2 * Common time routines among all ppc machines.
5 * Paul Mackerras' version and mine for PReP and Pmac.
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
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 <linux/processor.h>
63 #include <asm/trace.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_oneshot_stopped = decrementer_shutdown,
115 .set_state_shutdown = decrementer_shutdown,
116 .tick_resume = decrementer_shutdown,
117 .features = CLOCK_EVT_FEAT_ONESHOT |
118 CLOCK_EVT_FEAT_C3STOP,
120 EXPORT_SYMBOL(decrementer_clockevent);
122 DEFINE_PER_CPU(u64, decrementers_next_tb);
123 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
125 #define XSEC_PER_SEC (1024*1024)
128 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
130 /* compute ((xsec << 12) * max) >> 32 */
131 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
134 unsigned long tb_ticks_per_jiffy;
135 unsigned long tb_ticks_per_usec = 100; /* sane default */
136 EXPORT_SYMBOL(tb_ticks_per_usec);
137 unsigned long tb_ticks_per_sec;
138 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
140 DEFINE_SPINLOCK(rtc_lock);
141 EXPORT_SYMBOL_GPL(rtc_lock);
143 static u64 tb_to_ns_scale __read_mostly;
144 static unsigned tb_to_ns_shift __read_mostly;
145 static u64 boot_tb __read_mostly;
147 extern struct timezone sys_tz;
148 static long timezone_offset;
150 unsigned long ppc_proc_freq;
151 EXPORT_SYMBOL_GPL(ppc_proc_freq);
152 unsigned long ppc_tb_freq;
153 EXPORT_SYMBOL_GPL(ppc_tb_freq);
155 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
157 * Factor for converting from cputime_t (timebase ticks) to
158 * microseconds. This is stored as 0.64 fixed-point binary fraction.
160 u64 __cputime_usec_factor;
161 EXPORT_SYMBOL(__cputime_usec_factor);
163 #ifdef CONFIG_PPC_SPLPAR
164 void (*dtl_consumer)(struct dtl_entry *, u64);
167 static void calc_cputime_factors(void)
169 struct div_result res;
171 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
172 __cputime_usec_factor = res.result_low;
176 * Read the SPURR on systems that have it, otherwise the PURR,
177 * or if that doesn't exist return the timebase value passed in.
179 static inline unsigned long read_spurr(unsigned long tb)
181 if (cpu_has_feature(CPU_FTR_SPURR))
182 return mfspr(SPRN_SPURR);
183 if (cpu_has_feature(CPU_FTR_PURR))
184 return mfspr(SPRN_PURR);
188 #ifdef CONFIG_PPC_SPLPAR
191 * Scan the dispatch trace log and count up the stolen time.
192 * Should be called with interrupts disabled.
194 static u64 scan_dispatch_log(u64 stop_tb)
196 u64 i = local_paca->dtl_ridx;
197 struct dtl_entry *dtl = local_paca->dtl_curr;
198 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
199 struct lppaca *vpa = local_paca->lppaca_ptr;
207 if (i == be64_to_cpu(vpa->dtl_idx))
209 while (i < be64_to_cpu(vpa->dtl_idx)) {
210 dtb = be64_to_cpu(dtl->timebase);
211 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
212 be32_to_cpu(dtl->ready_to_enqueue_time);
214 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
215 /* buffer has overflowed */
216 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
217 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
223 dtl_consumer(dtl, i);
228 dtl = local_paca->dispatch_log;
230 local_paca->dtl_ridx = i;
231 local_paca->dtl_curr = dtl;
236 * Accumulate stolen time by scanning the dispatch trace log.
237 * Called on entry from user mode.
239 void accumulate_stolen_time(void)
242 unsigned long save_irq_soft_mask = irq_soft_mask_return();
243 struct cpu_accounting_data *acct = &local_paca->accounting;
245 /* We are called early in the exception entry, before
246 * soft/hard_enabled are sync'ed to the expected state
247 * for the exception. We are hard disabled but the PACA
248 * needs to reflect that so various debug stuff doesn't
251 irq_soft_mask_set(IRQS_DISABLED);
253 sst = scan_dispatch_log(acct->starttime_user);
254 ust = scan_dispatch_log(acct->starttime);
257 acct->steal_time += ust + sst;
259 irq_soft_mask_set(save_irq_soft_mask);
262 static inline u64 calculate_stolen_time(u64 stop_tb)
264 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
267 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
268 return scan_dispatch_log(stop_tb);
273 #else /* CONFIG_PPC_SPLPAR */
274 static inline u64 calculate_stolen_time(u64 stop_tb)
279 #endif /* CONFIG_PPC_SPLPAR */
282 * Account time for a transition between system, hard irq
285 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
286 unsigned long now, unsigned long stime)
288 unsigned long stime_scaled = 0;
289 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
290 unsigned long nowscaled, deltascaled;
291 unsigned long utime, utime_scaled;
293 nowscaled = read_spurr(now);
294 deltascaled = nowscaled - acct->startspurr;
295 acct->startspurr = nowscaled;
296 utime = acct->utime - acct->utime_sspurr;
297 acct->utime_sspurr = acct->utime;
300 * Because we don't read the SPURR on every kernel entry/exit,
301 * deltascaled includes both user and system SPURR ticks.
302 * Apportion these ticks to system SPURR ticks and user
303 * SPURR ticks in the same ratio as the system time (delta)
304 * and user time (udelta) values obtained from the timebase
305 * over the same interval. The system ticks get accounted here;
306 * the user ticks get saved up in paca->user_time_scaled to be
307 * used by account_process_tick.
309 stime_scaled = stime;
310 utime_scaled = utime;
311 if (deltascaled != stime + utime) {
313 stime_scaled = deltascaled * stime / (stime + utime);
314 utime_scaled = deltascaled - stime_scaled;
316 stime_scaled = deltascaled;
319 acct->utime_scaled += utime_scaled;
325 static unsigned long vtime_delta(struct task_struct *tsk,
326 unsigned long *stime_scaled,
327 unsigned long *steal_time)
329 unsigned long now, stime;
330 struct cpu_accounting_data *acct = get_accounting(tsk);
332 WARN_ON_ONCE(!irqs_disabled());
335 stime = now - acct->starttime;
336 acct->starttime = now;
338 *stime_scaled = vtime_delta_scaled(acct, now, stime);
340 *steal_time = calculate_stolen_time(now);
345 void vtime_account_system(struct task_struct *tsk)
347 unsigned long stime, stime_scaled, steal_time;
348 struct cpu_accounting_data *acct = get_accounting(tsk);
350 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
352 stime -= min(stime, steal_time);
353 acct->steal_time += steal_time;
355 if ((tsk->flags & PF_VCPU) && !irq_count()) {
356 acct->gtime += stime;
357 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
358 acct->utime_scaled += stime_scaled;
362 acct->hardirq_time += stime;
363 else if (in_serving_softirq())
364 acct->softirq_time += stime;
366 acct->stime += stime;
368 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
369 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;
384 static void vtime_flush_scaled(struct task_struct *tsk,
385 struct cpu_accounting_data *acct)
387 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
388 if (acct->utime_scaled)
389 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
390 if (acct->stime_scaled)
391 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
393 acct->utime_scaled = 0;
394 acct->utime_sspurr = 0;
395 acct->stime_scaled = 0;
400 * Account the whole cputime accumulated in the paca
401 * Must be called with interrupts disabled.
402 * Assumes that vtime_account_system/idle() has been called
403 * recently (i.e. since the last entry from usermode) so that
404 * get_paca()->user_time_scaled is up to date.
406 void vtime_flush(struct task_struct *tsk)
408 struct cpu_accounting_data *acct = get_accounting(tsk);
411 account_user_time(tsk, cputime_to_nsecs(acct->utime));
414 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
416 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
417 account_steal_time(cputime_to_nsecs(acct->steal_time));
418 acct->steal_time = 0;
422 account_idle_time(cputime_to_nsecs(acct->idle_time));
425 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
428 if (acct->hardirq_time)
429 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
431 if (acct->softirq_time)
432 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
435 vtime_flush_scaled(tsk, acct);
441 acct->hardirq_time = 0;
442 acct->softirq_time = 0;
445 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
446 #define calc_cputime_factors()
449 void __delay(unsigned long loops)
458 /* the RTCL register wraps at 1000000000 */
459 diff = get_rtcl() - start;
463 } while (diff < loops);
466 while (get_tbl() - start < loops)
471 EXPORT_SYMBOL(__delay);
473 void udelay(unsigned long usecs)
475 __delay(tb_ticks_per_usec * usecs);
477 EXPORT_SYMBOL(udelay);
480 unsigned long profile_pc(struct pt_regs *regs)
482 unsigned long pc = instruction_pointer(regs);
484 if (in_lock_functions(pc))
489 EXPORT_SYMBOL(profile_pc);
492 #ifdef CONFIG_IRQ_WORK
495 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
498 static inline unsigned long test_irq_work_pending(void)
502 asm volatile("lbz %0,%1(13)"
504 : "i" (offsetof(struct paca_struct, irq_work_pending)));
508 static inline void set_irq_work_pending_flag(void)
510 asm volatile("stb %0,%1(13)" : :
512 "i" (offsetof(struct paca_struct, irq_work_pending)));
515 static inline void clear_irq_work_pending(void)
517 asm volatile("stb %0,%1(13)" : :
519 "i" (offsetof(struct paca_struct, irq_work_pending)));
522 void arch_irq_work_raise(void)
525 set_irq_work_pending_flag();
527 * Non-nmi code running with interrupts disabled will replay
528 * irq_happened before it re-enables interrupts, so setthe
529 * decrementer there instead of causing a hardware exception
530 * which would immediately hit the masked interrupt handler
531 * and have the net effect of setting the decrementer in
534 * NMI interrupts can not check this when they return, so the
535 * decrementer hardware exception is raised, which will fire
536 * when interrupts are next enabled.
538 * BookE does not support this yet, it must audit all NMI
539 * interrupt handlers to ensure they call nmi_enter() so this
540 * check would be correct.
542 if (IS_ENABLED(CONFIG_BOOKE) || !irqs_disabled() || in_nmi()) {
546 local_paca->irq_happened |= PACA_IRQ_DEC;
553 DEFINE_PER_CPU(u8, irq_work_pending);
555 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
556 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
557 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
559 void arch_irq_work_raise(void)
562 set_irq_work_pending_flag();
567 #endif /* 32 vs 64 bit */
569 #else /* CONFIG_IRQ_WORK */
571 #define test_irq_work_pending() 0
572 #define clear_irq_work_pending()
574 #endif /* CONFIG_IRQ_WORK */
577 * timer_interrupt - gets called when the decrementer overflows,
578 * with interrupts disabled.
580 void timer_interrupt(struct pt_regs *regs)
582 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
583 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
584 struct pt_regs *old_regs;
587 /* Some implementations of hotplug will get timer interrupts while
588 * offline, just ignore these and we also need to set
589 * decrementers_next_tb as MAX to make sure __check_irq_replay
590 * don't replay timer interrupt when return, otherwise we'll trap
593 if (unlikely(!cpu_online(smp_processor_id()))) {
595 set_dec(decrementer_max);
599 /* Ensure a positive value is written to the decrementer, or else
600 * some CPUs will continue to take decrementer exceptions. When the
601 * PPC_WATCHDOG (decrementer based) is configured, keep this at most
602 * 31 bits, which is about 4 seconds on most systems, which gives
603 * the watchdog a chance of catching timer interrupt hard lockups.
605 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
608 set_dec(decrementer_max);
610 /* Conditionally hard-enable interrupts now that the DEC has been
611 * bumped to its maximum value
613 may_hard_irq_enable();
616 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
617 if (atomic_read(&ppc_n_lost_interrupts) != 0)
621 old_regs = set_irq_regs(regs);
623 trace_timer_interrupt_entry(regs);
625 if (test_irq_work_pending()) {
626 clear_irq_work_pending();
630 now = get_tb_or_rtc();
631 if (now >= *next_tb) {
633 if (evt->event_handler)
634 evt->event_handler(evt);
635 __this_cpu_inc(irq_stat.timer_irqs_event);
637 now = *next_tb - now;
638 if (now <= decrementer_max)
640 /* We may have raced with new irq work */
641 if (test_irq_work_pending())
643 __this_cpu_inc(irq_stat.timer_irqs_others);
646 trace_timer_interrupt_exit(regs);
648 set_irq_regs(old_regs);
650 EXPORT_SYMBOL(timer_interrupt);
652 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
653 void timer_broadcast_interrupt(void)
655 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
658 tick_receive_broadcast();
659 __this_cpu_inc(irq_stat.broadcast_irqs_event);
664 * Hypervisor decrementer interrupts shouldn't occur but are sometimes
665 * left pending on exit from a KVM guest. We don't need to do anything
666 * to clear them, as they are edge-triggered.
668 void hdec_interrupt(struct pt_regs *regs)
672 #ifdef CONFIG_SUSPEND
673 static void generic_suspend_disable_irqs(void)
675 /* Disable the decrementer, so that it doesn't interfere
679 set_dec(decrementer_max);
681 set_dec(decrementer_max);
684 static void generic_suspend_enable_irqs(void)
689 /* Overrides the weak version in kernel/power/main.c */
690 void arch_suspend_disable_irqs(void)
692 if (ppc_md.suspend_disable_irqs)
693 ppc_md.suspend_disable_irqs();
694 generic_suspend_disable_irqs();
697 /* Overrides the weak version in kernel/power/main.c */
698 void arch_suspend_enable_irqs(void)
700 generic_suspend_enable_irqs();
701 if (ppc_md.suspend_enable_irqs)
702 ppc_md.suspend_enable_irqs();
706 unsigned long long tb_to_ns(unsigned long long ticks)
708 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
710 EXPORT_SYMBOL_GPL(tb_to_ns);
713 * Scheduler clock - returns current time in nanosec units.
715 * Note: mulhdu(a, b) (multiply high double unsigned) returns
716 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
717 * are 64-bit unsigned numbers.
719 notrace unsigned long long sched_clock(void)
723 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
727 #ifdef CONFIG_PPC_PSERIES
730 * Running clock - attempts to give a view of time passing for a virtualised
732 * Uses the VTB register if available otherwise a next best guess.
734 unsigned long long running_clock(void)
737 * Don't read the VTB as a host since KVM does not switch in host
738 * timebase into the VTB when it takes a guest off the CPU, reading the
739 * VTB would result in reading 'last switched out' guest VTB.
741 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
742 * would be unsafe to rely only on the #ifdef above.
744 if (firmware_has_feature(FW_FEATURE_LPAR) &&
745 cpu_has_feature(CPU_FTR_ARCH_207S))
746 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
749 * This is a next best approximation without a VTB.
750 * On a host which is running bare metal there should never be any stolen
751 * time and on a host which doesn't do any virtualisation TB *should* equal
752 * VTB so it makes no difference anyway.
754 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
758 static int __init get_freq(char *name, int cells, unsigned long *val)
760 struct device_node *cpu;
764 /* The cpu node should have timebase and clock frequency properties */
765 cpu = of_find_node_by_type(NULL, "cpu");
768 fp = of_get_property(cpu, name, NULL);
771 *val = of_read_ulong(fp, cells);
780 static void start_cpu_decrementer(void)
782 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
785 /* Clear any pending timer interrupts */
786 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
788 tcr = mfspr(SPRN_TCR);
790 * The watchdog may have already been enabled by u-boot. So leave
791 * TRC[WP] (Watchdog Period) alone.
793 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
794 tcr |= TCR_DIE; /* Enable decrementer */
795 mtspr(SPRN_TCR, tcr);
799 void __init generic_calibrate_decr(void)
801 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
803 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
804 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
806 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
810 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
812 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
813 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
815 printk(KERN_ERR "WARNING: Estimating processor frequency "
820 int update_persistent_clock64(struct timespec64 now)
824 if (!ppc_md.set_rtc_time)
827 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
829 return ppc_md.set_rtc_time(&tm);
832 static void __read_persistent_clock(struct timespec64 *ts)
835 static int first = 1;
838 /* XXX this is a litle fragile but will work okay in the short term */
841 if (ppc_md.time_init)
842 timezone_offset = ppc_md.time_init();
844 /* get_boot_time() isn't guaranteed to be safe to call late */
845 if (ppc_md.get_boot_time) {
846 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
850 if (!ppc_md.get_rtc_time) {
854 ppc_md.get_rtc_time(&tm);
856 ts->tv_sec = rtc_tm_to_time64(&tm);
859 void read_persistent_clock64(struct timespec64 *ts)
861 __read_persistent_clock(ts);
863 /* Sanitize it in case real time clock is set below EPOCH */
864 if (ts->tv_sec < 0) {
871 /* clocksource code */
872 static notrace u64 rtc_read(struct clocksource *cs)
874 return (u64)get_rtc();
877 static notrace u64 timebase_read(struct clocksource *cs)
879 return (u64)get_tb();
883 void update_vsyscall(struct timekeeper *tk)
886 struct clocksource *clock = tk->tkr_mono.clock;
887 u32 mult = tk->tkr_mono.mult;
888 u32 shift = tk->tkr_mono.shift;
889 u64 cycle_last = tk->tkr_mono.cycle_last;
890 u64 new_tb_to_xs, new_stamp_xsec;
893 if (clock != &clocksource_timebase)
896 xt.tv_sec = tk->xtime_sec;
897 xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
899 /* Make userspace gettimeofday spin until we're done. */
900 ++vdso_data->tb_update_count;
904 * This computes ((2^20 / 1e9) * mult) >> shift as a
905 * 0.64 fixed-point fraction.
906 * The computation in the else clause below won't overflow
907 * (as long as the timebase frequency is >= 1.049 MHz)
908 * but loses precision because we lose the low bits of the constant
909 * in the shift. Note that 19342813113834067 ~= 2^(20+64) / 1e9.
910 * For a shift of 24 the error is about 0.5e-9, or about 0.5ns
911 * over a second. (Shift values are usually 22, 23 or 24.)
912 * For high frequency clocks such as the 512MHz timebase clock
913 * on POWER[6789], the mult value is small (e.g. 32768000)
914 * and so we can shift the constant by 16 initially
915 * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the
916 * remaining shifts after the multiplication, which gives a
917 * more accurate result (e.g. with mult = 32768000, shift = 24,
918 * the error is only about 1.2e-12, or 0.7ns over 10 minutes).
920 if (mult <= 62500000 && clock->shift >= 16)
921 new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16);
923 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
926 * Compute the fractional second in units of 2^-32 seconds.
927 * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift
928 * in nanoseconds, so multiplying that by 2^32 / 1e9 gives
929 * it in units of 2^-32 seconds.
930 * We assume shift <= 32 because clocks_calc_mult_shift()
931 * generates shift values in the range 0 - 32.
933 frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift);
934 do_div(frac_sec, NSEC_PER_SEC);
937 * Work out new stamp_xsec value for any legacy users of systemcfg.
938 * stamp_xsec is in units of 2^-20 seconds.
940 new_stamp_xsec = frac_sec >> 12;
941 new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC;
944 * tb_update_count is used to allow the userspace gettimeofday code
945 * to assure itself that it sees a consistent view of the tb_to_xs and
946 * stamp_xsec variables. It reads the tb_update_count, then reads
947 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
948 * the two values of tb_update_count match and are even then the
949 * tb_to_xs and stamp_xsec values are consistent. If not, then it
950 * loops back and reads them again until this criteria is met.
952 vdso_data->tb_orig_stamp = cycle_last;
953 vdso_data->stamp_xsec = new_stamp_xsec;
954 vdso_data->tb_to_xs = new_tb_to_xs;
955 vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec;
956 vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec;
957 vdso_data->stamp_xtime = xt;
958 vdso_data->stamp_sec_fraction = frac_sec;
960 ++(vdso_data->tb_update_count);
963 void update_vsyscall_tz(void)
965 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
966 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
969 static void __init clocksource_init(void)
971 struct clocksource *clock;
974 clock = &clocksource_rtc;
976 clock = &clocksource_timebase;
978 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
979 printk(KERN_ERR "clocksource: %s is already registered\n",
984 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
985 clock->name, clock->mult, clock->shift);
988 static int decrementer_set_next_event(unsigned long evt,
989 struct clock_event_device *dev)
991 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
994 /* We may have raced with new irq work */
995 if (test_irq_work_pending())
1001 static int decrementer_shutdown(struct clock_event_device *dev)
1003 decrementer_set_next_event(decrementer_max, dev);
1007 static void register_decrementer_clockevent(int cpu)
1009 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
1011 *dec = decrementer_clockevent;
1012 dec->cpumask = cpumask_of(cpu);
1014 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
1016 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
1017 dec->name, dec->mult, dec->shift, cpu);
1019 /* Set values for KVM, see kvm_emulate_dec() */
1020 decrementer_clockevent.mult = dec->mult;
1021 decrementer_clockevent.shift = dec->shift;
1024 static void enable_large_decrementer(void)
1026 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1029 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
1033 * If we're running as the hypervisor we need to enable the LD manually
1034 * otherwise firmware should have done it for us.
1036 if (cpu_has_feature(CPU_FTR_HVMODE))
1037 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
1040 static void __init set_decrementer_max(void)
1042 struct device_node *cpu;
1045 /* Prior to ISAv3 the decrementer is always 32 bit */
1046 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1049 cpu = of_find_node_by_type(NULL, "cpu");
1051 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
1052 if (bits > 64 || bits < 32) {
1053 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
1057 /* calculate the signed maximum given this many bits */
1058 decrementer_max = (1ul << (bits - 1)) - 1;
1063 pr_info("time_init: %u bit decrementer (max: %llx)\n",
1064 bits, decrementer_max);
1067 static void __init init_decrementer_clockevent(void)
1069 register_decrementer_clockevent(smp_processor_id());
1072 void secondary_cpu_time_init(void)
1074 /* Enable and test the large decrementer for this cpu */
1075 enable_large_decrementer();
1077 /* Start the decrementer on CPUs that have manual control
1080 start_cpu_decrementer();
1082 /* FIME: Should make unrelatred change to move snapshot_timebase
1084 register_decrementer_clockevent(smp_processor_id());
1087 /* This function is only called on the boot processor */
1088 void __init time_init(void)
1090 struct div_result res;
1095 /* 601 processor: dec counts down by 128 every 128ns */
1096 ppc_tb_freq = 1000000000;
1098 /* Normal PowerPC with timebase register */
1099 ppc_md.calibrate_decr();
1100 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1101 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1102 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1103 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1106 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1107 tb_ticks_per_sec = ppc_tb_freq;
1108 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1109 calc_cputime_factors();
1112 * Compute scale factor for sched_clock.
1113 * The calibrate_decr() function has set tb_ticks_per_sec,
1114 * which is the timebase frequency.
1115 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1116 * the 128-bit result as a 64.64 fixed-point number.
1117 * We then shift that number right until it is less than 1.0,
1118 * giving us the scale factor and shift count to use in
1121 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1122 scale = res.result_low;
1123 for (shift = 0; res.result_high != 0; ++shift) {
1124 scale = (scale >> 1) | (res.result_high << 63);
1125 res.result_high >>= 1;
1127 tb_to_ns_scale = scale;
1128 tb_to_ns_shift = shift;
1129 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1130 boot_tb = get_tb_or_rtc();
1132 /* If platform provided a timezone (pmac), we correct the time */
1133 if (timezone_offset) {
1134 sys_tz.tz_minuteswest = -timezone_offset / 60;
1135 sys_tz.tz_dsttime = 0;
1138 vdso_data->tb_update_count = 0;
1139 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1141 /* initialise and enable the large decrementer (if we have one) */
1142 set_decrementer_max();
1143 enable_large_decrementer();
1145 /* Start the decrementer on CPUs that have manual control
1148 start_cpu_decrementer();
1150 /* Register the clocksource */
1153 init_decrementer_clockevent();
1154 tick_setup_hrtimer_broadcast();
1156 #ifdef CONFIG_COMMON_CLK
1162 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1165 void div128_by_32(u64 dividend_high, u64 dividend_low,
1166 unsigned divisor, struct div_result *dr)
1168 unsigned long a, b, c, d;
1169 unsigned long w, x, y, z;
1172 a = dividend_high >> 32;
1173 b = dividend_high & 0xffffffff;
1174 c = dividend_low >> 32;
1175 d = dividend_low & 0xffffffff;
1178 ra = ((u64)(a - (w * divisor)) << 32) + b;
1180 rb = ((u64) do_div(ra, divisor) << 32) + c;
1183 rc = ((u64) do_div(rb, divisor) << 32) + d;
1186 do_div(rc, divisor);
1189 dr->result_high = ((u64)w << 32) + x;
1190 dr->result_low = ((u64)y << 32) + z;
1194 /* We don't need to calibrate delay, we use the CPU timebase for that */
1195 void calibrate_delay(void)
1197 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1198 * as the number of __delay(1) in a jiffy, so make it so
1200 loops_per_jiffy = tb_ticks_per_jiffy;
1203 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1204 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1206 ppc_md.get_rtc_time(tm);
1210 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1212 if (!ppc_md.set_rtc_time)
1215 if (ppc_md.set_rtc_time(tm) < 0)
1221 static const struct rtc_class_ops rtc_generic_ops = {
1222 .read_time = rtc_generic_get_time,
1223 .set_time = rtc_generic_set_time,
1226 static int __init rtc_init(void)
1228 struct platform_device *pdev;
1230 if (!ppc_md.get_rtc_time)
1233 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1235 sizeof(rtc_generic_ops));
1237 return PTR_ERR_OR_ZERO(pdev);
1240 device_initcall(rtc_init);