projects/linux·blob master · refs

1107 lines · 27.8 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Common time routines among all ppc machines.
4 *
5 * Written by Cort Dougan ([email protected]) to merge
6 * Paul Mackerras' version and mine for PReP and Pmac.
7 * MPC8xx/MBX changes by Dan Malek ([email protected]).
8 * Converted for 64-bit by Mike Corrigan ([email protected])
9 *
10 * First round of bugfixes by Gabriel Paubert ([email protected])
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).
15 *
16 * Speeded up do_gettimeofday by getting rid of references to
17 * xtime (which required locks for consistency). ([email protected])
18 *
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.
25 *
26 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
27 * "A Kernel Model for Precision Timekeeping" by Dave Mills
28 */
29
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>
38#include <linux/mm.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 <linux/mc146818rtc.h>
58#include <linux/platform_device.h>
59
60#include <asm/trace.h>
61#include <asm/interrupt.h>
62#include <asm/io.h>
63#include <asm/nvram.h>
64#include <asm/cache.h>
65#include <asm/machdep.h>
66#include <linux/uaccess.h>
67#include <asm/time.h>
68#include <asm/irq.h>
69#include <asm/div64.h>
70#include <asm/smp.h>
71#include <asm/vdso_datapage.h>
72#include <asm/firmware.h>
73#include <asm/mce.h>
74#include <asm/systemcfg.h>
75
76/* powerpc clocksource/clockevent code */
77
78#include <linux/clockchips.h>
79
80static u64 timebase_read(struct clocksource *);
81static struct clocksource clocksource_timebase = {
82 .name = "timebase",
83 .rating = 400,
84 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
85 .mask = CLOCKSOURCE_MASK(64),
86 .read = timebase_read,
87 .vdso_clock_mode = VDSO_CLOCKMODE_ARCHTIMER,
88};
89
90#define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
91u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
92EXPORT_SYMBOL_GPL(decrementer_max); /* for KVM HDEC */
93
94static int decrementer_set_next_event(unsigned long evt,
95 struct clock_event_device *dev);
96static int decrementer_shutdown(struct clock_event_device *evt);
97
98struct clock_event_device decrementer_clockevent = {
99 .name = "decrementer",
100 .rating = 200,
101 .irq = 0,
102 .set_next_event = decrementer_set_next_event,
103 .set_state_oneshot_stopped = decrementer_shutdown,
104 .set_state_shutdown = decrementer_shutdown,
105 .tick_resume = decrementer_shutdown,
106 .features = CLOCK_EVT_FEAT_ONESHOT |
107 CLOCK_EVT_FEAT_C3STOP,
108};
109EXPORT_SYMBOL(decrementer_clockevent);
110
111/*
112 * This always puts next_tb beyond now, so the clock event will never fire
113 * with the usual comparison, no need for a separate test for stopped.
114 */
115#define DEC_CLOCKEVENT_STOPPED ~0ULL
116DEFINE_PER_CPU(u64, decrementers_next_tb) = DEC_CLOCKEVENT_STOPPED;
117EXPORT_SYMBOL_GPL(decrementers_next_tb);
118static DEFINE_PER_CPU(struct clock_event_device, decrementers);
119
120#define XSEC_PER_SEC (1024*1024)
121
122#ifdef CONFIG_PPC64
123#define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
124#else
125/* compute ((xsec << 12) * max) >> 32 */
126#define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
127#endif
128
129unsigned long tb_ticks_per_jiffy;
130unsigned long tb_ticks_per_usec = 100; /* sane default */
131EXPORT_SYMBOL(tb_ticks_per_usec);
132unsigned long tb_ticks_per_sec;
133EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime conversions */
134
135DEFINE_SPINLOCK(rtc_lock);
136EXPORT_SYMBOL_GPL(rtc_lock);
137
138static u64 tb_to_ns_scale __read_mostly;
139static unsigned tb_to_ns_shift __read_mostly;
140static u64 boot_tb __ro_after_init;
141
142extern struct timezone sys_tz;
143static long timezone_offset;
144
145unsigned long ppc_proc_freq;
146EXPORT_SYMBOL_GPL(ppc_proc_freq);
147unsigned long ppc_tb_freq;
148EXPORT_SYMBOL_GPL(ppc_tb_freq);
149
150bool tb_invalid;
151
152#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
153/*
154 * Read the SPURR on systems that have it, otherwise the PURR,
155 * or if that doesn't exist return the timebase value passed in.
156 */
157static inline unsigned long read_spurr(unsigned long tb)
158{
159 if (cpu_has_feature(CPU_FTR_SPURR))
160 return mfspr(SPRN_SPURR);
161 if (cpu_has_feature(CPU_FTR_PURR))
162 return mfspr(SPRN_PURR);
163 return tb;
164}
165
166/*
167 * Account time for a transition between system, hard irq
168 * or soft irq state.
169 */
170static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
171 unsigned long now, unsigned long stime)
172{
173 unsigned long stime_scaled = 0;
174#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
175 unsigned long nowscaled, deltascaled;
176 unsigned long utime, utime_scaled;
177
178 nowscaled = read_spurr(now);
179 deltascaled = nowscaled - acct->startspurr;
180 acct->startspurr = nowscaled;
181 utime = acct->utime - acct->utime_sspurr;
182 acct->utime_sspurr = acct->utime;
183
184 /*
185 * Because we don't read the SPURR on every kernel entry/exit,
186 * deltascaled includes both user and system SPURR ticks.
187 * Apportion these ticks to system SPURR ticks and user
188 * SPURR ticks in the same ratio as the system time (delta)
189 * and user time (udelta) values obtained from the timebase
190 * over the same interval. The system ticks get accounted here;
191 * the user ticks get saved up in paca->user_time_scaled to be
192 * used by account_process_tick.
193 */
194 stime_scaled = stime;
195 utime_scaled = utime;
196 if (deltascaled != stime + utime) {
197 if (utime) {
198 stime_scaled = deltascaled * stime / (stime + utime);
199 utime_scaled = deltascaled - stime_scaled;
200 } else {
201 stime_scaled = deltascaled;
202 }
203 }
204 acct->utime_scaled += utime_scaled;
205#endif
206
207 return stime_scaled;
208}
209
210static unsigned long vtime_delta(struct cpu_accounting_data *acct,
211 unsigned long *stime_scaled,
212 unsigned long *steal_time)
213{
214 unsigned long now, stime;
215
216 WARN_ON_ONCE(!irqs_disabled());
217
218 now = mftb();
219 stime = now - acct->starttime;
220 acct->starttime = now;
221
222 *stime_scaled = vtime_delta_scaled(acct, now, stime);
223
224 if (IS_ENABLED(CONFIG_PPC_SPLPAR) &&
225 firmware_has_feature(FW_FEATURE_SPLPAR))
226 *steal_time = pseries_calculate_stolen_time(now);
227 else
228 *steal_time = 0;
229
230 return stime;
231}
232
233static void vtime_delta_kernel(struct cpu_accounting_data *acct,
234 unsigned long *stime, unsigned long *stime_scaled)
235{
236 unsigned long steal_time;
237
238 *stime = vtime_delta(acct, stime_scaled, &steal_time);
239 *stime -= min(*stime, steal_time);
240 acct->steal_time += steal_time;
241}
242
243void vtime_account_kernel(struct task_struct *tsk)
244{
245 struct cpu_accounting_data *acct = get_accounting(tsk);
246 unsigned long stime, stime_scaled;
247
248 vtime_delta_kernel(acct, &stime, &stime_scaled);
249
250 if (tsk->flags & PF_VCPU) {
251 acct->gtime += stime;
252#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
253 acct->utime_scaled += stime_scaled;
254#endif
255 } else {
256 acct->stime += stime;
257#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
258 acct->stime_scaled += stime_scaled;
259#endif
260 }
261}
262EXPORT_SYMBOL_GPL(vtime_account_kernel);
263
264void vtime_account_idle(struct task_struct *tsk)
265{
266 unsigned long stime, stime_scaled, steal_time;
267 struct cpu_accounting_data *acct = get_accounting(tsk);
268
269 stime = vtime_delta(acct, &stime_scaled, &steal_time);
270 acct->idle_time += stime + steal_time;
271}
272
273static void vtime_account_irq_field(struct cpu_accounting_data *acct,
274 unsigned long *field)
275{
276 unsigned long stime, stime_scaled;
277
278 vtime_delta_kernel(acct, &stime, &stime_scaled);
279 *field += stime;
280#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
281 acct->stime_scaled += stime_scaled;
282#endif
283}
284
285void vtime_account_softirq(struct task_struct *tsk)
286{
287 struct cpu_accounting_data *acct = get_accounting(tsk);
288 vtime_account_irq_field(acct, &acct->softirq_time);
289}
290
291void vtime_account_hardirq(struct task_struct *tsk)
292{
293 struct cpu_accounting_data *acct = get_accounting(tsk);
294 vtime_account_irq_field(acct, &acct->hardirq_time);
295}
296
297static void vtime_flush_scaled(struct task_struct *tsk,
298 struct cpu_accounting_data *acct)
299{
300#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
301 if (acct->utime_scaled)
302 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
303 if (acct->stime_scaled)
304 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
305
306 acct->utime_scaled = 0;
307 acct->utime_sspurr = 0;
308 acct->stime_scaled = 0;
309#endif
310}
311
312/*
313 * Account the whole cputime accumulated in the paca
314 * Must be called with interrupts disabled.
315 * Assumes that vtime_account_kernel/idle() has been called
316 * recently (i.e. since the last entry from usermode) so that
317 * get_paca()->user_time_scaled is up to date.
318 */
319void vtime_flush(struct task_struct *tsk)
320{
321 struct cpu_accounting_data *acct = get_accounting(tsk);
322
323 if (acct->utime)
324 account_user_time(tsk, cputime_to_nsecs(acct->utime));
325
326 if (acct->gtime)
327 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
328
329 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
330 account_steal_time(cputime_to_nsecs(acct->steal_time));
331 acct->steal_time = 0;
332 }
333
334 if (acct->idle_time)
335 account_idle_time(cputime_to_nsecs(acct->idle_time));
336
337 if (acct->stime)
338 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
339 CPUTIME_SYSTEM);
340
341 if (acct->hardirq_time)
342 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
343 CPUTIME_IRQ);
344 if (acct->softirq_time)
345 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
346 CPUTIME_SOFTIRQ);
347
348 vtime_flush_scaled(tsk, acct);
349
350 acct->utime = 0;
351 acct->gtime = 0;
352 acct->idle_time = 0;
353 acct->stime = 0;
354 acct->hardirq_time = 0;
355 acct->softirq_time = 0;
356}
357
358/*
359 * Called from the context switch with interrupts disabled, to charge all
360 * accumulated times to the current process, and to prepare accounting on
361 * the next process.
362 */
363void vtime_task_switch(struct task_struct *prev)
364{
365 if (is_idle_task(prev))
366 vtime_account_idle(prev);
367 else
368 vtime_account_kernel(prev);
369
370 vtime_flush(prev);
371
372 if (!IS_ENABLED(CONFIG_PPC64)) {
373 struct cpu_accounting_data *acct = get_accounting(current);
374 struct cpu_accounting_data *acct0 = get_accounting(prev);
375
376 acct->starttime = acct0->starttime;
377 }
378}
379
380/**
381 * vtime_reset - Fast forward vtime entry clocks
382 *
383 * Called from dynticks idle IRQ entry to fast-forward the clocks to current time
384 * so that the IRQ time is still accounted by vtime while nohz cputime is paused.
385 */
386void vtime_reset(void)
387{
388 struct cpu_accounting_data *acct = get_accounting(current);
389
390 acct->starttime = mftb();
391#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
392 acct->startspurr = read_spurr(acct->starttime);
393#endif
394}
395
396#ifdef CONFIG_NO_HZ_COMMON
397/**
398 * vtime_dyntick_start - Inform vtime about entry to idle-dynticks
399 *
400 * Called when idle enters in dyntick mode. The idle cputime that elapsed so far
401 * is accumulated and the tick subsystem takes over the idle cputime accounting.
402 */
403void vtime_dyntick_start(void)
404{
405 vtime_account_idle(current);
406}
407
408/**
409 * vtime_dyntick_stop - Inform vtime about exit from idle-dynticks
410 *
411 * Called when idle exits from dyntick mode. The vtime entry clocks are
412 * fast-forward to current time so that idle accounting restarts elapsing from
413 * now.
414 */
415void vtime_dyntick_stop(void)
416{
417 vtime_reset();
418}
419#endif /* CONFIG_NO_HZ_COMMON */
420#endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
421
422void __no_kcsan __delay(unsigned long loops)
423{
424 unsigned long start;
425
426 spin_begin();
427 if (tb_invalid) {
428 /*
429 * TB is in error state and isn't ticking anymore.
430 * HMI handler was unable to recover from TB error.
431 * Return immediately, so that kernel won't get stuck here.
432 */
433 spin_cpu_relax();
434 } else {
435 start = mftb();
436 while (mftb() - start < loops)
437 spin_cpu_relax();
438 }
439 spin_end();
440}
441EXPORT_SYMBOL(__delay);
442
443void __no_kcsan udelay(unsigned long usecs)
444{
445 __delay(tb_ticks_per_usec * usecs);
446}
447EXPORT_SYMBOL(udelay);
448
449#ifdef CONFIG_SMP
450unsigned long profile_pc(struct pt_regs *regs)
451{
452 unsigned long pc = instruction_pointer(regs);
453
454 if (in_lock_functions(pc))
455 return regs->link;
456
457 return pc;
458}
459EXPORT_SYMBOL(profile_pc);
460#endif
461
462#ifdef CONFIG_IRQ_WORK
463
464/*
465 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
466 */
467#ifdef CONFIG_PPC64
468static inline unsigned long test_irq_work_pending(void)
469{
470 unsigned long x;
471
472 asm volatile("lbz %0,%1(13)"
473 : "=r" (x)
474 : "i" (offsetof(struct paca_struct, irq_work_pending)));
475 return x;
476}
477
478static inline void set_irq_work_pending_flag(void)
479{
480 asm volatile("stb %0,%1(13)" : :
481 "r" (1),
482 "i" (offsetof(struct paca_struct, irq_work_pending)));
483}
484
485static inline void clear_irq_work_pending(void)
486{
487 asm volatile("stb %0,%1(13)" : :
488 "r" (0),
489 "i" (offsetof(struct paca_struct, irq_work_pending)));
490}
491
492#else /* 32-bit */
493
494DEFINE_PER_CPU(u8, irq_work_pending);
495
496#define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
497#define test_irq_work_pending() __this_cpu_read(irq_work_pending)
498#define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
499
500#endif /* 32 vs 64 bit */
501
502/*
503 * Must be called with preemption disabled since it updates
504 * per-CPU irq_work state and programs the local CPU decrementer.
505 */
506void arch_irq_work_raise(void)
507{
508 /*
509 * 64-bit code that uses irq soft-mask can just cause an immediate
510 * interrupt here that gets soft masked, if this is called under
511 * local_irq_disable(). It might be possible to prevent that happening
512 * by noticing interrupts are disabled and setting decrementer pending
513 * to be replayed when irqs are enabled. The problem there is that
514 * tracing can call irq_work_raise, including in code that does low
515 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
516 * which could get tangled up if we're messing with the same state
517 * here.
518 */
519 set_irq_work_pending_flag();
520 set_dec(1);
521}
522
523static void set_dec_or_work(u64 val)
524{
525 set_dec(val);
526 /* We may have raced with new irq work */
527 if (unlikely(test_irq_work_pending()))
528 set_dec(1);
529}
530
531#else /* CONFIG_IRQ_WORK */
532
533#define test_irq_work_pending() 0
534#define clear_irq_work_pending()
535
536static void set_dec_or_work(u64 val)
537{
538 set_dec(val);
539}
540#endif /* CONFIG_IRQ_WORK */
541
542#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
543void timer_rearm_host_dec(u64 now)
544{
545 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
546
547 WARN_ON_ONCE(!arch_irqs_disabled());
548 WARN_ON_ONCE(mfmsr() & MSR_EE);
549
550 if (now >= *next_tb) {
551 local_paca->irq_happened |= PACA_IRQ_DEC;
552 } else {
553 now = *next_tb - now;
554 if (now > decrementer_max)
555 now = decrementer_max;
556 set_dec_or_work(now);
557 }
558}
559EXPORT_SYMBOL_GPL(timer_rearm_host_dec);
560#endif
561
562/*
563 * timer_interrupt - gets called when the decrementer overflows,
564 * with interrupts disabled.
565 */
566DEFINE_INTERRUPT_HANDLER_ASYNC(timer_interrupt)
567{
568 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
569 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
570 struct pt_regs *old_regs;
571 u64 now;
572
573 /*
574 * Some implementations of hotplug will get timer interrupts while
575 * offline, just ignore these.
576 */
577 if (unlikely(!cpu_online(smp_processor_id()))) {
578 set_dec(decrementer_max);
579 return;
580 }
581
582 /* Conditionally hard-enable interrupts. */
583 if (should_hard_irq_enable(regs)) {
584 /*
585 * Ensure a positive value is written to the decrementer, or
586 * else some CPUs will continue to take decrementer exceptions.
587 * When the PPC_WATCHDOG (decrementer based) is configured,
588 * keep this at most 31 bits, which is about 4 seconds on most
589 * systems, which gives the watchdog a chance of catching timer
590 * interrupt hard lockups.
591 */
592 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
593 set_dec(0x7fffffff);
594 else
595 set_dec(decrementer_max);
596
597 do_hard_irq_enable();
598 }
599
600#if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
601 if (atomic_read(&ppc_n_lost_interrupts) != 0)
602 __do_IRQ(regs);
603#endif
604
605 old_regs = set_irq_regs(regs);
606
607 trace_timer_interrupt_entry(regs);
608
609 if (test_irq_work_pending()) {
610 clear_irq_work_pending();
611 mce_run_irq_context_handlers();
612 irq_work_run();
613 }
614
615 now = get_tb();
616 if (now >= *next_tb) {
617 evt->event_handler(evt);
618 __this_cpu_inc(irq_stat.timer_irqs_event);
619 } else {
620 now = *next_tb - now;
621 if (now > decrementer_max)
622 now = decrementer_max;
623 set_dec_or_work(now);
624 __this_cpu_inc(irq_stat.timer_irqs_others);
625 }
626
627 trace_timer_interrupt_exit(regs);
628
629 set_irq_regs(old_regs);
630}
631EXPORT_SYMBOL(timer_interrupt);
632
633#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
634void timer_broadcast_interrupt(void)
635{
636 tick_receive_broadcast();
637 __this_cpu_inc(irq_stat.broadcast_irqs_event);
638}
639#endif
640
641#ifdef CONFIG_SUSPEND
642/* Overrides the weak version in kernel/power/main.c */
643void arch_suspend_disable_irqs(void)
644{
645 if (ppc_md.suspend_disable_irqs)
646 ppc_md.suspend_disable_irqs();
647
648 /* Disable the decrementer, so that it doesn't interfere
649 * with suspending.
650 */
651
652 set_dec(decrementer_max);
653 local_irq_disable();
654 set_dec(decrementer_max);
655}
656
657/* Overrides the weak version in kernel/power/main.c */
658void arch_suspend_enable_irqs(void)
659{
660 local_irq_enable();
661
662 if (ppc_md.suspend_enable_irqs)
663 ppc_md.suspend_enable_irqs();
664}
665#endif
666
667unsigned long long tb_to_ns(unsigned long long ticks)
668{
669 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
670}
671EXPORT_SYMBOL_GPL(tb_to_ns);
672
673/*
674 * Scheduler clock - returns current time in nanosec units.
675 *
676 * Note: mulhdu(a, b) (multiply high double unsigned) returns
677 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
678 * are 64-bit unsigned numbers.
679 */
680notrace unsigned long long sched_clock(void)
681{
682 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
683}
684
685#ifdef CONFIG_PPC_SPLPAR
686u64 get_boot_tb(void)
687{
688 return boot_tb;
689}
690#endif
691
692#ifdef CONFIG_PPC_PSERIES
693
694/*
695 * Running clock - attempts to give a view of time passing for a virtualised
696 * kernels.
697 * Uses the VTB register if available otherwise a next best guess.
698 */
699unsigned long long running_clock(void)
700{
701 /*
702 * Don't read the VTB as a host since KVM does not switch in host
703 * timebase into the VTB when it takes a guest off the CPU, reading the
704 * VTB would result in reading 'last switched out' guest VTB.
705 *
706 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
707 * would be unsafe to rely only on the #ifdef above.
708 */
709 if (firmware_has_feature(FW_FEATURE_LPAR) &&
710 cpu_has_feature(CPU_FTR_ARCH_207S))
711 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
712
713 /*
714 * This is a next best approximation without a VTB.
715 * On a host which is running bare metal there should never be any stolen
716 * time and on a host which doesn't do any virtualisation TB *should* equal
717 * VTB so it makes no difference anyway.
718 */
719 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
720}
721#endif
722
723static int __init get_freq(char *name, int cells, unsigned long *val)
724{
725 struct device_node *cpu;
726 const __be32 *fp;
727 int found = 0;
728
729 /* The cpu node should have timebase and clock frequency properties */
730 cpu = of_find_node_by_type(NULL, "cpu");
731
732 if (cpu) {
733 fp = of_get_property(cpu, name, NULL);
734 if (fp) {
735 found = 1;
736 *val = of_read_ulong(fp, cells);
737 }
738
739 of_node_put(cpu);
740 }
741
742 return found;
743}
744
745static void start_cpu_decrementer(void)
746{
747#ifdef CONFIG_BOOKE
748 unsigned int tcr;
749
750 /* Clear any pending timer interrupts */
751 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
752
753 tcr = mfspr(SPRN_TCR);
754 /*
755 * The watchdog may have already been enabled by u-boot. So leave
756 * TRC[WP] (Watchdog Period) alone.
757 */
758 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
759 tcr |= TCR_DIE; /* Enable decrementer */
760 mtspr(SPRN_TCR, tcr);
761#endif
762}
763
764void __init generic_calibrate_decr(void)
765{
766 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
767
768 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
769 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
770
771 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
772 "(not found)\n");
773 }
774
775 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
776
777 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
778 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
779
780 printk(KERN_ERR "WARNING: Estimating processor frequency "
781 "(not found)\n");
782 }
783}
784
785int update_persistent_clock64(struct timespec64 now)
786{
787 struct rtc_time tm;
788
789 if (!ppc_md.set_rtc_time)
790 return -ENODEV;
791
792 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
793
794 return ppc_md.set_rtc_time(&tm);
795}
796
797static void __read_persistent_clock(struct timespec64 *ts)
798{
799 struct rtc_time tm;
800 static int first = 1;
801
802 ts->tv_nsec = 0;
803 /* XXX this is a little fragile but will work okay in the short term */
804 if (first) {
805 first = 0;
806 if (ppc_md.time_init)
807 timezone_offset = ppc_md.time_init();
808
809 /* get_boot_time() isn't guaranteed to be safe to call late */
810 if (ppc_md.get_boot_time) {
811 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
812 return;
813 }
814 }
815 if (!ppc_md.get_rtc_time) {
816 ts->tv_sec = 0;
817 return;
818 }
819 ppc_md.get_rtc_time(&tm);
820
821 ts->tv_sec = rtc_tm_to_time64(&tm);
822}
823
824void read_persistent_clock64(struct timespec64 *ts)
825{
826 __read_persistent_clock(ts);
827
828 /* Sanitize it in case real time clock is set below EPOCH */
829 if (ts->tv_sec < 0) {
830 ts->tv_sec = 0;
831 ts->tv_nsec = 0;
832 }
833
834}
835
836/* clocksource code */
837static notrace u64 timebase_read(struct clocksource *cs)
838{
839 return (u64)get_tb();
840}
841
842static void __init clocksource_init(void)
843{
844 struct clocksource *clock = &clocksource_timebase;
845
846 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
847 printk(KERN_ERR "clocksource: %s is already registered\n",
848 clock->name);
849 return;
850 }
851
852 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
853 clock->name, clock->mult, clock->shift);
854}
855
856static int decrementer_set_next_event(unsigned long evt,
857 struct clock_event_device *dev)
858{
859 __this_cpu_write(decrementers_next_tb, get_tb() + evt);
860 set_dec_or_work(evt);
861
862 return 0;
863}
864
865static int decrementer_shutdown(struct clock_event_device *dev)
866{
867 __this_cpu_write(decrementers_next_tb, DEC_CLOCKEVENT_STOPPED);
868 set_dec_or_work(decrementer_max);
869
870 return 0;
871}
872
873static void register_decrementer_clockevent(int cpu)
874{
875 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
876
877 *dec = decrementer_clockevent;
878 dec->cpumask = cpumask_of(cpu);
879
880 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
881
882 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
883 dec->name, dec->mult, dec->shift, cpu);
884
885 /* Set values for KVM, see kvm_emulate_dec() */
886 decrementer_clockevent.mult = dec->mult;
887 decrementer_clockevent.shift = dec->shift;
888}
889
890static void enable_large_decrementer(void)
891{
892 if (!cpu_has_feature(CPU_FTR_ARCH_300))
893 return;
894
895 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
896 return;
897
898 /*
899 * If we're running as the hypervisor we need to enable the LD manually
900 * otherwise firmware should have done it for us.
901 */
902 if (cpu_has_feature(CPU_FTR_HVMODE))
903 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
904}
905
906static void __init set_decrementer_max(void)
907{
908 struct device_node *cpu;
909 u32 bits = 32;
910
911 /* Prior to ISAv3 the decrementer is always 32 bit */
912 if (!cpu_has_feature(CPU_FTR_ARCH_300))
913 return;
914
915 cpu = of_find_node_by_type(NULL, "cpu");
916
917 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
918 if (bits > 64 || bits < 32) {
919 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
920 bits = 32;
921 }
922
923 /* calculate the signed maximum given this many bits */
924 decrementer_max = (1ul << (bits - 1)) - 1;
925 }
926
927 of_node_put(cpu);
928
929 pr_info("time_init: %u bit decrementer (max: %llx)\n",
930 bits, decrementer_max);
931}
932
933static void __init init_decrementer_clockevent(void)
934{
935 register_decrementer_clockevent(smp_processor_id());
936 vtime_reset();
937}
938
939void secondary_cpu_time_init(void)
940{
941 /* Enable and test the large decrementer for this cpu */
942 enable_large_decrementer();
943
944 /* Start the decrementer on CPUs that have manual control
945 * such as BookE
946 */
947 start_cpu_decrementer();
948
949 /* FIME: Should make unrelated change to move snapshot_timebase
950 * call here ! */
951 register_decrementer_clockevent(smp_processor_id());
952 vtime_reset();
953}
954
955/*
956 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
957 * result.
958 */
959static __init void div128_by_32(u64 dividend_high, u64 dividend_low,
960 unsigned int divisor, struct div_result *dr)
961{
962 unsigned long a, b, c, d;
963 unsigned long w, x, y, z;
964 u64 ra, rb, rc;
965
966 a = dividend_high >> 32;
967 b = dividend_high & 0xffffffff;
968 c = dividend_low >> 32;
969 d = dividend_low & 0xffffffff;
970
971 w = a / divisor;
972 ra = ((u64)(a - (w * divisor)) << 32) + b;
973
974 rb = ((u64)do_div(ra, divisor) << 32) + c;
975 x = ra;
976
977 rc = ((u64)do_div(rb, divisor) << 32) + d;
978 y = rb;
979
980 do_div(rc, divisor);
981 z = rc;
982
983 dr->result_high = ((u64)w << 32) + x;
984 dr->result_low = ((u64)y << 32) + z;
985}
986
987/* This function is only called on the boot processor */
988void __init time_init(void)
989{
990 struct div_result res;
991 u64 scale;
992 unsigned shift;
993
994 /* Normal PowerPC with timebase register */
995 if (ppc_md.calibrate_decr)
996 ppc_md.calibrate_decr();
997 else
998 generic_calibrate_decr();
999
1000 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1001 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1002 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1003 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1004
1005 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1006 tb_ticks_per_sec = ppc_tb_freq;
1007 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1008
1009 /*
1010 * Compute scale factor for sched_clock.
1011 * The calibrate_decr() function has set tb_ticks_per_sec,
1012 * which is the timebase frequency.
1013 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1014 * the 128-bit result as a 64.64 fixed-point number.
1015 * We then shift that number right until it is less than 1.0,
1016 * giving us the scale factor and shift count to use in
1017 * sched_clock().
1018 */
1019 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1020 scale = res.result_low;
1021 for (shift = 0; res.result_high != 0; ++shift) {
1022 scale = (scale >> 1) | (res.result_high << 63);
1023 res.result_high >>= 1;
1024 }
1025 tb_to_ns_scale = scale;
1026 tb_to_ns_shift = shift;
1027 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1028 boot_tb = get_tb();
1029
1030 /* If platform provided a timezone (pmac), we correct the time */
1031 if (timezone_offset) {
1032 sys_tz.tz_minuteswest = -timezone_offset / 60;
1033 sys_tz.tz_dsttime = 0;
1034 }
1035
1036 vdso_k_arch_data->tb_ticks_per_sec = tb_ticks_per_sec;
1037#ifdef CONFIG_PPC64_PROC_SYSTEMCFG
1038 systemcfg->tb_ticks_per_sec = tb_ticks_per_sec;
1039#endif
1040
1041 /* initialise and enable the large decrementer (if we have one) */
1042 set_decrementer_max();
1043 enable_large_decrementer();
1044
1045 /* Start the decrementer on CPUs that have manual control
1046 * such as BookE
1047 */
1048 start_cpu_decrementer();
1049
1050 /* Register the clocksource */
1051 clocksource_init();
1052
1053 init_decrementer_clockevent();
1054 tick_setup_hrtimer_broadcast();
1055
1056 of_clk_init(NULL);
1057 enable_sched_clock_irqtime();
1058}
1059
1060/* We don't need to calibrate delay, we use the CPU timebase for that */
1061void calibrate_delay(void)
1062{
1063 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1064 * as the number of __delay(1) in a jiffy, so make it so
1065 */
1066 loops_per_jiffy = tb_ticks_per_jiffy;
1067}
1068
1069#if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1070static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1071{
1072 ppc_md.get_rtc_time(tm);
1073 return 0;
1074}
1075
1076static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1077{
1078 if (!ppc_md.set_rtc_time)
1079 return -EOPNOTSUPP;
1080
1081 if (ppc_md.set_rtc_time(tm) < 0)
1082 return -EOPNOTSUPP;
1083
1084 return 0;
1085}
1086
1087static const struct rtc_class_ops rtc_generic_ops = {
1088 .read_time = rtc_generic_get_time,
1089 .set_time = rtc_generic_set_time,
1090};
1091
1092static int __init rtc_init(void)
1093{
1094 struct platform_device *pdev;
1095
1096 if (!ppc_md.get_rtc_time)
1097 return -ENODEV;
1098
1099 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1100 &rtc_generic_ops,
1101 sizeof(rtc_generic_ops));
1102
1103 return PTR_ERR_OR_ZERO(pdev);
1104}
1105
1106device_initcall(rtc_init);
1107#endif