1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Common time routines among all ppc machines.
5 * Written by Cort Dougan (cort@cs.nmt.edu) to merge
6 * Paul Mackerras' version and mine for PReP and Pmac.
7 * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net).
8 * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com)
10 * First round of bugfixes by Gabriel Paubert (paubert@iram.es)
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
17 * xtime (which required locks for consistency). (mikejc@us.ibm.com)
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/sched/clock.h>
57 #include <linux/processor.h>
58 #include <asm/trace.h>
60 #include <asm/interrupt.h>
62 #include <asm/nvram.h>
63 #include <asm/cache.h>
64 #include <asm/machdep.h>
65 #include <linux/uaccess.h>
69 #include <asm/div64.h>
71 #include <asm/vdso_datapage.h>
72 #include <asm/firmware.h>
73 #include <asm/asm-prototypes.h>
75 /* powerpc clocksource/clockevent code */
77 #include <linux/clockchips.h>
78 #include <linux/timekeeper_internal.h>
80 static u64 timebase_read(struct clocksource *);
81 static struct clocksource clocksource_timebase = {
84 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
85 .mask = CLOCKSOURCE_MASK(64),
86 .read = timebase_read,
87 .vdso_clock_mode = VDSO_CLOCKMODE_ARCHTIMER,
90 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
91 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
93 static int decrementer_set_next_event(unsigned long evt,
94 struct clock_event_device *dev);
95 static int decrementer_shutdown(struct clock_event_device *evt);
97 struct clock_event_device decrementer_clockevent = {
98 .name = "decrementer",
101 .set_next_event = decrementer_set_next_event,
102 .set_state_oneshot_stopped = decrementer_shutdown,
103 .set_state_shutdown = decrementer_shutdown,
104 .tick_resume = decrementer_shutdown,
105 .features = CLOCK_EVT_FEAT_ONESHOT |
106 CLOCK_EVT_FEAT_C3STOP,
108 EXPORT_SYMBOL(decrementer_clockevent);
110 DEFINE_PER_CPU(u64, decrementers_next_tb);
111 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
113 #define XSEC_PER_SEC (1024*1024)
116 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
118 /* compute ((xsec << 12) * max) >> 32 */
119 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
122 unsigned long tb_ticks_per_jiffy;
123 unsigned long tb_ticks_per_usec = 100; /* sane default */
124 EXPORT_SYMBOL(tb_ticks_per_usec);
125 unsigned long tb_ticks_per_sec;
126 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
128 DEFINE_SPINLOCK(rtc_lock);
129 EXPORT_SYMBOL_GPL(rtc_lock);
131 static u64 tb_to_ns_scale __read_mostly;
132 static unsigned tb_to_ns_shift __read_mostly;
133 static u64 boot_tb __read_mostly;
135 extern struct timezone sys_tz;
136 static long timezone_offset;
138 unsigned long ppc_proc_freq;
139 EXPORT_SYMBOL_GPL(ppc_proc_freq);
140 unsigned long ppc_tb_freq;
141 EXPORT_SYMBOL_GPL(ppc_tb_freq);
145 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
147 * Factor for converting from cputime_t (timebase ticks) to
148 * microseconds. This is stored as 0.64 fixed-point binary fraction.
150 u64 __cputime_usec_factor;
151 EXPORT_SYMBOL(__cputime_usec_factor);
153 #ifdef CONFIG_PPC_SPLPAR
154 void (*dtl_consumer)(struct dtl_entry *, u64);
157 static void calc_cputime_factors(void)
159 struct div_result res;
161 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
162 __cputime_usec_factor = res.result_low;
166 * Read the SPURR on systems that have it, otherwise the PURR,
167 * or if that doesn't exist return the timebase value passed in.
169 static inline unsigned long read_spurr(unsigned long tb)
171 if (cpu_has_feature(CPU_FTR_SPURR))
172 return mfspr(SPRN_SPURR);
173 if (cpu_has_feature(CPU_FTR_PURR))
174 return mfspr(SPRN_PURR);
178 #ifdef CONFIG_PPC_SPLPAR
183 * Scan the dispatch trace log and count up the stolen time.
184 * Should be called with interrupts disabled.
186 static u64 scan_dispatch_log(u64 stop_tb)
188 u64 i = local_paca->dtl_ridx;
189 struct dtl_entry *dtl = local_paca->dtl_curr;
190 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
191 struct lppaca *vpa = local_paca->lppaca_ptr;
199 if (i == be64_to_cpu(vpa->dtl_idx))
201 while (i < be64_to_cpu(vpa->dtl_idx)) {
202 dtb = be64_to_cpu(dtl->timebase);
203 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
204 be32_to_cpu(dtl->ready_to_enqueue_time);
206 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
207 /* buffer has overflowed */
208 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
209 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
215 dtl_consumer(dtl, i);
220 dtl = local_paca->dispatch_log;
222 local_paca->dtl_ridx = i;
223 local_paca->dtl_curr = dtl;
228 * Accumulate stolen time by scanning the dispatch trace log.
229 * Called on entry from user mode.
231 void notrace accumulate_stolen_time(void)
234 unsigned long save_irq_soft_mask = irq_soft_mask_return();
235 struct cpu_accounting_data *acct = &local_paca->accounting;
237 /* We are called early in the exception entry, before
238 * soft/hard_enabled are sync'ed to the expected state
239 * for the exception. We are hard disabled but the PACA
240 * needs to reflect that so various debug stuff doesn't
243 irq_soft_mask_set(IRQS_DISABLED);
245 sst = scan_dispatch_log(acct->starttime_user);
246 ust = scan_dispatch_log(acct->starttime);
249 acct->steal_time += ust + sst;
251 irq_soft_mask_set(save_irq_soft_mask);
254 static inline u64 calculate_stolen_time(u64 stop_tb)
256 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
259 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
260 return scan_dispatch_log(stop_tb);
265 #else /* CONFIG_PPC_SPLPAR */
266 static inline u64 calculate_stolen_time(u64 stop_tb)
271 #endif /* CONFIG_PPC_SPLPAR */
274 * Account time for a transition between system, hard irq
277 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
278 unsigned long now, unsigned long stime)
280 unsigned long stime_scaled = 0;
281 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
282 unsigned long nowscaled, deltascaled;
283 unsigned long utime, utime_scaled;
285 nowscaled = read_spurr(now);
286 deltascaled = nowscaled - acct->startspurr;
287 acct->startspurr = nowscaled;
288 utime = acct->utime - acct->utime_sspurr;
289 acct->utime_sspurr = acct->utime;
292 * Because we don't read the SPURR on every kernel entry/exit,
293 * deltascaled includes both user and system SPURR ticks.
294 * Apportion these ticks to system SPURR ticks and user
295 * SPURR ticks in the same ratio as the system time (delta)
296 * and user time (udelta) values obtained from the timebase
297 * over the same interval. The system ticks get accounted here;
298 * the user ticks get saved up in paca->user_time_scaled to be
299 * used by account_process_tick.
301 stime_scaled = stime;
302 utime_scaled = utime;
303 if (deltascaled != stime + utime) {
305 stime_scaled = deltascaled * stime / (stime + utime);
306 utime_scaled = deltascaled - stime_scaled;
308 stime_scaled = deltascaled;
311 acct->utime_scaled += utime_scaled;
317 static unsigned long vtime_delta(struct cpu_accounting_data *acct,
318 unsigned long *stime_scaled,
319 unsigned long *steal_time)
321 unsigned long now, stime;
323 WARN_ON_ONCE(!irqs_disabled());
326 stime = now - acct->starttime;
327 acct->starttime = now;
329 *stime_scaled = vtime_delta_scaled(acct, now, stime);
331 *steal_time = calculate_stolen_time(now);
336 static void vtime_delta_kernel(struct cpu_accounting_data *acct,
337 unsigned long *stime, unsigned long *stime_scaled)
339 unsigned long steal_time;
341 *stime = vtime_delta(acct, stime_scaled, &steal_time);
342 *stime -= min(*stime, steal_time);
343 acct->steal_time += steal_time;
346 void vtime_account_kernel(struct task_struct *tsk)
348 struct cpu_accounting_data *acct = get_accounting(tsk);
349 unsigned long stime, stime_scaled;
351 vtime_delta_kernel(acct, &stime, &stime_scaled);
353 if (tsk->flags & PF_VCPU) {
354 acct->gtime += stime;
355 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
356 acct->utime_scaled += stime_scaled;
359 acct->stime += stime;
360 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
361 acct->stime_scaled += stime_scaled;
365 EXPORT_SYMBOL_GPL(vtime_account_kernel);
367 void vtime_account_idle(struct task_struct *tsk)
369 unsigned long stime, stime_scaled, steal_time;
370 struct cpu_accounting_data *acct = get_accounting(tsk);
372 stime = vtime_delta(acct, &stime_scaled, &steal_time);
373 acct->idle_time += stime + steal_time;
376 static void vtime_account_irq_field(struct cpu_accounting_data *acct,
377 unsigned long *field)
379 unsigned long stime, stime_scaled;
381 vtime_delta_kernel(acct, &stime, &stime_scaled);
383 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
384 acct->stime_scaled += stime_scaled;
388 void vtime_account_softirq(struct task_struct *tsk)
390 struct cpu_accounting_data *acct = get_accounting(tsk);
391 vtime_account_irq_field(acct, &acct->softirq_time);
394 void vtime_account_hardirq(struct task_struct *tsk)
396 struct cpu_accounting_data *acct = get_accounting(tsk);
397 vtime_account_irq_field(acct, &acct->hardirq_time);
400 static void vtime_flush_scaled(struct task_struct *tsk,
401 struct cpu_accounting_data *acct)
403 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
404 if (acct->utime_scaled)
405 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
406 if (acct->stime_scaled)
407 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
409 acct->utime_scaled = 0;
410 acct->utime_sspurr = 0;
411 acct->stime_scaled = 0;
416 * Account the whole cputime accumulated in the paca
417 * Must be called with interrupts disabled.
418 * Assumes that vtime_account_kernel/idle() has been called
419 * recently (i.e. since the last entry from usermode) so that
420 * get_paca()->user_time_scaled is up to date.
422 void vtime_flush(struct task_struct *tsk)
424 struct cpu_accounting_data *acct = get_accounting(tsk);
427 account_user_time(tsk, cputime_to_nsecs(acct->utime));
430 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
432 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
433 account_steal_time(cputime_to_nsecs(acct->steal_time));
434 acct->steal_time = 0;
438 account_idle_time(cputime_to_nsecs(acct->idle_time));
441 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
444 if (acct->hardirq_time)
445 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
447 if (acct->softirq_time)
448 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
451 vtime_flush_scaled(tsk, acct);
457 acct->hardirq_time = 0;
458 acct->softirq_time = 0;
461 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
462 #define calc_cputime_factors()
465 void __delay(unsigned long loops)
472 * TB is in error state and isn't ticking anymore.
473 * HMI handler was unable to recover from TB error.
474 * Return immediately, so that kernel won't get stuck here.
479 while (mftb() - start < loops)
484 EXPORT_SYMBOL(__delay);
486 void udelay(unsigned long usecs)
488 __delay(tb_ticks_per_usec * usecs);
490 EXPORT_SYMBOL(udelay);
493 unsigned long profile_pc(struct pt_regs *regs)
495 unsigned long pc = instruction_pointer(regs);
497 if (in_lock_functions(pc))
502 EXPORT_SYMBOL(profile_pc);
505 #ifdef CONFIG_IRQ_WORK
508 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
511 static inline void set_irq_work_pending_flag(void)
513 asm volatile("stb %0,%1(13)" : :
515 "i" (offsetof(struct paca_struct, irq_work_pending)));
518 static inline void clear_irq_work_pending(void)
520 asm volatile("stb %0,%1(13)" : :
522 "i" (offsetof(struct paca_struct, irq_work_pending)));
527 DEFINE_PER_CPU(u8, irq_work_pending);
529 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
530 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
531 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
533 #endif /* 32 vs 64 bit */
535 void arch_irq_work_raise(void)
538 * 64-bit code that uses irq soft-mask can just cause an immediate
539 * interrupt here that gets soft masked, if this is called under
540 * local_irq_disable(). It might be possible to prevent that happening
541 * by noticing interrupts are disabled and setting decrementer pending
542 * to be replayed when irqs are enabled. The problem there is that
543 * tracing can call irq_work_raise, including in code that does low
544 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
545 * which could get tangled up if we're messing with the same state
549 set_irq_work_pending_flag();
554 #else /* CONFIG_IRQ_WORK */
556 #define test_irq_work_pending() 0
557 #define clear_irq_work_pending()
559 #endif /* CONFIG_IRQ_WORK */
562 * timer_interrupt - gets called when the decrementer overflows,
563 * with interrupts disabled.
565 DEFINE_INTERRUPT_HANDLER_ASYNC(timer_interrupt)
567 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
568 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
569 struct pt_regs *old_regs;
573 * Some implementations of hotplug will get timer interrupts while
574 * offline, just ignore these.
576 if (unlikely(!cpu_online(smp_processor_id()))) {
577 set_dec(decrementer_max);
581 /* Ensure a positive value is written to the decrementer, or else
582 * some CPUs will continue to take decrementer exceptions. When the
583 * PPC_WATCHDOG (decrementer based) is configured, keep this at most
584 * 31 bits, which is about 4 seconds on most systems, which gives
585 * the watchdog a chance of catching timer interrupt hard lockups.
587 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
590 set_dec(decrementer_max);
592 /* Conditionally hard-enable interrupts now that the DEC has been
593 * bumped to its maximum value
595 may_hard_irq_enable();
598 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
599 if (atomic_read(&ppc_n_lost_interrupts) != 0)
603 old_regs = set_irq_regs(regs);
605 trace_timer_interrupt_entry(regs);
607 if (test_irq_work_pending()) {
608 clear_irq_work_pending();
613 if (now >= *next_tb) {
615 if (evt->event_handler)
616 evt->event_handler(evt);
617 __this_cpu_inc(irq_stat.timer_irqs_event);
619 now = *next_tb - now;
620 if (now <= decrementer_max)
622 /* We may have raced with new irq work */
623 if (test_irq_work_pending())
625 __this_cpu_inc(irq_stat.timer_irqs_others);
628 trace_timer_interrupt_exit(regs);
630 set_irq_regs(old_regs);
632 EXPORT_SYMBOL(timer_interrupt);
634 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
635 void timer_broadcast_interrupt(void)
637 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
640 tick_receive_broadcast();
641 __this_cpu_inc(irq_stat.broadcast_irqs_event);
645 #ifdef CONFIG_SUSPEND
646 static void generic_suspend_disable_irqs(void)
648 /* Disable the decrementer, so that it doesn't interfere
652 set_dec(decrementer_max);
654 set_dec(decrementer_max);
657 static void generic_suspend_enable_irqs(void)
662 /* Overrides the weak version in kernel/power/main.c */
663 void arch_suspend_disable_irqs(void)
665 if (ppc_md.suspend_disable_irqs)
666 ppc_md.suspend_disable_irqs();
667 generic_suspend_disable_irqs();
670 /* Overrides the weak version in kernel/power/main.c */
671 void arch_suspend_enable_irqs(void)
673 generic_suspend_enable_irqs();
674 if (ppc_md.suspend_enable_irqs)
675 ppc_md.suspend_enable_irqs();
679 unsigned long long tb_to_ns(unsigned long long ticks)
681 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
683 EXPORT_SYMBOL_GPL(tb_to_ns);
686 * Scheduler clock - returns current time in nanosec units.
688 * Note: mulhdu(a, b) (multiply high double unsigned) returns
689 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
690 * are 64-bit unsigned numbers.
692 notrace unsigned long long sched_clock(void)
694 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
698 #ifdef CONFIG_PPC_PSERIES
701 * Running clock - attempts to give a view of time passing for a virtualised
703 * Uses the VTB register if available otherwise a next best guess.
705 unsigned long long running_clock(void)
708 * Don't read the VTB as a host since KVM does not switch in host
709 * timebase into the VTB when it takes a guest off the CPU, reading the
710 * VTB would result in reading 'last switched out' guest VTB.
712 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
713 * would be unsafe to rely only on the #ifdef above.
715 if (firmware_has_feature(FW_FEATURE_LPAR) &&
716 cpu_has_feature(CPU_FTR_ARCH_207S))
717 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
720 * This is a next best approximation without a VTB.
721 * On a host which is running bare metal there should never be any stolen
722 * time and on a host which doesn't do any virtualisation TB *should* equal
723 * VTB so it makes no difference anyway.
725 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
729 static int __init get_freq(char *name, int cells, unsigned long *val)
731 struct device_node *cpu;
735 /* The cpu node should have timebase and clock frequency properties */
736 cpu = of_find_node_by_type(NULL, "cpu");
739 fp = of_get_property(cpu, name, NULL);
742 *val = of_read_ulong(fp, cells);
751 static void start_cpu_decrementer(void)
753 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
756 /* Clear any pending timer interrupts */
757 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
759 tcr = mfspr(SPRN_TCR);
761 * The watchdog may have already been enabled by u-boot. So leave
762 * TRC[WP] (Watchdog Period) alone.
764 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
765 tcr |= TCR_DIE; /* Enable decrementer */
766 mtspr(SPRN_TCR, tcr);
770 void __init generic_calibrate_decr(void)
772 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
774 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
775 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
777 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
781 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
783 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
784 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
786 printk(KERN_ERR "WARNING: Estimating processor frequency "
791 int update_persistent_clock64(struct timespec64 now)
795 if (!ppc_md.set_rtc_time)
798 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
800 return ppc_md.set_rtc_time(&tm);
803 static void __read_persistent_clock(struct timespec64 *ts)
806 static int first = 1;
809 /* XXX this is a litle fragile but will work okay in the short term */
812 if (ppc_md.time_init)
813 timezone_offset = ppc_md.time_init();
815 /* get_boot_time() isn't guaranteed to be safe to call late */
816 if (ppc_md.get_boot_time) {
817 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
821 if (!ppc_md.get_rtc_time) {
825 ppc_md.get_rtc_time(&tm);
827 ts->tv_sec = rtc_tm_to_time64(&tm);
830 void read_persistent_clock64(struct timespec64 *ts)
832 __read_persistent_clock(ts);
834 /* Sanitize it in case real time clock is set below EPOCH */
835 if (ts->tv_sec < 0) {
842 /* clocksource code */
843 static notrace u64 timebase_read(struct clocksource *cs)
845 return (u64)get_tb();
848 static void __init clocksource_init(void)
850 struct clocksource *clock = &clocksource_timebase;
852 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
853 printk(KERN_ERR "clocksource: %s is already registered\n",
858 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
859 clock->name, clock->mult, clock->shift);
862 static int decrementer_set_next_event(unsigned long evt,
863 struct clock_event_device *dev)
865 __this_cpu_write(decrementers_next_tb, get_tb() + evt);
868 /* We may have raced with new irq work */
869 if (test_irq_work_pending())
875 static int decrementer_shutdown(struct clock_event_device *dev)
877 decrementer_set_next_event(decrementer_max, dev);
881 static void register_decrementer_clockevent(int cpu)
883 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
885 *dec = decrementer_clockevent;
886 dec->cpumask = cpumask_of(cpu);
888 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
890 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
891 dec->name, dec->mult, dec->shift, cpu);
893 /* Set values for KVM, see kvm_emulate_dec() */
894 decrementer_clockevent.mult = dec->mult;
895 decrementer_clockevent.shift = dec->shift;
898 static void enable_large_decrementer(void)
900 if (!cpu_has_feature(CPU_FTR_ARCH_300))
903 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
907 * If we're running as the hypervisor we need to enable the LD manually
908 * otherwise firmware should have done it for us.
910 if (cpu_has_feature(CPU_FTR_HVMODE))
911 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
914 static void __init set_decrementer_max(void)
916 struct device_node *cpu;
919 /* Prior to ISAv3 the decrementer is always 32 bit */
920 if (!cpu_has_feature(CPU_FTR_ARCH_300))
923 cpu = of_find_node_by_type(NULL, "cpu");
925 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
926 if (bits > 64 || bits < 32) {
927 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
931 /* calculate the signed maximum given this many bits */
932 decrementer_max = (1ul << (bits - 1)) - 1;
937 pr_info("time_init: %u bit decrementer (max: %llx)\n",
938 bits, decrementer_max);
941 static void __init init_decrementer_clockevent(void)
943 register_decrementer_clockevent(smp_processor_id());
946 void secondary_cpu_time_init(void)
948 /* Enable and test the large decrementer for this cpu */
949 enable_large_decrementer();
951 /* Start the decrementer on CPUs that have manual control
954 start_cpu_decrementer();
956 /* FIME: Should make unrelatred change to move snapshot_timebase
958 register_decrementer_clockevent(smp_processor_id());
961 /* This function is only called on the boot processor */
962 void __init time_init(void)
964 struct div_result res;
968 /* Normal PowerPC with timebase register */
969 ppc_md.calibrate_decr();
970 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
971 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
972 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
973 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
975 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
976 tb_ticks_per_sec = ppc_tb_freq;
977 tb_ticks_per_usec = ppc_tb_freq / 1000000;
978 calc_cputime_factors();
981 * Compute scale factor for sched_clock.
982 * The calibrate_decr() function has set tb_ticks_per_sec,
983 * which is the timebase frequency.
984 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
985 * the 128-bit result as a 64.64 fixed-point number.
986 * We then shift that number right until it is less than 1.0,
987 * giving us the scale factor and shift count to use in
990 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
991 scale = res.result_low;
992 for (shift = 0; res.result_high != 0; ++shift) {
993 scale = (scale >> 1) | (res.result_high << 63);
994 res.result_high >>= 1;
996 tb_to_ns_scale = scale;
997 tb_to_ns_shift = shift;
998 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1001 /* If platform provided a timezone (pmac), we correct the time */
1002 if (timezone_offset) {
1003 sys_tz.tz_minuteswest = -timezone_offset / 60;
1004 sys_tz.tz_dsttime = 0;
1007 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1009 /* initialise and enable the large decrementer (if we have one) */
1010 set_decrementer_max();
1011 enable_large_decrementer();
1013 /* Start the decrementer on CPUs that have manual control
1016 start_cpu_decrementer();
1018 /* Register the clocksource */
1021 init_decrementer_clockevent();
1022 tick_setup_hrtimer_broadcast();
1025 enable_sched_clock_irqtime();
1029 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1032 void div128_by_32(u64 dividend_high, u64 dividend_low,
1033 unsigned divisor, struct div_result *dr)
1035 unsigned long a, b, c, d;
1036 unsigned long w, x, y, z;
1039 a = dividend_high >> 32;
1040 b = dividend_high & 0xffffffff;
1041 c = dividend_low >> 32;
1042 d = dividend_low & 0xffffffff;
1045 ra = ((u64)(a - (w * divisor)) << 32) + b;
1047 rb = ((u64) do_div(ra, divisor) << 32) + c;
1050 rc = ((u64) do_div(rb, divisor) << 32) + d;
1053 do_div(rc, divisor);
1056 dr->result_high = ((u64)w << 32) + x;
1057 dr->result_low = ((u64)y << 32) + z;
1061 /* We don't need to calibrate delay, we use the CPU timebase for that */
1062 void calibrate_delay(void)
1064 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1065 * as the number of __delay(1) in a jiffy, so make it so
1067 loops_per_jiffy = tb_ticks_per_jiffy;
1070 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1071 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1073 ppc_md.get_rtc_time(tm);
1077 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1079 if (!ppc_md.set_rtc_time)
1082 if (ppc_md.set_rtc_time(tm) < 0)
1088 static const struct rtc_class_ops rtc_generic_ops = {
1089 .read_time = rtc_generic_get_time,
1090 .set_time = rtc_generic_set_time,
1093 static int __init rtc_init(void)
1095 struct platform_device *pdev;
1097 if (!ppc_md.get_rtc_time)
1100 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1102 sizeof(rtc_generic_ops));
1104 return PTR_ERR_OR_ZERO(pdev);
1107 device_initcall(rtc_init);