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 <asm/uaccess.h>
71 #include <asm/div64.h>
73 #include <asm/vdso_datapage.h>
74 #include <asm/firmware.h>
75 #include <asm/cputime.h>
77 /* powerpc clocksource/clockevent code */
79 #include <linux/clockchips.h>
80 #include <linux/timekeeper_internal.h>
82 static cycle_t rtc_read(struct clocksource *);
83 static struct clocksource clocksource_rtc = {
86 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
87 .mask = CLOCKSOURCE_MASK(64),
91 static cycle_t timebase_read(struct clocksource *);
92 static struct clocksource clocksource_timebase = {
95 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
96 .mask = CLOCKSOURCE_MASK(64),
97 .read = timebase_read,
100 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
101 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
103 static int decrementer_set_next_event(unsigned long evt,
104 struct clock_event_device *dev);
105 static int decrementer_shutdown(struct clock_event_device *evt);
107 struct clock_event_device decrementer_clockevent = {
108 .name = "decrementer",
111 .set_next_event = decrementer_set_next_event,
112 .set_state_shutdown = decrementer_shutdown,
113 .tick_resume = decrementer_shutdown,
114 .features = CLOCK_EVT_FEAT_ONESHOT |
115 CLOCK_EVT_FEAT_C3STOP,
117 EXPORT_SYMBOL(decrementer_clockevent);
119 DEFINE_PER_CPU(u64, decrementers_next_tb);
120 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
122 #define XSEC_PER_SEC (1024*1024)
125 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
127 /* compute ((xsec << 12) * max) >> 32 */
128 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
131 unsigned long tb_ticks_per_jiffy;
132 unsigned long tb_ticks_per_usec = 100; /* sane default */
133 EXPORT_SYMBOL(tb_ticks_per_usec);
134 unsigned long tb_ticks_per_sec;
135 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
137 DEFINE_SPINLOCK(rtc_lock);
138 EXPORT_SYMBOL_GPL(rtc_lock);
140 static u64 tb_to_ns_scale __read_mostly;
141 static unsigned tb_to_ns_shift __read_mostly;
142 static u64 boot_tb __read_mostly;
144 extern struct timezone sys_tz;
145 static long timezone_offset;
147 unsigned long ppc_proc_freq;
148 EXPORT_SYMBOL_GPL(ppc_proc_freq);
149 unsigned long ppc_tb_freq;
150 EXPORT_SYMBOL_GPL(ppc_tb_freq);
152 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
154 * Factors for converting from cputime_t (timebase ticks) to
155 * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
156 * These are all stored as 0.64 fixed-point binary fractions.
158 u64 __cputime_jiffies_factor;
159 EXPORT_SYMBOL(__cputime_jiffies_factor);
160 u64 __cputime_usec_factor;
161 EXPORT_SYMBOL(__cputime_usec_factor);
162 u64 __cputime_sec_factor;
163 EXPORT_SYMBOL(__cputime_sec_factor);
164 u64 __cputime_clockt_factor;
165 EXPORT_SYMBOL(__cputime_clockt_factor);
166 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
167 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
169 cputime_t cputime_one_jiffy;
171 #ifdef CONFIG_PPC_SPLPAR
172 void (*dtl_consumer)(struct dtl_entry *, u64);
176 #define get_accounting(tsk) (&get_paca()->accounting)
178 #define get_accounting(tsk) (&task_thread_info(tsk)->accounting)
181 static void calc_cputime_factors(void)
183 struct div_result res;
185 div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
186 __cputime_jiffies_factor = res.result_low;
187 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
188 __cputime_usec_factor = res.result_low;
189 div128_by_32(1, 0, tb_ticks_per_sec, &res);
190 __cputime_sec_factor = res.result_low;
191 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
192 __cputime_clockt_factor = res.result_low;
196 * Read the SPURR on systems that have it, otherwise the PURR,
197 * or if that doesn't exist return the timebase value passed in.
199 static unsigned long read_spurr(unsigned long tb)
201 if (cpu_has_feature(CPU_FTR_SPURR))
202 return mfspr(SPRN_SPURR);
203 if (cpu_has_feature(CPU_FTR_PURR))
204 return mfspr(SPRN_PURR);
208 #ifdef CONFIG_PPC_SPLPAR
211 * Scan the dispatch trace log and count up the stolen time.
212 * Should be called with interrupts disabled.
214 static u64 scan_dispatch_log(u64 stop_tb)
216 u64 i = local_paca->dtl_ridx;
217 struct dtl_entry *dtl = local_paca->dtl_curr;
218 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
219 struct lppaca *vpa = local_paca->lppaca_ptr;
227 if (i == be64_to_cpu(vpa->dtl_idx))
229 while (i < be64_to_cpu(vpa->dtl_idx)) {
230 dtb = be64_to_cpu(dtl->timebase);
231 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
232 be32_to_cpu(dtl->ready_to_enqueue_time);
234 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
235 /* buffer has overflowed */
236 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
237 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
243 dtl_consumer(dtl, i);
248 dtl = local_paca->dispatch_log;
250 local_paca->dtl_ridx = i;
251 local_paca->dtl_curr = dtl;
256 * Accumulate stolen time by scanning the dispatch trace log.
257 * Called on entry from user mode.
259 void accumulate_stolen_time(void)
262 u8 save_soft_enabled = local_paca->soft_enabled;
263 struct cpu_accounting_data *acct = &local_paca->accounting;
265 /* We are called early in the exception entry, before
266 * soft/hard_enabled are sync'ed to the expected state
267 * for the exception. We are hard disabled but the PACA
268 * needs to reflect that so various debug stuff doesn't
271 local_paca->soft_enabled = 0;
273 sst = scan_dispatch_log(acct->starttime_user);
274 ust = scan_dispatch_log(acct->starttime);
275 acct->system_time -= sst;
276 acct->user_time -= ust;
277 local_paca->stolen_time += ust + sst;
279 local_paca->soft_enabled = save_soft_enabled;
282 static inline u64 calculate_stolen_time(u64 stop_tb)
286 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
287 stolen = scan_dispatch_log(stop_tb);
288 get_paca()->accounting.system_time -= stolen;
291 stolen += get_paca()->stolen_time;
292 get_paca()->stolen_time = 0;
296 #else /* CONFIG_PPC_SPLPAR */
297 static inline u64 calculate_stolen_time(u64 stop_tb)
302 #endif /* CONFIG_PPC_SPLPAR */
305 * Account time for a transition between system, hard irq
308 static unsigned long vtime_delta(struct task_struct *tsk,
309 unsigned long *sys_scaled,
310 unsigned long *stolen)
312 unsigned long now, nowscaled, deltascaled;
313 unsigned long udelta, delta, user_scaled;
314 struct cpu_accounting_data *acct = get_accounting(tsk);
316 WARN_ON_ONCE(!irqs_disabled());
319 nowscaled = read_spurr(now);
320 acct->system_time += now - acct->starttime;
321 acct->starttime = now;
322 deltascaled = nowscaled - acct->startspurr;
323 acct->startspurr = nowscaled;
325 *stolen = calculate_stolen_time(now);
327 delta = acct->system_time;
328 acct->system_time = 0;
329 udelta = acct->user_time - acct->utime_sspurr;
330 acct->utime_sspurr = acct->user_time;
333 * Because we don't read the SPURR on every kernel entry/exit,
334 * deltascaled includes both user and system SPURR ticks.
335 * Apportion these ticks to system SPURR ticks and user
336 * SPURR ticks in the same ratio as the system time (delta)
337 * and user time (udelta) values obtained from the timebase
338 * over the same interval. The system ticks get accounted here;
339 * the user ticks get saved up in paca->user_time_scaled to be
340 * used by account_process_tick.
343 user_scaled = udelta;
344 if (deltascaled != delta + udelta) {
346 *sys_scaled = deltascaled * delta / (delta + udelta);
347 user_scaled = deltascaled - *sys_scaled;
349 *sys_scaled = deltascaled;
352 acct->user_time_scaled += user_scaled;
357 void vtime_account_system(struct task_struct *tsk)
359 unsigned long delta, sys_scaled, stolen;
361 delta = vtime_delta(tsk, &sys_scaled, &stolen);
362 account_system_time(tsk, 0, delta, sys_scaled);
364 account_steal_time(stolen);
366 EXPORT_SYMBOL_GPL(vtime_account_system);
368 void vtime_account_idle(struct task_struct *tsk)
370 unsigned long delta, sys_scaled, stolen;
372 delta = vtime_delta(tsk, &sys_scaled, &stolen);
373 account_idle_time(delta + stolen);
377 * Transfer the user time accumulated in the paca
378 * by the exception entry and exit code to the generic
379 * process user time records.
380 * Must be called with interrupts disabled.
381 * Assumes that vtime_account_system/idle() has been called
382 * recently (i.e. since the last entry from usermode) so that
383 * get_paca()->user_time_scaled is up to date.
385 void vtime_account_user(struct task_struct *tsk)
387 cputime_t utime, utimescaled;
388 struct cpu_accounting_data *acct = get_accounting(tsk);
390 utime = acct->user_time;
391 utimescaled = acct->user_time_scaled;
393 acct->user_time_scaled = 0;
394 acct->utime_sspurr = 0;
395 account_user_time(tsk, utime, utimescaled);
400 * Called from the context switch with interrupts disabled, to charge all
401 * accumulated times to the current process, and to prepare accounting on
404 void arch_vtime_task_switch(struct task_struct *prev)
406 struct cpu_accounting_data *acct = get_accounting(current);
408 acct->starttime = get_accounting(prev)->starttime;
409 acct->system_time = 0;
412 #endif /* CONFIG_PPC32 */
414 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
415 #define calc_cputime_factors()
418 void __delay(unsigned long loops)
426 /* the RTCL register wraps at 1000000000 */
427 diff = get_rtcl() - start;
430 } while (diff < loops);
433 while (get_tbl() - start < loops)
438 EXPORT_SYMBOL(__delay);
440 void udelay(unsigned long usecs)
442 __delay(tb_ticks_per_usec * usecs);
444 EXPORT_SYMBOL(udelay);
447 unsigned long profile_pc(struct pt_regs *regs)
449 unsigned long pc = instruction_pointer(regs);
451 if (in_lock_functions(pc))
456 EXPORT_SYMBOL(profile_pc);
459 #ifdef CONFIG_IRQ_WORK
462 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
465 static inline unsigned long test_irq_work_pending(void)
469 asm volatile("lbz %0,%1(13)"
471 : "i" (offsetof(struct paca_struct, irq_work_pending)));
475 static inline void set_irq_work_pending_flag(void)
477 asm volatile("stb %0,%1(13)" : :
479 "i" (offsetof(struct paca_struct, irq_work_pending)));
482 static inline void clear_irq_work_pending(void)
484 asm volatile("stb %0,%1(13)" : :
486 "i" (offsetof(struct paca_struct, irq_work_pending)));
491 DEFINE_PER_CPU(u8, irq_work_pending);
493 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
494 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
495 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
497 #endif /* 32 vs 64 bit */
499 void arch_irq_work_raise(void)
502 set_irq_work_pending_flag();
507 #else /* CONFIG_IRQ_WORK */
509 #define test_irq_work_pending() 0
510 #define clear_irq_work_pending()
512 #endif /* CONFIG_IRQ_WORK */
514 static void __timer_interrupt(void)
516 struct pt_regs *regs = get_irq_regs();
517 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
518 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
521 trace_timer_interrupt_entry(regs);
523 if (test_irq_work_pending()) {
524 clear_irq_work_pending();
528 now = get_tb_or_rtc();
529 if (now >= *next_tb) {
531 if (evt->event_handler)
532 evt->event_handler(evt);
533 __this_cpu_inc(irq_stat.timer_irqs_event);
535 now = *next_tb - now;
536 if (now <= decrementer_max)
538 /* We may have raced with new irq work */
539 if (test_irq_work_pending())
541 __this_cpu_inc(irq_stat.timer_irqs_others);
545 /* collect purr register values often, for accurate calculations */
546 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
547 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
548 cu->current_tb = mfspr(SPRN_PURR);
552 trace_timer_interrupt_exit(regs);
556 * timer_interrupt - gets called when the decrementer overflows,
557 * with interrupts disabled.
559 void timer_interrupt(struct pt_regs * regs)
561 struct pt_regs *old_regs;
562 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
564 /* Ensure a positive value is written to the decrementer, or else
565 * some CPUs will continue to take decrementer exceptions.
567 set_dec(decrementer_max);
569 /* Some implementations of hotplug will get timer interrupts while
570 * offline, just ignore these and we also need to set
571 * decrementers_next_tb as MAX to make sure __check_irq_replay
572 * don't replay timer interrupt when return, otherwise we'll trap
575 if (!cpu_online(smp_processor_id())) {
580 /* Conditionally hard-enable interrupts now that the DEC has been
581 * bumped to its maximum value
583 may_hard_irq_enable();
586 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
587 if (atomic_read(&ppc_n_lost_interrupts) != 0)
591 old_regs = set_irq_regs(regs);
596 set_irq_regs(old_regs);
600 * Hypervisor decrementer interrupts shouldn't occur but are sometimes
601 * left pending on exit from a KVM guest. We don't need to do anything
602 * to clear them, as they are edge-triggered.
604 void hdec_interrupt(struct pt_regs *regs)
608 #ifdef CONFIG_SUSPEND
609 static void generic_suspend_disable_irqs(void)
611 /* Disable the decrementer, so that it doesn't interfere
615 set_dec(decrementer_max);
617 set_dec(decrementer_max);
620 static void generic_suspend_enable_irqs(void)
625 /* Overrides the weak version in kernel/power/main.c */
626 void arch_suspend_disable_irqs(void)
628 if (ppc_md.suspend_disable_irqs)
629 ppc_md.suspend_disable_irqs();
630 generic_suspend_disable_irqs();
633 /* Overrides the weak version in kernel/power/main.c */
634 void arch_suspend_enable_irqs(void)
636 generic_suspend_enable_irqs();
637 if (ppc_md.suspend_enable_irqs)
638 ppc_md.suspend_enable_irqs();
642 unsigned long long tb_to_ns(unsigned long long ticks)
644 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
646 EXPORT_SYMBOL_GPL(tb_to_ns);
649 * Scheduler clock - returns current time in nanosec units.
651 * Note: mulhdu(a, b) (multiply high double unsigned) returns
652 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
653 * are 64-bit unsigned numbers.
655 unsigned long long sched_clock(void)
659 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
663 #ifdef CONFIG_PPC_PSERIES
666 * Running clock - attempts to give a view of time passing for a virtualised
668 * Uses the VTB register if available otherwise a next best guess.
670 unsigned long long running_clock(void)
673 * Don't read the VTB as a host since KVM does not switch in host
674 * timebase into the VTB when it takes a guest off the CPU, reading the
675 * VTB would result in reading 'last switched out' guest VTB.
677 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
678 * would be unsafe to rely only on the #ifdef above.
680 if (firmware_has_feature(FW_FEATURE_LPAR) &&
681 cpu_has_feature(CPU_FTR_ARCH_207S))
682 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
685 * This is a next best approximation without a VTB.
686 * On a host which is running bare metal there should never be any stolen
687 * time and on a host which doesn't do any virtualisation TB *should* equal
688 * VTB so it makes no difference anyway.
690 return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
694 static int __init get_freq(char *name, int cells, unsigned long *val)
696 struct device_node *cpu;
700 /* The cpu node should have timebase and clock frequency properties */
701 cpu = of_find_node_by_type(NULL, "cpu");
704 fp = of_get_property(cpu, name, NULL);
707 *val = of_read_ulong(fp, cells);
716 static void start_cpu_decrementer(void)
718 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
719 /* Clear any pending timer interrupts */
720 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
722 /* Enable decrementer interrupt */
723 mtspr(SPRN_TCR, TCR_DIE);
724 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
727 void __init generic_calibrate_decr(void)
729 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
731 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
732 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
734 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
738 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
740 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
741 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
743 printk(KERN_ERR "WARNING: Estimating processor frequency "
748 int update_persistent_clock(struct timespec now)
752 if (!ppc_md.set_rtc_time)
755 to_tm(now.tv_sec + 1 + timezone_offset, &tm);
759 return ppc_md.set_rtc_time(&tm);
762 static void __read_persistent_clock(struct timespec *ts)
765 static int first = 1;
768 /* XXX this is a litle fragile but will work okay in the short term */
771 if (ppc_md.time_init)
772 timezone_offset = ppc_md.time_init();
774 /* get_boot_time() isn't guaranteed to be safe to call late */
775 if (ppc_md.get_boot_time) {
776 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
780 if (!ppc_md.get_rtc_time) {
784 ppc_md.get_rtc_time(&tm);
786 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
787 tm.tm_hour, tm.tm_min, tm.tm_sec);
790 void read_persistent_clock(struct timespec *ts)
792 __read_persistent_clock(ts);
794 /* Sanitize it in case real time clock is set below EPOCH */
795 if (ts->tv_sec < 0) {
802 /* clocksource code */
803 static cycle_t rtc_read(struct clocksource *cs)
805 return (cycle_t)get_rtc();
808 static cycle_t timebase_read(struct clocksource *cs)
810 return (cycle_t)get_tb();
813 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
814 struct clocksource *clock, u32 mult, cycle_t cycle_last)
816 u64 new_tb_to_xs, new_stamp_xsec;
819 if (clock != &clocksource_timebase)
822 /* Make userspace gettimeofday spin until we're done. */
823 ++vdso_data->tb_update_count;
826 /* 19342813113834067 ~= 2^(20+64) / 1e9 */
827 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
828 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
829 do_div(new_stamp_xsec, 1000000000);
830 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
832 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
833 /* this is tv_nsec / 1e9 as a 0.32 fraction */
834 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
837 * tb_update_count is used to allow the userspace gettimeofday code
838 * to assure itself that it sees a consistent view of the tb_to_xs and
839 * stamp_xsec variables. It reads the tb_update_count, then reads
840 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
841 * the two values of tb_update_count match and are even then the
842 * tb_to_xs and stamp_xsec values are consistent. If not, then it
843 * loops back and reads them again until this criteria is met.
844 * We expect the caller to have done the first increment of
845 * vdso_data->tb_update_count already.
847 vdso_data->tb_orig_stamp = cycle_last;
848 vdso_data->stamp_xsec = new_stamp_xsec;
849 vdso_data->tb_to_xs = new_tb_to_xs;
850 vdso_data->wtom_clock_sec = wtm->tv_sec;
851 vdso_data->wtom_clock_nsec = wtm->tv_nsec;
852 vdso_data->stamp_xtime = *wall_time;
853 vdso_data->stamp_sec_fraction = frac_sec;
855 ++(vdso_data->tb_update_count);
858 void update_vsyscall_tz(void)
860 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
861 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
864 static void __init clocksource_init(void)
866 struct clocksource *clock;
869 clock = &clocksource_rtc;
871 clock = &clocksource_timebase;
873 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
874 printk(KERN_ERR "clocksource: %s is already registered\n",
879 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
880 clock->name, clock->mult, clock->shift);
883 static int decrementer_set_next_event(unsigned long evt,
884 struct clock_event_device *dev)
886 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
889 /* We may have raced with new irq work */
890 if (test_irq_work_pending())
896 static int decrementer_shutdown(struct clock_event_device *dev)
898 decrementer_set_next_event(decrementer_max, dev);
902 /* Interrupt handler for the timer broadcast IPI */
903 void tick_broadcast_ipi_handler(void)
905 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
907 *next_tb = get_tb_or_rtc();
911 static void register_decrementer_clockevent(int cpu)
913 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
915 *dec = decrementer_clockevent;
916 dec->cpumask = cpumask_of(cpu);
918 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
919 dec->name, dec->mult, dec->shift, cpu);
921 clockevents_register_device(dec);
924 static void enable_large_decrementer(void)
926 if (!cpu_has_feature(CPU_FTR_ARCH_300))
929 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
933 * If we're running as the hypervisor we need to enable the LD manually
934 * otherwise firmware should have done it for us.
936 if (cpu_has_feature(CPU_FTR_HVMODE))
937 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
940 static void __init set_decrementer_max(void)
942 struct device_node *cpu;
945 /* Prior to ISAv3 the decrementer is always 32 bit */
946 if (!cpu_has_feature(CPU_FTR_ARCH_300))
949 cpu = of_find_node_by_type(NULL, "cpu");
951 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
952 if (bits > 64 || bits < 32) {
953 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
957 /* calculate the signed maximum given this many bits */
958 decrementer_max = (1ul << (bits - 1)) - 1;
963 pr_info("time_init: %u bit decrementer (max: %llx)\n",
964 bits, decrementer_max);
967 static void __init init_decrementer_clockevent(void)
969 int cpu = smp_processor_id();
971 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
973 decrementer_clockevent.max_delta_ns =
974 clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
975 decrementer_clockevent.min_delta_ns =
976 clockevent_delta2ns(2, &decrementer_clockevent);
978 register_decrementer_clockevent(cpu);
981 void secondary_cpu_time_init(void)
983 /* Enable and test the large decrementer for this cpu */
984 enable_large_decrementer();
986 /* Start the decrementer on CPUs that have manual control
989 start_cpu_decrementer();
991 /* FIME: Should make unrelatred change to move snapshot_timebase
993 register_decrementer_clockevent(smp_processor_id());
996 /* This function is only called on the boot processor */
997 void __init time_init(void)
999 struct div_result res;
1004 /* 601 processor: dec counts down by 128 every 128ns */
1005 ppc_tb_freq = 1000000000;
1007 /* Normal PowerPC with timebase register */
1008 ppc_md.calibrate_decr();
1009 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1010 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1011 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1012 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1015 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1016 tb_ticks_per_sec = ppc_tb_freq;
1017 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1018 calc_cputime_factors();
1019 setup_cputime_one_jiffy();
1022 * Compute scale factor for sched_clock.
1023 * The calibrate_decr() function has set tb_ticks_per_sec,
1024 * which is the timebase frequency.
1025 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1026 * the 128-bit result as a 64.64 fixed-point number.
1027 * We then shift that number right until it is less than 1.0,
1028 * giving us the scale factor and shift count to use in
1031 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1032 scale = res.result_low;
1033 for (shift = 0; res.result_high != 0; ++shift) {
1034 scale = (scale >> 1) | (res.result_high << 63);
1035 res.result_high >>= 1;
1037 tb_to_ns_scale = scale;
1038 tb_to_ns_shift = shift;
1039 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1040 boot_tb = get_tb_or_rtc();
1042 /* If platform provided a timezone (pmac), we correct the time */
1043 if (timezone_offset) {
1044 sys_tz.tz_minuteswest = -timezone_offset / 60;
1045 sys_tz.tz_dsttime = 0;
1048 vdso_data->tb_update_count = 0;
1049 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1051 /* initialise and enable the large decrementer (if we have one) */
1052 set_decrementer_max();
1053 enable_large_decrementer();
1055 /* Start the decrementer on CPUs that have manual control
1058 start_cpu_decrementer();
1060 /* Register the clocksource */
1063 init_decrementer_clockevent();
1064 tick_setup_hrtimer_broadcast();
1066 #ifdef CONFIG_COMMON_CLK
1073 #define STARTOFTIME 1970
1074 #define SECDAY 86400L
1075 #define SECYR (SECDAY * 365)
1076 #define leapyear(year) ((year) % 4 == 0 && \
1077 ((year) % 100 != 0 || (year) % 400 == 0))
1078 #define days_in_year(a) (leapyear(a) ? 366 : 365)
1079 #define days_in_month(a) (month_days[(a) - 1])
1081 static int month_days[12] = {
1082 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1085 void to_tm(int tim, struct rtc_time * tm)
1088 register long hms, day;
1093 /* Hours, minutes, seconds are easy */
1094 tm->tm_hour = hms / 3600;
1095 tm->tm_min = (hms % 3600) / 60;
1096 tm->tm_sec = (hms % 3600) % 60;
1098 /* Number of years in days */
1099 for (i = STARTOFTIME; day >= days_in_year(i); i++)
1100 day -= days_in_year(i);
1103 /* Number of months in days left */
1104 if (leapyear(tm->tm_year))
1105 days_in_month(FEBRUARY) = 29;
1106 for (i = 1; day >= days_in_month(i); i++)
1107 day -= days_in_month(i);
1108 days_in_month(FEBRUARY) = 28;
1111 /* Days are what is left over (+1) from all that. */
1112 tm->tm_mday = day + 1;
1115 * No-one uses the day of the week.
1119 EXPORT_SYMBOL(to_tm);
1122 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1125 void div128_by_32(u64 dividend_high, u64 dividend_low,
1126 unsigned divisor, struct div_result *dr)
1128 unsigned long a, b, c, d;
1129 unsigned long w, x, y, z;
1132 a = dividend_high >> 32;
1133 b = dividend_high & 0xffffffff;
1134 c = dividend_low >> 32;
1135 d = dividend_low & 0xffffffff;
1138 ra = ((u64)(a - (w * divisor)) << 32) + b;
1140 rb = ((u64) do_div(ra, divisor) << 32) + c;
1143 rc = ((u64) do_div(rb, divisor) << 32) + d;
1146 do_div(rc, divisor);
1149 dr->result_high = ((u64)w << 32) + x;
1150 dr->result_low = ((u64)y << 32) + z;
1154 /* We don't need to calibrate delay, we use the CPU timebase for that */
1155 void calibrate_delay(void)
1157 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1158 * as the number of __delay(1) in a jiffy, so make it so
1160 loops_per_jiffy = tb_ticks_per_jiffy;
1163 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1164 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1166 ppc_md.get_rtc_time(tm);
1167 return rtc_valid_tm(tm);
1170 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1172 if (!ppc_md.set_rtc_time)
1175 if (ppc_md.set_rtc_time(tm) < 0)
1181 static const struct rtc_class_ops rtc_generic_ops = {
1182 .read_time = rtc_generic_get_time,
1183 .set_time = rtc_generic_set_time,
1186 static int __init rtc_init(void)
1188 struct platform_device *pdev;
1190 if (!ppc_md.get_rtc_time)
1193 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1195 sizeof(rtc_generic_ops));
1197 return PTR_ERR_OR_ZERO(pdev);
1200 device_initcall(rtc_init);