1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Common time routines among all ppc machines.
6 * Paul Mackerras' version and mine for PReP and Pmac.
11 * to make clock more stable (2.4.0-test5). The only thing
12 * that this code assumes is that the timebases have been synchronized
13 * by firmware on SMP and are never stopped (never do sleep
14 * on SMP then, nap and doze are OK).
16 * Speeded up do_gettimeofday by getting rid of references to
19 * TODO (not necessarily in this file):
20 * - improve precision and reproducibility of timebase frequency
21 * measurement at boot time.
22 * - for astronomical applications: add a new function to get
23 * non ambiguous timestamps even around leap seconds. This needs
24 * a new timestamp format and a good name.
26 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
27 * "A Kernel Model for Precision Timekeeping" by Dave Mills
30 #include <linux/errno.h>
31 #include <linux/export.h>
32 #include <linux/sched.h>
33 #include <linux/sched/clock.h>
34 #include <linux/kernel.h>
35 #include <linux/param.h>
36 #include <linux/string.h>
38 #include <linux/interrupt.h>
39 #include <linux/timex.h>
40 #include <linux/kernel_stat.h>
41 #include <linux/time.h>
42 #include <linux/init.h>
43 #include <linux/profile.h>
44 #include <linux/cpu.h>
45 #include <linux/security.h>
46 #include <linux/percpu.h>
47 #include <linux/rtc.h>
48 #include <linux/jiffies.h>
49 #include <linux/posix-timers.h>
50 #include <linux/irq.h>
51 #include <linux/delay.h>
52 #include <linux/irq_work.h>
53 #include <linux/of_clk.h>
54 #include <linux/suspend.h>
55 #include <linux/sched/cputime.h>
56 #include <linux/processor.h>
57 #include <asm/trace.h>
60 #include <asm/nvram.h>
61 #include <asm/cache.h>
62 #include <asm/machdep.h>
63 #include <linux/uaccess.h>
67 #include <asm/div64.h>
69 #include <asm/vdso_datapage.h>
70 #include <asm/firmware.h>
71 #include <asm/asm-prototypes.h>
73 /* powerpc clocksource/clockevent code */
75 #include <linux/clockchips.h>
76 #include <linux/timekeeper_internal.h>
78 static u64 timebase_read(struct clocksource *);
79 static struct clocksource clocksource_timebase = {
82 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
83 .mask = CLOCKSOURCE_MASK(64),
84 .read = timebase_read,
85 .vdso_clock_mode = VDSO_CLOCKMODE_ARCHTIMER,
88 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
89 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
91 static int decrementer_set_next_event(unsigned long evt,
92 struct clock_event_device *dev);
93 static int decrementer_shutdown(struct clock_event_device *evt);
95 struct clock_event_device decrementer_clockevent = {
96 .name = "decrementer",
99 .set_next_event = decrementer_set_next_event,
100 .set_state_oneshot_stopped = decrementer_shutdown,
101 .set_state_shutdown = decrementer_shutdown,
102 .tick_resume = decrementer_shutdown,
103 .features = CLOCK_EVT_FEAT_ONESHOT |
104 CLOCK_EVT_FEAT_C3STOP,
106 EXPORT_SYMBOL(decrementer_clockevent);
108 DEFINE_PER_CPU(u64, decrementers_next_tb);
109 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
111 #define XSEC_PER_SEC (1024*1024)
114 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
116 /* compute ((xsec << 12) * max) >> 32 */
117 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
120 unsigned long tb_ticks_per_jiffy;
121 unsigned long tb_ticks_per_usec = 100; /* sane default */
122 EXPORT_SYMBOL(tb_ticks_per_usec);
123 unsigned long tb_ticks_per_sec;
124 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
126 DEFINE_SPINLOCK(rtc_lock);
127 EXPORT_SYMBOL_GPL(rtc_lock);
129 static u64 tb_to_ns_scale __read_mostly;
130 static unsigned tb_to_ns_shift __read_mostly;
131 static u64 boot_tb __read_mostly;
133 extern struct timezone sys_tz;
134 static long timezone_offset;
136 unsigned long ppc_proc_freq;
137 EXPORT_SYMBOL_GPL(ppc_proc_freq);
138 unsigned long ppc_tb_freq;
139 EXPORT_SYMBOL_GPL(ppc_tb_freq);
143 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
145 * Factor for converting from cputime_t (timebase ticks) to
146 * microseconds. This is stored as 0.64 fixed-point binary fraction.
148 u64 __cputime_usec_factor;
149 EXPORT_SYMBOL(__cputime_usec_factor);
151 #ifdef CONFIG_PPC_SPLPAR
152 void (*dtl_consumer)(struct dtl_entry *, u64);
155 static void calc_cputime_factors(void)
157 struct div_result res;
159 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
160 __cputime_usec_factor = res.result_low;
164 * Read the SPURR on systems that have it, otherwise the PURR,
165 * or if that doesn't exist return the timebase value passed in.
167 static inline unsigned long read_spurr(unsigned long tb)
169 if (cpu_has_feature(CPU_FTR_SPURR))
170 return mfspr(SPRN_SPURR);
171 if (cpu_has_feature(CPU_FTR_PURR))
172 return mfspr(SPRN_PURR);
176 #ifdef CONFIG_PPC_SPLPAR
181 * Scan the dispatch trace log and count up the stolen time.
182 * Should be called with interrupts disabled.
184 static u64 scan_dispatch_log(u64 stop_tb)
186 u64 i = local_paca->dtl_ridx;
187 struct dtl_entry *dtl = local_paca->dtl_curr;
188 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
189 struct lppaca *vpa = local_paca->lppaca_ptr;
197 if (i == be64_to_cpu(vpa->dtl_idx))
199 while (i < be64_to_cpu(vpa->dtl_idx)) {
200 dtb = be64_to_cpu(dtl->timebase);
201 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
202 be32_to_cpu(dtl->ready_to_enqueue_time);
204 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
205 /* buffer has overflowed */
206 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
207 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
213 dtl_consumer(dtl, i);
218 dtl = local_paca->dispatch_log;
220 local_paca->dtl_ridx = i;
221 local_paca->dtl_curr = dtl;
226 * Accumulate stolen time by scanning the dispatch trace log.
227 * Called on entry from user mode.
229 void notrace accumulate_stolen_time(void)
232 unsigned long save_irq_soft_mask = irq_soft_mask_return();
233 struct cpu_accounting_data *acct = &local_paca->accounting;
235 /* We are called early in the exception entry, before
236 * soft/hard_enabled are sync'ed to the expected state
237 * for the exception. We are hard disabled but the PACA
238 * needs to reflect that so various debug stuff doesn't
241 irq_soft_mask_set(IRQS_DISABLED);
243 sst = scan_dispatch_log(acct->starttime_user);
244 ust = scan_dispatch_log(acct->starttime);
247 acct->steal_time += ust + sst;
249 irq_soft_mask_set(save_irq_soft_mask);
252 static inline u64 calculate_stolen_time(u64 stop_tb)
254 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
257 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
258 return scan_dispatch_log(stop_tb);
263 #else /* CONFIG_PPC_SPLPAR */
264 static inline u64 calculate_stolen_time(u64 stop_tb)
269 #endif /* CONFIG_PPC_SPLPAR */
272 * Account time for a transition between system, hard irq
275 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
276 unsigned long now, unsigned long stime)
278 unsigned long stime_scaled = 0;
279 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
280 unsigned long nowscaled, deltascaled;
281 unsigned long utime, utime_scaled;
283 nowscaled = read_spurr(now);
284 deltascaled = nowscaled - acct->startspurr;
285 acct->startspurr = nowscaled;
286 utime = acct->utime - acct->utime_sspurr;
287 acct->utime_sspurr = acct->utime;
290 * Because we don't read the SPURR on every kernel entry/exit,
291 * deltascaled includes both user and system SPURR ticks.
292 * Apportion these ticks to system SPURR ticks and user
293 * SPURR ticks in the same ratio as the system time (delta)
294 * and user time (udelta) values obtained from the timebase
295 * over the same interval. The system ticks get accounted here;
296 * the user ticks get saved up in paca->user_time_scaled to be
297 * used by account_process_tick.
299 stime_scaled = stime;
300 utime_scaled = utime;
301 if (deltascaled != stime + utime) {
303 stime_scaled = deltascaled * stime / (stime + utime);
304 utime_scaled = deltascaled - stime_scaled;
306 stime_scaled = deltascaled;
309 acct->utime_scaled += utime_scaled;
315 static unsigned long vtime_delta(struct cpu_accounting_data *acct,
316 unsigned long *stime_scaled,
317 unsigned long *steal_time)
319 unsigned long now, stime;
321 WARN_ON_ONCE(!irqs_disabled());
324 stime = now - acct->starttime;
325 acct->starttime = now;
327 *stime_scaled = vtime_delta_scaled(acct, now, stime);
329 *steal_time = calculate_stolen_time(now);
334 static void vtime_delta_kernel(struct cpu_accounting_data *acct,
335 unsigned long *stime, unsigned long *stime_scaled)
337 unsigned long steal_time;
339 *stime = vtime_delta(acct, stime_scaled, &steal_time);
340 *stime -= min(*stime, steal_time);
341 acct->steal_time += steal_time;
344 void vtime_account_kernel(struct task_struct *tsk)
346 struct cpu_accounting_data *acct = get_accounting(tsk);
347 unsigned long stime, stime_scaled;
349 vtime_delta_kernel(acct, &stime, &stime_scaled);
351 if (tsk->flags & PF_VCPU) {
352 acct->gtime += stime;
353 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
354 acct->utime_scaled += stime_scaled;
357 acct->stime += stime;
358 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
359 acct->stime_scaled += stime_scaled;
363 EXPORT_SYMBOL_GPL(vtime_account_kernel);
365 void vtime_account_idle(struct task_struct *tsk)
367 unsigned long stime, stime_scaled, steal_time;
368 struct cpu_accounting_data *acct = get_accounting(tsk);
370 stime = vtime_delta(acct, &stime_scaled, &steal_time);
371 acct->idle_time += stime + steal_time;
374 static void vtime_account_irq_field(struct cpu_accounting_data *acct,
375 unsigned long *field)
377 unsigned long stime, stime_scaled;
379 vtime_delta_kernel(acct, &stime, &stime_scaled);
381 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
382 acct->stime_scaled += stime_scaled;
386 void vtime_account_softirq(struct task_struct *tsk)
388 struct cpu_accounting_data *acct = get_accounting(tsk);
389 vtime_account_irq_field(acct, &acct->softirq_time);
392 void vtime_account_hardirq(struct task_struct *tsk)
394 struct cpu_accounting_data *acct = get_accounting(tsk);
395 vtime_account_irq_field(acct, &acct->hardirq_time);
398 static void vtime_flush_scaled(struct task_struct *tsk,
399 struct cpu_accounting_data *acct)
401 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
402 if (acct->utime_scaled)
403 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
404 if (acct->stime_scaled)
405 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
407 acct->utime_scaled = 0;
408 acct->utime_sspurr = 0;
409 acct->stime_scaled = 0;
414 * Account the whole cputime accumulated in the paca
415 * Must be called with interrupts disabled.
416 * Assumes that vtime_account_kernel/idle() has been called
417 * recently (i.e. since the last entry from usermode) so that
418 * get_paca()->user_time_scaled is up to date.
420 void vtime_flush(struct task_struct *tsk)
422 struct cpu_accounting_data *acct = get_accounting(tsk);
425 account_user_time(tsk, cputime_to_nsecs(acct->utime));
428 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
430 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
431 account_steal_time(cputime_to_nsecs(acct->steal_time));
432 acct->steal_time = 0;
436 account_idle_time(cputime_to_nsecs(acct->idle_time));
439 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
442 if (acct->hardirq_time)
443 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
445 if (acct->softirq_time)
446 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
449 vtime_flush_scaled(tsk, acct);
455 acct->hardirq_time = 0;
456 acct->softirq_time = 0;
459 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
460 #define calc_cputime_factors()
463 void __delay(unsigned long loops)
470 * TB is in error state and isn't ticking anymore.
471 * HMI handler was unable to recover from TB error.
472 * Return immediately, so that kernel won't get stuck here.
477 while (mftb() - start < loops)
482 EXPORT_SYMBOL(__delay);
484 void udelay(unsigned long usecs)
486 __delay(tb_ticks_per_usec * usecs);
488 EXPORT_SYMBOL(udelay);
491 unsigned long profile_pc(struct pt_regs *regs)
493 unsigned long pc = instruction_pointer(regs);
495 if (in_lock_functions(pc))
500 EXPORT_SYMBOL(profile_pc);
503 #ifdef CONFIG_IRQ_WORK
506 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
509 static inline unsigned long test_irq_work_pending(void)
513 asm volatile("lbz %0,%1(13)"
515 : "i" (offsetof(struct paca_struct, irq_work_pending)));
519 static inline void set_irq_work_pending_flag(void)
521 asm volatile("stb %0,%1(13)" : :
523 "i" (offsetof(struct paca_struct, irq_work_pending)));
526 static inline void clear_irq_work_pending(void)
528 asm volatile("stb %0,%1(13)" : :
530 "i" (offsetof(struct paca_struct, irq_work_pending)));
535 DEFINE_PER_CPU(u8, irq_work_pending);
537 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
538 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
539 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
541 #endif /* 32 vs 64 bit */
543 void arch_irq_work_raise(void)
546 * 64-bit code that uses irq soft-mask can just cause an immediate
547 * interrupt here that gets soft masked, if this is called under
548 * local_irq_disable(). It might be possible to prevent that happening
549 * by noticing interrupts are disabled and setting decrementer pending
550 * to be replayed when irqs are enabled. The problem there is that
551 * tracing can call irq_work_raise, including in code that does low
552 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
553 * which could get tangled up if we're messing with the same state
557 set_irq_work_pending_flag();
562 #else /* CONFIG_IRQ_WORK */
564 #define test_irq_work_pending() 0
565 #define clear_irq_work_pending()
567 #endif /* CONFIG_IRQ_WORK */
570 * timer_interrupt - gets called when the decrementer overflows,
571 * with interrupts disabled.
573 void timer_interrupt(struct pt_regs *regs)
575 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
576 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
577 struct pt_regs *old_regs;
581 * Some implementations of hotplug will get timer interrupts while
582 * offline, just ignore these.
584 if (unlikely(!cpu_online(smp_processor_id()))) {
585 set_dec(decrementer_max);
589 /* Ensure a positive value is written to the decrementer, or else
590 * some CPUs will continue to take decrementer exceptions. When the
591 * PPC_WATCHDOG (decrementer based) is configured, keep this at most
592 * 31 bits, which is about 4 seconds on most systems, which gives
593 * the watchdog a chance of catching timer interrupt hard lockups.
595 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
598 set_dec(decrementer_max);
600 /* Conditionally hard-enable interrupts now that the DEC has been
601 * bumped to its maximum value
603 may_hard_irq_enable();
606 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
607 if (atomic_read(&ppc_n_lost_interrupts) != 0)
611 old_regs = set_irq_regs(regs);
613 trace_timer_interrupt_entry(regs);
615 if (test_irq_work_pending()) {
616 clear_irq_work_pending();
621 if (now >= *next_tb) {
623 if (evt->event_handler)
624 evt->event_handler(evt);
625 __this_cpu_inc(irq_stat.timer_irqs_event);
627 now = *next_tb - now;
628 if (now <= decrementer_max)
630 /* We may have raced with new irq work */
631 if (test_irq_work_pending())
633 __this_cpu_inc(irq_stat.timer_irqs_others);
636 trace_timer_interrupt_exit(regs);
638 set_irq_regs(old_regs);
640 EXPORT_SYMBOL(timer_interrupt);
642 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
643 void timer_broadcast_interrupt(void)
645 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
648 tick_receive_broadcast();
649 __this_cpu_inc(irq_stat.broadcast_irqs_event);
653 #ifdef CONFIG_SUSPEND
654 static void generic_suspend_disable_irqs(void)
656 /* Disable the decrementer, so that it doesn't interfere
660 set_dec(decrementer_max);
662 set_dec(decrementer_max);
665 static void generic_suspend_enable_irqs(void)
670 /* Overrides the weak version in kernel/power/main.c */
671 void arch_suspend_disable_irqs(void)
673 if (ppc_md.suspend_disable_irqs)
674 ppc_md.suspend_disable_irqs();
675 generic_suspend_disable_irqs();
678 /* Overrides the weak version in kernel/power/main.c */
679 void arch_suspend_enable_irqs(void)
681 generic_suspend_enable_irqs();
682 if (ppc_md.suspend_enable_irqs)
683 ppc_md.suspend_enable_irqs();
687 unsigned long long tb_to_ns(unsigned long long ticks)
689 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
691 EXPORT_SYMBOL_GPL(tb_to_ns);
694 * Scheduler clock - returns current time in nanosec units.
696 * Note: mulhdu(a, b) (multiply high double unsigned) returns
697 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
698 * are 64-bit unsigned numbers.
700 notrace unsigned long long sched_clock(void)
702 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
706 #ifdef CONFIG_PPC_PSERIES
709 * Running clock - attempts to give a view of time passing for a virtualised
711 * Uses the VTB register if available otherwise a next best guess.
713 unsigned long long running_clock(void)
716 * Don't read the VTB as a host since KVM does not switch in host
717 * timebase into the VTB when it takes a guest off the CPU, reading the
718 * VTB would result in reading 'last switched out' guest VTB.
720 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
721 * would be unsafe to rely only on the #ifdef above.
723 if (firmware_has_feature(FW_FEATURE_LPAR) &&
724 cpu_has_feature(CPU_FTR_ARCH_207S))
725 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
728 * This is a next best approximation without a VTB.
729 * On a host which is running bare metal there should never be any stolen
730 * time and on a host which doesn't do any virtualisation TB *should* equal
731 * VTB so it makes no difference anyway.
733 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
737 static int __init get_freq(char *name, int cells, unsigned long *val)
739 struct device_node *cpu;
743 /* The cpu node should have timebase and clock frequency properties */
744 cpu = of_find_node_by_type(NULL, "cpu");
747 fp = of_get_property(cpu, name, NULL);
750 *val = of_read_ulong(fp, cells);
759 static void start_cpu_decrementer(void)
761 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
764 /* Clear any pending timer interrupts */
765 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
767 tcr = mfspr(SPRN_TCR);
769 * The watchdog may have already been enabled by u-boot. So leave
770 * TRC[WP] (Watchdog Period) alone.
772 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
773 tcr |= TCR_DIE; /* Enable decrementer */
774 mtspr(SPRN_TCR, tcr);
778 void __init generic_calibrate_decr(void)
780 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
782 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
783 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
785 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
789 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
791 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
792 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
794 printk(KERN_ERR "WARNING: Estimating processor frequency "
799 int update_persistent_clock64(struct timespec64 now)
803 if (!ppc_md.set_rtc_time)
806 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
808 return ppc_md.set_rtc_time(&tm);
811 static void __read_persistent_clock(struct timespec64 *ts)
814 static int first = 1;
817 /* XXX this is a litle fragile but will work okay in the short term */
820 if (ppc_md.time_init)
821 timezone_offset = ppc_md.time_init();
823 /* get_boot_time() isn't guaranteed to be safe to call late */
824 if (ppc_md.get_boot_time) {
825 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
829 if (!ppc_md.get_rtc_time) {
833 ppc_md.get_rtc_time(&tm);
835 ts->tv_sec = rtc_tm_to_time64(&tm);
838 void read_persistent_clock64(struct timespec64 *ts)
840 __read_persistent_clock(ts);
842 /* Sanitize it in case real time clock is set below EPOCH */
843 if (ts->tv_sec < 0) {
850 /* clocksource code */
851 static notrace u64 timebase_read(struct clocksource *cs)
853 return (u64)get_tb();
856 static void __init clocksource_init(void)
858 struct clocksource *clock = &clocksource_timebase;
860 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
861 printk(KERN_ERR "clocksource: %s is already registered\n",
866 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
867 clock->name, clock->mult, clock->shift);
870 static int decrementer_set_next_event(unsigned long evt,
871 struct clock_event_device *dev)
873 __this_cpu_write(decrementers_next_tb, get_tb() + evt);
876 /* We may have raced with new irq work */
877 if (test_irq_work_pending())
883 static int decrementer_shutdown(struct clock_event_device *dev)
885 decrementer_set_next_event(decrementer_max, dev);
889 static void register_decrementer_clockevent(int cpu)
891 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
893 *dec = decrementer_clockevent;
894 dec->cpumask = cpumask_of(cpu);
896 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
898 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
899 dec->name, dec->mult, dec->shift, cpu);
901 /* Set values for KVM, see kvm_emulate_dec() */
902 decrementer_clockevent.mult = dec->mult;
903 decrementer_clockevent.shift = dec->shift;
906 static void enable_large_decrementer(void)
908 if (!cpu_has_feature(CPU_FTR_ARCH_300))
911 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
915 * If we're running as the hypervisor we need to enable the LD manually
916 * otherwise firmware should have done it for us.
918 if (cpu_has_feature(CPU_FTR_HVMODE))
919 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
922 static void __init set_decrementer_max(void)
924 struct device_node *cpu;
927 /* Prior to ISAv3 the decrementer is always 32 bit */
928 if (!cpu_has_feature(CPU_FTR_ARCH_300))
931 cpu = of_find_node_by_type(NULL, "cpu");
933 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
934 if (bits > 64 || bits < 32) {
935 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
939 /* calculate the signed maximum given this many bits */
940 decrementer_max = (1ul << (bits - 1)) - 1;
945 pr_info("time_init: %u bit decrementer (max: %llx)\n",
946 bits, decrementer_max);
949 static void __init init_decrementer_clockevent(void)
951 register_decrementer_clockevent(smp_processor_id());
954 void secondary_cpu_time_init(void)
956 /* Enable and test the large decrementer for this cpu */
957 enable_large_decrementer();
959 /* Start the decrementer on CPUs that have manual control
962 start_cpu_decrementer();
964 /* FIME: Should make unrelatred change to move snapshot_timebase
966 register_decrementer_clockevent(smp_processor_id());
969 /* This function is only called on the boot processor */
970 void __init time_init(void)
972 struct div_result res;
976 /* Normal PowerPC with timebase register */
977 ppc_md.calibrate_decr();
978 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
979 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
980 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
981 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
983 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
984 tb_ticks_per_sec = ppc_tb_freq;
985 tb_ticks_per_usec = ppc_tb_freq / 1000000;
986 calc_cputime_factors();
989 * Compute scale factor for sched_clock.
990 * The calibrate_decr() function has set tb_ticks_per_sec,
991 * which is the timebase frequency.
992 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
993 * the 128-bit result as a 64.64 fixed-point number.
994 * We then shift that number right until it is less than 1.0,
995 * giving us the scale factor and shift count to use in
998 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
999 scale = res.result_low;
1000 for (shift = 0; res.result_high != 0; ++shift) {
1001 scale = (scale >> 1) | (res.result_high << 63);
1002 res.result_high >>= 1;
1004 tb_to_ns_scale = scale;
1005 tb_to_ns_shift = shift;
1006 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1009 /* If platform provided a timezone (pmac), we correct the time */
1010 if (timezone_offset) {
1011 sys_tz.tz_minuteswest = -timezone_offset / 60;
1012 sys_tz.tz_dsttime = 0;
1015 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1017 /* initialise and enable the large decrementer (if we have one) */
1018 set_decrementer_max();
1019 enable_large_decrementer();
1021 /* Start the decrementer on CPUs that have manual control
1024 start_cpu_decrementer();
1026 /* Register the clocksource */
1029 init_decrementer_clockevent();
1030 tick_setup_hrtimer_broadcast();
1036 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1039 void div128_by_32(u64 dividend_high, u64 dividend_low,
1040 unsigned divisor, struct div_result *dr)
1042 unsigned long a, b, c, d;
1043 unsigned long w, x, y, z;
1046 a = dividend_high >> 32;
1047 b = dividend_high & 0xffffffff;
1048 c = dividend_low >> 32;
1049 d = dividend_low & 0xffffffff;
1052 ra = ((u64)(a - (w * divisor)) << 32) + b;
1054 rb = ((u64) do_div(ra, divisor) << 32) + c;
1057 rc = ((u64) do_div(rb, divisor) << 32) + d;
1060 do_div(rc, divisor);
1063 dr->result_high = ((u64)w << 32) + x;
1064 dr->result_low = ((u64)y << 32) + z;
1068 /* We don't need to calibrate delay, we use the CPU timebase for that */
1069 void calibrate_delay(void)
1071 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1072 * as the number of __delay(1) in a jiffy, so make it so
1074 loops_per_jiffy = tb_ticks_per_jiffy;
1077 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1078 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1080 ppc_md.get_rtc_time(tm);
1084 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1086 if (!ppc_md.set_rtc_time)
1089 if (ppc_md.set_rtc_time(tm) < 0)
1095 static const struct rtc_class_ops rtc_generic_ops = {
1096 .read_time = rtc_generic_get_time,
1097 .set_time = rtc_generic_set_time,
1100 static int __init rtc_init(void)
1102 struct platform_device *pdev;
1104 if (!ppc_md.get_rtc_time)
1107 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1109 sizeof(rtc_generic_ops));
1111 return PTR_ERR_OR_ZERO(pdev);
1114 device_initcall(rtc_init);