2 * Common time routines among all ppc machines.
4 * Written by Cort Dougan (cort@cs.nmt.edu) to merge
5 * Paul Mackerras' version and mine for PReP and Pmac.
6 * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net).
7 * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com)
9 * First round of bugfixes by Gabriel Paubert (paubert@iram.es)
10 * to make clock more stable (2.4.0-test5). The only thing
11 * that this code assumes is that the timebases have been synchronized
12 * by firmware on SMP and are never stopped (never do sleep
13 * on SMP then, nap and doze are OK).
15 * Speeded up do_gettimeofday by getting rid of references to
16 * xtime (which required locks for consistency). (mikejc@us.ibm.com)
18 * TODO (not necessarily in this file):
19 * - improve precision and reproducibility of timebase frequency
20 * measurement at boot time.
21 * - for astronomical applications: add a new function to get
22 * non ambiguous timestamps even around leap seconds. This needs
23 * a new timestamp format and a good name.
25 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
26 * "A Kernel Model for Precision Timekeeping" by Dave Mills
28 * This program is free software; you can redistribute it and/or
29 * modify it under the terms of the GNU General Public License
30 * as published by the Free Software Foundation; either version
31 * 2 of the License, or (at your option) any later version.
34 #include <linux/errno.h>
35 #include <linux/export.h>
36 #include <linux/sched.h>
37 #include <linux/sched/clock.h>
38 #include <linux/kernel.h>
39 #include <linux/param.h>
40 #include <linux/string.h>
42 #include <linux/interrupt.h>
43 #include <linux/timex.h>
44 #include <linux/kernel_stat.h>
45 #include <linux/time.h>
46 #include <linux/clockchips.h>
47 #include <linux/init.h>
48 #include <linux/profile.h>
49 #include <linux/cpu.h>
50 #include <linux/security.h>
51 #include <linux/percpu.h>
52 #include <linux/rtc.h>
53 #include <linux/jiffies.h>
54 #include <linux/posix-timers.h>
55 #include <linux/irq.h>
56 #include <linux/delay.h>
57 #include <linux/irq_work.h>
58 #include <linux/clk-provider.h>
59 #include <linux/suspend.h>
60 #include <linux/rtc.h>
61 #include <linux/sched/cputime.h>
62 #include <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_shutdown = decrementer_shutdown,
115 .tick_resume = decrementer_shutdown,
116 .features = CLOCK_EVT_FEAT_ONESHOT |
117 CLOCK_EVT_FEAT_C3STOP,
119 EXPORT_SYMBOL(decrementer_clockevent);
121 DEFINE_PER_CPU(u64, decrementers_next_tb);
122 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
124 #define XSEC_PER_SEC (1024*1024)
127 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
129 /* compute ((xsec << 12) * max) >> 32 */
130 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
133 unsigned long tb_ticks_per_jiffy;
134 unsigned long tb_ticks_per_usec = 100; /* sane default */
135 EXPORT_SYMBOL(tb_ticks_per_usec);
136 unsigned long tb_ticks_per_sec;
137 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
139 DEFINE_SPINLOCK(rtc_lock);
140 EXPORT_SYMBOL_GPL(rtc_lock);
142 static u64 tb_to_ns_scale __read_mostly;
143 static unsigned tb_to_ns_shift __read_mostly;
144 static u64 boot_tb __read_mostly;
146 extern struct timezone sys_tz;
147 static long timezone_offset;
149 unsigned long ppc_proc_freq;
150 EXPORT_SYMBOL_GPL(ppc_proc_freq);
151 unsigned long ppc_tb_freq;
152 EXPORT_SYMBOL_GPL(ppc_tb_freq);
154 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
156 * Factor for converting from cputime_t (timebase ticks) to
157 * microseconds. This is stored as 0.64 fixed-point binary fraction.
159 u64 __cputime_usec_factor;
160 EXPORT_SYMBOL(__cputime_usec_factor);
162 #ifdef CONFIG_PPC_SPLPAR
163 void (*dtl_consumer)(struct dtl_entry *, u64);
166 static void calc_cputime_factors(void)
168 struct div_result res;
170 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
171 __cputime_usec_factor = res.result_low;
175 * Read the SPURR on systems that have it, otherwise the PURR,
176 * or if that doesn't exist return the timebase value passed in.
178 static unsigned long read_spurr(unsigned long tb)
180 if (cpu_has_feature(CPU_FTR_SPURR))
181 return mfspr(SPRN_SPURR);
182 if (cpu_has_feature(CPU_FTR_PURR))
183 return mfspr(SPRN_PURR);
187 #ifdef CONFIG_PPC_SPLPAR
190 * Scan the dispatch trace log and count up the stolen time.
191 * Should be called with interrupts disabled.
193 static u64 scan_dispatch_log(u64 stop_tb)
195 u64 i = local_paca->dtl_ridx;
196 struct dtl_entry *dtl = local_paca->dtl_curr;
197 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
198 struct lppaca *vpa = local_paca->lppaca_ptr;
206 if (i == be64_to_cpu(vpa->dtl_idx))
208 while (i < be64_to_cpu(vpa->dtl_idx)) {
209 dtb = be64_to_cpu(dtl->timebase);
210 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
211 be32_to_cpu(dtl->ready_to_enqueue_time);
213 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
214 /* buffer has overflowed */
215 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
216 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
222 dtl_consumer(dtl, i);
227 dtl = local_paca->dispatch_log;
229 local_paca->dtl_ridx = i;
230 local_paca->dtl_curr = dtl;
235 * Accumulate stolen time by scanning the dispatch trace log.
236 * Called on entry from user mode.
238 void notrace accumulate_stolen_time(void)
241 unsigned long save_irq_soft_mask = irq_soft_mask_return();
242 struct cpu_accounting_data *acct = &local_paca->accounting;
244 /* We are called early in the exception entry, before
245 * soft/hard_enabled are sync'ed to the expected state
246 * for the exception. We are hard disabled but the PACA
247 * needs to reflect that so various debug stuff doesn't
250 irq_soft_mask_set(IRQS_DISABLED);
252 sst = scan_dispatch_log(acct->starttime_user);
253 ust = scan_dispatch_log(acct->starttime);
256 acct->steal_time += ust + sst;
258 irq_soft_mask_set(save_irq_soft_mask);
261 static inline u64 calculate_stolen_time(u64 stop_tb)
263 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
266 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
267 return scan_dispatch_log(stop_tb);
272 #else /* CONFIG_PPC_SPLPAR */
273 static inline u64 calculate_stolen_time(u64 stop_tb)
278 #endif /* CONFIG_PPC_SPLPAR */
281 * Account time for a transition between system, hard irq
284 static unsigned long vtime_delta(struct task_struct *tsk,
285 unsigned long *stime_scaled,
286 unsigned long *steal_time)
288 unsigned long now, nowscaled, deltascaled;
290 unsigned long utime, utime_scaled;
291 struct cpu_accounting_data *acct = get_accounting(tsk);
293 WARN_ON_ONCE(!irqs_disabled());
296 nowscaled = read_spurr(now);
297 stime = now - acct->starttime;
298 acct->starttime = now;
299 deltascaled = nowscaled - acct->startspurr;
300 acct->startspurr = nowscaled;
302 *steal_time = calculate_stolen_time(now);
304 utime = acct->utime - acct->utime_sspurr;
305 acct->utime_sspurr = acct->utime;
308 * Because we don't read the SPURR on every kernel entry/exit,
309 * deltascaled includes both user and system SPURR ticks.
310 * Apportion these ticks to system SPURR ticks and user
311 * SPURR ticks in the same ratio as the system time (delta)
312 * and user time (udelta) values obtained from the timebase
313 * over the same interval. The system ticks get accounted here;
314 * the user ticks get saved up in paca->user_time_scaled to be
315 * used by account_process_tick.
317 *stime_scaled = stime;
318 utime_scaled = utime;
319 if (deltascaled != stime + utime) {
321 *stime_scaled = deltascaled * stime / (stime + utime);
322 utime_scaled = deltascaled - *stime_scaled;
324 *stime_scaled = deltascaled;
327 acct->utime_scaled += utime_scaled;
332 void vtime_account_system(struct task_struct *tsk)
334 unsigned long stime, stime_scaled, steal_time;
335 struct cpu_accounting_data *acct = get_accounting(tsk);
337 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
339 stime -= min(stime, steal_time);
340 acct->steal_time += steal_time;
342 if ((tsk->flags & PF_VCPU) && !irq_count()) {
343 acct->gtime += stime;
344 acct->utime_scaled += stime_scaled;
347 acct->hardirq_time += stime;
348 else if (in_serving_softirq())
349 acct->softirq_time += stime;
351 acct->stime += stime;
353 acct->stime_scaled += stime_scaled;
356 EXPORT_SYMBOL_GPL(vtime_account_system);
358 void vtime_account_idle(struct task_struct *tsk)
360 unsigned long stime, stime_scaled, steal_time;
361 struct cpu_accounting_data *acct = get_accounting(tsk);
363 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
364 acct->idle_time += stime + steal_time;
368 * Account the whole cputime accumulated in the paca
369 * Must be called with interrupts disabled.
370 * Assumes that vtime_account_system/idle() has been called
371 * recently (i.e. since the last entry from usermode) so that
372 * get_paca()->user_time_scaled is up to date.
374 void vtime_flush(struct task_struct *tsk)
376 struct cpu_accounting_data *acct = get_accounting(tsk);
379 account_user_time(tsk, cputime_to_nsecs(acct->utime));
381 if (acct->utime_scaled)
382 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
385 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
387 if (acct->steal_time)
388 account_steal_time(cputime_to_nsecs(acct->steal_time));
391 account_idle_time(cputime_to_nsecs(acct->idle_time));
394 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
396 if (acct->stime_scaled)
397 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
399 if (acct->hardirq_time)
400 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
402 if (acct->softirq_time)
403 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
407 acct->utime_scaled = 0;
408 acct->utime_sspurr = 0;
410 acct->steal_time = 0;
413 acct->stime_scaled = 0;
414 acct->hardirq_time = 0;
415 acct->softirq_time = 0;
418 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
419 #define calc_cputime_factors()
422 void __delay(unsigned long loops)
431 /* the RTCL register wraps at 1000000000 */
432 diff = get_rtcl() - start;
436 } while (diff < loops);
439 while (get_tbl() - start < loops)
444 EXPORT_SYMBOL(__delay);
446 void udelay(unsigned long usecs)
448 __delay(tb_ticks_per_usec * usecs);
450 EXPORT_SYMBOL(udelay);
453 unsigned long profile_pc(struct pt_regs *regs)
455 unsigned long pc = instruction_pointer(regs);
457 if (in_lock_functions(pc))
462 EXPORT_SYMBOL(profile_pc);
465 #ifdef CONFIG_IRQ_WORK
468 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
471 static inline unsigned long test_irq_work_pending(void)
475 asm volatile("lbz %0,%1(13)"
477 : "i" (offsetof(struct paca_struct, irq_work_pending)));
481 static inline void set_irq_work_pending_flag(void)
483 asm volatile("stb %0,%1(13)" : :
485 "i" (offsetof(struct paca_struct, irq_work_pending)));
488 static inline void clear_irq_work_pending(void)
490 asm volatile("stb %0,%1(13)" : :
492 "i" (offsetof(struct paca_struct, irq_work_pending)));
497 DEFINE_PER_CPU(u8, irq_work_pending);
499 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
500 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
501 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
503 #endif /* 32 vs 64 bit */
505 void arch_irq_work_raise(void)
508 * 64-bit code that uses irq soft-mask can just cause an immediate
509 * interrupt here that gets soft masked, if this is called under
510 * local_irq_disable(). It might be possible to prevent that happening
511 * by noticing interrupts are disabled and setting decrementer pending
512 * to be replayed when irqs are enabled. The problem there is that
513 * tracing can call irq_work_raise, including in code that does low
514 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
515 * which could get tangled up if we're messing with the same state
519 set_irq_work_pending_flag();
524 #else /* CONFIG_IRQ_WORK */
526 #define test_irq_work_pending() 0
527 #define clear_irq_work_pending()
529 #endif /* CONFIG_IRQ_WORK */
532 * timer_interrupt - gets called when the decrementer overflows,
533 * with interrupts disabled.
535 void timer_interrupt(struct pt_regs *regs)
537 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
538 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
539 struct pt_regs *old_regs;
542 /* Some implementations of hotplug will get timer interrupts while
543 * offline, just ignore these and we also need to set
544 * decrementers_next_tb as MAX to make sure __check_irq_replay
545 * don't replay timer interrupt when return, otherwise we'll trap
548 if (unlikely(!cpu_online(smp_processor_id()))) {
550 set_dec(decrementer_max);
554 /* Ensure a positive value is written to the decrementer, or else
555 * some CPUs will continue to take decrementer exceptions. When the
556 * PPC_WATCHDOG (decrementer based) is configured, keep this at most
557 * 31 bits, which is about 4 seconds on most systems, which gives
558 * the watchdog a chance of catching timer interrupt hard lockups.
560 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
563 set_dec(decrementer_max);
565 /* Conditionally hard-enable interrupts now that the DEC has been
566 * bumped to its maximum value
568 may_hard_irq_enable();
571 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
572 if (atomic_read(&ppc_n_lost_interrupts) != 0)
576 old_regs = set_irq_regs(regs);
578 trace_timer_interrupt_entry(regs);
580 if (test_irq_work_pending()) {
581 clear_irq_work_pending();
585 now = get_tb_or_rtc();
586 if (now >= *next_tb) {
588 if (evt->event_handler)
589 evt->event_handler(evt);
590 __this_cpu_inc(irq_stat.timer_irqs_event);
592 now = *next_tb - now;
593 if (now <= decrementer_max)
595 /* We may have raced with new irq work */
596 if (test_irq_work_pending())
598 __this_cpu_inc(irq_stat.timer_irqs_others);
601 trace_timer_interrupt_exit(regs);
603 set_irq_regs(old_regs);
605 EXPORT_SYMBOL(timer_interrupt);
607 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
608 void timer_broadcast_interrupt(void)
610 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
613 tick_receive_broadcast();
614 __this_cpu_inc(irq_stat.broadcast_irqs_event);
619 * Hypervisor decrementer interrupts shouldn't occur but are sometimes
620 * left pending on exit from a KVM guest. We don't need to do anything
621 * to clear them, as they are edge-triggered.
623 void hdec_interrupt(struct pt_regs *regs)
627 #ifdef CONFIG_SUSPEND
628 static void generic_suspend_disable_irqs(void)
630 /* Disable the decrementer, so that it doesn't interfere
634 set_dec(decrementer_max);
636 set_dec(decrementer_max);
639 static void generic_suspend_enable_irqs(void)
644 /* Overrides the weak version in kernel/power/main.c */
645 void arch_suspend_disable_irqs(void)
647 if (ppc_md.suspend_disable_irqs)
648 ppc_md.suspend_disable_irqs();
649 generic_suspend_disable_irqs();
652 /* Overrides the weak version in kernel/power/main.c */
653 void arch_suspend_enable_irqs(void)
655 generic_suspend_enable_irqs();
656 if (ppc_md.suspend_enable_irqs)
657 ppc_md.suspend_enable_irqs();
661 unsigned long long tb_to_ns(unsigned long long ticks)
663 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
665 EXPORT_SYMBOL_GPL(tb_to_ns);
668 * Scheduler clock - returns current time in nanosec units.
670 * Note: mulhdu(a, b) (multiply high double unsigned) returns
671 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
672 * are 64-bit unsigned numbers.
674 notrace unsigned long long sched_clock(void)
678 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
682 #ifdef CONFIG_PPC_PSERIES
685 * Running clock - attempts to give a view of time passing for a virtualised
687 * Uses the VTB register if available otherwise a next best guess.
689 unsigned long long running_clock(void)
692 * Don't read the VTB as a host since KVM does not switch in host
693 * timebase into the VTB when it takes a guest off the CPU, reading the
694 * VTB would result in reading 'last switched out' guest VTB.
696 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
697 * would be unsafe to rely only on the #ifdef above.
699 if (firmware_has_feature(FW_FEATURE_LPAR) &&
700 cpu_has_feature(CPU_FTR_ARCH_207S))
701 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
704 * This is a next best approximation without a VTB.
705 * On a host which is running bare metal there should never be any stolen
706 * time and on a host which doesn't do any virtualisation TB *should* equal
707 * VTB so it makes no difference anyway.
709 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
713 static int __init get_freq(char *name, int cells, unsigned long *val)
715 struct device_node *cpu;
719 /* The cpu node should have timebase and clock frequency properties */
720 cpu = of_find_node_by_type(NULL, "cpu");
723 fp = of_get_property(cpu, name, NULL);
726 *val = of_read_ulong(fp, cells);
735 static void start_cpu_decrementer(void)
737 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
740 /* Clear any pending timer interrupts */
741 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
743 tcr = mfspr(SPRN_TCR);
745 * The watchdog may have already been enabled by u-boot. So leave
746 * TRC[WP] (Watchdog Period) alone.
748 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
749 tcr |= TCR_DIE; /* Enable decrementer */
750 mtspr(SPRN_TCR, tcr);
754 void __init generic_calibrate_decr(void)
756 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
758 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
759 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
761 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
765 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
767 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
768 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
770 printk(KERN_ERR "WARNING: Estimating processor frequency "
775 int update_persistent_clock64(struct timespec64 now)
779 if (!ppc_md.set_rtc_time)
782 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
784 return ppc_md.set_rtc_time(&tm);
787 static void __read_persistent_clock(struct timespec64 *ts)
790 static int first = 1;
793 /* XXX this is a litle fragile but will work okay in the short term */
796 if (ppc_md.time_init)
797 timezone_offset = ppc_md.time_init();
799 /* get_boot_time() isn't guaranteed to be safe to call late */
800 if (ppc_md.get_boot_time) {
801 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
805 if (!ppc_md.get_rtc_time) {
809 ppc_md.get_rtc_time(&tm);
811 ts->tv_sec = rtc_tm_to_time64(&tm);
814 void read_persistent_clock64(struct timespec64 *ts)
816 __read_persistent_clock(ts);
818 /* Sanitize it in case real time clock is set below EPOCH */
819 if (ts->tv_sec < 0) {
826 /* clocksource code */
827 static notrace u64 rtc_read(struct clocksource *cs)
829 return (u64)get_rtc();
832 static notrace u64 timebase_read(struct clocksource *cs)
834 return (u64)get_tb();
838 void update_vsyscall(struct timekeeper *tk)
841 struct clocksource *clock = tk->tkr_mono.clock;
842 u32 mult = tk->tkr_mono.mult;
843 u32 shift = tk->tkr_mono.shift;
844 u64 cycle_last = tk->tkr_mono.cycle_last;
845 u64 new_tb_to_xs, new_stamp_xsec;
848 if (clock != &clocksource_timebase)
851 xt.tv_sec = tk->xtime_sec;
852 xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
854 /* Make userspace gettimeofday spin until we're done. */
855 ++vdso_data->tb_update_count;
859 * This computes ((2^20 / 1e9) * mult) >> shift as a
860 * 0.64 fixed-point fraction.
861 * The computation in the else clause below won't overflow
862 * (as long as the timebase frequency is >= 1.049 MHz)
863 * but loses precision because we lose the low bits of the constant
864 * in the shift. Note that 19342813113834067 ~= 2^(20+64) / 1e9.
865 * For a shift of 24 the error is about 0.5e-9, or about 0.5ns
866 * over a second. (Shift values are usually 22, 23 or 24.)
867 * For high frequency clocks such as the 512MHz timebase clock
868 * on POWER[6789], the mult value is small (e.g. 32768000)
869 * and so we can shift the constant by 16 initially
870 * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the
871 * remaining shifts after the multiplication, which gives a
872 * more accurate result (e.g. with mult = 32768000, shift = 24,
873 * the error is only about 1.2e-12, or 0.7ns over 10 minutes).
875 if (mult <= 62500000 && clock->shift >= 16)
876 new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16);
878 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
881 * Compute the fractional second in units of 2^-32 seconds.
882 * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift
883 * in nanoseconds, so multiplying that by 2^32 / 1e9 gives
884 * it in units of 2^-32 seconds.
885 * We assume shift <= 32 because clocks_calc_mult_shift()
886 * generates shift values in the range 0 - 32.
888 frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift);
889 do_div(frac_sec, NSEC_PER_SEC);
892 * Work out new stamp_xsec value for any legacy users of systemcfg.
893 * stamp_xsec is in units of 2^-20 seconds.
895 new_stamp_xsec = frac_sec >> 12;
896 new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC;
899 * tb_update_count is used to allow the userspace gettimeofday code
900 * to assure itself that it sees a consistent view of the tb_to_xs and
901 * stamp_xsec variables. It reads the tb_update_count, then reads
902 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
903 * the two values of tb_update_count match and are even then the
904 * tb_to_xs and stamp_xsec values are consistent. If not, then it
905 * loops back and reads them again until this criteria is met.
907 vdso_data->tb_orig_stamp = cycle_last;
908 vdso_data->stamp_xsec = new_stamp_xsec;
909 vdso_data->tb_to_xs = new_tb_to_xs;
910 vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec;
911 vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec;
912 vdso_data->stamp_xtime = xt;
913 vdso_data->stamp_sec_fraction = frac_sec;
914 vdso_data->hrtimer_res = hrtimer_resolution;
916 ++(vdso_data->tb_update_count);
919 void update_vsyscall_tz(void)
921 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
922 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
925 static void __init clocksource_init(void)
927 struct clocksource *clock;
930 clock = &clocksource_rtc;
932 clock = &clocksource_timebase;
934 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
935 printk(KERN_ERR "clocksource: %s is already registered\n",
940 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
941 clock->name, clock->mult, clock->shift);
944 static int decrementer_set_next_event(unsigned long evt,
945 struct clock_event_device *dev)
947 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
950 /* We may have raced with new irq work */
951 if (test_irq_work_pending())
957 static int decrementer_shutdown(struct clock_event_device *dev)
959 decrementer_set_next_event(decrementer_max, dev);
963 static void register_decrementer_clockevent(int cpu)
965 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
967 *dec = decrementer_clockevent;
968 dec->cpumask = cpumask_of(cpu);
970 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
972 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
973 dec->name, dec->mult, dec->shift, cpu);
975 /* Set values for KVM, see kvm_emulate_dec() */
976 decrementer_clockevent.mult = dec->mult;
977 decrementer_clockevent.shift = dec->shift;
980 static void enable_large_decrementer(void)
982 if (!cpu_has_feature(CPU_FTR_ARCH_300))
985 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
989 * If we're running as the hypervisor we need to enable the LD manually
990 * otherwise firmware should have done it for us.
992 if (cpu_has_feature(CPU_FTR_HVMODE))
993 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
996 static void __init set_decrementer_max(void)
998 struct device_node *cpu;
1001 /* Prior to ISAv3 the decrementer is always 32 bit */
1002 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1005 cpu = of_find_node_by_type(NULL, "cpu");
1007 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
1008 if (bits > 64 || bits < 32) {
1009 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
1013 /* calculate the signed maximum given this many bits */
1014 decrementer_max = (1ul << (bits - 1)) - 1;
1019 pr_info("time_init: %u bit decrementer (max: %llx)\n",
1020 bits, decrementer_max);
1023 static void __init init_decrementer_clockevent(void)
1025 register_decrementer_clockevent(smp_processor_id());
1028 void secondary_cpu_time_init(void)
1030 /* Enable and test the large decrementer for this cpu */
1031 enable_large_decrementer();
1033 /* Start the decrementer on CPUs that have manual control
1036 start_cpu_decrementer();
1038 /* FIME: Should make unrelatred change to move snapshot_timebase
1040 register_decrementer_clockevent(smp_processor_id());
1043 /* This function is only called on the boot processor */
1044 void __init time_init(void)
1046 struct div_result res;
1051 /* 601 processor: dec counts down by 128 every 128ns */
1052 ppc_tb_freq = 1000000000;
1054 /* Normal PowerPC with timebase register */
1055 ppc_md.calibrate_decr();
1056 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1057 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1058 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1059 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1062 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1063 tb_ticks_per_sec = ppc_tb_freq;
1064 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1065 calc_cputime_factors();
1068 * Compute scale factor for sched_clock.
1069 * The calibrate_decr() function has set tb_ticks_per_sec,
1070 * which is the timebase frequency.
1071 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1072 * the 128-bit result as a 64.64 fixed-point number.
1073 * We then shift that number right until it is less than 1.0,
1074 * giving us the scale factor and shift count to use in
1077 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1078 scale = res.result_low;
1079 for (shift = 0; res.result_high != 0; ++shift) {
1080 scale = (scale >> 1) | (res.result_high << 63);
1081 res.result_high >>= 1;
1083 tb_to_ns_scale = scale;
1084 tb_to_ns_shift = shift;
1085 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1086 boot_tb = get_tb_or_rtc();
1088 /* If platform provided a timezone (pmac), we correct the time */
1089 if (timezone_offset) {
1090 sys_tz.tz_minuteswest = -timezone_offset / 60;
1091 sys_tz.tz_dsttime = 0;
1094 vdso_data->tb_update_count = 0;
1095 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1097 /* initialise and enable the large decrementer (if we have one) */
1098 set_decrementer_max();
1099 enable_large_decrementer();
1101 /* Start the decrementer on CPUs that have manual control
1104 start_cpu_decrementer();
1106 /* Register the clocksource */
1109 init_decrementer_clockevent();
1110 tick_setup_hrtimer_broadcast();
1112 #ifdef CONFIG_COMMON_CLK
1118 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1121 void div128_by_32(u64 dividend_high, u64 dividend_low,
1122 unsigned divisor, struct div_result *dr)
1124 unsigned long a, b, c, d;
1125 unsigned long w, x, y, z;
1128 a = dividend_high >> 32;
1129 b = dividend_high & 0xffffffff;
1130 c = dividend_low >> 32;
1131 d = dividend_low & 0xffffffff;
1134 ra = ((u64)(a - (w * divisor)) << 32) + b;
1136 rb = ((u64) do_div(ra, divisor) << 32) + c;
1139 rc = ((u64) do_div(rb, divisor) << 32) + d;
1142 do_div(rc, divisor);
1145 dr->result_high = ((u64)w << 32) + x;
1146 dr->result_low = ((u64)y << 32) + z;
1150 /* We don't need to calibrate delay, we use the CPU timebase for that */
1151 void calibrate_delay(void)
1153 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1154 * as the number of __delay(1) in a jiffy, so make it so
1156 loops_per_jiffy = tb_ticks_per_jiffy;
1159 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1160 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1162 ppc_md.get_rtc_time(tm);
1166 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1168 if (!ppc_md.set_rtc_time)
1171 if (ppc_md.set_rtc_time(tm) < 0)
1177 static const struct rtc_class_ops rtc_generic_ops = {
1178 .read_time = rtc_generic_get_time,
1179 .set_time = rtc_generic_set_time,
1182 static int __init rtc_init(void)
1184 struct platform_device *pdev;
1186 if (!ppc_md.get_rtc_time)
1189 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1191 sizeof(rtc_generic_ops));
1193 return PTR_ERR_OR_ZERO(pdev);
1196 device_initcall(rtc_init);