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/kernel.h>
38 #include <linux/param.h>
39 #include <linux/string.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>
63 #include <asm/processor.h>
64 #include <asm/nvram.h>
65 #include <asm/cache.h>
66 #include <asm/machdep.h>
67 #include <linux/uaccess.h>
71 #include <asm/div64.h>
73 #include <asm/vdso_datapage.h>
74 #include <asm/firmware.h>
75 #include <asm/cputime.h>
76 #include <asm/asm-prototypes.h>
78 /* powerpc clocksource/clockevent code */
80 #include <linux/clockchips.h>
81 #include <linux/timekeeper_internal.h>
83 static u64 rtc_read(struct clocksource *);
84 static struct clocksource clocksource_rtc = {
87 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
88 .mask = CLOCKSOURCE_MASK(64),
92 static u64 timebase_read(struct clocksource *);
93 static struct clocksource clocksource_timebase = {
96 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
97 .mask = CLOCKSOURCE_MASK(64),
98 .read = timebase_read,
101 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
102 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
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);
108 struct clock_event_device decrementer_clockevent = {
109 .name = "decrementer",
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,
118 EXPORT_SYMBOL(decrementer_clockevent);
120 DEFINE_PER_CPU(u64, decrementers_next_tb);
121 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
123 #define XSEC_PER_SEC (1024*1024)
126 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
128 /* compute ((xsec << 12) * max) >> 32 */
129 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
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 */
138 DEFINE_SPINLOCK(rtc_lock);
139 EXPORT_SYMBOL_GPL(rtc_lock);
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;
145 extern struct timezone sys_tz;
146 static long timezone_offset;
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);
153 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
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.
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);
168 cputime_t cputime_one_jiffy;
170 #ifdef CONFIG_PPC_SPLPAR
171 void (*dtl_consumer)(struct dtl_entry *, u64);
175 #define get_accounting(tsk) (&get_paca()->accounting)
177 #define get_accounting(tsk) (&task_thread_info(tsk)->accounting)
180 static void calc_cputime_factors(void)
182 struct div_result res;
184 div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
185 __cputime_jiffies_factor = res.result_low;
186 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
187 __cputime_usec_factor = res.result_low;
188 div128_by_32(1, 0, tb_ticks_per_sec, &res);
189 __cputime_sec_factor = res.result_low;
190 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
191 __cputime_clockt_factor = res.result_low;
195 * Read the SPURR on systems that have it, otherwise the PURR,
196 * or if that doesn't exist return the timebase value passed in.
198 static unsigned long read_spurr(unsigned long tb)
200 if (cpu_has_feature(CPU_FTR_SPURR))
201 return mfspr(SPRN_SPURR);
202 if (cpu_has_feature(CPU_FTR_PURR))
203 return mfspr(SPRN_PURR);
207 #ifdef CONFIG_PPC_SPLPAR
210 * Scan the dispatch trace log and count up the stolen time.
211 * Should be called with interrupts disabled.
213 static u64 scan_dispatch_log(u64 stop_tb)
215 u64 i = local_paca->dtl_ridx;
216 struct dtl_entry *dtl = local_paca->dtl_curr;
217 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
218 struct lppaca *vpa = local_paca->lppaca_ptr;
226 if (i == be64_to_cpu(vpa->dtl_idx))
228 while (i < be64_to_cpu(vpa->dtl_idx)) {
229 dtb = be64_to_cpu(dtl->timebase);
230 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
231 be32_to_cpu(dtl->ready_to_enqueue_time);
233 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
234 /* buffer has overflowed */
235 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
236 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
242 dtl_consumer(dtl, i);
247 dtl = local_paca->dispatch_log;
249 local_paca->dtl_ridx = i;
250 local_paca->dtl_curr = dtl;
255 * Accumulate stolen time by scanning the dispatch trace log.
256 * Called on entry from user mode.
258 void accumulate_stolen_time(void)
261 u8 save_soft_enabled = local_paca->soft_enabled;
262 struct cpu_accounting_data *acct = &local_paca->accounting;
264 /* We are called early in the exception entry, before
265 * soft/hard_enabled are sync'ed to the expected state
266 * for the exception. We are hard disabled but the PACA
267 * needs to reflect that so various debug stuff doesn't
270 local_paca->soft_enabled = 0;
272 sst = scan_dispatch_log(acct->starttime_user);
273 ust = scan_dispatch_log(acct->starttime);
276 acct->steal_time += ust + sst;
278 local_paca->soft_enabled = save_soft_enabled;
281 static inline u64 calculate_stolen_time(u64 stop_tb)
283 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
284 return scan_dispatch_log(stop_tb);
289 #else /* CONFIG_PPC_SPLPAR */
290 static inline u64 calculate_stolen_time(u64 stop_tb)
295 #endif /* CONFIG_PPC_SPLPAR */
298 * Account time for a transition between system, hard irq
301 static unsigned long vtime_delta(struct task_struct *tsk,
302 unsigned long *stime_scaled,
303 unsigned long *steal_time)
305 unsigned long now, nowscaled, deltascaled;
307 unsigned long utime, utime_scaled;
308 struct cpu_accounting_data *acct = get_accounting(tsk);
310 WARN_ON_ONCE(!irqs_disabled());
313 nowscaled = read_spurr(now);
314 stime = now - acct->starttime;
315 acct->starttime = now;
316 deltascaled = nowscaled - acct->startspurr;
317 acct->startspurr = nowscaled;
319 *steal_time = calculate_stolen_time(now);
321 utime = acct->utime - acct->utime_sspurr;
322 acct->utime_sspurr = acct->utime;
325 * Because we don't read the SPURR on every kernel entry/exit,
326 * deltascaled includes both user and system SPURR ticks.
327 * Apportion these ticks to system SPURR ticks and user
328 * SPURR ticks in the same ratio as the system time (delta)
329 * and user time (udelta) values obtained from the timebase
330 * over the same interval. The system ticks get accounted here;
331 * the user ticks get saved up in paca->user_time_scaled to be
332 * used by account_process_tick.
334 *stime_scaled = stime;
335 utime_scaled = utime;
336 if (deltascaled != stime + utime) {
338 *stime_scaled = deltascaled * stime / (stime + utime);
339 utime_scaled = deltascaled - *stime_scaled;
341 *stime_scaled = deltascaled;
344 acct->utime_scaled += utime_scaled;
349 void vtime_account_system(struct task_struct *tsk)
351 unsigned long stime, stime_scaled, steal_time;
352 struct cpu_accounting_data *acct = get_accounting(tsk);
354 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
356 stime -= min(stime, steal_time);
357 acct->steal_time += steal_time;
359 if ((tsk->flags & PF_VCPU) && !irq_count()) {
360 acct->gtime += stime;
361 acct->utime_scaled += stime_scaled;
364 acct->hardirq_time += stime;
365 else if (in_serving_softirq())
366 acct->softirq_time += stime;
368 acct->stime += stime;
370 acct->stime_scaled += stime_scaled;
373 EXPORT_SYMBOL_GPL(vtime_account_system);
375 void vtime_account_idle(struct task_struct *tsk)
377 unsigned long stime, stime_scaled, steal_time;
378 struct cpu_accounting_data *acct = get_accounting(tsk);
380 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
381 acct->idle_time += stime + steal_time;
385 * Transfer the user time accumulated in the paca
386 * by the exception entry and exit code to the generic
387 * process user time records.
388 * Must be called with interrupts disabled.
389 * Assumes that vtime_account_system/idle() has been called
390 * recently (i.e. since the last entry from usermode) so that
391 * get_paca()->user_time_scaled is up to date.
393 void vtime_account_user(struct task_struct *tsk)
395 struct cpu_accounting_data *acct = get_accounting(tsk);
398 account_user_time(tsk, acct->utime);
400 if (acct->utime_scaled)
401 tsk->utimescaled += acct->utime_scaled;
404 account_guest_time(tsk, acct->gtime);
406 if (acct->steal_time)
407 account_steal_time(acct->steal_time);
410 account_idle_time(acct->idle_time);
413 account_system_index_time(tsk, acct->stime, CPUTIME_SYSTEM);
415 if (acct->stime_scaled)
416 tsk->stimescaled += acct->stime_scaled;
418 if (acct->hardirq_time)
419 account_system_index_time(tsk, acct->hardirq_time, CPUTIME_IRQ);
421 if (acct->softirq_time)
422 account_system_index_time(tsk, acct->softirq_time, CPUTIME_SOFTIRQ);
425 acct->utime_scaled = 0;
426 acct->utime_sspurr = 0;
428 acct->steal_time = 0;
431 acct->stime_scaled = 0;
432 acct->hardirq_time = 0;
433 acct->softirq_time = 0;
438 * Called from the context switch with interrupts disabled, to charge all
439 * accumulated times to the current process, and to prepare accounting on
442 void arch_vtime_task_switch(struct task_struct *prev)
444 struct cpu_accounting_data *acct = get_accounting(current);
446 acct->starttime = get_accounting(prev)->starttime;
447 acct->startspurr = get_accounting(prev)->startspurr;
449 #endif /* CONFIG_PPC32 */
451 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
452 #define calc_cputime_factors()
455 void __delay(unsigned long loops)
463 /* the RTCL register wraps at 1000000000 */
464 diff = get_rtcl() - start;
467 } while (diff < loops);
470 while (get_tbl() - start < loops)
475 EXPORT_SYMBOL(__delay);
477 void udelay(unsigned long usecs)
479 __delay(tb_ticks_per_usec * usecs);
481 EXPORT_SYMBOL(udelay);
484 unsigned long profile_pc(struct pt_regs *regs)
486 unsigned long pc = instruction_pointer(regs);
488 if (in_lock_functions(pc))
493 EXPORT_SYMBOL(profile_pc);
496 #ifdef CONFIG_IRQ_WORK
499 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
502 static inline unsigned long test_irq_work_pending(void)
506 asm volatile("lbz %0,%1(13)"
508 : "i" (offsetof(struct paca_struct, irq_work_pending)));
512 static inline void set_irq_work_pending_flag(void)
514 asm volatile("stb %0,%1(13)" : :
516 "i" (offsetof(struct paca_struct, irq_work_pending)));
519 static inline void clear_irq_work_pending(void)
521 asm volatile("stb %0,%1(13)" : :
523 "i" (offsetof(struct paca_struct, irq_work_pending)));
528 DEFINE_PER_CPU(u8, irq_work_pending);
530 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
531 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
532 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
534 #endif /* 32 vs 64 bit */
536 void arch_irq_work_raise(void)
539 set_irq_work_pending_flag();
544 #else /* CONFIG_IRQ_WORK */
546 #define test_irq_work_pending() 0
547 #define clear_irq_work_pending()
549 #endif /* CONFIG_IRQ_WORK */
551 static void __timer_interrupt(void)
553 struct pt_regs *regs = get_irq_regs();
554 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
555 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
558 trace_timer_interrupt_entry(regs);
560 if (test_irq_work_pending()) {
561 clear_irq_work_pending();
565 now = get_tb_or_rtc();
566 if (now >= *next_tb) {
568 if (evt->event_handler)
569 evt->event_handler(evt);
570 __this_cpu_inc(irq_stat.timer_irqs_event);
572 now = *next_tb - now;
573 if (now <= decrementer_max)
575 /* We may have raced with new irq work */
576 if (test_irq_work_pending())
578 __this_cpu_inc(irq_stat.timer_irqs_others);
582 /* collect purr register values often, for accurate calculations */
583 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
584 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
585 cu->current_tb = mfspr(SPRN_PURR);
589 trace_timer_interrupt_exit(regs);
593 * timer_interrupt - gets called when the decrementer overflows,
594 * with interrupts disabled.
596 void timer_interrupt(struct pt_regs * regs)
598 struct pt_regs *old_regs;
599 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
601 /* Ensure a positive value is written to the decrementer, or else
602 * some CPUs will continue to take decrementer exceptions.
604 set_dec(decrementer_max);
606 /* Some implementations of hotplug will get timer interrupts while
607 * offline, just ignore these and we also need to set
608 * decrementers_next_tb as MAX to make sure __check_irq_replay
609 * don't replay timer interrupt when return, otherwise we'll trap
612 if (!cpu_online(smp_processor_id())) {
617 /* Conditionally hard-enable interrupts now that the DEC has been
618 * bumped to its maximum value
620 may_hard_irq_enable();
623 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
624 if (atomic_read(&ppc_n_lost_interrupts) != 0)
628 old_regs = set_irq_regs(regs);
633 set_irq_regs(old_regs);
635 EXPORT_SYMBOL(timer_interrupt);
638 * Hypervisor decrementer interrupts shouldn't occur but are sometimes
639 * left pending on exit from a KVM guest. We don't need to do anything
640 * to clear them, as they are edge-triggered.
642 void hdec_interrupt(struct pt_regs *regs)
646 #ifdef CONFIG_SUSPEND
647 static void generic_suspend_disable_irqs(void)
649 /* Disable the decrementer, so that it doesn't interfere
653 set_dec(decrementer_max);
655 set_dec(decrementer_max);
658 static void generic_suspend_enable_irqs(void)
663 /* Overrides the weak version in kernel/power/main.c */
664 void arch_suspend_disable_irqs(void)
666 if (ppc_md.suspend_disable_irqs)
667 ppc_md.suspend_disable_irqs();
668 generic_suspend_disable_irqs();
671 /* Overrides the weak version in kernel/power/main.c */
672 void arch_suspend_enable_irqs(void)
674 generic_suspend_enable_irqs();
675 if (ppc_md.suspend_enable_irqs)
676 ppc_md.suspend_enable_irqs();
680 unsigned long long tb_to_ns(unsigned long long ticks)
682 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
684 EXPORT_SYMBOL_GPL(tb_to_ns);
687 * Scheduler clock - returns current time in nanosec units.
689 * Note: mulhdu(a, b) (multiply high double unsigned) returns
690 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
691 * are 64-bit unsigned numbers.
693 unsigned long long sched_clock(void)
697 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
701 #ifdef CONFIG_PPC_PSERIES
704 * Running clock - attempts to give a view of time passing for a virtualised
706 * Uses the VTB register if available otherwise a next best guess.
708 unsigned long long running_clock(void)
711 * Don't read the VTB as a host since KVM does not switch in host
712 * timebase into the VTB when it takes a guest off the CPU, reading the
713 * VTB would result in reading 'last switched out' guest VTB.
715 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
716 * would be unsafe to rely only on the #ifdef above.
718 if (firmware_has_feature(FW_FEATURE_LPAR) &&
719 cpu_has_feature(CPU_FTR_ARCH_207S))
720 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
723 * This is a next best approximation without a VTB.
724 * On a host which is running bare metal there should never be any stolen
725 * time and on a host which doesn't do any virtualisation TB *should* equal
726 * VTB so it makes no difference anyway.
728 return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
732 static int __init get_freq(char *name, int cells, unsigned long *val)
734 struct device_node *cpu;
738 /* The cpu node should have timebase and clock frequency properties */
739 cpu = of_find_node_by_type(NULL, "cpu");
742 fp = of_get_property(cpu, name, NULL);
745 *val = of_read_ulong(fp, cells);
754 static void start_cpu_decrementer(void)
756 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
757 /* Clear any pending timer interrupts */
758 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
760 /* Enable decrementer interrupt */
761 mtspr(SPRN_TCR, TCR_DIE);
762 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
765 void __init generic_calibrate_decr(void)
767 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
769 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
770 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
772 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
776 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
778 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
779 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
781 printk(KERN_ERR "WARNING: Estimating processor frequency "
786 int update_persistent_clock(struct timespec now)
790 if (!ppc_md.set_rtc_time)
793 to_tm(now.tv_sec + 1 + timezone_offset, &tm);
797 return ppc_md.set_rtc_time(&tm);
800 static void __read_persistent_clock(struct timespec *ts)
803 static int first = 1;
806 /* XXX this is a litle fragile but will work okay in the short term */
809 if (ppc_md.time_init)
810 timezone_offset = ppc_md.time_init();
812 /* get_boot_time() isn't guaranteed to be safe to call late */
813 if (ppc_md.get_boot_time) {
814 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
818 if (!ppc_md.get_rtc_time) {
822 ppc_md.get_rtc_time(&tm);
824 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
825 tm.tm_hour, tm.tm_min, tm.tm_sec);
828 void read_persistent_clock(struct timespec *ts)
830 __read_persistent_clock(ts);
832 /* Sanitize it in case real time clock is set below EPOCH */
833 if (ts->tv_sec < 0) {
840 /* clocksource code */
841 static u64 rtc_read(struct clocksource *cs)
843 return (u64)get_rtc();
846 static u64 timebase_read(struct clocksource *cs)
848 return (u64)get_tb();
851 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
852 struct clocksource *clock, u32 mult, u64 cycle_last)
854 u64 new_tb_to_xs, new_stamp_xsec;
857 if (clock != &clocksource_timebase)
860 /* Make userspace gettimeofday spin until we're done. */
861 ++vdso_data->tb_update_count;
864 /* 19342813113834067 ~= 2^(20+64) / 1e9 */
865 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
866 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
867 do_div(new_stamp_xsec, 1000000000);
868 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
870 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
871 /* this is tv_nsec / 1e9 as a 0.32 fraction */
872 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
875 * tb_update_count is used to allow the userspace gettimeofday code
876 * to assure itself that it sees a consistent view of the tb_to_xs and
877 * stamp_xsec variables. It reads the tb_update_count, then reads
878 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
879 * the two values of tb_update_count match and are even then the
880 * tb_to_xs and stamp_xsec values are consistent. If not, then it
881 * loops back and reads them again until this criteria is met.
882 * We expect the caller to have done the first increment of
883 * vdso_data->tb_update_count already.
885 vdso_data->tb_orig_stamp = cycle_last;
886 vdso_data->stamp_xsec = new_stamp_xsec;
887 vdso_data->tb_to_xs = new_tb_to_xs;
888 vdso_data->wtom_clock_sec = wtm->tv_sec;
889 vdso_data->wtom_clock_nsec = wtm->tv_nsec;
890 vdso_data->stamp_xtime = *wall_time;
891 vdso_data->stamp_sec_fraction = frac_sec;
893 ++(vdso_data->tb_update_count);
896 void update_vsyscall_tz(void)
898 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
899 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
902 static void __init clocksource_init(void)
904 struct clocksource *clock;
907 clock = &clocksource_rtc;
909 clock = &clocksource_timebase;
911 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
912 printk(KERN_ERR "clocksource: %s is already registered\n",
917 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
918 clock->name, clock->mult, clock->shift);
921 static int decrementer_set_next_event(unsigned long evt,
922 struct clock_event_device *dev)
924 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
927 /* We may have raced with new irq work */
928 if (test_irq_work_pending())
934 static int decrementer_shutdown(struct clock_event_device *dev)
936 decrementer_set_next_event(decrementer_max, dev);
940 /* Interrupt handler for the timer broadcast IPI */
941 void tick_broadcast_ipi_handler(void)
943 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
945 *next_tb = get_tb_or_rtc();
949 static void register_decrementer_clockevent(int cpu)
951 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
953 *dec = decrementer_clockevent;
954 dec->cpumask = cpumask_of(cpu);
956 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
957 dec->name, dec->mult, dec->shift, cpu);
959 clockevents_register_device(dec);
962 static void enable_large_decrementer(void)
964 if (!cpu_has_feature(CPU_FTR_ARCH_300))
967 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
971 * If we're running as the hypervisor we need to enable the LD manually
972 * otherwise firmware should have done it for us.
974 if (cpu_has_feature(CPU_FTR_HVMODE))
975 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
978 static void __init set_decrementer_max(void)
980 struct device_node *cpu;
983 /* Prior to ISAv3 the decrementer is always 32 bit */
984 if (!cpu_has_feature(CPU_FTR_ARCH_300))
987 cpu = of_find_node_by_type(NULL, "cpu");
989 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
990 if (bits > 64 || bits < 32) {
991 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
995 /* calculate the signed maximum given this many bits */
996 decrementer_max = (1ul << (bits - 1)) - 1;
1001 pr_info("time_init: %u bit decrementer (max: %llx)\n",
1002 bits, decrementer_max);
1005 static void __init init_decrementer_clockevent(void)
1007 int cpu = smp_processor_id();
1009 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
1011 decrementer_clockevent.max_delta_ns =
1012 clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
1013 decrementer_clockevent.min_delta_ns =
1014 clockevent_delta2ns(2, &decrementer_clockevent);
1016 register_decrementer_clockevent(cpu);
1019 void secondary_cpu_time_init(void)
1021 /* Enable and test the large decrementer for this cpu */
1022 enable_large_decrementer();
1024 /* Start the decrementer on CPUs that have manual control
1027 start_cpu_decrementer();
1029 /* FIME: Should make unrelatred change to move snapshot_timebase
1031 register_decrementer_clockevent(smp_processor_id());
1034 /* This function is only called on the boot processor */
1035 void __init time_init(void)
1037 struct div_result res;
1042 /* 601 processor: dec counts down by 128 every 128ns */
1043 ppc_tb_freq = 1000000000;
1045 /* Normal PowerPC with timebase register */
1046 ppc_md.calibrate_decr();
1047 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1048 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1049 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1050 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1053 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1054 tb_ticks_per_sec = ppc_tb_freq;
1055 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1056 calc_cputime_factors();
1057 setup_cputime_one_jiffy();
1060 * Compute scale factor for sched_clock.
1061 * The calibrate_decr() function has set tb_ticks_per_sec,
1062 * which is the timebase frequency.
1063 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1064 * the 128-bit result as a 64.64 fixed-point number.
1065 * We then shift that number right until it is less than 1.0,
1066 * giving us the scale factor and shift count to use in
1069 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1070 scale = res.result_low;
1071 for (shift = 0; res.result_high != 0; ++shift) {
1072 scale = (scale >> 1) | (res.result_high << 63);
1073 res.result_high >>= 1;
1075 tb_to_ns_scale = scale;
1076 tb_to_ns_shift = shift;
1077 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1078 boot_tb = get_tb_or_rtc();
1080 /* If platform provided a timezone (pmac), we correct the time */
1081 if (timezone_offset) {
1082 sys_tz.tz_minuteswest = -timezone_offset / 60;
1083 sys_tz.tz_dsttime = 0;
1086 vdso_data->tb_update_count = 0;
1087 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1089 /* initialise and enable the large decrementer (if we have one) */
1090 set_decrementer_max();
1091 enable_large_decrementer();
1093 /* Start the decrementer on CPUs that have manual control
1096 start_cpu_decrementer();
1098 /* Register the clocksource */
1101 init_decrementer_clockevent();
1102 tick_setup_hrtimer_broadcast();
1104 #ifdef CONFIG_COMMON_CLK
1111 #define STARTOFTIME 1970
1112 #define SECDAY 86400L
1113 #define SECYR (SECDAY * 365)
1114 #define leapyear(year) ((year) % 4 == 0 && \
1115 ((year) % 100 != 0 || (year) % 400 == 0))
1116 #define days_in_year(a) (leapyear(a) ? 366 : 365)
1117 #define days_in_month(a) (month_days[(a) - 1])
1119 static int month_days[12] = {
1120 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1123 void to_tm(int tim, struct rtc_time * tm)
1126 register long hms, day;
1131 /* Hours, minutes, seconds are easy */
1132 tm->tm_hour = hms / 3600;
1133 tm->tm_min = (hms % 3600) / 60;
1134 tm->tm_sec = (hms % 3600) % 60;
1136 /* Number of years in days */
1137 for (i = STARTOFTIME; day >= days_in_year(i); i++)
1138 day -= days_in_year(i);
1141 /* Number of months in days left */
1142 if (leapyear(tm->tm_year))
1143 days_in_month(FEBRUARY) = 29;
1144 for (i = 1; day >= days_in_month(i); i++)
1145 day -= days_in_month(i);
1146 days_in_month(FEBRUARY) = 28;
1149 /* Days are what is left over (+1) from all that. */
1150 tm->tm_mday = day + 1;
1153 * No-one uses the day of the week.
1157 EXPORT_SYMBOL(to_tm);
1160 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1163 void div128_by_32(u64 dividend_high, u64 dividend_low,
1164 unsigned divisor, struct div_result *dr)
1166 unsigned long a, b, c, d;
1167 unsigned long w, x, y, z;
1170 a = dividend_high >> 32;
1171 b = dividend_high & 0xffffffff;
1172 c = dividend_low >> 32;
1173 d = dividend_low & 0xffffffff;
1176 ra = ((u64)(a - (w * divisor)) << 32) + b;
1178 rb = ((u64) do_div(ra, divisor) << 32) + c;
1181 rc = ((u64) do_div(rb, divisor) << 32) + d;
1184 do_div(rc, divisor);
1187 dr->result_high = ((u64)w << 32) + x;
1188 dr->result_low = ((u64)y << 32) + z;
1192 /* We don't need to calibrate delay, we use the CPU timebase for that */
1193 void calibrate_delay(void)
1195 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1196 * as the number of __delay(1) in a jiffy, so make it so
1198 loops_per_jiffy = tb_ticks_per_jiffy;
1201 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1202 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1204 ppc_md.get_rtc_time(tm);
1205 return rtc_valid_tm(tm);
1208 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1210 if (!ppc_md.set_rtc_time)
1213 if (ppc_md.set_rtc_time(tm) < 0)
1219 static const struct rtc_class_ops rtc_generic_ops = {
1220 .read_time = rtc_generic_get_time,
1221 .set_time = rtc_generic_set_time,
1224 static int __init rtc_init(void)
1226 struct platform_device *pdev;
1228 if (!ppc_md.get_rtc_time)
1231 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1233 sizeof(rtc_generic_ops));
1235 return PTR_ERR_OR_ZERO(pdev);
1238 device_initcall(rtc_init);