result_high >>= 1;
1047 }
1048 tb_to_ns_scale = scale;
1049 tb_to_ns_shift = shift;
1050 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1051 boot_tb = get_tb_or_rtc();
1052
1053 /* If platform provided a timezone (pmac), we correct the time */
1054 if (timezone_offset) {
1055 sys_tz.tz_minuteswest = -timezone_offset / 60;
1056 sys_tz.tz_dsttime = 0;
1057 }
1058
1059 vdso_data->tb_update_count = 0;
1060 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1061
1062 /* initialise and enable the large decrementer (if we have one) */
1063 set_decrementer_max();
1064 enable_large_decrementer();
1065
1066 /* Start the decrementer on CPUs that have manual control
1067 * such as BookE
1068 */
1069 start_cpu_decrementer();
1070
1071 /* Register the clocksource */
1072 clocksource_init();
1073
1074 init_decrementer_clockevent();
1075 tick_setup_hrtimer_broadcast();
1076
1077 #ifdef CONFIG_COMMON_CLK
1078 of_clk_init(NULL);
1079 #endif
1080 }
1081
1082
1083 #define FEBRUARY 2
1084 #define STARTOFTIME 1970
1085 #define SECDAY 86400L
1086 #define SECYR (SECDAY * 365)
1087 #define leapyear(year) ((year) % 4 == 0 && \
1088 ((year) % 100 != 0 || (year) % 400 == 0))
1089 #define days_in_year(a) (leapyear(a) ? 366 : 365)
1090 #define days_in_month(a) (month_days[(a) - 1])
1091
1092 static int month_days[12] = {
1093 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1094 };
1095
to_tm(int tim,struct rtc_time * tm)1096 void to_tm(int tim, struct rtc_time * tm)
1097 {
1098 register int i;
1099 register long hms, day;
1100
1101 day = tim / SECDAY;
1102 hms = tim % SECDAY;
1103
1104 /* Hours, minutes, seconds are easy */
1105 tm->tm_hour = hms / 3600;
1106 tm->tm_min = (hms % 3600) / 60;
1107 tm->tm_sec = (hms % 3600) % 60;
1108
1109 /* Number of years in days */
1110 for (i = STARTOFTIME; day >= days_in_year(i); i++)
1111 day -= days_in_year(i);
1112 tm->tm_year = i;
1113
1114 /* Number of months in days left */
1115 if (leapyear(tm->tm_year))
1116 days_in_month(FEBRUARY) = 29;
1117 for (i = 1; day >= days_in_month(i); i++)
1118 day -= days_in_month(i);
1119 days_in_month(FEBRUARY) = 28;
1120 tm->tm_mon = i;
1121
1122 /* Days are what is left over (+1) from all that. */
1123 tm->tm_mday = day + 1;
1124
1125 /*
1126 * No-one uses the day of the week.
1127 */
1128 tm->tm_wday = -1;
1129 }
1130 EXPORT_SYMBOL(to_tm);
1131
1132 /*
1133 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1134 * result.
1135 */
div128_by_32(u64 dividend_high,u64 dividend_low,unsigned divisor,struct div_result * dr)1136 void div128_by_32(u64 dividend_high, u64 dividend_low,
1137 unsigned divisor, struct div_result *dr)
1138 {
1139 unsigned long a, b, c, d;
1140 unsigned long w, x, y, z;
1141 u64 ra, rb, rc;
1142
1143 a = dividend_high >> 32;
1144 b = dividend_high & 0xffffffff;
1145 c = dividend_low >> 32;
1146 d = dividend_low & 0xffffffff;
1147
1148 w = a / divisor;
1149 ra = ((u64)(a - (w * divisor)) << 32) + b;
1150
1151 rb = ((u64) do_div(ra, divisor) << 32) + c;
1152 x = ra;
1153
1154 rc = ((u64) do_div(rb, divisor) << 32) + d;
1155 y = rb;
1156
1157 do_div(rc, divisor);
1158 z = rc;
1159
1160 dr->result_high = ((u64)w << 32) + x;
1161 dr->result_low = ((u64)y << 32) + z;
1162
1163 }
1164
1165 /* We don't need to calibrate delay, we use the CPU timebase for that */
calibrate_delay(void)1166 void calibrate_delay(void)
1167 {
1168 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1169 * as the number of __delay(1) in a jiffy, so make it so
1170 */
1171 loops_per_jiffy = tb_ticks_per_jiffy;
1172 }
1173
1174 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
rtc_generic_get_time(struct device * dev,struct rtc_time * tm)1175 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1176 {
1177 ppc_md.get_rtc_time(tm);
1178 return rtc_valid_tm(tm);
1179 }
1180
rtc_generic_set_time(struct device * dev,struct rtc_time * tm)1181 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1182 {
1183 if (!ppc_md.set_rtc_time)
1184 return -EOPNOTSUPP;
1185
1186 if (ppc_md.set_rtc_time(tm) < 0)
1187 return -EOPNOTSUPP;
1188
1189 return 0;
1190 }
1191
1192 static const struct rtc_class_ops rtc_generic_ops = {
1193 .read_time = rtc_generic_get_time,
1194 .set_time = rtc_generic_set_time,
1195 };
1196
rtc_init(void)1197 static int __init rtc_init(void)
1198 {
1199 struct platform_device *pdev;
1200
1201 if (!ppc_md.get_rtc_time)
1202 return -ENODEV;
1203
1204 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1205 &rtc_generic_ops,
1206 sizeof(rtc_generic_ops));
1207
1208 return PTR_ERR_OR_ZERO(pdev);
1209 }
1210
1211 device_initcall(rtc_init);
1212 #endif
1213