2 * Common time routines among all ppc machines.
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)
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).
15 * Speeded up do_gettimeofday by getting rid of references to
16 * xtime (which required locks for consistency). (mikejc@us.ibm.com)
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.
25 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
26 * "A Kernel Model for Precision Timekeeping" by Dave Mills
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.
34 #include <linux/errno.h>
35 #include <linux/export.h>
36 #include <linux/sched.h>
37 #include <linux/sched/clock.h>
38 #include <linux/kernel.h>
39 #include <linux/param.h>
40 #include <linux/string.h>
42 #include <linux/interrupt.h>
43 #include <linux/timex.h>
44 #include <linux/kernel_stat.h>
45 #include <linux/time.h>
46 #include <linux/clockchips.h>
47 #include <linux/init.h>
48 #include <linux/profile.h>
49 #include <linux/cpu.h>
50 #include <linux/security.h>
51 #include <linux/percpu.h>
52 #include <linux/rtc.h>
53 #include <linux/jiffies.h>
54 #include <linux/posix-timers.h>
55 #include <linux/irq.h>
56 #include <linux/delay.h>
57 #include <linux/irq_work.h>
58 #include <linux/clk-provider.h>
59 #include <linux/suspend.h>
60 #include <linux/rtc.h>
61 #include <linux/sched/cputime.h>
62 #include <asm/trace.h>
65 #include <asm/processor.h>
66 #include <asm/nvram.h>
67 #include <asm/cache.h>
68 #include <asm/machdep.h>
69 #include <linux/uaccess.h>
73 #include <asm/div64.h>
75 #include <asm/vdso_datapage.h>
76 #include <asm/firmware.h>
77 #include <asm/asm-prototypes.h>
79 /* powerpc clocksource/clockevent code */
81 #include <linux/clockchips.h>
82 #include <linux/timekeeper_internal.h>
84 static u64 rtc_read(struct clocksource *);
85 static struct clocksource clocksource_rtc = {
88 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
89 .mask = CLOCKSOURCE_MASK(64),
93 static u64 timebase_read(struct clocksource *);
94 static struct clocksource clocksource_timebase = {
97 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
98 .mask = CLOCKSOURCE_MASK(64),
99 .read = timebase_read,
102 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
103 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
105 static int decrementer_set_next_event(unsigned long evt,
106 struct clock_event_device *dev);
107 static int decrementer_shutdown(struct clock_event_device *evt);
109 struct clock_event_device decrementer_clockevent = {
110 .name = "decrementer",
113 .set_next_event = decrementer_set_next_event,
114 .set_state_shutdown = decrementer_shutdown,
115 .tick_resume = decrementer_shutdown,
116 .features = CLOCK_EVT_FEAT_ONESHOT |
117 CLOCK_EVT_FEAT_C3STOP,
119 EXPORT_SYMBOL(decrementer_clockevent);
121 DEFINE_PER_CPU(u64, decrementers_next_tb);
122 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
124 #define XSEC_PER_SEC (1024*1024)
127 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
129 /* compute ((xsec << 12) * max) >> 32 */
130 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
133 unsigned long tb_ticks_per_jiffy;
134 unsigned long tb_ticks_per_usec = 100; /* sane default */
135 EXPORT_SYMBOL(tb_ticks_per_usec);
136 unsigned long tb_ticks_per_sec;
137 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
139 DEFINE_SPINLOCK(rtc_lock);
140 EXPORT_SYMBOL_GPL(rtc_lock);
142 static u64 tb_to_ns_scale __read_mostly;
143 static unsigned tb_to_ns_shift __read_mostly;
144 static u64 boot_tb __read_mostly;
146 extern struct timezone sys_tz;
147 static long timezone_offset;
149 unsigned long ppc_proc_freq;
150 EXPORT_SYMBOL_GPL(ppc_proc_freq);
151 unsigned long ppc_tb_freq;
152 EXPORT_SYMBOL_GPL(ppc_tb_freq);
154 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
156 * Factor for converting from cputime_t (timebase ticks) to
157 * microseconds. This is stored as 0.64 fixed-point binary fraction.
159 u64 __cputime_usec_factor;
160 EXPORT_SYMBOL(__cputime_usec_factor);
162 #ifdef CONFIG_PPC_SPLPAR
163 void (*dtl_consumer)(struct dtl_entry *, u64);
167 #define get_accounting(tsk) (&get_paca()->accounting)
169 #define get_accounting(tsk) (&task_thread_info(tsk)->accounting)
172 static void calc_cputime_factors(void)
174 struct div_result res;
176 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
177 __cputime_usec_factor = res.result_low;
181 * Read the SPURR on systems that have it, otherwise the PURR,
182 * or if that doesn't exist return the timebase value passed in.
184 static unsigned long read_spurr(unsigned long tb)
186 if (cpu_has_feature(CPU_FTR_SPURR))
187 return mfspr(SPRN_SPURR);
188 if (cpu_has_feature(CPU_FTR_PURR))
189 return mfspr(SPRN_PURR);
193 #ifdef CONFIG_PPC_SPLPAR
196 * Scan the dispatch trace log and count up the stolen time.
197 * Should be called with interrupts disabled.
199 static u64 scan_dispatch_log(u64 stop_tb)
201 u64 i = local_paca->dtl_ridx;
202 struct dtl_entry *dtl = local_paca->dtl_curr;
203 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
204 struct lppaca *vpa = local_paca->lppaca_ptr;
212 if (i == be64_to_cpu(vpa->dtl_idx))
214 while (i < be64_to_cpu(vpa->dtl_idx)) {
215 dtb = be64_to_cpu(dtl->timebase);
216 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
217 be32_to_cpu(dtl->ready_to_enqueue_time);
219 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
220 /* buffer has overflowed */
221 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
222 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
228 dtl_consumer(dtl, i);
233 dtl = local_paca->dispatch_log;
235 local_paca->dtl_ridx = i;
236 local_paca->dtl_curr = dtl;
241 * Accumulate stolen time by scanning the dispatch trace log.
242 * Called on entry from user mode.
244 void accumulate_stolen_time(void)
247 u8 save_soft_enabled = local_paca->soft_enabled;
248 struct cpu_accounting_data *acct = &local_paca->accounting;
250 /* We are called early in the exception entry, before
251 * soft/hard_enabled are sync'ed to the expected state
252 * for the exception. We are hard disabled but the PACA
253 * needs to reflect that so various debug stuff doesn't
256 local_paca->soft_enabled = 0;
258 sst = scan_dispatch_log(acct->starttime_user);
259 ust = scan_dispatch_log(acct->starttime);
262 acct->steal_time += ust + sst;
264 local_paca->soft_enabled = save_soft_enabled;
267 static inline u64 calculate_stolen_time(u64 stop_tb)
269 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
270 return scan_dispatch_log(stop_tb);
275 #else /* CONFIG_PPC_SPLPAR */
276 static inline u64 calculate_stolen_time(u64 stop_tb)
281 #endif /* CONFIG_PPC_SPLPAR */
284 * Account time for a transition between system, hard irq
287 static unsigned long vtime_delta(struct task_struct *tsk,
288 unsigned long *stime_scaled,
289 unsigned long *steal_time)
291 unsigned long now, nowscaled, deltascaled;
293 unsigned long utime, utime_scaled;
294 struct cpu_accounting_data *acct = get_accounting(tsk);
296 WARN_ON_ONCE(!irqs_disabled());
299 nowscaled = read_spurr(now);
300 stime = now - acct->starttime;
301 acct->starttime = now;
302 deltascaled = nowscaled - acct->startspurr;
303 acct->startspurr = nowscaled;
305 *steal_time = calculate_stolen_time(now);
307 utime = acct->utime - acct->utime_sspurr;
308 acct->utime_sspurr = acct->utime;
311 * Because we don't read the SPURR on every kernel entry/exit,
312 * deltascaled includes both user and system SPURR ticks.
313 * Apportion these ticks to system SPURR ticks and user
314 * SPURR ticks in the same ratio as the system time (delta)
315 * and user time (udelta) values obtained from the timebase
316 * over the same interval. The system ticks get accounted here;
317 * the user ticks get saved up in paca->user_time_scaled to be
318 * used by account_process_tick.
320 *stime_scaled = stime;
321 utime_scaled = utime;
322 if (deltascaled != stime + utime) {
324 *stime_scaled = deltascaled * stime / (stime + utime);
325 utime_scaled = deltascaled - *stime_scaled;
327 *stime_scaled = deltascaled;
330 acct->utime_scaled += utime_scaled;
335 void vtime_account_system(struct task_struct *tsk)
337 unsigned long stime, stime_scaled, steal_time;
338 struct cpu_accounting_data *acct = get_accounting(tsk);
340 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
342 stime -= min(stime, steal_time);
343 acct->steal_time += steal_time;
345 if ((tsk->flags & PF_VCPU) && !irq_count()) {
346 acct->gtime += stime;
347 acct->utime_scaled += stime_scaled;
350 acct->hardirq_time += stime;
351 else if (in_serving_softirq())
352 acct->softirq_time += stime;
354 acct->stime += stime;
356 acct->stime_scaled += stime_scaled;
359 EXPORT_SYMBOL_GPL(vtime_account_system);
361 void vtime_account_idle(struct task_struct *tsk)
363 unsigned long stime, stime_scaled, steal_time;
364 struct cpu_accounting_data *acct = get_accounting(tsk);
366 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
367 acct->idle_time += stime + steal_time;
371 * Account the whole cputime accumulated in the paca
372 * Must be called with interrupts disabled.
373 * Assumes that vtime_account_system/idle() has been called
374 * recently (i.e. since the last entry from usermode) so that
375 * get_paca()->user_time_scaled is up to date.
377 void vtime_flush(struct task_struct *tsk)
379 struct cpu_accounting_data *acct = get_accounting(tsk);
382 account_user_time(tsk, cputime_to_nsecs(acct->utime));
384 if (acct->utime_scaled)
385 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
388 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
390 if (acct->steal_time)
391 account_steal_time(cputime_to_nsecs(acct->steal_time));
394 account_idle_time(cputime_to_nsecs(acct->idle_time));
397 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
399 if (acct->stime_scaled)
400 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
402 if (acct->hardirq_time)
403 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
405 if (acct->softirq_time)
406 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
410 acct->utime_scaled = 0;
411 acct->utime_sspurr = 0;
413 acct->steal_time = 0;
416 acct->stime_scaled = 0;
417 acct->hardirq_time = 0;
418 acct->softirq_time = 0;
423 * Called from the context switch with interrupts disabled, to charge all
424 * accumulated times to the current process, and to prepare accounting on
427 void arch_vtime_task_switch(struct task_struct *prev)
429 struct cpu_accounting_data *acct = get_accounting(current);
431 acct->starttime = get_accounting(prev)->starttime;
432 acct->startspurr = get_accounting(prev)->startspurr;
434 #endif /* CONFIG_PPC32 */
436 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
437 #define calc_cputime_factors()
440 void __delay(unsigned long loops)
448 /* the RTCL register wraps at 1000000000 */
449 diff = get_rtcl() - start;
452 } while (diff < loops);
455 while (get_tbl() - start < loops)
460 EXPORT_SYMBOL(__delay);
462 void udelay(unsigned long usecs)
464 __delay(tb_ticks_per_usec * usecs);
466 EXPORT_SYMBOL(udelay);
469 unsigned long profile_pc(struct pt_regs *regs)
471 unsigned long pc = instruction_pointer(regs);
473 if (in_lock_functions(pc))
478 EXPORT_SYMBOL(profile_pc);
481 #ifdef CONFIG_IRQ_WORK
484 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
487 static inline unsigned long test_irq_work_pending(void)
491 asm volatile("lbz %0,%1(13)"
493 : "i" (offsetof(struct paca_struct, irq_work_pending)));
497 static inline void set_irq_work_pending_flag(void)
499 asm volatile("stb %0,%1(13)" : :
501 "i" (offsetof(struct paca_struct, irq_work_pending)));
504 static inline void clear_irq_work_pending(void)
506 asm volatile("stb %0,%1(13)" : :
508 "i" (offsetof(struct paca_struct, irq_work_pending)));
513 DEFINE_PER_CPU(u8, irq_work_pending);
515 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
516 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
517 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
519 #endif /* 32 vs 64 bit */
521 void arch_irq_work_raise(void)
524 set_irq_work_pending_flag();
529 #else /* CONFIG_IRQ_WORK */
531 #define test_irq_work_pending() 0
532 #define clear_irq_work_pending()
534 #endif /* CONFIG_IRQ_WORK */
536 static void __timer_interrupt(void)
538 struct pt_regs *regs = get_irq_regs();
539 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
540 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
543 trace_timer_interrupt_entry(regs);
545 if (test_irq_work_pending()) {
546 clear_irq_work_pending();
550 now = get_tb_or_rtc();
551 if (now >= *next_tb) {
553 if (evt->event_handler)
554 evt->event_handler(evt);
555 __this_cpu_inc(irq_stat.timer_irqs_event);
557 now = *next_tb - now;
558 if (now <= decrementer_max)
560 /* We may have raced with new irq work */
561 if (test_irq_work_pending())
563 __this_cpu_inc(irq_stat.timer_irqs_others);
567 /* collect purr register values often, for accurate calculations */
568 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
569 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
570 cu->current_tb = mfspr(SPRN_PURR);
574 trace_timer_interrupt_exit(regs);
578 * timer_interrupt - gets called when the decrementer overflows,
579 * with interrupts disabled.
581 void timer_interrupt(struct pt_regs * regs)
583 struct pt_regs *old_regs;
584 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
586 /* Ensure a positive value is written to the decrementer, or else
587 * some CPUs will continue to take decrementer exceptions.
589 set_dec(decrementer_max);
591 /* Some implementations of hotplug will get timer interrupts while
592 * offline, just ignore these and we also need to set
593 * decrementers_next_tb as MAX to make sure __check_irq_replay
594 * don't replay timer interrupt when return, otherwise we'll trap
597 if (!cpu_online(smp_processor_id())) {
602 /* Conditionally hard-enable interrupts now that the DEC has been
603 * bumped to its maximum value
605 may_hard_irq_enable();
608 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
609 if (atomic_read(&ppc_n_lost_interrupts) != 0)
613 old_regs = set_irq_regs(regs);
618 set_irq_regs(old_regs);
620 EXPORT_SYMBOL(timer_interrupt);
623 * Hypervisor decrementer interrupts shouldn't occur but are sometimes
624 * left pending on exit from a KVM guest. We don't need to do anything
625 * to clear them, as they are edge-triggered.
627 void hdec_interrupt(struct pt_regs *regs)
631 #ifdef CONFIG_SUSPEND
632 static void generic_suspend_disable_irqs(void)
634 /* Disable the decrementer, so that it doesn't interfere
638 set_dec(decrementer_max);
640 set_dec(decrementer_max);
643 static void generic_suspend_enable_irqs(void)
648 /* Overrides the weak version in kernel/power/main.c */
649 void arch_suspend_disable_irqs(void)
651 if (ppc_md.suspend_disable_irqs)
652 ppc_md.suspend_disable_irqs();
653 generic_suspend_disable_irqs();
656 /* Overrides the weak version in kernel/power/main.c */
657 void arch_suspend_enable_irqs(void)
659 generic_suspend_enable_irqs();
660 if (ppc_md.suspend_enable_irqs)
661 ppc_md.suspend_enable_irqs();
665 unsigned long long tb_to_ns(unsigned long long ticks)
667 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
669 EXPORT_SYMBOL_GPL(tb_to_ns);
672 * Scheduler clock - returns current time in nanosec units.
674 * Note: mulhdu(a, b) (multiply high double unsigned) returns
675 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
676 * are 64-bit unsigned numbers.
678 unsigned long long sched_clock(void)
682 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
686 #ifdef CONFIG_PPC_PSERIES
689 * Running clock - attempts to give a view of time passing for a virtualised
691 * Uses the VTB register if available otherwise a next best guess.
693 unsigned long long running_clock(void)
696 * Don't read the VTB as a host since KVM does not switch in host
697 * timebase into the VTB when it takes a guest off the CPU, reading the
698 * VTB would result in reading 'last switched out' guest VTB.
700 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
701 * would be unsafe to rely only on the #ifdef above.
703 if (firmware_has_feature(FW_FEATURE_LPAR) &&
704 cpu_has_feature(CPU_FTR_ARCH_207S))
705 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
708 * This is a next best approximation without a VTB.
709 * On a host which is running bare metal there should never be any stolen
710 * time and on a host which doesn't do any virtualisation TB *should* equal
711 * VTB so it makes no difference anyway.
713 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
717 static int __init get_freq(char *name, int cells, unsigned long *val)
719 struct device_node *cpu;
723 /* The cpu node should have timebase and clock frequency properties */
724 cpu = of_find_node_by_type(NULL, "cpu");
727 fp = of_get_property(cpu, name, NULL);
730 *val = of_read_ulong(fp, cells);
739 static void start_cpu_decrementer(void)
741 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
742 /* Clear any pending timer interrupts */
743 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
745 /* Enable decrementer interrupt */
746 mtspr(SPRN_TCR, TCR_DIE);
747 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
750 void __init generic_calibrate_decr(void)
752 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
754 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
755 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
757 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
761 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
763 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
764 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
766 printk(KERN_ERR "WARNING: Estimating processor frequency "
771 int update_persistent_clock(struct timespec now)
775 if (!ppc_md.set_rtc_time)
778 to_tm(now.tv_sec + 1 + timezone_offset, &tm);
782 return ppc_md.set_rtc_time(&tm);
785 static void __read_persistent_clock(struct timespec *ts)
788 static int first = 1;
791 /* XXX this is a litle fragile but will work okay in the short term */
794 if (ppc_md.time_init)
795 timezone_offset = ppc_md.time_init();
797 /* get_boot_time() isn't guaranteed to be safe to call late */
798 if (ppc_md.get_boot_time) {
799 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
803 if (!ppc_md.get_rtc_time) {
807 ppc_md.get_rtc_time(&tm);
809 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
810 tm.tm_hour, tm.tm_min, tm.tm_sec);
813 void read_persistent_clock(struct timespec *ts)
815 __read_persistent_clock(ts);
817 /* Sanitize it in case real time clock is set below EPOCH */
818 if (ts->tv_sec < 0) {
825 /* clocksource code */
826 static u64 rtc_read(struct clocksource *cs)
828 return (u64)get_rtc();
831 static u64 timebase_read(struct clocksource *cs)
833 return (u64)get_tb();
836 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
837 struct clocksource *clock, u32 mult, u64 cycle_last)
839 u64 new_tb_to_xs, new_stamp_xsec;
842 if (clock != &clocksource_timebase)
845 /* Make userspace gettimeofday spin until we're done. */
846 ++vdso_data->tb_update_count;
849 /* 19342813113834067 ~= 2^(20+64) / 1e9 */
850 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
851 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
852 do_div(new_stamp_xsec, 1000000000);
853 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
855 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
856 /* this is tv_nsec / 1e9 as a 0.32 fraction */
857 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
860 * tb_update_count is used to allow the userspace gettimeofday code
861 * to assure itself that it sees a consistent view of the tb_to_xs and
862 * stamp_xsec variables. It reads the tb_update_count, then reads
863 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
864 * the two values of tb_update_count match and are even then the
865 * tb_to_xs and stamp_xsec values are consistent. If not, then it
866 * loops back and reads them again until this criteria is met.
867 * We expect the caller to have done the first increment of
868 * vdso_data->tb_update_count already.
870 vdso_data->tb_orig_stamp = cycle_last;
871 vdso_data->stamp_xsec = new_stamp_xsec;
872 vdso_data->tb_to_xs = new_tb_to_xs;
873 vdso_data->wtom_clock_sec = wtm->tv_sec;
874 vdso_data->wtom_clock_nsec = wtm->tv_nsec;
875 vdso_data->stamp_xtime = *wall_time;
876 vdso_data->stamp_sec_fraction = frac_sec;
878 ++(vdso_data->tb_update_count);
881 void update_vsyscall_tz(void)
883 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
884 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
887 static void __init clocksource_init(void)
889 struct clocksource *clock;
892 clock = &clocksource_rtc;
894 clock = &clocksource_timebase;
896 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
897 printk(KERN_ERR "clocksource: %s is already registered\n",
902 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
903 clock->name, clock->mult, clock->shift);
906 static int decrementer_set_next_event(unsigned long evt,
907 struct clock_event_device *dev)
909 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
912 /* We may have raced with new irq work */
913 if (test_irq_work_pending())
919 static int decrementer_shutdown(struct clock_event_device *dev)
921 decrementer_set_next_event(decrementer_max, dev);
925 /* Interrupt handler for the timer broadcast IPI */
926 void tick_broadcast_ipi_handler(void)
928 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
930 *next_tb = get_tb_or_rtc();
934 static void register_decrementer_clockevent(int cpu)
936 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
938 *dec = decrementer_clockevent;
939 dec->cpumask = cpumask_of(cpu);
941 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
942 dec->name, dec->mult, dec->shift, cpu);
944 clockevents_register_device(dec);
947 static void enable_large_decrementer(void)
949 if (!cpu_has_feature(CPU_FTR_ARCH_300))
952 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
956 * If we're running as the hypervisor we need to enable the LD manually
957 * otherwise firmware should have done it for us.
959 if (cpu_has_feature(CPU_FTR_HVMODE))
960 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
963 static void __init set_decrementer_max(void)
965 struct device_node *cpu;
968 /* Prior to ISAv3 the decrementer is always 32 bit */
969 if (!cpu_has_feature(CPU_FTR_ARCH_300))
972 cpu = of_find_node_by_type(NULL, "cpu");
974 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
975 if (bits > 64 || bits < 32) {
976 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
980 /* calculate the signed maximum given this many bits */
981 decrementer_max = (1ul << (bits - 1)) - 1;
986 pr_info("time_init: %u bit decrementer (max: %llx)\n",
987 bits, decrementer_max);
990 static void __init init_decrementer_clockevent(void)
992 int cpu = smp_processor_id();
994 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
996 decrementer_clockevent.max_delta_ns =
997 clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
998 decrementer_clockevent.max_delta_ticks = decrementer_max;
999 decrementer_clockevent.min_delta_ns =
1000 clockevent_delta2ns(2, &decrementer_clockevent);
1001 decrementer_clockevent.min_delta_ticks = 2;
1003 register_decrementer_clockevent(cpu);
1006 void secondary_cpu_time_init(void)
1008 /* Enable and test the large decrementer for this cpu */
1009 enable_large_decrementer();
1011 /* Start the decrementer on CPUs that have manual control
1014 start_cpu_decrementer();
1016 /* FIME: Should make unrelatred change to move snapshot_timebase
1018 register_decrementer_clockevent(smp_processor_id());
1021 /* This function is only called on the boot processor */
1022 void __init time_init(void)
1024 struct div_result res;
1029 /* 601 processor: dec counts down by 128 every 128ns */
1030 ppc_tb_freq = 1000000000;
1032 /* Normal PowerPC with timebase register */
1033 ppc_md.calibrate_decr();
1034 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1035 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1036 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1037 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1040 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1041 tb_ticks_per_sec = ppc_tb_freq;
1042 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1043 calc_cputime_factors();
1046 * Compute scale factor for sched_clock.
1047 * The calibrate_decr() function has set tb_ticks_per_sec,
1048 * which is the timebase frequency.
1049 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1050 * the 128-bit result as a 64.64 fixed-point number.
1051 * We then shift that number right until it is less than 1.0,
1052 * giving us the scale factor and shift count to use in
1055 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1056 scale = res.result_low;
1057 for (shift = 0; res.result_high != 0; ++shift) {
1058 scale = (scale >> 1) | (res.result_high << 63);
1059 res.result_high >>= 1;
1061 tb_to_ns_scale = scale;
1062 tb_to_ns_shift = shift;
1063 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1064 boot_tb = get_tb_or_rtc();
1066 /* If platform provided a timezone (pmac), we correct the time */
1067 if (timezone_offset) {
1068 sys_tz.tz_minuteswest = -timezone_offset / 60;
1069 sys_tz.tz_dsttime = 0;
1072 vdso_data->tb_update_count = 0;
1073 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1075 /* initialise and enable the large decrementer (if we have one) */
1076 set_decrementer_max();
1077 enable_large_decrementer();
1079 /* Start the decrementer on CPUs that have manual control
1082 start_cpu_decrementer();
1084 /* Register the clocksource */
1087 init_decrementer_clockevent();
1088 tick_setup_hrtimer_broadcast();
1090 #ifdef CONFIG_COMMON_CLK
1097 #define STARTOFTIME 1970
1098 #define SECDAY 86400L
1099 #define SECYR (SECDAY * 365)
1100 #define leapyear(year) ((year) % 4 == 0 && \
1101 ((year) % 100 != 0 || (year) % 400 == 0))
1102 #define days_in_year(a) (leapyear(a) ? 366 : 365)
1103 #define days_in_month(a) (month_days[(a) - 1])
1105 static int month_days[12] = {
1106 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1109 void to_tm(int tim, struct rtc_time * tm)
1112 register long hms, day;
1117 /* Hours, minutes, seconds are easy */
1118 tm->tm_hour = hms / 3600;
1119 tm->tm_min = (hms % 3600) / 60;
1120 tm->tm_sec = (hms % 3600) % 60;
1122 /* Number of years in days */
1123 for (i = STARTOFTIME; day >= days_in_year(i); i++)
1124 day -= days_in_year(i);
1127 /* Number of months in days left */
1128 if (leapyear(tm->tm_year))
1129 days_in_month(FEBRUARY) = 29;
1130 for (i = 1; day >= days_in_month(i); i++)
1131 day -= days_in_month(i);
1132 days_in_month(FEBRUARY) = 28;
1135 /* Days are what is left over (+1) from all that. */
1136 tm->tm_mday = day + 1;
1139 * No-one uses the day of the week.
1143 EXPORT_SYMBOL(to_tm);
1146 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1149 void div128_by_32(u64 dividend_high, u64 dividend_low,
1150 unsigned divisor, struct div_result *dr)
1152 unsigned long a, b, c, d;
1153 unsigned long w, x, y, z;
1156 a = dividend_high >> 32;
1157 b = dividend_high & 0xffffffff;
1158 c = dividend_low >> 32;
1159 d = dividend_low & 0xffffffff;
1162 ra = ((u64)(a - (w * divisor)) << 32) + b;
1164 rb = ((u64) do_div(ra, divisor) << 32) + c;
1167 rc = ((u64) do_div(rb, divisor) << 32) + d;
1170 do_div(rc, divisor);
1173 dr->result_high = ((u64)w << 32) + x;
1174 dr->result_low = ((u64)y << 32) + z;
1178 /* We don't need to calibrate delay, we use the CPU timebase for that */
1179 void calibrate_delay(void)
1181 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1182 * as the number of __delay(1) in a jiffy, so make it so
1184 loops_per_jiffy = tb_ticks_per_jiffy;
1187 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1188 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1190 ppc_md.get_rtc_time(tm);
1191 return rtc_valid_tm(tm);
1194 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1196 if (!ppc_md.set_rtc_time)
1199 if (ppc_md.set_rtc_time(tm) < 0)
1205 static const struct rtc_class_ops rtc_generic_ops = {
1206 .read_time = rtc_generic_get_time,
1207 .set_time = rtc_generic_set_time,
1210 static int __init rtc_init(void)
1212 struct platform_device *pdev;
1214 if (!ppc_md.get_rtc_time)
1217 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1219 sizeof(rtc_generic_ops));
1221 return PTR_ERR_OR_ZERO(pdev);
1224 device_initcall(rtc_init);