Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Common time routines among all ppc machines.
4 *
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)
9 *
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).
15 *
16 * Speeded up do_gettimeofday by getting rid of references to
17 * xtime (which required locks for consistency). (mikejc@us.ibm.com)
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/kernel.h>
35#include <linux/param.h>
36#include <linux/string.h>
37#include <linux/mm.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>
58
59#include <asm/io.h>
60#include <asm/nvram.h>
61#include <asm/cache.h>
62#include <asm/machdep.h>
63#include <linux/uaccess.h>
64#include <asm/time.h>
65#include <asm/prom.h>
66#include <asm/irq.h>
67#include <asm/div64.h>
68#include <asm/smp.h>
69#include <asm/vdso_datapage.h>
70#include <asm/firmware.h>
71#include <asm/asm-prototypes.h>
72
73/* powerpc clocksource/clockevent code */
74
75#include <linux/clockchips.h>
76#include <linux/timekeeper_internal.h>
77
78static u64 rtc_read(struct clocksource *);
79static struct clocksource clocksource_rtc = {
80 .name = "rtc",
81 .rating = 400,
82 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
83 .mask = CLOCKSOURCE_MASK(64),
84 .read = rtc_read,
85};
86
87static u64 timebase_read(struct clocksource *);
88static struct clocksource clocksource_timebase = {
89 .name = "timebase",
90 .rating = 400,
91 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
92 .mask = CLOCKSOURCE_MASK(64),
93 .read = timebase_read,
94};
95
96#define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
97u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
98
99static int decrementer_set_next_event(unsigned long evt,
100 struct clock_event_device *dev);
101static int decrementer_shutdown(struct clock_event_device *evt);
102
103struct clock_event_device decrementer_clockevent = {
104 .name = "decrementer",
105 .rating = 200,
106 .irq = 0,
107 .set_next_event = decrementer_set_next_event,
108 .set_state_oneshot_stopped = decrementer_shutdown,
109 .set_state_shutdown = decrementer_shutdown,
110 .tick_resume = decrementer_shutdown,
111 .features = CLOCK_EVT_FEAT_ONESHOT |
112 CLOCK_EVT_FEAT_C3STOP,
113};
114EXPORT_SYMBOL(decrementer_clockevent);
115
116DEFINE_PER_CPU(u64, decrementers_next_tb);
117static DEFINE_PER_CPU(struct clock_event_device, decrementers);
118
119#define XSEC_PER_SEC (1024*1024)
120
121#ifdef CONFIG_PPC64
122#define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
123#else
124/* compute ((xsec << 12) * max) >> 32 */
125#define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
126#endif
127
128unsigned long tb_ticks_per_jiffy;
129unsigned long tb_ticks_per_usec = 100; /* sane default */
130EXPORT_SYMBOL(tb_ticks_per_usec);
131unsigned long tb_ticks_per_sec;
132EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
133
134DEFINE_SPINLOCK(rtc_lock);
135EXPORT_SYMBOL_GPL(rtc_lock);
136
137static u64 tb_to_ns_scale __read_mostly;
138static unsigned tb_to_ns_shift __read_mostly;
139static u64 boot_tb __read_mostly;
140
141extern struct timezone sys_tz;
142static long timezone_offset;
143
144unsigned long ppc_proc_freq;
145EXPORT_SYMBOL_GPL(ppc_proc_freq);
146unsigned long ppc_tb_freq;
147EXPORT_SYMBOL_GPL(ppc_tb_freq);
148
149bool tb_invalid;
150
151#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
152/*
153 * Factor for converting from cputime_t (timebase ticks) to
154 * microseconds. This is stored as 0.64 fixed-point binary fraction.
155 */
156u64 __cputime_usec_factor;
157EXPORT_SYMBOL(__cputime_usec_factor);
158
159#ifdef CONFIG_PPC_SPLPAR
160void (*dtl_consumer)(struct dtl_entry *, u64);
161#endif
162
163static void calc_cputime_factors(void)
164{
165 struct div_result res;
166
167 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
168 __cputime_usec_factor = res.result_low;
169}
170
171/*
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.
174 */
175static inline unsigned long read_spurr(unsigned long tb)
176{
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);
181 return tb;
182}
183
184#ifdef CONFIG_PPC_SPLPAR
185
186#include <asm/dtl.h>
187
188/*
189 * Scan the dispatch trace log and count up the stolen time.
190 * Should be called with interrupts disabled.
191 */
192static u64 scan_dispatch_log(u64 stop_tb)
193{
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;
198 u64 tb_delta;
199 u64 stolen = 0;
200 u64 dtb;
201
202 if (!dtl)
203 return 0;
204
205 if (i == be64_to_cpu(vpa->dtl_idx))
206 return 0;
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);
211 barrier();
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);
216 continue;
217 }
218 if (dtb > stop_tb)
219 break;
220 if (dtl_consumer)
221 dtl_consumer(dtl, i);
222 stolen += tb_delta;
223 ++i;
224 ++dtl;
225 if (dtl == dtl_end)
226 dtl = local_paca->dispatch_log;
227 }
228 local_paca->dtl_ridx = i;
229 local_paca->dtl_curr = dtl;
230 return stolen;
231}
232
233/*
234 * Accumulate stolen time by scanning the dispatch trace log.
235 * Called on entry from user mode.
236 */
237void notrace accumulate_stolen_time(void)
238{
239 u64 sst, ust;
240 unsigned long save_irq_soft_mask = irq_soft_mask_return();
241 struct cpu_accounting_data *acct = &local_paca->accounting;
242
243 /* We are called early in the exception entry, before
244 * soft/hard_enabled are sync'ed to the expected state
245 * for the exception. We are hard disabled but the PACA
246 * needs to reflect that so various debug stuff doesn't
247 * complain
248 */
249 irq_soft_mask_set(IRQS_DISABLED);
250
251 sst = scan_dispatch_log(acct->starttime_user);
252 ust = scan_dispatch_log(acct->starttime);
253 acct->stime -= sst;
254 acct->utime -= ust;
255 acct->steal_time += ust + sst;
256
257 irq_soft_mask_set(save_irq_soft_mask);
258}
259
260static inline u64 calculate_stolen_time(u64 stop_tb)
261{
262 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
263 return 0;
264
265 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
266 return scan_dispatch_log(stop_tb);
267
268 return 0;
269}
270
271#else /* CONFIG_PPC_SPLPAR */
272static inline u64 calculate_stolen_time(u64 stop_tb)
273{
274 return 0;
275}
276
277#endif /* CONFIG_PPC_SPLPAR */
278
279/*
280 * Account time for a transition between system, hard irq
281 * or soft irq state.
282 */
283static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
284 unsigned long now, unsigned long stime)
285{
286 unsigned long stime_scaled = 0;
287#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
288 unsigned long nowscaled, deltascaled;
289 unsigned long utime, utime_scaled;
290
291 nowscaled = read_spurr(now);
292 deltascaled = nowscaled - acct->startspurr;
293 acct->startspurr = nowscaled;
294 utime = acct->utime - acct->utime_sspurr;
295 acct->utime_sspurr = acct->utime;
296
297 /*
298 * Because we don't read the SPURR on every kernel entry/exit,
299 * deltascaled includes both user and system SPURR ticks.
300 * Apportion these ticks to system SPURR ticks and user
301 * SPURR ticks in the same ratio as the system time (delta)
302 * and user time (udelta) values obtained from the timebase
303 * over the same interval. The system ticks get accounted here;
304 * the user ticks get saved up in paca->user_time_scaled to be
305 * used by account_process_tick.
306 */
307 stime_scaled = stime;
308 utime_scaled = utime;
309 if (deltascaled != stime + utime) {
310 if (utime) {
311 stime_scaled = deltascaled * stime / (stime + utime);
312 utime_scaled = deltascaled - stime_scaled;
313 } else {
314 stime_scaled = deltascaled;
315 }
316 }
317 acct->utime_scaled += utime_scaled;
318#endif
319
320 return stime_scaled;
321}
322
323static unsigned long vtime_delta(struct task_struct *tsk,
324 unsigned long *stime_scaled,
325 unsigned long *steal_time)
326{
327 unsigned long now, stime;
328 struct cpu_accounting_data *acct = get_accounting(tsk);
329
330 WARN_ON_ONCE(!irqs_disabled());
331
332 now = mftb();
333 stime = now - acct->starttime;
334 acct->starttime = now;
335
336 *stime_scaled = vtime_delta_scaled(acct, now, stime);
337
338 *steal_time = calculate_stolen_time(now);
339
340 return stime;
341}
342
343void vtime_account_kernel(struct task_struct *tsk)
344{
345 unsigned long stime, stime_scaled, steal_time;
346 struct cpu_accounting_data *acct = get_accounting(tsk);
347
348 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
349
350 stime -= min(stime, steal_time);
351 acct->steal_time += steal_time;
352
353 if ((tsk->flags & PF_VCPU) && !irq_count()) {
354 acct->gtime += stime;
355#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
356 acct->utime_scaled += stime_scaled;
357#endif
358 } else {
359 if (hardirq_count())
360 acct->hardirq_time += stime;
361 else if (in_serving_softirq())
362 acct->softirq_time += stime;
363 else
364 acct->stime += stime;
365
366#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
367 acct->stime_scaled += stime_scaled;
368#endif
369 }
370}
371EXPORT_SYMBOL_GPL(vtime_account_kernel);
372
373void vtime_account_idle(struct task_struct *tsk)
374{
375 unsigned long stime, stime_scaled, steal_time;
376 struct cpu_accounting_data *acct = get_accounting(tsk);
377
378 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
379 acct->idle_time += stime + steal_time;
380}
381
382static void vtime_flush_scaled(struct task_struct *tsk,
383 struct cpu_accounting_data *acct)
384{
385#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
386 if (acct->utime_scaled)
387 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
388 if (acct->stime_scaled)
389 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
390
391 acct->utime_scaled = 0;
392 acct->utime_sspurr = 0;
393 acct->stime_scaled = 0;
394#endif
395}
396
397/*
398 * Account the whole cputime accumulated in the paca
399 * Must be called with interrupts disabled.
400 * Assumes that vtime_account_kernel/idle() has been called
401 * recently (i.e. since the last entry from usermode) so that
402 * get_paca()->user_time_scaled is up to date.
403 */
404void vtime_flush(struct task_struct *tsk)
405{
406 struct cpu_accounting_data *acct = get_accounting(tsk);
407
408 if (acct->utime)
409 account_user_time(tsk, cputime_to_nsecs(acct->utime));
410
411 if (acct->gtime)
412 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
413
414 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
415 account_steal_time(cputime_to_nsecs(acct->steal_time));
416 acct->steal_time = 0;
417 }
418
419 if (acct->idle_time)
420 account_idle_time(cputime_to_nsecs(acct->idle_time));
421
422 if (acct->stime)
423 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
424 CPUTIME_SYSTEM);
425
426 if (acct->hardirq_time)
427 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
428 CPUTIME_IRQ);
429 if (acct->softirq_time)
430 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
431 CPUTIME_SOFTIRQ);
432
433 vtime_flush_scaled(tsk, acct);
434
435 acct->utime = 0;
436 acct->gtime = 0;
437 acct->idle_time = 0;
438 acct->stime = 0;
439 acct->hardirq_time = 0;
440 acct->softirq_time = 0;
441}
442
443#else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
444#define calc_cputime_factors()
445#endif
446
447void __delay(unsigned long loops)
448{
449 unsigned long start;
450 int diff;
451
452 spin_begin();
453 if (__USE_RTC()) {
454 start = get_rtcl();
455 do {
456 /* the RTCL register wraps at 1000000000 */
457 diff = get_rtcl() - start;
458 if (diff < 0)
459 diff += 1000000000;
460 spin_cpu_relax();
461 } while (diff < loops);
462 } else if (tb_invalid) {
463 /*
464 * TB is in error state and isn't ticking anymore.
465 * HMI handler was unable to recover from TB error.
466 * Return immediately, so that kernel won't get stuck here.
467 */
468 spin_cpu_relax();
469 } else {
470 start = get_tbl();
471 while (get_tbl() - start < loops)
472 spin_cpu_relax();
473 }
474 spin_end();
475}
476EXPORT_SYMBOL(__delay);
477
478void udelay(unsigned long usecs)
479{
480 __delay(tb_ticks_per_usec * usecs);
481}
482EXPORT_SYMBOL(udelay);
483
484#ifdef CONFIG_SMP
485unsigned long profile_pc(struct pt_regs *regs)
486{
487 unsigned long pc = instruction_pointer(regs);
488
489 if (in_lock_functions(pc))
490 return regs->link;
491
492 return pc;
493}
494EXPORT_SYMBOL(profile_pc);
495#endif
496
497#ifdef CONFIG_IRQ_WORK
498
499/*
500 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
501 */
502#ifdef CONFIG_PPC64
503static inline unsigned long test_irq_work_pending(void)
504{
505 unsigned long x;
506
507 asm volatile("lbz %0,%1(13)"
508 : "=r" (x)
509 : "i" (offsetof(struct paca_struct, irq_work_pending)));
510 return x;
511}
512
513static inline void set_irq_work_pending_flag(void)
514{
515 asm volatile("stb %0,%1(13)" : :
516 "r" (1),
517 "i" (offsetof(struct paca_struct, irq_work_pending)));
518}
519
520static inline void clear_irq_work_pending(void)
521{
522 asm volatile("stb %0,%1(13)" : :
523 "r" (0),
524 "i" (offsetof(struct paca_struct, irq_work_pending)));
525}
526
527#else /* 32-bit */
528
529DEFINE_PER_CPU(u8, irq_work_pending);
530
531#define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
532#define test_irq_work_pending() __this_cpu_read(irq_work_pending)
533#define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
534
535#endif /* 32 vs 64 bit */
536
537void arch_irq_work_raise(void)
538{
539 /*
540 * 64-bit code that uses irq soft-mask can just cause an immediate
541 * interrupt here that gets soft masked, if this is called under
542 * local_irq_disable(). It might be possible to prevent that happening
543 * by noticing interrupts are disabled and setting decrementer pending
544 * to be replayed when irqs are enabled. The problem there is that
545 * tracing can call irq_work_raise, including in code that does low
546 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
547 * which could get tangled up if we're messing with the same state
548 * here.
549 */
550 preempt_disable();
551 set_irq_work_pending_flag();
552 set_dec(1);
553 preempt_enable();
554}
555
556#else /* CONFIG_IRQ_WORK */
557
558#define test_irq_work_pending() 0
559#define clear_irq_work_pending()
560
561#endif /* CONFIG_IRQ_WORK */
562
563/*
564 * timer_interrupt - gets called when the decrementer overflows,
565 * with interrupts disabled.
566 */
567void timer_interrupt(struct pt_regs *regs)
568{
569 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
570 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
571 struct pt_regs *old_regs;
572 u64 now;
573
574 /* Some implementations of hotplug will get timer interrupts while
575 * offline, just ignore these and we also need to set
576 * decrementers_next_tb as MAX to make sure __check_irq_replay
577 * don't replay timer interrupt when return, otherwise we'll trap
578 * here infinitely :(
579 */
580 if (unlikely(!cpu_online(smp_processor_id()))) {
581 *next_tb = ~(u64)0;
582 set_dec(decrementer_max);
583 return;
584 }
585
586 /* Ensure a positive value is written to the decrementer, or else
587 * some CPUs will continue to take decrementer exceptions. When the
588 * PPC_WATCHDOG (decrementer based) is configured, keep this at most
589 * 31 bits, which is about 4 seconds on most systems, which gives
590 * the watchdog a chance of catching timer interrupt hard lockups.
591 */
592 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
593 set_dec(0x7fffffff);
594 else
595 set_dec(decrementer_max);
596
597 /* Conditionally hard-enable interrupts now that the DEC has been
598 * bumped to its maximum value
599 */
600 may_hard_irq_enable();
601
602
603#if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
604 if (atomic_read(&ppc_n_lost_interrupts) != 0)
605 do_IRQ(regs);
606#endif
607
608 old_regs = set_irq_regs(regs);
609 irq_enter();
610 trace_timer_interrupt_entry(regs);
611
612 if (test_irq_work_pending()) {
613 clear_irq_work_pending();
614 irq_work_run();
615 }
616
617 now = get_tb_or_rtc();
618 if (now >= *next_tb) {
619 *next_tb = ~(u64)0;
620 if (evt->event_handler)
621 evt->event_handler(evt);
622 __this_cpu_inc(irq_stat.timer_irqs_event);
623 } else {
624 now = *next_tb - now;
625 if (now <= decrementer_max)
626 set_dec(now);
627 /* We may have raced with new irq work */
628 if (test_irq_work_pending())
629 set_dec(1);
630 __this_cpu_inc(irq_stat.timer_irqs_others);
631 }
632
633 trace_timer_interrupt_exit(regs);
634 irq_exit();
635 set_irq_regs(old_regs);
636}
637EXPORT_SYMBOL(timer_interrupt);
638
639#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
640void timer_broadcast_interrupt(void)
641{
642 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
643
644 *next_tb = ~(u64)0;
645 tick_receive_broadcast();
646 __this_cpu_inc(irq_stat.broadcast_irqs_event);
647}
648#endif
649
650#ifdef CONFIG_SUSPEND
651static void generic_suspend_disable_irqs(void)
652{
653 /* Disable the decrementer, so that it doesn't interfere
654 * with suspending.
655 */
656
657 set_dec(decrementer_max);
658 local_irq_disable();
659 set_dec(decrementer_max);
660}
661
662static void generic_suspend_enable_irqs(void)
663{
664 local_irq_enable();
665}
666
667/* Overrides the weak version in kernel/power/main.c */
668void arch_suspend_disable_irqs(void)
669{
670 if (ppc_md.suspend_disable_irqs)
671 ppc_md.suspend_disable_irqs();
672 generic_suspend_disable_irqs();
673}
674
675/* Overrides the weak version in kernel/power/main.c */
676void arch_suspend_enable_irqs(void)
677{
678 generic_suspend_enable_irqs();
679 if (ppc_md.suspend_enable_irqs)
680 ppc_md.suspend_enable_irqs();
681}
682#endif
683
684unsigned long long tb_to_ns(unsigned long long ticks)
685{
686 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
687}
688EXPORT_SYMBOL_GPL(tb_to_ns);
689
690/*
691 * Scheduler clock - returns current time in nanosec units.
692 *
693 * Note: mulhdu(a, b) (multiply high double unsigned) returns
694 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
695 * are 64-bit unsigned numbers.
696 */
697notrace unsigned long long sched_clock(void)
698{
699 if (__USE_RTC())
700 return get_rtc();
701 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
702}
703
704
705#ifdef CONFIG_PPC_PSERIES
706
707/*
708 * Running clock - attempts to give a view of time passing for a virtualised
709 * kernels.
710 * Uses the VTB register if available otherwise a next best guess.
711 */
712unsigned long long running_clock(void)
713{
714 /*
715 * Don't read the VTB as a host since KVM does not switch in host
716 * timebase into the VTB when it takes a guest off the CPU, reading the
717 * VTB would result in reading 'last switched out' guest VTB.
718 *
719 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
720 * would be unsafe to rely only on the #ifdef above.
721 */
722 if (firmware_has_feature(FW_FEATURE_LPAR) &&
723 cpu_has_feature(CPU_FTR_ARCH_207S))
724 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
725
726 /*
727 * This is a next best approximation without a VTB.
728 * On a host which is running bare metal there should never be any stolen
729 * time and on a host which doesn't do any virtualisation TB *should* equal
730 * VTB so it makes no difference anyway.
731 */
732 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
733}
734#endif
735
736static int __init get_freq(char *name, int cells, unsigned long *val)
737{
738 struct device_node *cpu;
739 const __be32 *fp;
740 int found = 0;
741
742 /* The cpu node should have timebase and clock frequency properties */
743 cpu = of_find_node_by_type(NULL, "cpu");
744
745 if (cpu) {
746 fp = of_get_property(cpu, name, NULL);
747 if (fp) {
748 found = 1;
749 *val = of_read_ulong(fp, cells);
750 }
751
752 of_node_put(cpu);
753 }
754
755 return found;
756}
757
758static void start_cpu_decrementer(void)
759{
760#if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
761 unsigned int tcr;
762
763 /* Clear any pending timer interrupts */
764 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
765
766 tcr = mfspr(SPRN_TCR);
767 /*
768 * The watchdog may have already been enabled by u-boot. So leave
769 * TRC[WP] (Watchdog Period) alone.
770 */
771 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
772 tcr |= TCR_DIE; /* Enable decrementer */
773 mtspr(SPRN_TCR, tcr);
774#endif
775}
776
777void __init generic_calibrate_decr(void)
778{
779 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
780
781 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
782 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
783
784 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
785 "(not found)\n");
786 }
787
788 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
789
790 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
791 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
792
793 printk(KERN_ERR "WARNING: Estimating processor frequency "
794 "(not found)\n");
795 }
796}
797
798int update_persistent_clock64(struct timespec64 now)
799{
800 struct rtc_time tm;
801
802 if (!ppc_md.set_rtc_time)
803 return -ENODEV;
804
805 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
806
807 return ppc_md.set_rtc_time(&tm);
808}
809
810static void __read_persistent_clock(struct timespec64 *ts)
811{
812 struct rtc_time tm;
813 static int first = 1;
814
815 ts->tv_nsec = 0;
816 /* XXX this is a litle fragile but will work okay in the short term */
817 if (first) {
818 first = 0;
819 if (ppc_md.time_init)
820 timezone_offset = ppc_md.time_init();
821
822 /* get_boot_time() isn't guaranteed to be safe to call late */
823 if (ppc_md.get_boot_time) {
824 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
825 return;
826 }
827 }
828 if (!ppc_md.get_rtc_time) {
829 ts->tv_sec = 0;
830 return;
831 }
832 ppc_md.get_rtc_time(&tm);
833
834 ts->tv_sec = rtc_tm_to_time64(&tm);
835}
836
837void read_persistent_clock64(struct timespec64 *ts)
838{
839 __read_persistent_clock(ts);
840
841 /* Sanitize it in case real time clock is set below EPOCH */
842 if (ts->tv_sec < 0) {
843 ts->tv_sec = 0;
844 ts->tv_nsec = 0;
845 }
846
847}
848
849/* clocksource code */
850static notrace u64 rtc_read(struct clocksource *cs)
851{
852 return (u64)get_rtc();
853}
854
855static notrace u64 timebase_read(struct clocksource *cs)
856{
857 return (u64)get_tb();
858}
859
860
861void update_vsyscall(struct timekeeper *tk)
862{
863 struct timespec64 xt;
864 struct clocksource *clock = tk->tkr_mono.clock;
865 u32 mult = tk->tkr_mono.mult;
866 u32 shift = tk->tkr_mono.shift;
867 u64 cycle_last = tk->tkr_mono.cycle_last;
868 u64 new_tb_to_xs, new_stamp_xsec;
869 u64 frac_sec;
870
871 if (clock != &clocksource_timebase)
872 return;
873
874 xt.tv_sec = tk->xtime_sec;
875 xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
876
877 /* Make userspace gettimeofday spin until we're done. */
878 ++vdso_data->tb_update_count;
879 smp_mb();
880
881 /*
882 * This computes ((2^20 / 1e9) * mult) >> shift as a
883 * 0.64 fixed-point fraction.
884 * The computation in the else clause below won't overflow
885 * (as long as the timebase frequency is >= 1.049 MHz)
886 * but loses precision because we lose the low bits of the constant
887 * in the shift. Note that 19342813113834067 ~= 2^(20+64) / 1e9.
888 * For a shift of 24 the error is about 0.5e-9, or about 0.5ns
889 * over a second. (Shift values are usually 22, 23 or 24.)
890 * For high frequency clocks such as the 512MHz timebase clock
891 * on POWER[6789], the mult value is small (e.g. 32768000)
892 * and so we can shift the constant by 16 initially
893 * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the
894 * remaining shifts after the multiplication, which gives a
895 * more accurate result (e.g. with mult = 32768000, shift = 24,
896 * the error is only about 1.2e-12, or 0.7ns over 10 minutes).
897 */
898 if (mult <= 62500000 && clock->shift >= 16)
899 new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16);
900 else
901 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
902
903 /*
904 * Compute the fractional second in units of 2^-32 seconds.
905 * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift
906 * in nanoseconds, so multiplying that by 2^32 / 1e9 gives
907 * it in units of 2^-32 seconds.
908 * We assume shift <= 32 because clocks_calc_mult_shift()
909 * generates shift values in the range 0 - 32.
910 */
911 frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift);
912 do_div(frac_sec, NSEC_PER_SEC);
913
914 /*
915 * Work out new stamp_xsec value for any legacy users of systemcfg.
916 * stamp_xsec is in units of 2^-20 seconds.
917 */
918 new_stamp_xsec = frac_sec >> 12;
919 new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC;
920
921 /*
922 * tb_update_count is used to allow the userspace gettimeofday code
923 * to assure itself that it sees a consistent view of the tb_to_xs and
924 * stamp_xsec variables. It reads the tb_update_count, then reads
925 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
926 * the two values of tb_update_count match and are even then the
927 * tb_to_xs and stamp_xsec values are consistent. If not, then it
928 * loops back and reads them again until this criteria is met.
929 */
930 vdso_data->tb_orig_stamp = cycle_last;
931 vdso_data->stamp_xsec = new_stamp_xsec;
932 vdso_data->tb_to_xs = new_tb_to_xs;
933 vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec;
934 vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec;
935 vdso_data->stamp_xtime_sec = xt.tv_sec;
936 vdso_data->stamp_xtime_nsec = xt.tv_nsec;
937 vdso_data->stamp_sec_fraction = frac_sec;
938 vdso_data->hrtimer_res = hrtimer_resolution;
939 smp_wmb();
940 ++(vdso_data->tb_update_count);
941}
942
943void update_vsyscall_tz(void)
944{
945 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
946 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
947}
948
949static void __init clocksource_init(void)
950{
951 struct clocksource *clock;
952
953 if (__USE_RTC())
954 clock = &clocksource_rtc;
955 else
956 clock = &clocksource_timebase;
957
958 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
959 printk(KERN_ERR "clocksource: %s is already registered\n",
960 clock->name);
961 return;
962 }
963
964 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
965 clock->name, clock->mult, clock->shift);
966}
967
968static int decrementer_set_next_event(unsigned long evt,
969 struct clock_event_device *dev)
970{
971 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
972 set_dec(evt);
973
974 /* We may have raced with new irq work */
975 if (test_irq_work_pending())
976 set_dec(1);
977
978 return 0;
979}
980
981static int decrementer_shutdown(struct clock_event_device *dev)
982{
983 decrementer_set_next_event(decrementer_max, dev);
984 return 0;
985}
986
987static void register_decrementer_clockevent(int cpu)
988{
989 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
990
991 *dec = decrementer_clockevent;
992 dec->cpumask = cpumask_of(cpu);
993
994 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
995
996 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
997 dec->name, dec->mult, dec->shift, cpu);
998
999 /* Set values for KVM, see kvm_emulate_dec() */
1000 decrementer_clockevent.mult = dec->mult;
1001 decrementer_clockevent.shift = dec->shift;
1002}
1003
1004static void enable_large_decrementer(void)
1005{
1006 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1007 return;
1008
1009 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
1010 return;
1011
1012 /*
1013 * If we're running as the hypervisor we need to enable the LD manually
1014 * otherwise firmware should have done it for us.
1015 */
1016 if (cpu_has_feature(CPU_FTR_HVMODE))
1017 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
1018}
1019
1020static void __init set_decrementer_max(void)
1021{
1022 struct device_node *cpu;
1023 u32 bits = 32;
1024
1025 /* Prior to ISAv3 the decrementer is always 32 bit */
1026 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1027 return;
1028
1029 cpu = of_find_node_by_type(NULL, "cpu");
1030
1031 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
1032 if (bits > 64 || bits < 32) {
1033 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
1034 bits = 32;
1035 }
1036
1037 /* calculate the signed maximum given this many bits */
1038 decrementer_max = (1ul << (bits - 1)) - 1;
1039 }
1040
1041 of_node_put(cpu);
1042
1043 pr_info("time_init: %u bit decrementer (max: %llx)\n",
1044 bits, decrementer_max);
1045}
1046
1047static void __init init_decrementer_clockevent(void)
1048{
1049 register_decrementer_clockevent(smp_processor_id());
1050}
1051
1052void secondary_cpu_time_init(void)
1053{
1054 /* Enable and test the large decrementer for this cpu */
1055 enable_large_decrementer();
1056
1057 /* Start the decrementer on CPUs that have manual control
1058 * such as BookE
1059 */
1060 start_cpu_decrementer();
1061
1062 /* FIME: Should make unrelatred change to move snapshot_timebase
1063 * call here ! */
1064 register_decrementer_clockevent(smp_processor_id());
1065}
1066
1067/* This function is only called on the boot processor */
1068void __init time_init(void)
1069{
1070 struct div_result res;
1071 u64 scale;
1072 unsigned shift;
1073
1074 if (__USE_RTC()) {
1075 /* 601 processor: dec counts down by 128 every 128ns */
1076 ppc_tb_freq = 1000000000;
1077 } else {
1078 /* Normal PowerPC with timebase register */
1079 ppc_md.calibrate_decr();
1080 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1081 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1082 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1083 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1084 }
1085
1086 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1087 tb_ticks_per_sec = ppc_tb_freq;
1088 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1089 calc_cputime_factors();
1090
1091 /*
1092 * Compute scale factor for sched_clock.
1093 * The calibrate_decr() function has set tb_ticks_per_sec,
1094 * which is the timebase frequency.
1095 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1096 * the 128-bit result as a 64.64 fixed-point number.
1097 * We then shift that number right until it is less than 1.0,
1098 * giving us the scale factor and shift count to use in
1099 * sched_clock().
1100 */
1101 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1102 scale = res.result_low;
1103 for (shift = 0; res.result_high != 0; ++shift) {
1104 scale = (scale >> 1) | (res.result_high << 63);
1105 res.result_high >>= 1;
1106 }
1107 tb_to_ns_scale = scale;
1108 tb_to_ns_shift = shift;
1109 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1110 boot_tb = get_tb_or_rtc();
1111
1112 /* If platform provided a timezone (pmac), we correct the time */
1113 if (timezone_offset) {
1114 sys_tz.tz_minuteswest = -timezone_offset / 60;
1115 sys_tz.tz_dsttime = 0;
1116 }
1117
1118 vdso_data->tb_update_count = 0;
1119 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1120
1121 /* initialise and enable the large decrementer (if we have one) */
1122 set_decrementer_max();
1123 enable_large_decrementer();
1124
1125 /* Start the decrementer on CPUs that have manual control
1126 * such as BookE
1127 */
1128 start_cpu_decrementer();
1129
1130 /* Register the clocksource */
1131 clocksource_init();
1132
1133 init_decrementer_clockevent();
1134 tick_setup_hrtimer_broadcast();
1135
1136 of_clk_init(NULL);
1137}
1138
1139/*
1140 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1141 * result.
1142 */
1143void div128_by_32(u64 dividend_high, u64 dividend_low,
1144 unsigned divisor, struct div_result *dr)
1145{
1146 unsigned long a, b, c, d;
1147 unsigned long w, x, y, z;
1148 u64 ra, rb, rc;
1149
1150 a = dividend_high >> 32;
1151 b = dividend_high & 0xffffffff;
1152 c = dividend_low >> 32;
1153 d = dividend_low & 0xffffffff;
1154
1155 w = a / divisor;
1156 ra = ((u64)(a - (w * divisor)) << 32) + b;
1157
1158 rb = ((u64) do_div(ra, divisor) << 32) + c;
1159 x = ra;
1160
1161 rc = ((u64) do_div(rb, divisor) << 32) + d;
1162 y = rb;
1163
1164 do_div(rc, divisor);
1165 z = rc;
1166
1167 dr->result_high = ((u64)w << 32) + x;
1168 dr->result_low = ((u64)y << 32) + z;
1169
1170}
1171
1172/* We don't need to calibrate delay, we use the CPU timebase for that */
1173void calibrate_delay(void)
1174{
1175 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1176 * as the number of __delay(1) in a jiffy, so make it so
1177 */
1178 loops_per_jiffy = tb_ticks_per_jiffy;
1179}
1180
1181#if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1182static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1183{
1184 ppc_md.get_rtc_time(tm);
1185 return 0;
1186}
1187
1188static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1189{
1190 if (!ppc_md.set_rtc_time)
1191 return -EOPNOTSUPP;
1192
1193 if (ppc_md.set_rtc_time(tm) < 0)
1194 return -EOPNOTSUPP;
1195
1196 return 0;
1197}
1198
1199static const struct rtc_class_ops rtc_generic_ops = {
1200 .read_time = rtc_generic_get_time,
1201 .set_time = rtc_generic_set_time,
1202};
1203
1204static int __init rtc_init(void)
1205{
1206 struct platform_device *pdev;
1207
1208 if (!ppc_md.get_rtc_time)
1209 return -ENODEV;
1210
1211 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1212 &rtc_generic_ops,
1213 sizeof(rtc_generic_ops));
1214
1215 return PTR_ERR_OR_ZERO(pdev);
1216}
1217
1218device_initcall(rtc_init);
1219#endif
1/*
2 * Common time routines among all ppc machines.
3 *
4 * Written by Cort Dougan (cort@cs.nmt.edu) to merge
5 * Paul Mackerras' version and mine for PReP and Pmac.
6 * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net).
7 * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com)
8 *
9 * First round of bugfixes by Gabriel Paubert (paubert@iram.es)
10 * to make clock more stable (2.4.0-test5). The only thing
11 * that this code assumes is that the timebases have been synchronized
12 * by firmware on SMP and are never stopped (never do sleep
13 * on SMP then, nap and doze are OK).
14 *
15 * Speeded up do_gettimeofday by getting rid of references to
16 * xtime (which required locks for consistency). (mikejc@us.ibm.com)
17 *
18 * TODO (not necessarily in this file):
19 * - improve precision and reproducibility of timebase frequency
20 * measurement at boot time.
21 * - for astronomical applications: add a new function to get
22 * non ambiguous timestamps even around leap seconds. This needs
23 * a new timestamp format and a good name.
24 *
25 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
26 * "A Kernel Model for Precision Timekeeping" by Dave Mills
27 *
28 * This program is free software; you can redistribute it and/or
29 * modify it under the terms of the GNU General Public License
30 * as published by the Free Software Foundation; either version
31 * 2 of the License, or (at your option) any later version.
32 */
33
34#include <linux/errno.h>
35#include <linux/export.h>
36#include <linux/sched.h>
37#include <linux/kernel.h>
38#include <linux/param.h>
39#include <linux/string.h>
40#include <linux/mm.h>
41#include <linux/interrupt.h>
42#include <linux/timex.h>
43#include <linux/kernel_stat.h>
44#include <linux/time.h>
45#include <linux/clockchips.h>
46#include <linux/init.h>
47#include <linux/profile.h>
48#include <linux/cpu.h>
49#include <linux/security.h>
50#include <linux/percpu.h>
51#include <linux/rtc.h>
52#include <linux/jiffies.h>
53#include <linux/posix-timers.h>
54#include <linux/irq.h>
55#include <linux/delay.h>
56#include <linux/irq_work.h>
57#include <linux/clk-provider.h>
58#include <linux/suspend.h>
59#include <linux/rtc.h>
60#include <asm/trace.h>
61
62#include <asm/io.h>
63#include <asm/processor.h>
64#include <asm/nvram.h>
65#include <asm/cache.h>
66#include <asm/machdep.h>
67#include <linux/uaccess.h>
68#include <asm/time.h>
69#include <asm/prom.h>
70#include <asm/irq.h>
71#include <asm/div64.h>
72#include <asm/smp.h>
73#include <asm/vdso_datapage.h>
74#include <asm/firmware.h>
75#include <asm/cputime.h>
76#include <asm/asm-prototypes.h>
77
78/* powerpc clocksource/clockevent code */
79
80#include <linux/clockchips.h>
81#include <linux/timekeeper_internal.h>
82
83static u64 rtc_read(struct clocksource *);
84static struct clocksource clocksource_rtc = {
85 .name = "rtc",
86 .rating = 400,
87 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
88 .mask = CLOCKSOURCE_MASK(64),
89 .read = rtc_read,
90};
91
92static u64 timebase_read(struct clocksource *);
93static struct clocksource clocksource_timebase = {
94 .name = "timebase",
95 .rating = 400,
96 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
97 .mask = CLOCKSOURCE_MASK(64),
98 .read = timebase_read,
99};
100
101#define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
102u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
103
104static int decrementer_set_next_event(unsigned long evt,
105 struct clock_event_device *dev);
106static int decrementer_shutdown(struct clock_event_device *evt);
107
108struct clock_event_device decrementer_clockevent = {
109 .name = "decrementer",
110 .rating = 200,
111 .irq = 0,
112 .set_next_event = decrementer_set_next_event,
113 .set_state_shutdown = decrementer_shutdown,
114 .tick_resume = decrementer_shutdown,
115 .features = CLOCK_EVT_FEAT_ONESHOT |
116 CLOCK_EVT_FEAT_C3STOP,
117};
118EXPORT_SYMBOL(decrementer_clockevent);
119
120DEFINE_PER_CPU(u64, decrementers_next_tb);
121static DEFINE_PER_CPU(struct clock_event_device, decrementers);
122
123#define XSEC_PER_SEC (1024*1024)
124
125#ifdef CONFIG_PPC64
126#define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
127#else
128/* compute ((xsec << 12) * max) >> 32 */
129#define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
130#endif
131
132unsigned long tb_ticks_per_jiffy;
133unsigned long tb_ticks_per_usec = 100; /* sane default */
134EXPORT_SYMBOL(tb_ticks_per_usec);
135unsigned long tb_ticks_per_sec;
136EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
137
138DEFINE_SPINLOCK(rtc_lock);
139EXPORT_SYMBOL_GPL(rtc_lock);
140
141static u64 tb_to_ns_scale __read_mostly;
142static unsigned tb_to_ns_shift __read_mostly;
143static u64 boot_tb __read_mostly;
144
145extern struct timezone sys_tz;
146static long timezone_offset;
147
148unsigned long ppc_proc_freq;
149EXPORT_SYMBOL_GPL(ppc_proc_freq);
150unsigned long ppc_tb_freq;
151EXPORT_SYMBOL_GPL(ppc_tb_freq);
152
153#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
154/*
155 * Factors for converting from cputime_t (timebase ticks) to
156 * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
157 * These are all stored as 0.64 fixed-point binary fractions.
158 */
159u64 __cputime_jiffies_factor;
160EXPORT_SYMBOL(__cputime_jiffies_factor);
161u64 __cputime_usec_factor;
162EXPORT_SYMBOL(__cputime_usec_factor);
163u64 __cputime_sec_factor;
164EXPORT_SYMBOL(__cputime_sec_factor);
165u64 __cputime_clockt_factor;
166EXPORT_SYMBOL(__cputime_clockt_factor);
167
168cputime_t cputime_one_jiffy;
169
170#ifdef CONFIG_PPC_SPLPAR
171void (*dtl_consumer)(struct dtl_entry *, u64);
172#endif
173
174#ifdef CONFIG_PPC64
175#define get_accounting(tsk) (&get_paca()->accounting)
176#else
177#define get_accounting(tsk) (&task_thread_info(tsk)->accounting)
178#endif
179
180static void calc_cputime_factors(void)
181{
182 struct div_result res;
183
184 div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
185 __cputime_jiffies_factor = res.result_low;
186 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
187 __cputime_usec_factor = res.result_low;
188 div128_by_32(1, 0, tb_ticks_per_sec, &res);
189 __cputime_sec_factor = res.result_low;
190 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
191 __cputime_clockt_factor = res.result_low;
192}
193
194/*
195 * Read the SPURR on systems that have it, otherwise the PURR,
196 * or if that doesn't exist return the timebase value passed in.
197 */
198static unsigned long read_spurr(unsigned long tb)
199{
200 if (cpu_has_feature(CPU_FTR_SPURR))
201 return mfspr(SPRN_SPURR);
202 if (cpu_has_feature(CPU_FTR_PURR))
203 return mfspr(SPRN_PURR);
204 return tb;
205}
206
207#ifdef CONFIG_PPC_SPLPAR
208
209/*
210 * Scan the dispatch trace log and count up the stolen time.
211 * Should be called with interrupts disabled.
212 */
213static u64 scan_dispatch_log(u64 stop_tb)
214{
215 u64 i = local_paca->dtl_ridx;
216 struct dtl_entry *dtl = local_paca->dtl_curr;
217 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
218 struct lppaca *vpa = local_paca->lppaca_ptr;
219 u64 tb_delta;
220 u64 stolen = 0;
221 u64 dtb;
222
223 if (!dtl)
224 return 0;
225
226 if (i == be64_to_cpu(vpa->dtl_idx))
227 return 0;
228 while (i < be64_to_cpu(vpa->dtl_idx)) {
229 dtb = be64_to_cpu(dtl->timebase);
230 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
231 be32_to_cpu(dtl->ready_to_enqueue_time);
232 barrier();
233 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
234 /* buffer has overflowed */
235 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
236 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
237 continue;
238 }
239 if (dtb > stop_tb)
240 break;
241 if (dtl_consumer)
242 dtl_consumer(dtl, i);
243 stolen += tb_delta;
244 ++i;
245 ++dtl;
246 if (dtl == dtl_end)
247 dtl = local_paca->dispatch_log;
248 }
249 local_paca->dtl_ridx = i;
250 local_paca->dtl_curr = dtl;
251 return stolen;
252}
253
254/*
255 * Accumulate stolen time by scanning the dispatch trace log.
256 * Called on entry from user mode.
257 */
258void accumulate_stolen_time(void)
259{
260 u64 sst, ust;
261 u8 save_soft_enabled = local_paca->soft_enabled;
262 struct cpu_accounting_data *acct = &local_paca->accounting;
263
264 /* We are called early in the exception entry, before
265 * soft/hard_enabled are sync'ed to the expected state
266 * for the exception. We are hard disabled but the PACA
267 * needs to reflect that so various debug stuff doesn't
268 * complain
269 */
270 local_paca->soft_enabled = 0;
271
272 sst = scan_dispatch_log(acct->starttime_user);
273 ust = scan_dispatch_log(acct->starttime);
274 acct->system_time -= sst;
275 acct->user_time -= ust;
276 local_paca->stolen_time += ust + sst;
277
278 local_paca->soft_enabled = save_soft_enabled;
279}
280
281static inline u64 calculate_stolen_time(u64 stop_tb)
282{
283 u64 stolen = 0;
284
285 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
286 stolen = scan_dispatch_log(stop_tb);
287 get_paca()->accounting.system_time -= stolen;
288 }
289
290 stolen += get_paca()->stolen_time;
291 get_paca()->stolen_time = 0;
292 return stolen;
293}
294
295#else /* CONFIG_PPC_SPLPAR */
296static inline u64 calculate_stolen_time(u64 stop_tb)
297{
298 return 0;
299}
300
301#endif /* CONFIG_PPC_SPLPAR */
302
303/*
304 * Account time for a transition between system, hard irq
305 * or soft irq state.
306 */
307static unsigned long vtime_delta(struct task_struct *tsk,
308 unsigned long *sys_scaled,
309 unsigned long *stolen)
310{
311 unsigned long now, nowscaled, deltascaled;
312 unsigned long udelta, delta, user_scaled;
313 struct cpu_accounting_data *acct = get_accounting(tsk);
314
315 WARN_ON_ONCE(!irqs_disabled());
316
317 now = mftb();
318 nowscaled = read_spurr(now);
319 acct->system_time += now - acct->starttime;
320 acct->starttime = now;
321 deltascaled = nowscaled - acct->startspurr;
322 acct->startspurr = nowscaled;
323
324 *stolen = calculate_stolen_time(now);
325
326 delta = acct->system_time;
327 acct->system_time = 0;
328 udelta = acct->user_time - acct->utime_sspurr;
329 acct->utime_sspurr = acct->user_time;
330
331 /*
332 * Because we don't read the SPURR on every kernel entry/exit,
333 * deltascaled includes both user and system SPURR ticks.
334 * Apportion these ticks to system SPURR ticks and user
335 * SPURR ticks in the same ratio as the system time (delta)
336 * and user time (udelta) values obtained from the timebase
337 * over the same interval. The system ticks get accounted here;
338 * the user ticks get saved up in paca->user_time_scaled to be
339 * used by account_process_tick.
340 */
341 *sys_scaled = delta;
342 user_scaled = udelta;
343 if (deltascaled != delta + udelta) {
344 if (udelta) {
345 *sys_scaled = deltascaled * delta / (delta + udelta);
346 user_scaled = deltascaled - *sys_scaled;
347 } else {
348 *sys_scaled = deltascaled;
349 }
350 }
351 acct->user_time_scaled += user_scaled;
352
353 return delta;
354}
355
356void vtime_account_system(struct task_struct *tsk)
357{
358 unsigned long delta, sys_scaled, stolen;
359
360 delta = vtime_delta(tsk, &sys_scaled, &stolen);
361 account_system_time(tsk, 0, delta);
362 tsk->stimescaled += sys_scaled;
363 if (stolen)
364 account_steal_time(stolen);
365}
366EXPORT_SYMBOL_GPL(vtime_account_system);
367
368void vtime_account_idle(struct task_struct *tsk)
369{
370 unsigned long delta, sys_scaled, stolen;
371
372 delta = vtime_delta(tsk, &sys_scaled, &stolen);
373 account_idle_time(delta + stolen);
374}
375
376/*
377 * Transfer the user time accumulated in the paca
378 * by the exception entry and exit code to the generic
379 * process user time records.
380 * Must be called with interrupts disabled.
381 * Assumes that vtime_account_system/idle() has been called
382 * recently (i.e. since the last entry from usermode) so that
383 * get_paca()->user_time_scaled is up to date.
384 */
385void vtime_account_user(struct task_struct *tsk)
386{
387 cputime_t utime, utimescaled;
388 struct cpu_accounting_data *acct = get_accounting(tsk);
389
390 utime = acct->user_time;
391 utimescaled = acct->user_time_scaled;
392 acct->user_time = 0;
393 acct->user_time_scaled = 0;
394 acct->utime_sspurr = 0;
395 account_user_time(tsk, utime);
396 tsk->utimescaled += utimescaled;
397}
398
399#ifdef CONFIG_PPC32
400/*
401 * Called from the context switch with interrupts disabled, to charge all
402 * accumulated times to the current process, and to prepare accounting on
403 * the next process.
404 */
405void arch_vtime_task_switch(struct task_struct *prev)
406{
407 struct cpu_accounting_data *acct = get_accounting(current);
408
409 acct->starttime = get_accounting(prev)->starttime;
410 acct->system_time = 0;
411 acct->user_time = 0;
412}
413#endif /* CONFIG_PPC32 */
414
415#else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
416#define calc_cputime_factors()
417#endif
418
419void __delay(unsigned long loops)
420{
421 unsigned long start;
422 int diff;
423
424 if (__USE_RTC()) {
425 start = get_rtcl();
426 do {
427 /* the RTCL register wraps at 1000000000 */
428 diff = get_rtcl() - start;
429 if (diff < 0)
430 diff += 1000000000;
431 } while (diff < loops);
432 } else {
433 start = get_tbl();
434 while (get_tbl() - start < loops)
435 HMT_low();
436 HMT_medium();
437 }
438}
439EXPORT_SYMBOL(__delay);
440
441void udelay(unsigned long usecs)
442{
443 __delay(tb_ticks_per_usec * usecs);
444}
445EXPORT_SYMBOL(udelay);
446
447#ifdef CONFIG_SMP
448unsigned long profile_pc(struct pt_regs *regs)
449{
450 unsigned long pc = instruction_pointer(regs);
451
452 if (in_lock_functions(pc))
453 return regs->link;
454
455 return pc;
456}
457EXPORT_SYMBOL(profile_pc);
458#endif
459
460#ifdef CONFIG_IRQ_WORK
461
462/*
463 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
464 */
465#ifdef CONFIG_PPC64
466static inline unsigned long test_irq_work_pending(void)
467{
468 unsigned long x;
469
470 asm volatile("lbz %0,%1(13)"
471 : "=r" (x)
472 : "i" (offsetof(struct paca_struct, irq_work_pending)));
473 return x;
474}
475
476static inline void set_irq_work_pending_flag(void)
477{
478 asm volatile("stb %0,%1(13)" : :
479 "r" (1),
480 "i" (offsetof(struct paca_struct, irq_work_pending)));
481}
482
483static inline void clear_irq_work_pending(void)
484{
485 asm volatile("stb %0,%1(13)" : :
486 "r" (0),
487 "i" (offsetof(struct paca_struct, irq_work_pending)));
488}
489
490#else /* 32-bit */
491
492DEFINE_PER_CPU(u8, irq_work_pending);
493
494#define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
495#define test_irq_work_pending() __this_cpu_read(irq_work_pending)
496#define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
497
498#endif /* 32 vs 64 bit */
499
500void arch_irq_work_raise(void)
501{
502 preempt_disable();
503 set_irq_work_pending_flag();
504 set_dec(1);
505 preempt_enable();
506}
507
508#else /* CONFIG_IRQ_WORK */
509
510#define test_irq_work_pending() 0
511#define clear_irq_work_pending()
512
513#endif /* CONFIG_IRQ_WORK */
514
515static void __timer_interrupt(void)
516{
517 struct pt_regs *regs = get_irq_regs();
518 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
519 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
520 u64 now;
521
522 trace_timer_interrupt_entry(regs);
523
524 if (test_irq_work_pending()) {
525 clear_irq_work_pending();
526 irq_work_run();
527 }
528
529 now = get_tb_or_rtc();
530 if (now >= *next_tb) {
531 *next_tb = ~(u64)0;
532 if (evt->event_handler)
533 evt->event_handler(evt);
534 __this_cpu_inc(irq_stat.timer_irqs_event);
535 } else {
536 now = *next_tb - now;
537 if (now <= decrementer_max)
538 set_dec(now);
539 /* We may have raced with new irq work */
540 if (test_irq_work_pending())
541 set_dec(1);
542 __this_cpu_inc(irq_stat.timer_irqs_others);
543 }
544
545#ifdef CONFIG_PPC64
546 /* collect purr register values often, for accurate calculations */
547 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
548 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
549 cu->current_tb = mfspr(SPRN_PURR);
550 }
551#endif
552
553 trace_timer_interrupt_exit(regs);
554}
555
556/*
557 * timer_interrupt - gets called when the decrementer overflows,
558 * with interrupts disabled.
559 */
560void timer_interrupt(struct pt_regs * regs)
561{
562 struct pt_regs *old_regs;
563 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
564
565 /* Ensure a positive value is written to the decrementer, or else
566 * some CPUs will continue to take decrementer exceptions.
567 */
568 set_dec(decrementer_max);
569
570 /* Some implementations of hotplug will get timer interrupts while
571 * offline, just ignore these and we also need to set
572 * decrementers_next_tb as MAX to make sure __check_irq_replay
573 * don't replay timer interrupt when return, otherwise we'll trap
574 * here infinitely :(
575 */
576 if (!cpu_online(smp_processor_id())) {
577 *next_tb = ~(u64)0;
578 return;
579 }
580
581 /* Conditionally hard-enable interrupts now that the DEC has been
582 * bumped to its maximum value
583 */
584 may_hard_irq_enable();
585
586
587#if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
588 if (atomic_read(&ppc_n_lost_interrupts) != 0)
589 do_IRQ(regs);
590#endif
591
592 old_regs = set_irq_regs(regs);
593 irq_enter();
594
595 __timer_interrupt();
596 irq_exit();
597 set_irq_regs(old_regs);
598}
599EXPORT_SYMBOL(timer_interrupt);
600
601/*
602 * Hypervisor decrementer interrupts shouldn't occur but are sometimes
603 * left pending on exit from a KVM guest. We don't need to do anything
604 * to clear them, as they are edge-triggered.
605 */
606void hdec_interrupt(struct pt_regs *regs)
607{
608}
609
610#ifdef CONFIG_SUSPEND
611static void generic_suspend_disable_irqs(void)
612{
613 /* Disable the decrementer, so that it doesn't interfere
614 * with suspending.
615 */
616
617 set_dec(decrementer_max);
618 local_irq_disable();
619 set_dec(decrementer_max);
620}
621
622static void generic_suspend_enable_irqs(void)
623{
624 local_irq_enable();
625}
626
627/* Overrides the weak version in kernel/power/main.c */
628void arch_suspend_disable_irqs(void)
629{
630 if (ppc_md.suspend_disable_irqs)
631 ppc_md.suspend_disable_irqs();
632 generic_suspend_disable_irqs();
633}
634
635/* Overrides the weak version in kernel/power/main.c */
636void arch_suspend_enable_irqs(void)
637{
638 generic_suspend_enable_irqs();
639 if (ppc_md.suspend_enable_irqs)
640 ppc_md.suspend_enable_irqs();
641}
642#endif
643
644unsigned long long tb_to_ns(unsigned long long ticks)
645{
646 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
647}
648EXPORT_SYMBOL_GPL(tb_to_ns);
649
650/*
651 * Scheduler clock - returns current time in nanosec units.
652 *
653 * Note: mulhdu(a, b) (multiply high double unsigned) returns
654 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
655 * are 64-bit unsigned numbers.
656 */
657unsigned long long sched_clock(void)
658{
659 if (__USE_RTC())
660 return get_rtc();
661 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
662}
663
664
665#ifdef CONFIG_PPC_PSERIES
666
667/*
668 * Running clock - attempts to give a view of time passing for a virtualised
669 * kernels.
670 * Uses the VTB register if available otherwise a next best guess.
671 */
672unsigned long long running_clock(void)
673{
674 /*
675 * Don't read the VTB as a host since KVM does not switch in host
676 * timebase into the VTB when it takes a guest off the CPU, reading the
677 * VTB would result in reading 'last switched out' guest VTB.
678 *
679 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
680 * would be unsafe to rely only on the #ifdef above.
681 */
682 if (firmware_has_feature(FW_FEATURE_LPAR) &&
683 cpu_has_feature(CPU_FTR_ARCH_207S))
684 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
685
686 /*
687 * This is a next best approximation without a VTB.
688 * On a host which is running bare metal there should never be any stolen
689 * time and on a host which doesn't do any virtualisation TB *should* equal
690 * VTB so it makes no difference anyway.
691 */
692 return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
693}
694#endif
695
696static int __init get_freq(char *name, int cells, unsigned long *val)
697{
698 struct device_node *cpu;
699 const __be32 *fp;
700 int found = 0;
701
702 /* The cpu node should have timebase and clock frequency properties */
703 cpu = of_find_node_by_type(NULL, "cpu");
704
705 if (cpu) {
706 fp = of_get_property(cpu, name, NULL);
707 if (fp) {
708 found = 1;
709 *val = of_read_ulong(fp, cells);
710 }
711
712 of_node_put(cpu);
713 }
714
715 return found;
716}
717
718static void start_cpu_decrementer(void)
719{
720#if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
721 /* Clear any pending timer interrupts */
722 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
723
724 /* Enable decrementer interrupt */
725 mtspr(SPRN_TCR, TCR_DIE);
726#endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
727}
728
729void __init generic_calibrate_decr(void)
730{
731 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
732
733 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
734 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
735
736 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
737 "(not found)\n");
738 }
739
740 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
741
742 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
743 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
744
745 printk(KERN_ERR "WARNING: Estimating processor frequency "
746 "(not found)\n");
747 }
748}
749
750int update_persistent_clock(struct timespec now)
751{
752 struct rtc_time tm;
753
754 if (!ppc_md.set_rtc_time)
755 return -ENODEV;
756
757 to_tm(now.tv_sec + 1 + timezone_offset, &tm);
758 tm.tm_year -= 1900;
759 tm.tm_mon -= 1;
760
761 return ppc_md.set_rtc_time(&tm);
762}
763
764static void __read_persistent_clock(struct timespec *ts)
765{
766 struct rtc_time tm;
767 static int first = 1;
768
769 ts->tv_nsec = 0;
770 /* XXX this is a litle fragile but will work okay in the short term */
771 if (first) {
772 first = 0;
773 if (ppc_md.time_init)
774 timezone_offset = ppc_md.time_init();
775
776 /* get_boot_time() isn't guaranteed to be safe to call late */
777 if (ppc_md.get_boot_time) {
778 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
779 return;
780 }
781 }
782 if (!ppc_md.get_rtc_time) {
783 ts->tv_sec = 0;
784 return;
785 }
786 ppc_md.get_rtc_time(&tm);
787
788 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
789 tm.tm_hour, tm.tm_min, tm.tm_sec);
790}
791
792void read_persistent_clock(struct timespec *ts)
793{
794 __read_persistent_clock(ts);
795
796 /* Sanitize it in case real time clock is set below EPOCH */
797 if (ts->tv_sec < 0) {
798 ts->tv_sec = 0;
799 ts->tv_nsec = 0;
800 }
801
802}
803
804/* clocksource code */
805static u64 rtc_read(struct clocksource *cs)
806{
807 return (u64)get_rtc();
808}
809
810static u64 timebase_read(struct clocksource *cs)
811{
812 return (u64)get_tb();
813}
814
815void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
816 struct clocksource *clock, u32 mult, u64 cycle_last)
817{
818 u64 new_tb_to_xs, new_stamp_xsec;
819 u32 frac_sec;
820
821 if (clock != &clocksource_timebase)
822 return;
823
824 /* Make userspace gettimeofday spin until we're done. */
825 ++vdso_data->tb_update_count;
826 smp_mb();
827
828 /* 19342813113834067 ~= 2^(20+64) / 1e9 */
829 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
830 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
831 do_div(new_stamp_xsec, 1000000000);
832 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
833
834 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
835 /* this is tv_nsec / 1e9 as a 0.32 fraction */
836 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
837
838 /*
839 * tb_update_count is used to allow the userspace gettimeofday code
840 * to assure itself that it sees a consistent view of the tb_to_xs and
841 * stamp_xsec variables. It reads the tb_update_count, then reads
842 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
843 * the two values of tb_update_count match and are even then the
844 * tb_to_xs and stamp_xsec values are consistent. If not, then it
845 * loops back and reads them again until this criteria is met.
846 * We expect the caller to have done the first increment of
847 * vdso_data->tb_update_count already.
848 */
849 vdso_data->tb_orig_stamp = cycle_last;
850 vdso_data->stamp_xsec = new_stamp_xsec;
851 vdso_data->tb_to_xs = new_tb_to_xs;
852 vdso_data->wtom_clock_sec = wtm->tv_sec;
853 vdso_data->wtom_clock_nsec = wtm->tv_nsec;
854 vdso_data->stamp_xtime = *wall_time;
855 vdso_data->stamp_sec_fraction = frac_sec;
856 smp_wmb();
857 ++(vdso_data->tb_update_count);
858}
859
860void update_vsyscall_tz(void)
861{
862 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
863 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
864}
865
866static void __init clocksource_init(void)
867{
868 struct clocksource *clock;
869
870 if (__USE_RTC())
871 clock = &clocksource_rtc;
872 else
873 clock = &clocksource_timebase;
874
875 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
876 printk(KERN_ERR "clocksource: %s is already registered\n",
877 clock->name);
878 return;
879 }
880
881 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
882 clock->name, clock->mult, clock->shift);
883}
884
885static int decrementer_set_next_event(unsigned long evt,
886 struct clock_event_device *dev)
887{
888 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
889 set_dec(evt);
890
891 /* We may have raced with new irq work */
892 if (test_irq_work_pending())
893 set_dec(1);
894
895 return 0;
896}
897
898static int decrementer_shutdown(struct clock_event_device *dev)
899{
900 decrementer_set_next_event(decrementer_max, dev);
901 return 0;
902}
903
904/* Interrupt handler for the timer broadcast IPI */
905void tick_broadcast_ipi_handler(void)
906{
907 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
908
909 *next_tb = get_tb_or_rtc();
910 __timer_interrupt();
911}
912
913static void register_decrementer_clockevent(int cpu)
914{
915 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
916
917 *dec = decrementer_clockevent;
918 dec->cpumask = cpumask_of(cpu);
919
920 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
921 dec->name, dec->mult, dec->shift, cpu);
922
923 clockevents_register_device(dec);
924}
925
926static void enable_large_decrementer(void)
927{
928 if (!cpu_has_feature(CPU_FTR_ARCH_300))
929 return;
930
931 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
932 return;
933
934 /*
935 * If we're running as the hypervisor we need to enable the LD manually
936 * otherwise firmware should have done it for us.
937 */
938 if (cpu_has_feature(CPU_FTR_HVMODE))
939 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
940}
941
942static void __init set_decrementer_max(void)
943{
944 struct device_node *cpu;
945 u32 bits = 32;
946
947 /* Prior to ISAv3 the decrementer is always 32 bit */
948 if (!cpu_has_feature(CPU_FTR_ARCH_300))
949 return;
950
951 cpu = of_find_node_by_type(NULL, "cpu");
952
953 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
954 if (bits > 64 || bits < 32) {
955 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
956 bits = 32;
957 }
958
959 /* calculate the signed maximum given this many bits */
960 decrementer_max = (1ul << (bits - 1)) - 1;
961 }
962
963 of_node_put(cpu);
964
965 pr_info("time_init: %u bit decrementer (max: %llx)\n",
966 bits, decrementer_max);
967}
968
969static void __init init_decrementer_clockevent(void)
970{
971 int cpu = smp_processor_id();
972
973 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
974
975 decrementer_clockevent.max_delta_ns =
976 clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
977 decrementer_clockevent.min_delta_ns =
978 clockevent_delta2ns(2, &decrementer_clockevent);
979
980 register_decrementer_clockevent(cpu);
981}
982
983void secondary_cpu_time_init(void)
984{
985 /* Enable and test the large decrementer for this cpu */
986 enable_large_decrementer();
987
988 /* Start the decrementer on CPUs that have manual control
989 * such as BookE
990 */
991 start_cpu_decrementer();
992
993 /* FIME: Should make unrelatred change to move snapshot_timebase
994 * call here ! */
995 register_decrementer_clockevent(smp_processor_id());
996}
997
998/* This function is only called on the boot processor */
999void __init time_init(void)
1000{
1001 struct div_result res;
1002 u64 scale;
1003 unsigned shift;
1004
1005 if (__USE_RTC()) {
1006 /* 601 processor: dec counts down by 128 every 128ns */
1007 ppc_tb_freq = 1000000000;
1008 } else {
1009 /* Normal PowerPC with timebase register */
1010 ppc_md.calibrate_decr();
1011 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1012 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1013 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1014 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1015 }
1016
1017 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1018 tb_ticks_per_sec = ppc_tb_freq;
1019 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1020 calc_cputime_factors();
1021 setup_cputime_one_jiffy();
1022
1023 /*
1024 * Compute scale factor for sched_clock.
1025 * The calibrate_decr() function has set tb_ticks_per_sec,
1026 * which is the timebase frequency.
1027 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1028 * the 128-bit result as a 64.64 fixed-point number.
1029 * We then shift that number right until it is less than 1.0,
1030 * giving us the scale factor and shift count to use in
1031 * sched_clock().
1032 */
1033 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1034 scale = res.result_low;
1035 for (shift = 0; res.result_high != 0; ++shift) {
1036 scale = (scale >> 1) | (res.result_high << 63);
1037 res.result_high >>= 1;
1038 }
1039 tb_to_ns_scale = scale;
1040 tb_to_ns_shift = shift;
1041 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1042 boot_tb = get_tb_or_rtc();
1043
1044 /* If platform provided a timezone (pmac), we correct the time */
1045 if (timezone_offset) {
1046 sys_tz.tz_minuteswest = -timezone_offset / 60;
1047 sys_tz.tz_dsttime = 0;
1048 }
1049
1050 vdso_data->tb_update_count = 0;
1051 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1052
1053 /* initialise and enable the large decrementer (if we have one) */
1054 set_decrementer_max();
1055 enable_large_decrementer();
1056
1057 /* Start the decrementer on CPUs that have manual control
1058 * such as BookE
1059 */
1060 start_cpu_decrementer();
1061
1062 /* Register the clocksource */
1063 clocksource_init();
1064
1065 init_decrementer_clockevent();
1066 tick_setup_hrtimer_broadcast();
1067
1068#ifdef CONFIG_COMMON_CLK
1069 of_clk_init(NULL);
1070#endif
1071}
1072
1073
1074#define FEBRUARY 2
1075#define STARTOFTIME 1970
1076#define SECDAY 86400L
1077#define SECYR (SECDAY * 365)
1078#define leapyear(year) ((year) % 4 == 0 && \
1079 ((year) % 100 != 0 || (year) % 400 == 0))
1080#define days_in_year(a) (leapyear(a) ? 366 : 365)
1081#define days_in_month(a) (month_days[(a) - 1])
1082
1083static int month_days[12] = {
1084 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1085};
1086
1087void to_tm(int tim, struct rtc_time * tm)
1088{
1089 register int i;
1090 register long hms, day;
1091
1092 day = tim / SECDAY;
1093 hms = tim % SECDAY;
1094
1095 /* Hours, minutes, seconds are easy */
1096 tm->tm_hour = hms / 3600;
1097 tm->tm_min = (hms % 3600) / 60;
1098 tm->tm_sec = (hms % 3600) % 60;
1099
1100 /* Number of years in days */
1101 for (i = STARTOFTIME; day >= days_in_year(i); i++)
1102 day -= days_in_year(i);
1103 tm->tm_year = i;
1104
1105 /* Number of months in days left */
1106 if (leapyear(tm->tm_year))
1107 days_in_month(FEBRUARY) = 29;
1108 for (i = 1; day >= days_in_month(i); i++)
1109 day -= days_in_month(i);
1110 days_in_month(FEBRUARY) = 28;
1111 tm->tm_mon = i;
1112
1113 /* Days are what is left over (+1) from all that. */
1114 tm->tm_mday = day + 1;
1115
1116 /*
1117 * No-one uses the day of the week.
1118 */
1119 tm->tm_wday = -1;
1120}
1121EXPORT_SYMBOL(to_tm);
1122
1123/*
1124 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1125 * result.
1126 */
1127void div128_by_32(u64 dividend_high, u64 dividend_low,
1128 unsigned divisor, struct div_result *dr)
1129{
1130 unsigned long a, b, c, d;
1131 unsigned long w, x, y, z;
1132 u64 ra, rb, rc;
1133
1134 a = dividend_high >> 32;
1135 b = dividend_high & 0xffffffff;
1136 c = dividend_low >> 32;
1137 d = dividend_low & 0xffffffff;
1138
1139 w = a / divisor;
1140 ra = ((u64)(a - (w * divisor)) << 32) + b;
1141
1142 rb = ((u64) do_div(ra, divisor) << 32) + c;
1143 x = ra;
1144
1145 rc = ((u64) do_div(rb, divisor) << 32) + d;
1146 y = rb;
1147
1148 do_div(rc, divisor);
1149 z = rc;
1150
1151 dr->result_high = ((u64)w << 32) + x;
1152 dr->result_low = ((u64)y << 32) + z;
1153
1154}
1155
1156/* We don't need to calibrate delay, we use the CPU timebase for that */
1157void calibrate_delay(void)
1158{
1159 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1160 * as the number of __delay(1) in a jiffy, so make it so
1161 */
1162 loops_per_jiffy = tb_ticks_per_jiffy;
1163}
1164
1165#if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1166static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1167{
1168 ppc_md.get_rtc_time(tm);
1169 return rtc_valid_tm(tm);
1170}
1171
1172static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1173{
1174 if (!ppc_md.set_rtc_time)
1175 return -EOPNOTSUPP;
1176
1177 if (ppc_md.set_rtc_time(tm) < 0)
1178 return -EOPNOTSUPP;
1179
1180 return 0;
1181}
1182
1183static const struct rtc_class_ops rtc_generic_ops = {
1184 .read_time = rtc_generic_get_time,
1185 .set_time = rtc_generic_set_time,
1186};
1187
1188static int __init rtc_init(void)
1189{
1190 struct platform_device *pdev;
1191
1192 if (!ppc_md.get_rtc_time)
1193 return -ENODEV;
1194
1195 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1196 &rtc_generic_ops,
1197 sizeof(rtc_generic_ops));
1198
1199 return PTR_ERR_OR_ZERO(pdev);
1200}
1201
1202device_initcall(rtc_init);
1203#endif