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/sched/cputime.h>
35 #include <linux/kernel.h>
36 #include <linux/param.h>
37 #include <linux/string.h>
39 #include <linux/interrupt.h>
40 #include <linux/timex.h>
41 #include <linux/kernel_stat.h>
42 #include <linux/time.h>
43 #include <linux/init.h>
44 #include <linux/profile.h>
45 #include <linux/cpu.h>
46 #include <linux/security.h>
47 #include <linux/percpu.h>
48 #include <linux/rtc.h>
49 #include <linux/jiffies.h>
50 #include <linux/posix-timers.h>
51 #include <linux/irq.h>
52 #include <linux/delay.h>
53 #include <linux/irq_work.h>
54 #include <linux/of_clk.h>
55 #include <linux/suspend.h>
56 #include <linux/processor.h>
57 #include <asm/trace.h>
59 #include <asm/interrupt.h>
61 #include <asm/nvram.h>
62 #include <asm/cache.h>
63 #include <asm/machdep.h>
64 #include <linux/uaccess.h>
68 #include <asm/div64.h>
70 #include <asm/vdso_datapage.h>
71 #include <asm/firmware.h>
74 /* powerpc clocksource/clockevent code */
76 #include <linux/clockchips.h>
77 #include <linux/timekeeper_internal.h>
79 static u64 timebase_read(struct clocksource *);
80 static struct clocksource clocksource_timebase = {
83 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
84 .mask = CLOCKSOURCE_MASK(64),
85 .read = timebase_read,
86 .vdso_clock_mode = VDSO_CLOCKMODE_ARCHTIMER,
89 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
90 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
91 EXPORT_SYMBOL_GPL(decrementer_max); /* for KVM HDEC */
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);
111 * This always puts next_tb beyond now, so the clock event will never fire
112 * with the usual comparison, no need for a separate test for stopped.
114 #define DEC_CLOCKEVENT_STOPPED ~0ULL
115 DEFINE_PER_CPU(u64, decrementers_next_tb) = DEC_CLOCKEVENT_STOPPED;
116 EXPORT_SYMBOL_GPL(decrementers_next_tb);
117 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
119 #define XSEC_PER_SEC (1024*1024)
122 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
124 /* compute ((xsec << 12) * max) >> 32 */
125 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
128 unsigned long tb_ticks_per_jiffy;
129 unsigned long tb_ticks_per_usec = 100; /* sane default */
130 EXPORT_SYMBOL(tb_ticks_per_usec);
131 unsigned long tb_ticks_per_sec;
132 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
134 DEFINE_SPINLOCK(rtc_lock);
135 EXPORT_SYMBOL_GPL(rtc_lock);
137 static u64 tb_to_ns_scale __read_mostly;
138 static unsigned tb_to_ns_shift __read_mostly;
139 static u64 boot_tb __read_mostly;
141 extern struct timezone sys_tz;
142 static long timezone_offset;
144 unsigned long ppc_proc_freq;
145 EXPORT_SYMBOL_GPL(ppc_proc_freq);
146 unsigned long ppc_tb_freq;
147 EXPORT_SYMBOL_GPL(ppc_tb_freq);
151 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
153 * Factor for converting from cputime_t (timebase ticks) to
154 * microseconds. This is stored as 0.64 fixed-point binary fraction.
156 u64 __cputime_usec_factor;
157 EXPORT_SYMBOL(__cputime_usec_factor);
159 #ifdef CONFIG_PPC_SPLPAR
160 void (*dtl_consumer)(struct dtl_entry *, u64);
163 static void calc_cputime_factors(void)
165 struct div_result res;
167 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
168 __cputime_usec_factor = res.result_low;
172 * Read the SPURR on systems that have it, otherwise the PURR,
173 * or if that doesn't exist return the timebase value passed in.
175 static inline unsigned long read_spurr(unsigned long tb)
177 if (cpu_has_feature(CPU_FTR_SPURR))
178 return mfspr(SPRN_SPURR);
179 if (cpu_has_feature(CPU_FTR_PURR))
180 return mfspr(SPRN_PURR);
184 #ifdef CONFIG_PPC_SPLPAR
189 * Scan the dispatch trace log and count up the stolen time.
190 * Should be called with interrupts disabled.
192 static u64 scan_dispatch_log(u64 stop_tb)
194 u64 i = local_paca->dtl_ridx;
195 struct dtl_entry *dtl = local_paca->dtl_curr;
196 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
197 struct lppaca *vpa = local_paca->lppaca_ptr;
205 if (i == be64_to_cpu(vpa->dtl_idx))
207 while (i < be64_to_cpu(vpa->dtl_idx)) {
208 dtb = be64_to_cpu(dtl->timebase);
209 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
210 be32_to_cpu(dtl->ready_to_enqueue_time);
212 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
213 /* buffer has overflowed */
214 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
215 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
221 dtl_consumer(dtl, i);
226 dtl = local_paca->dispatch_log;
228 local_paca->dtl_ridx = i;
229 local_paca->dtl_curr = dtl;
234 * Accumulate stolen time by scanning the dispatch trace log.
235 * Called on entry from user mode.
237 void notrace accumulate_stolen_time(void)
240 struct cpu_accounting_data *acct = &local_paca->accounting;
242 sst = scan_dispatch_log(acct->starttime_user);
243 ust = scan_dispatch_log(acct->starttime);
246 acct->steal_time += ust + sst;
249 static inline u64 calculate_stolen_time(u64 stop_tb)
251 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
254 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
255 return scan_dispatch_log(stop_tb);
260 #else /* CONFIG_PPC_SPLPAR */
261 static inline u64 calculate_stolen_time(u64 stop_tb)
266 #endif /* CONFIG_PPC_SPLPAR */
269 * Account time for a transition between system, hard irq
272 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
273 unsigned long now, unsigned long stime)
275 unsigned long stime_scaled = 0;
276 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
277 unsigned long nowscaled, deltascaled;
278 unsigned long utime, utime_scaled;
280 nowscaled = read_spurr(now);
281 deltascaled = nowscaled - acct->startspurr;
282 acct->startspurr = nowscaled;
283 utime = acct->utime - acct->utime_sspurr;
284 acct->utime_sspurr = acct->utime;
287 * Because we don't read the SPURR on every kernel entry/exit,
288 * deltascaled includes both user and system SPURR ticks.
289 * Apportion these ticks to system SPURR ticks and user
290 * SPURR ticks in the same ratio as the system time (delta)
291 * and user time (udelta) values obtained from the timebase
292 * over the same interval. The system ticks get accounted here;
293 * the user ticks get saved up in paca->user_time_scaled to be
294 * used by account_process_tick.
296 stime_scaled = stime;
297 utime_scaled = utime;
298 if (deltascaled != stime + utime) {
300 stime_scaled = deltascaled * stime / (stime + utime);
301 utime_scaled = deltascaled - stime_scaled;
303 stime_scaled = deltascaled;
306 acct->utime_scaled += utime_scaled;
312 static unsigned long vtime_delta(struct cpu_accounting_data *acct,
313 unsigned long *stime_scaled,
314 unsigned long *steal_time)
316 unsigned long now, stime;
318 WARN_ON_ONCE(!irqs_disabled());
321 stime = now - acct->starttime;
322 acct->starttime = now;
324 *stime_scaled = vtime_delta_scaled(acct, now, stime);
326 *steal_time = calculate_stolen_time(now);
331 static void vtime_delta_kernel(struct cpu_accounting_data *acct,
332 unsigned long *stime, unsigned long *stime_scaled)
334 unsigned long steal_time;
336 *stime = vtime_delta(acct, stime_scaled, &steal_time);
337 *stime -= min(*stime, steal_time);
338 acct->steal_time += steal_time;
341 void vtime_account_kernel(struct task_struct *tsk)
343 struct cpu_accounting_data *acct = get_accounting(tsk);
344 unsigned long stime, stime_scaled;
346 vtime_delta_kernel(acct, &stime, &stime_scaled);
348 if (tsk->flags & PF_VCPU) {
349 acct->gtime += stime;
350 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
351 acct->utime_scaled += stime_scaled;
354 acct->stime += stime;
355 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
356 acct->stime_scaled += stime_scaled;
360 EXPORT_SYMBOL_GPL(vtime_account_kernel);
362 void vtime_account_idle(struct task_struct *tsk)
364 unsigned long stime, stime_scaled, steal_time;
365 struct cpu_accounting_data *acct = get_accounting(tsk);
367 stime = vtime_delta(acct, &stime_scaled, &steal_time);
368 acct->idle_time += stime + steal_time;
371 static void vtime_account_irq_field(struct cpu_accounting_data *acct,
372 unsigned long *field)
374 unsigned long stime, stime_scaled;
376 vtime_delta_kernel(acct, &stime, &stime_scaled);
378 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
379 acct->stime_scaled += stime_scaled;
383 void vtime_account_softirq(struct task_struct *tsk)
385 struct cpu_accounting_data *acct = get_accounting(tsk);
386 vtime_account_irq_field(acct, &acct->softirq_time);
389 void vtime_account_hardirq(struct task_struct *tsk)
391 struct cpu_accounting_data *acct = get_accounting(tsk);
392 vtime_account_irq_field(acct, &acct->hardirq_time);
395 static void vtime_flush_scaled(struct task_struct *tsk,
396 struct cpu_accounting_data *acct)
398 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
399 if (acct->utime_scaled)
400 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
401 if (acct->stime_scaled)
402 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
404 acct->utime_scaled = 0;
405 acct->utime_sspurr = 0;
406 acct->stime_scaled = 0;
411 * Account the whole cputime accumulated in the paca
412 * Must be called with interrupts disabled.
413 * Assumes that vtime_account_kernel/idle() has been called
414 * recently (i.e. since the last entry from usermode) so that
415 * get_paca()->user_time_scaled is up to date.
417 void vtime_flush(struct task_struct *tsk)
419 struct cpu_accounting_data *acct = get_accounting(tsk);
422 account_user_time(tsk, cputime_to_nsecs(acct->utime));
425 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
427 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
428 account_steal_time(cputime_to_nsecs(acct->steal_time));
429 acct->steal_time = 0;
433 account_idle_time(cputime_to_nsecs(acct->idle_time));
436 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
439 if (acct->hardirq_time)
440 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
442 if (acct->softirq_time)
443 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
446 vtime_flush_scaled(tsk, acct);
452 acct->hardirq_time = 0;
453 acct->softirq_time = 0;
456 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
457 #define calc_cputime_factors()
460 void __delay(unsigned long loops)
467 * TB is in error state and isn't ticking anymore.
468 * HMI handler was unable to recover from TB error.
469 * Return immediately, so that kernel won't get stuck here.
474 while (mftb() - start < loops)
479 EXPORT_SYMBOL(__delay);
481 void udelay(unsigned long usecs)
483 __delay(tb_ticks_per_usec * usecs);
485 EXPORT_SYMBOL(udelay);
488 unsigned long profile_pc(struct pt_regs *regs)
490 unsigned long pc = instruction_pointer(regs);
492 if (in_lock_functions(pc))
497 EXPORT_SYMBOL(profile_pc);
500 #ifdef CONFIG_IRQ_WORK
503 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
506 static inline unsigned long test_irq_work_pending(void)
510 asm volatile("lbz %0,%1(13)"
512 : "i" (offsetof(struct paca_struct, irq_work_pending)));
516 static inline void set_irq_work_pending_flag(void)
518 asm volatile("stb %0,%1(13)" : :
520 "i" (offsetof(struct paca_struct, irq_work_pending)));
523 static inline void clear_irq_work_pending(void)
525 asm volatile("stb %0,%1(13)" : :
527 "i" (offsetof(struct paca_struct, irq_work_pending)));
532 DEFINE_PER_CPU(u8, irq_work_pending);
534 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
535 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
536 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
538 #endif /* 32 vs 64 bit */
540 void arch_irq_work_raise(void)
543 * 64-bit code that uses irq soft-mask can just cause an immediate
544 * interrupt here that gets soft masked, if this is called under
545 * local_irq_disable(). It might be possible to prevent that happening
546 * by noticing interrupts are disabled and setting decrementer pending
547 * to be replayed when irqs are enabled. The problem there is that
548 * tracing can call irq_work_raise, including in code that does low
549 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
550 * which could get tangled up if we're messing with the same state
554 set_irq_work_pending_flag();
559 static void set_dec_or_work(u64 val)
562 /* We may have raced with new irq work */
563 if (unlikely(test_irq_work_pending()))
567 #else /* CONFIG_IRQ_WORK */
569 #define test_irq_work_pending() 0
570 #define clear_irq_work_pending()
572 static void set_dec_or_work(u64 val)
576 #endif /* CONFIG_IRQ_WORK */
578 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
579 void timer_rearm_host_dec(u64 now)
581 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
583 WARN_ON_ONCE(!arch_irqs_disabled());
584 WARN_ON_ONCE(mfmsr() & MSR_EE);
586 if (now >= *next_tb) {
587 local_paca->irq_happened |= PACA_IRQ_DEC;
589 now = *next_tb - now;
590 if (now > decrementer_max)
591 now = decrementer_max;
592 set_dec_or_work(now);
595 EXPORT_SYMBOL_GPL(timer_rearm_host_dec);
599 * timer_interrupt - gets called when the decrementer overflows,
600 * with interrupts disabled.
602 DEFINE_INTERRUPT_HANDLER_ASYNC(timer_interrupt)
604 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
605 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
606 struct pt_regs *old_regs;
610 * Some implementations of hotplug will get timer interrupts while
611 * offline, just ignore these.
613 if (unlikely(!cpu_online(smp_processor_id()))) {
614 set_dec(decrementer_max);
618 /* Conditionally hard-enable interrupts. */
619 if (should_hard_irq_enable()) {
621 * Ensure a positive value is written to the decrementer, or
622 * else some CPUs will continue to take decrementer exceptions.
623 * When the PPC_WATCHDOG (decrementer based) is configured,
624 * keep this at most 31 bits, which is about 4 seconds on most
625 * systems, which gives the watchdog a chance of catching timer
626 * interrupt hard lockups.
628 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
631 set_dec(decrementer_max);
633 do_hard_irq_enable();
636 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
637 if (atomic_read(&ppc_n_lost_interrupts) != 0)
641 old_regs = set_irq_regs(regs);
643 trace_timer_interrupt_entry(regs);
645 if (test_irq_work_pending()) {
646 clear_irq_work_pending();
647 mce_run_irq_context_handlers();
652 if (now >= *next_tb) {
653 evt->event_handler(evt);
654 __this_cpu_inc(irq_stat.timer_irqs_event);
656 now = *next_tb - now;
657 if (now > decrementer_max)
658 now = decrementer_max;
659 set_dec_or_work(now);
660 __this_cpu_inc(irq_stat.timer_irqs_others);
663 trace_timer_interrupt_exit(regs);
665 set_irq_regs(old_regs);
667 EXPORT_SYMBOL(timer_interrupt);
669 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
670 void timer_broadcast_interrupt(void)
672 tick_receive_broadcast();
673 __this_cpu_inc(irq_stat.broadcast_irqs_event);
677 #ifdef CONFIG_SUSPEND
678 /* Overrides the weak version in kernel/power/main.c */
679 void arch_suspend_disable_irqs(void)
681 if (ppc_md.suspend_disable_irqs)
682 ppc_md.suspend_disable_irqs();
684 /* Disable the decrementer, so that it doesn't interfere
688 set_dec(decrementer_max);
690 set_dec(decrementer_max);
693 /* Overrides the weak version in kernel/power/main.c */
694 void arch_suspend_enable_irqs(void)
698 if (ppc_md.suspend_enable_irqs)
699 ppc_md.suspend_enable_irqs();
703 unsigned long long tb_to_ns(unsigned long long ticks)
705 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
707 EXPORT_SYMBOL_GPL(tb_to_ns);
710 * Scheduler clock - returns current time in nanosec units.
712 * Note: mulhdu(a, b) (multiply high double unsigned) returns
713 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
714 * are 64-bit unsigned numbers.
716 notrace unsigned long long sched_clock(void)
718 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
722 #ifdef CONFIG_PPC_PSERIES
725 * Running clock - attempts to give a view of time passing for a virtualised
727 * Uses the VTB register if available otherwise a next best guess.
729 unsigned long long running_clock(void)
732 * Don't read the VTB as a host since KVM does not switch in host
733 * timebase into the VTB when it takes a guest off the CPU, reading the
734 * VTB would result in reading 'last switched out' guest VTB.
736 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
737 * would be unsafe to rely only on the #ifdef above.
739 if (firmware_has_feature(FW_FEATURE_LPAR) &&
740 cpu_has_feature(CPU_FTR_ARCH_207S))
741 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
744 * This is a next best approximation without a VTB.
745 * On a host which is running bare metal there should never be any stolen
746 * time and on a host which doesn't do any virtualisation TB *should* equal
747 * VTB so it makes no difference anyway.
749 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
753 static int __init get_freq(char *name, int cells, unsigned long *val)
755 struct device_node *cpu;
759 /* The cpu node should have timebase and clock frequency properties */
760 cpu = of_find_node_by_type(NULL, "cpu");
763 fp = of_get_property(cpu, name, NULL);
766 *val = of_read_ulong(fp, cells);
775 static void start_cpu_decrementer(void)
777 #ifdef CONFIG_BOOKE_OR_40x
780 /* Clear any pending timer interrupts */
781 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
783 tcr = mfspr(SPRN_TCR);
785 * The watchdog may have already been enabled by u-boot. So leave
786 * TRC[WP] (Watchdog Period) alone.
788 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
789 tcr |= TCR_DIE; /* Enable decrementer */
790 mtspr(SPRN_TCR, tcr);
794 void __init generic_calibrate_decr(void)
796 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
798 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
799 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
801 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
805 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
807 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
808 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
810 printk(KERN_ERR "WARNING: Estimating processor frequency "
815 int update_persistent_clock64(struct timespec64 now)
819 if (!ppc_md.set_rtc_time)
822 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
824 return ppc_md.set_rtc_time(&tm);
827 static void __read_persistent_clock(struct timespec64 *ts)
830 static int first = 1;
833 /* XXX this is a litle fragile but will work okay in the short term */
836 if (ppc_md.time_init)
837 timezone_offset = ppc_md.time_init();
839 /* get_boot_time() isn't guaranteed to be safe to call late */
840 if (ppc_md.get_boot_time) {
841 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
845 if (!ppc_md.get_rtc_time) {
849 ppc_md.get_rtc_time(&tm);
851 ts->tv_sec = rtc_tm_to_time64(&tm);
854 void read_persistent_clock64(struct timespec64 *ts)
856 __read_persistent_clock(ts);
858 /* Sanitize it in case real time clock is set below EPOCH */
859 if (ts->tv_sec < 0) {
866 /* clocksource code */
867 static notrace u64 timebase_read(struct clocksource *cs)
869 return (u64)get_tb();
872 static void __init clocksource_init(void)
874 struct clocksource *clock = &clocksource_timebase;
876 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
877 printk(KERN_ERR "clocksource: %s is already registered\n",
882 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
883 clock->name, clock->mult, clock->shift);
886 static int decrementer_set_next_event(unsigned long evt,
887 struct clock_event_device *dev)
889 __this_cpu_write(decrementers_next_tb, get_tb() + evt);
890 set_dec_or_work(evt);
895 static int decrementer_shutdown(struct clock_event_device *dev)
897 __this_cpu_write(decrementers_next_tb, DEC_CLOCKEVENT_STOPPED);
898 set_dec_or_work(decrementer_max);
903 static void register_decrementer_clockevent(int cpu)
905 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
907 *dec = decrementer_clockevent;
908 dec->cpumask = cpumask_of(cpu);
910 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
912 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
913 dec->name, dec->mult, dec->shift, cpu);
915 /* Set values for KVM, see kvm_emulate_dec() */
916 decrementer_clockevent.mult = dec->mult;
917 decrementer_clockevent.shift = dec->shift;
920 static void enable_large_decrementer(void)
922 if (!cpu_has_feature(CPU_FTR_ARCH_300))
925 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
929 * If we're running as the hypervisor we need to enable the LD manually
930 * otherwise firmware should have done it for us.
932 if (cpu_has_feature(CPU_FTR_HVMODE))
933 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
936 static void __init set_decrementer_max(void)
938 struct device_node *cpu;
941 /* Prior to ISAv3 the decrementer is always 32 bit */
942 if (!cpu_has_feature(CPU_FTR_ARCH_300))
945 cpu = of_find_node_by_type(NULL, "cpu");
947 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
948 if (bits > 64 || bits < 32) {
949 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
953 /* calculate the signed maximum given this many bits */
954 decrementer_max = (1ul << (bits - 1)) - 1;
959 pr_info("time_init: %u bit decrementer (max: %llx)\n",
960 bits, decrementer_max);
963 static void __init init_decrementer_clockevent(void)
965 register_decrementer_clockevent(smp_processor_id());
968 void secondary_cpu_time_init(void)
970 /* Enable and test the large decrementer for this cpu */
971 enable_large_decrementer();
973 /* Start the decrementer on CPUs that have manual control
976 start_cpu_decrementer();
978 /* FIME: Should make unrelatred change to move snapshot_timebase
980 register_decrementer_clockevent(smp_processor_id());
983 /* This function is only called on the boot processor */
984 void __init time_init(void)
986 struct div_result res;
990 /* Normal PowerPC with timebase register */
991 ppc_md.calibrate_decr();
992 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
993 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
994 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
995 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
997 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
998 tb_ticks_per_sec = ppc_tb_freq;
999 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1000 calc_cputime_factors();
1003 * Compute scale factor for sched_clock.
1004 * The calibrate_decr() function has set tb_ticks_per_sec,
1005 * which is the timebase frequency.
1006 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1007 * the 128-bit result as a 64.64 fixed-point number.
1008 * We then shift that number right until it is less than 1.0,
1009 * giving us the scale factor and shift count to use in
1012 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1013 scale = res.result_low;
1014 for (shift = 0; res.result_high != 0; ++shift) {
1015 scale = (scale >> 1) | (res.result_high << 63);
1016 res.result_high >>= 1;
1018 tb_to_ns_scale = scale;
1019 tb_to_ns_shift = shift;
1020 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1023 /* If platform provided a timezone (pmac), we correct the time */
1024 if (timezone_offset) {
1025 sys_tz.tz_minuteswest = -timezone_offset / 60;
1026 sys_tz.tz_dsttime = 0;
1029 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1031 /* initialise and enable the large decrementer (if we have one) */
1032 set_decrementer_max();
1033 enable_large_decrementer();
1035 /* Start the decrementer on CPUs that have manual control
1038 start_cpu_decrementer();
1040 /* Register the clocksource */
1043 init_decrementer_clockevent();
1044 tick_setup_hrtimer_broadcast();
1047 enable_sched_clock_irqtime();
1051 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1054 void div128_by_32(u64 dividend_high, u64 dividend_low,
1055 unsigned divisor, struct div_result *dr)
1057 unsigned long a, b, c, d;
1058 unsigned long w, x, y, z;
1061 a = dividend_high >> 32;
1062 b = dividend_high & 0xffffffff;
1063 c = dividend_low >> 32;
1064 d = dividend_low & 0xffffffff;
1067 ra = ((u64)(a - (w * divisor)) << 32) + b;
1069 rb = ((u64) do_div(ra, divisor) << 32) + c;
1072 rc = ((u64) do_div(rb, divisor) << 32) + d;
1075 do_div(rc, divisor);
1078 dr->result_high = ((u64)w << 32) + x;
1079 dr->result_low = ((u64)y << 32) + z;
1083 /* We don't need to calibrate delay, we use the CPU timebase for that */
1084 void calibrate_delay(void)
1086 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1087 * as the number of __delay(1) in a jiffy, so make it so
1089 loops_per_jiffy = tb_ticks_per_jiffy;
1092 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1093 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1095 ppc_md.get_rtc_time(tm);
1099 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1101 if (!ppc_md.set_rtc_time)
1104 if (ppc_md.set_rtc_time(tm) < 0)
1110 static const struct rtc_class_ops rtc_generic_ops = {
1111 .read_time = rtc_generic_get_time,
1112 .set_time = rtc_generic_set_time,
1115 static int __init rtc_init(void)
1117 struct platform_device *pdev;
1119 if (!ppc_md.get_rtc_time)
1122 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1124 sizeof(rtc_generic_ops));
1126 return PTR_ERR_OR_ZERO(pdev);
1129 device_initcall(rtc_init);