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);
274 acct->system_time -= sst;
275 acct->user_time -= ust;
276 local_paca->stolen_time += ust + sst;
278 local_paca->soft_enabled = save_soft_enabled;
281 static inline u64 calculate_stolen_time(u64 stop_tb)
285 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
286 stolen = scan_dispatch_log(stop_tb);
287 get_paca()->accounting.system_time -= stolen;
290 stolen += get_paca()->stolen_time;
291 get_paca()->stolen_time = 0;
295 #else /* CONFIG_PPC_SPLPAR */
296 static inline u64 calculate_stolen_time(u64 stop_tb)
301 #endif /* CONFIG_PPC_SPLPAR */
304 * Account time for a transition between system, hard irq
307 static unsigned long vtime_delta(struct task_struct *tsk,
308 unsigned long *sys_scaled,
309 unsigned long *stolen)
311 unsigned long now, nowscaled, deltascaled;
312 unsigned long udelta, delta, user_scaled;
313 struct cpu_accounting_data *acct = get_accounting(tsk);
315 WARN_ON_ONCE(!irqs_disabled());
318 nowscaled = read_spurr(now);
319 acct->system_time += now - acct->starttime;
320 acct->starttime = now;
321 deltascaled = nowscaled - acct->startspurr;
322 acct->startspurr = nowscaled;
324 *stolen = calculate_stolen_time(now);
326 delta = acct->system_time;
327 acct->system_time = 0;
328 udelta = acct->user_time - acct->utime_sspurr;
329 acct->utime_sspurr = acct->user_time;
332 * Because we don't read the SPURR on every kernel entry/exit,
333 * deltascaled includes both user and system SPURR ticks.
334 * Apportion these ticks to system SPURR ticks and user
335 * SPURR ticks in the same ratio as the system time (delta)
336 * and user time (udelta) values obtained from the timebase
337 * over the same interval. The system ticks get accounted here;
338 * the user ticks get saved up in paca->user_time_scaled to be
339 * used by account_process_tick.
342 user_scaled = udelta;
343 if (deltascaled != delta + udelta) {
345 *sys_scaled = deltascaled * delta / (delta + udelta);
346 user_scaled = deltascaled - *sys_scaled;
348 *sys_scaled = deltascaled;
351 acct->user_time_scaled += user_scaled;
356 void vtime_account_system(struct task_struct *tsk)
358 unsigned long delta, sys_scaled, stolen;
360 delta = vtime_delta(tsk, &sys_scaled, &stolen);
361 account_system_time(tsk, 0, delta);
362 tsk->stimescaled += 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);
396 tsk->utimescaled += utimescaled;
401 * Called from the context switch with interrupts disabled, to charge all
402 * accumulated times to the current process, and to prepare accounting on
405 void arch_vtime_task_switch(struct task_struct *prev)
407 struct cpu_accounting_data *acct = get_accounting(current);
409 acct->starttime = get_accounting(prev)->starttime;
410 acct->startspurr = get_accounting(prev)->startspurr;
411 acct->system_time = 0;
414 #endif /* CONFIG_PPC32 */
416 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
417 #define calc_cputime_factors()
420 void __delay(unsigned long loops)
428 /* the RTCL register wraps at 1000000000 */
429 diff = get_rtcl() - start;
432 } while (diff < loops);
435 while (get_tbl() - start < loops)
440 EXPORT_SYMBOL(__delay);
442 void udelay(unsigned long usecs)
444 __delay(tb_ticks_per_usec * usecs);
446 EXPORT_SYMBOL(udelay);
449 unsigned long profile_pc(struct pt_regs *regs)
451 unsigned long pc = instruction_pointer(regs);
453 if (in_lock_functions(pc))
458 EXPORT_SYMBOL(profile_pc);
461 #ifdef CONFIG_IRQ_WORK
464 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
467 static inline unsigned long test_irq_work_pending(void)
471 asm volatile("lbz %0,%1(13)"
473 : "i" (offsetof(struct paca_struct, irq_work_pending)));
477 static inline void set_irq_work_pending_flag(void)
479 asm volatile("stb %0,%1(13)" : :
481 "i" (offsetof(struct paca_struct, irq_work_pending)));
484 static inline void clear_irq_work_pending(void)
486 asm volatile("stb %0,%1(13)" : :
488 "i" (offsetof(struct paca_struct, irq_work_pending)));
493 DEFINE_PER_CPU(u8, irq_work_pending);
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)
499 #endif /* 32 vs 64 bit */
501 void arch_irq_work_raise(void)
504 set_irq_work_pending_flag();
509 #else /* CONFIG_IRQ_WORK */
511 #define test_irq_work_pending() 0
512 #define clear_irq_work_pending()
514 #endif /* CONFIG_IRQ_WORK */
516 static void __timer_interrupt(void)
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);
523 trace_timer_interrupt_entry(regs);
525 if (test_irq_work_pending()) {
526 clear_irq_work_pending();
530 now = get_tb_or_rtc();
531 if (now >= *next_tb) {
533 if (evt->event_handler)
534 evt->event_handler(evt);
535 __this_cpu_inc(irq_stat.timer_irqs_event);
537 now = *next_tb - now;
538 if (now <= decrementer_max)
540 /* We may have raced with new irq work */
541 if (test_irq_work_pending())
543 __this_cpu_inc(irq_stat.timer_irqs_others);
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);
554 trace_timer_interrupt_exit(regs);
558 * timer_interrupt - gets called when the decrementer overflows,
559 * with interrupts disabled.
561 void timer_interrupt(struct pt_regs * regs)
563 struct pt_regs *old_regs;
564 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
566 /* Ensure a positive value is written to the decrementer, or else
567 * some CPUs will continue to take decrementer exceptions.
569 set_dec(decrementer_max);
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
577 if (!cpu_online(smp_processor_id())) {
582 /* Conditionally hard-enable interrupts now that the DEC has been
583 * bumped to its maximum value
585 may_hard_irq_enable();
588 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
589 if (atomic_read(&ppc_n_lost_interrupts) != 0)
593 old_regs = set_irq_regs(regs);
598 set_irq_regs(old_regs);
600 EXPORT_SYMBOL(timer_interrupt);
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.
607 void hdec_interrupt(struct pt_regs *regs)
611 #ifdef CONFIG_SUSPEND
612 static void generic_suspend_disable_irqs(void)
614 /* Disable the decrementer, so that it doesn't interfere
618 set_dec(decrementer_max);
620 set_dec(decrementer_max);
623 static void generic_suspend_enable_irqs(void)
628 /* Overrides the weak version in kernel/power/main.c */
629 void arch_suspend_disable_irqs(void)
631 if (ppc_md.suspend_disable_irqs)
632 ppc_md.suspend_disable_irqs();
633 generic_suspend_disable_irqs();
636 /* Overrides the weak version in kernel/power/main.c */
637 void arch_suspend_enable_irqs(void)
639 generic_suspend_enable_irqs();
640 if (ppc_md.suspend_enable_irqs)
641 ppc_md.suspend_enable_irqs();
645 unsigned long long tb_to_ns(unsigned long long ticks)
647 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
649 EXPORT_SYMBOL_GPL(tb_to_ns);
652 * Scheduler clock - returns current time in nanosec units.
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.
658 unsigned long long sched_clock(void)
662 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
666 #ifdef CONFIG_PPC_PSERIES
669 * Running clock - attempts to give a view of time passing for a virtualised
671 * Uses the VTB register if available otherwise a next best guess.
673 unsigned long long running_clock(void)
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.
680 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
681 * would be unsafe to rely only on the #ifdef above.
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;
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.
693 return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
697 static int __init get_freq(char *name, int cells, unsigned long *val)
699 struct device_node *cpu;
703 /* The cpu node should have timebase and clock frequency properties */
704 cpu = of_find_node_by_type(NULL, "cpu");
707 fp = of_get_property(cpu, name, NULL);
710 *val = of_read_ulong(fp, cells);
719 static void start_cpu_decrementer(void)
721 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
722 /* Clear any pending timer interrupts */
723 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
725 /* Enable decrementer interrupt */
726 mtspr(SPRN_TCR, TCR_DIE);
727 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
730 void __init generic_calibrate_decr(void)
732 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
734 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
735 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
737 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
741 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
743 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
744 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
746 printk(KERN_ERR "WARNING: Estimating processor frequency "
751 int update_persistent_clock(struct timespec now)
755 if (!ppc_md.set_rtc_time)
758 to_tm(now.tv_sec + 1 + timezone_offset, &tm);
762 return ppc_md.set_rtc_time(&tm);
765 static void __read_persistent_clock(struct timespec *ts)
768 static int first = 1;
771 /* XXX this is a litle fragile but will work okay in the short term */
774 if (ppc_md.time_init)
775 timezone_offset = ppc_md.time_init();
777 /* get_boot_time() isn't guaranteed to be safe to call late */
778 if (ppc_md.get_boot_time) {
779 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
783 if (!ppc_md.get_rtc_time) {
787 ppc_md.get_rtc_time(&tm);
789 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
790 tm.tm_hour, tm.tm_min, tm.tm_sec);
793 void read_persistent_clock(struct timespec *ts)
795 __read_persistent_clock(ts);
797 /* Sanitize it in case real time clock is set below EPOCH */
798 if (ts->tv_sec < 0) {
805 /* clocksource code */
806 static u64 rtc_read(struct clocksource *cs)
808 return (u64)get_rtc();
811 static u64 timebase_read(struct clocksource *cs)
813 return (u64)get_tb();
816 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
817 struct clocksource *clock, u32 mult, u64 cycle_last)
819 u64 new_tb_to_xs, new_stamp_xsec;
822 if (clock != &clocksource_timebase)
825 /* Make userspace gettimeofday spin until we're done. */
826 ++vdso_data->tb_update_count;
829 /* 19342813113834067 ~= 2^(20+64) / 1e9 */
830 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
831 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
832 do_div(new_stamp_xsec, 1000000000);
833 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
835 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
836 /* this is tv_nsec / 1e9 as a 0.32 fraction */
837 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
840 * tb_update_count is used to allow the userspace gettimeofday code
841 * to assure itself that it sees a consistent view of the tb_to_xs and
842 * stamp_xsec variables. It reads the tb_update_count, then reads
843 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
844 * the two values of tb_update_count match and are even then the
845 * tb_to_xs and stamp_xsec values are consistent. If not, then it
846 * loops back and reads them again until this criteria is met.
847 * We expect the caller to have done the first increment of
848 * vdso_data->tb_update_count already.
850 vdso_data->tb_orig_stamp = cycle_last;
851 vdso_data->stamp_xsec = new_stamp_xsec;
852 vdso_data->tb_to_xs = new_tb_to_xs;
853 vdso_data->wtom_clock_sec = wtm->tv_sec;
854 vdso_data->wtom_clock_nsec = wtm->tv_nsec;
855 vdso_data->stamp_xtime = *wall_time;
856 vdso_data->stamp_sec_fraction = frac_sec;
858 ++(vdso_data->tb_update_count);
861 void update_vsyscall_tz(void)
863 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
864 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
867 static void __init clocksource_init(void)
869 struct clocksource *clock;
872 clock = &clocksource_rtc;
874 clock = &clocksource_timebase;
876 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
877 printk(KERN_ERR "clocksource: %s is already registered\n",
882 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
883 clock->name, clock->mult, clock->shift);
886 static int decrementer_set_next_event(unsigned long evt,
887 struct clock_event_device *dev)
889 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
892 /* We may have raced with new irq work */
893 if (test_irq_work_pending())
899 static int decrementer_shutdown(struct clock_event_device *dev)
901 decrementer_set_next_event(decrementer_max, dev);
905 /* Interrupt handler for the timer broadcast IPI */
906 void tick_broadcast_ipi_handler(void)
908 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
910 *next_tb = get_tb_or_rtc();
914 static void register_decrementer_clockevent(int cpu)
916 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
918 *dec = decrementer_clockevent;
919 dec->cpumask = cpumask_of(cpu);
921 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
922 dec->name, dec->mult, dec->shift, cpu);
924 clockevents_register_device(dec);
927 static void enable_large_decrementer(void)
929 if (!cpu_has_feature(CPU_FTR_ARCH_300))
932 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
936 * If we're running as the hypervisor we need to enable the LD manually
937 * otherwise firmware should have done it for us.
939 if (cpu_has_feature(CPU_FTR_HVMODE))
940 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
943 static void __init set_decrementer_max(void)
945 struct device_node *cpu;
948 /* Prior to ISAv3 the decrementer is always 32 bit */
949 if (!cpu_has_feature(CPU_FTR_ARCH_300))
952 cpu = of_find_node_by_type(NULL, "cpu");
954 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
955 if (bits > 64 || bits < 32) {
956 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
960 /* calculate the signed maximum given this many bits */
961 decrementer_max = (1ul << (bits - 1)) - 1;
966 pr_info("time_init: %u bit decrementer (max: %llx)\n",
967 bits, decrementer_max);
970 static void __init init_decrementer_clockevent(void)
972 int cpu = smp_processor_id();
974 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
976 decrementer_clockevent.max_delta_ns =
977 clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
978 decrementer_clockevent.min_delta_ns =
979 clockevent_delta2ns(2, &decrementer_clockevent);
981 register_decrementer_clockevent(cpu);
984 void secondary_cpu_time_init(void)
986 /* Enable and test the large decrementer for this cpu */
987 enable_large_decrementer();
989 /* Start the decrementer on CPUs that have manual control
992 start_cpu_decrementer();
994 /* FIME: Should make unrelatred change to move snapshot_timebase
996 register_decrementer_clockevent(smp_processor_id());
999 /* This function is only called on the boot processor */
1000 void __init time_init(void)
1002 struct div_result res;
1007 /* 601 processor: dec counts down by 128 every 128ns */
1008 ppc_tb_freq = 1000000000;
1010 /* Normal PowerPC with timebase register */
1011 ppc_md.calibrate_decr();
1012 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1013 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1014 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1015 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1018 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1019 tb_ticks_per_sec = ppc_tb_freq;
1020 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1021 calc_cputime_factors();
1022 setup_cputime_one_jiffy();
1025 * Compute scale factor for sched_clock.
1026 * The calibrate_decr() function has set tb_ticks_per_sec,
1027 * which is the timebase frequency.
1028 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1029 * the 128-bit result as a 64.64 fixed-point number.
1030 * We then shift that number right until it is less than 1.0,
1031 * giving us the scale factor and shift count to use in
1034 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1035 scale = res.result_low;
1036 for (shift = 0; res.result_high != 0; ++shift) {
1037 scale = (scale >> 1) | (res.result_high << 63);
1038 res.result_high >>= 1;
1040 tb_to_ns_scale = scale;
1041 tb_to_ns_shift = shift;
1042 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1043 boot_tb = get_tb_or_rtc();
1045 /* If platform provided a timezone (pmac), we correct the time */
1046 if (timezone_offset) {
1047 sys_tz.tz_minuteswest = -timezone_offset / 60;
1048 sys_tz.tz_dsttime = 0;
1051 vdso_data->tb_update_count = 0;
1052 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1054 /* initialise and enable the large decrementer (if we have one) */
1055 set_decrementer_max();
1056 enable_large_decrementer();
1058 /* Start the decrementer on CPUs that have manual control
1061 start_cpu_decrementer();
1063 /* Register the clocksource */
1066 init_decrementer_clockevent();
1067 tick_setup_hrtimer_broadcast();
1069 #ifdef CONFIG_COMMON_CLK
1076 #define STARTOFTIME 1970
1077 #define SECDAY 86400L
1078 #define SECYR (SECDAY * 365)
1079 #define leapyear(year) ((year) % 4 == 0 && \
1080 ((year) % 100 != 0 || (year) % 400 == 0))
1081 #define days_in_year(a) (leapyear(a) ? 366 : 365)
1082 #define days_in_month(a) (month_days[(a) - 1])
1084 static int month_days[12] = {
1085 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1088 void to_tm(int tim, struct rtc_time * tm)
1091 register long hms, day;
1096 /* Hours, minutes, seconds are easy */
1097 tm->tm_hour = hms / 3600;
1098 tm->tm_min = (hms % 3600) / 60;
1099 tm->tm_sec = (hms % 3600) % 60;
1101 /* Number of years in days */
1102 for (i = STARTOFTIME; day >= days_in_year(i); i++)
1103 day -= days_in_year(i);
1106 /* Number of months in days left */
1107 if (leapyear(tm->tm_year))
1108 days_in_month(FEBRUARY) = 29;
1109 for (i = 1; day >= days_in_month(i); i++)
1110 day -= days_in_month(i);
1111 days_in_month(FEBRUARY) = 28;
1114 /* Days are what is left over (+1) from all that. */
1115 tm->tm_mday = day + 1;
1118 * No-one uses the day of the week.
1122 EXPORT_SYMBOL(to_tm);
1125 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1128 void div128_by_32(u64 dividend_high, u64 dividend_low,
1129 unsigned divisor, struct div_result *dr)
1131 unsigned long a, b, c, d;
1132 unsigned long w, x, y, z;
1135 a = dividend_high >> 32;
1136 b = dividend_high & 0xffffffff;
1137 c = dividend_low >> 32;
1138 d = dividend_low & 0xffffffff;
1141 ra = ((u64)(a - (w * divisor)) << 32) + b;
1143 rb = ((u64) do_div(ra, divisor) << 32) + c;
1146 rc = ((u64) do_div(rb, divisor) << 32) + d;
1149 do_div(rc, divisor);
1152 dr->result_high = ((u64)w << 32) + x;
1153 dr->result_low = ((u64)y << 32) + z;
1157 /* We don't need to calibrate delay, we use the CPU timebase for that */
1158 void calibrate_delay(void)
1160 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1161 * as the number of __delay(1) in a jiffy, so make it so
1163 loops_per_jiffy = tb_ticks_per_jiffy;
1166 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1167 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1169 ppc_md.get_rtc_time(tm);
1170 return rtc_valid_tm(tm);
1173 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1175 if (!ppc_md.set_rtc_time)
1178 if (ppc_md.set_rtc_time(tm) < 0)
1184 static const struct rtc_class_ops rtc_generic_ops = {
1185 .read_time = rtc_generic_get_time,
1186 .set_time = rtc_generic_set_time,
1189 static int __init rtc_init(void)
1191 struct platform_device *pdev;
1193 if (!ppc_md.get_rtc_time)
1196 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1198 sizeof(rtc_generic_ops));
1200 return PTR_ERR_OR_ZERO(pdev);
1203 device_initcall(rtc_init);