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/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
75/* powerpc clocksource/clockevent code */
76
77#include <linux/clockchips.h>
78#include <linux/timekeeper_internal.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 __read_mostly;
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#endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
358
359void __delay(unsigned long loops)
360{
361 unsigned long start;
362
363 spin_begin();
364 if (tb_invalid) {
365 /*
366 * TB is in error state and isn't ticking anymore.
367 * HMI handler was unable to recover from TB error.
368 * Return immediately, so that kernel won't get stuck here.
369 */
370 spin_cpu_relax();
371 } else {
372 start = mftb();
373 while (mftb() - start < loops)
374 spin_cpu_relax();
375 }
376 spin_end();
377}
378EXPORT_SYMBOL(__delay);
379
380void udelay(unsigned long usecs)
381{
382 __delay(tb_ticks_per_usec * usecs);
383}
384EXPORT_SYMBOL(udelay);
385
386#ifdef CONFIG_SMP
387unsigned long profile_pc(struct pt_regs *regs)
388{
389 unsigned long pc = instruction_pointer(regs);
390
391 if (in_lock_functions(pc))
392 return regs->link;
393
394 return pc;
395}
396EXPORT_SYMBOL(profile_pc);
397#endif
398
399#ifdef CONFIG_IRQ_WORK
400
401/*
402 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
403 */
404#ifdef CONFIG_PPC64
405static inline unsigned long test_irq_work_pending(void)
406{
407 unsigned long x;
408
409 asm volatile("lbz %0,%1(13)"
410 : "=r" (x)
411 : "i" (offsetof(struct paca_struct, irq_work_pending)));
412 return x;
413}
414
415static inline void set_irq_work_pending_flag(void)
416{
417 asm volatile("stb %0,%1(13)" : :
418 "r" (1),
419 "i" (offsetof(struct paca_struct, irq_work_pending)));
420}
421
422static inline void clear_irq_work_pending(void)
423{
424 asm volatile("stb %0,%1(13)" : :
425 "r" (0),
426 "i" (offsetof(struct paca_struct, irq_work_pending)));
427}
428
429#else /* 32-bit */
430
431DEFINE_PER_CPU(u8, irq_work_pending);
432
433#define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
434#define test_irq_work_pending() __this_cpu_read(irq_work_pending)
435#define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
436
437#endif /* 32 vs 64 bit */
438
439void arch_irq_work_raise(void)
440{
441 /*
442 * 64-bit code that uses irq soft-mask can just cause an immediate
443 * interrupt here that gets soft masked, if this is called under
444 * local_irq_disable(). It might be possible to prevent that happening
445 * by noticing interrupts are disabled and setting decrementer pending
446 * to be replayed when irqs are enabled. The problem there is that
447 * tracing can call irq_work_raise, including in code that does low
448 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
449 * which could get tangled up if we're messing with the same state
450 * here.
451 */
452 preempt_disable();
453 set_irq_work_pending_flag();
454 set_dec(1);
455 preempt_enable();
456}
457
458static void set_dec_or_work(u64 val)
459{
460 set_dec(val);
461 /* We may have raced with new irq work */
462 if (unlikely(test_irq_work_pending()))
463 set_dec(1);
464}
465
466#else /* CONFIG_IRQ_WORK */
467
468#define test_irq_work_pending() 0
469#define clear_irq_work_pending()
470
471static void set_dec_or_work(u64 val)
472{
473 set_dec(val);
474}
475#endif /* CONFIG_IRQ_WORK */
476
477#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
478void timer_rearm_host_dec(u64 now)
479{
480 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
481
482 WARN_ON_ONCE(!arch_irqs_disabled());
483 WARN_ON_ONCE(mfmsr() & MSR_EE);
484
485 if (now >= *next_tb) {
486 local_paca->irq_happened |= PACA_IRQ_DEC;
487 } else {
488 now = *next_tb - now;
489 if (now > decrementer_max)
490 now = decrementer_max;
491 set_dec_or_work(now);
492 }
493}
494EXPORT_SYMBOL_GPL(timer_rearm_host_dec);
495#endif
496
497/*
498 * timer_interrupt - gets called when the decrementer overflows,
499 * with interrupts disabled.
500 */
501DEFINE_INTERRUPT_HANDLER_ASYNC(timer_interrupt)
502{
503 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
504 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
505 struct pt_regs *old_regs;
506 u64 now;
507
508 /*
509 * Some implementations of hotplug will get timer interrupts while
510 * offline, just ignore these.
511 */
512 if (unlikely(!cpu_online(smp_processor_id()))) {
513 set_dec(decrementer_max);
514 return;
515 }
516
517 /* Conditionally hard-enable interrupts. */
518 if (should_hard_irq_enable(regs)) {
519 /*
520 * Ensure a positive value is written to the decrementer, or
521 * else some CPUs will continue to take decrementer exceptions.
522 * When the PPC_WATCHDOG (decrementer based) is configured,
523 * keep this at most 31 bits, which is about 4 seconds on most
524 * systems, which gives the watchdog a chance of catching timer
525 * interrupt hard lockups.
526 */
527 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
528 set_dec(0x7fffffff);
529 else
530 set_dec(decrementer_max);
531
532 do_hard_irq_enable();
533 }
534
535#if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
536 if (atomic_read(&ppc_n_lost_interrupts) != 0)
537 __do_IRQ(regs);
538#endif
539
540 old_regs = set_irq_regs(regs);
541
542 trace_timer_interrupt_entry(regs);
543
544 if (test_irq_work_pending()) {
545 clear_irq_work_pending();
546 mce_run_irq_context_handlers();
547 irq_work_run();
548 }
549
550 now = get_tb();
551 if (now >= *next_tb) {
552 evt->event_handler(evt);
553 __this_cpu_inc(irq_stat.timer_irqs_event);
554 } else {
555 now = *next_tb - now;
556 if (now > decrementer_max)
557 now = decrementer_max;
558 set_dec_or_work(now);
559 __this_cpu_inc(irq_stat.timer_irqs_others);
560 }
561
562 trace_timer_interrupt_exit(regs);
563
564 set_irq_regs(old_regs);
565}
566EXPORT_SYMBOL(timer_interrupt);
567
568#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
569void timer_broadcast_interrupt(void)
570{
571 tick_receive_broadcast();
572 __this_cpu_inc(irq_stat.broadcast_irqs_event);
573}
574#endif
575
576#ifdef CONFIG_SUSPEND
577/* Overrides the weak version in kernel/power/main.c */
578void arch_suspend_disable_irqs(void)
579{
580 if (ppc_md.suspend_disable_irqs)
581 ppc_md.suspend_disable_irqs();
582
583 /* Disable the decrementer, so that it doesn't interfere
584 * with suspending.
585 */
586
587 set_dec(decrementer_max);
588 local_irq_disable();
589 set_dec(decrementer_max);
590}
591
592/* Overrides the weak version in kernel/power/main.c */
593void arch_suspend_enable_irqs(void)
594{
595 local_irq_enable();
596
597 if (ppc_md.suspend_enable_irqs)
598 ppc_md.suspend_enable_irqs();
599}
600#endif
601
602unsigned long long tb_to_ns(unsigned long long ticks)
603{
604 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
605}
606EXPORT_SYMBOL_GPL(tb_to_ns);
607
608/*
609 * Scheduler clock - returns current time in nanosec units.
610 *
611 * Note: mulhdu(a, b) (multiply high double unsigned) returns
612 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
613 * are 64-bit unsigned numbers.
614 */
615notrace unsigned long long sched_clock(void)
616{
617 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
618}
619
620
621#ifdef CONFIG_PPC_PSERIES
622
623/*
624 * Running clock - attempts to give a view of time passing for a virtualised
625 * kernels.
626 * Uses the VTB register if available otherwise a next best guess.
627 */
628unsigned long long running_clock(void)
629{
630 /*
631 * Don't read the VTB as a host since KVM does not switch in host
632 * timebase into the VTB when it takes a guest off the CPU, reading the
633 * VTB would result in reading 'last switched out' guest VTB.
634 *
635 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
636 * would be unsafe to rely only on the #ifdef above.
637 */
638 if (firmware_has_feature(FW_FEATURE_LPAR) &&
639 cpu_has_feature(CPU_FTR_ARCH_207S))
640 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
641
642 /*
643 * This is a next best approximation without a VTB.
644 * On a host which is running bare metal there should never be any stolen
645 * time and on a host which doesn't do any virtualisation TB *should* equal
646 * VTB so it makes no difference anyway.
647 */
648 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
649}
650#endif
651
652static int __init get_freq(char *name, int cells, unsigned long *val)
653{
654 struct device_node *cpu;
655 const __be32 *fp;
656 int found = 0;
657
658 /* The cpu node should have timebase and clock frequency properties */
659 cpu = of_find_node_by_type(NULL, "cpu");
660
661 if (cpu) {
662 fp = of_get_property(cpu, name, NULL);
663 if (fp) {
664 found = 1;
665 *val = of_read_ulong(fp, cells);
666 }
667
668 of_node_put(cpu);
669 }
670
671 return found;
672}
673
674static void start_cpu_decrementer(void)
675{
676#ifdef CONFIG_BOOKE_OR_40x
677 unsigned int tcr;
678
679 /* Clear any pending timer interrupts */
680 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
681
682 tcr = mfspr(SPRN_TCR);
683 /*
684 * The watchdog may have already been enabled by u-boot. So leave
685 * TRC[WP] (Watchdog Period) alone.
686 */
687 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
688 tcr |= TCR_DIE; /* Enable decrementer */
689 mtspr(SPRN_TCR, tcr);
690#endif
691}
692
693void __init generic_calibrate_decr(void)
694{
695 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
696
697 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
698 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
699
700 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
701 "(not found)\n");
702 }
703
704 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
705
706 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
707 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
708
709 printk(KERN_ERR "WARNING: Estimating processor frequency "
710 "(not found)\n");
711 }
712}
713
714int update_persistent_clock64(struct timespec64 now)
715{
716 struct rtc_time tm;
717
718 if (!ppc_md.set_rtc_time)
719 return -ENODEV;
720
721 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
722
723 return ppc_md.set_rtc_time(&tm);
724}
725
726static void __read_persistent_clock(struct timespec64 *ts)
727{
728 struct rtc_time tm;
729 static int first = 1;
730
731 ts->tv_nsec = 0;
732 /* XXX this is a little fragile but will work okay in the short term */
733 if (first) {
734 first = 0;
735 if (ppc_md.time_init)
736 timezone_offset = ppc_md.time_init();
737
738 /* get_boot_time() isn't guaranteed to be safe to call late */
739 if (ppc_md.get_boot_time) {
740 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
741 return;
742 }
743 }
744 if (!ppc_md.get_rtc_time) {
745 ts->tv_sec = 0;
746 return;
747 }
748 ppc_md.get_rtc_time(&tm);
749
750 ts->tv_sec = rtc_tm_to_time64(&tm);
751}
752
753void read_persistent_clock64(struct timespec64 *ts)
754{
755 __read_persistent_clock(ts);
756
757 /* Sanitize it in case real time clock is set below EPOCH */
758 if (ts->tv_sec < 0) {
759 ts->tv_sec = 0;
760 ts->tv_nsec = 0;
761 }
762
763}
764
765/* clocksource code */
766static notrace u64 timebase_read(struct clocksource *cs)
767{
768 return (u64)get_tb();
769}
770
771static void __init clocksource_init(void)
772{
773 struct clocksource *clock = &clocksource_timebase;
774
775 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
776 printk(KERN_ERR "clocksource: %s is already registered\n",
777 clock->name);
778 return;
779 }
780
781 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
782 clock->name, clock->mult, clock->shift);
783}
784
785static int decrementer_set_next_event(unsigned long evt,
786 struct clock_event_device *dev)
787{
788 __this_cpu_write(decrementers_next_tb, get_tb() + evt);
789 set_dec_or_work(evt);
790
791 return 0;
792}
793
794static int decrementer_shutdown(struct clock_event_device *dev)
795{
796 __this_cpu_write(decrementers_next_tb, DEC_CLOCKEVENT_STOPPED);
797 set_dec_or_work(decrementer_max);
798
799 return 0;
800}
801
802static void register_decrementer_clockevent(int cpu)
803{
804 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
805
806 *dec = decrementer_clockevent;
807 dec->cpumask = cpumask_of(cpu);
808
809 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
810
811 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
812 dec->name, dec->mult, dec->shift, cpu);
813
814 /* Set values for KVM, see kvm_emulate_dec() */
815 decrementer_clockevent.mult = dec->mult;
816 decrementer_clockevent.shift = dec->shift;
817}
818
819static void enable_large_decrementer(void)
820{
821 if (!cpu_has_feature(CPU_FTR_ARCH_300))
822 return;
823
824 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
825 return;
826
827 /*
828 * If we're running as the hypervisor we need to enable the LD manually
829 * otherwise firmware should have done it for us.
830 */
831 if (cpu_has_feature(CPU_FTR_HVMODE))
832 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
833}
834
835static void __init set_decrementer_max(void)
836{
837 struct device_node *cpu;
838 u32 bits = 32;
839
840 /* Prior to ISAv3 the decrementer is always 32 bit */
841 if (!cpu_has_feature(CPU_FTR_ARCH_300))
842 return;
843
844 cpu = of_find_node_by_type(NULL, "cpu");
845
846 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
847 if (bits > 64 || bits < 32) {
848 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
849 bits = 32;
850 }
851
852 /* calculate the signed maximum given this many bits */
853 decrementer_max = (1ul << (bits - 1)) - 1;
854 }
855
856 of_node_put(cpu);
857
858 pr_info("time_init: %u bit decrementer (max: %llx)\n",
859 bits, decrementer_max);
860}
861
862static void __init init_decrementer_clockevent(void)
863{
864 register_decrementer_clockevent(smp_processor_id());
865}
866
867void secondary_cpu_time_init(void)
868{
869 /* Enable and test the large decrementer for this cpu */
870 enable_large_decrementer();
871
872 /* Start the decrementer on CPUs that have manual control
873 * such as BookE
874 */
875 start_cpu_decrementer();
876
877 /* FIME: Should make unrelated change to move snapshot_timebase
878 * call here ! */
879 register_decrementer_clockevent(smp_processor_id());
880}
881
882/* This function is only called on the boot processor */
883void __init time_init(void)
884{
885 struct div_result res;
886 u64 scale;
887 unsigned shift;
888
889 /* Normal PowerPC with timebase register */
890 ppc_md.calibrate_decr();
891 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
892 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
893 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
894 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
895
896 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
897 tb_ticks_per_sec = ppc_tb_freq;
898 tb_ticks_per_usec = ppc_tb_freq / 1000000;
899
900 /*
901 * Compute scale factor for sched_clock.
902 * The calibrate_decr() function has set tb_ticks_per_sec,
903 * which is the timebase frequency.
904 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
905 * the 128-bit result as a 64.64 fixed-point number.
906 * We then shift that number right until it is less than 1.0,
907 * giving us the scale factor and shift count to use in
908 * sched_clock().
909 */
910 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
911 scale = res.result_low;
912 for (shift = 0; res.result_high != 0; ++shift) {
913 scale = (scale >> 1) | (res.result_high << 63);
914 res.result_high >>= 1;
915 }
916 tb_to_ns_scale = scale;
917 tb_to_ns_shift = shift;
918 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
919 boot_tb = get_tb();
920
921 /* If platform provided a timezone (pmac), we correct the time */
922 if (timezone_offset) {
923 sys_tz.tz_minuteswest = -timezone_offset / 60;
924 sys_tz.tz_dsttime = 0;
925 }
926
927 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
928
929 /* initialise and enable the large decrementer (if we have one) */
930 set_decrementer_max();
931 enable_large_decrementer();
932
933 /* Start the decrementer on CPUs that have manual control
934 * such as BookE
935 */
936 start_cpu_decrementer();
937
938 /* Register the clocksource */
939 clocksource_init();
940
941 init_decrementer_clockevent();
942 tick_setup_hrtimer_broadcast();
943
944 of_clk_init(NULL);
945 enable_sched_clock_irqtime();
946}
947
948/*
949 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
950 * result.
951 */
952void div128_by_32(u64 dividend_high, u64 dividend_low,
953 unsigned divisor, struct div_result *dr)
954{
955 unsigned long a, b, c, d;
956 unsigned long w, x, y, z;
957 u64 ra, rb, rc;
958
959 a = dividend_high >> 32;
960 b = dividend_high & 0xffffffff;
961 c = dividend_low >> 32;
962 d = dividend_low & 0xffffffff;
963
964 w = a / divisor;
965 ra = ((u64)(a - (w * divisor)) << 32) + b;
966
967 rb = ((u64) do_div(ra, divisor) << 32) + c;
968 x = ra;
969
970 rc = ((u64) do_div(rb, divisor) << 32) + d;
971 y = rb;
972
973 do_div(rc, divisor);
974 z = rc;
975
976 dr->result_high = ((u64)w << 32) + x;
977 dr->result_low = ((u64)y << 32) + z;
978
979}
980
981/* We don't need to calibrate delay, we use the CPU timebase for that */
982void calibrate_delay(void)
983{
984 /* Some generic code (such as spinlock debug) use loops_per_jiffy
985 * as the number of __delay(1) in a jiffy, so make it so
986 */
987 loops_per_jiffy = tb_ticks_per_jiffy;
988}
989
990#if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
991static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
992{
993 ppc_md.get_rtc_time(tm);
994 return 0;
995}
996
997static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
998{
999 if (!ppc_md.set_rtc_time)
1000 return -EOPNOTSUPP;
1001
1002 if (ppc_md.set_rtc_time(tm) < 0)
1003 return -EOPNOTSUPP;
1004
1005 return 0;
1006}
1007
1008static const struct rtc_class_ops rtc_generic_ops = {
1009 .read_time = rtc_generic_get_time,
1010 .set_time = rtc_generic_set_time,
1011};
1012
1013static int __init rtc_init(void)
1014{
1015 struct platform_device *pdev;
1016
1017 if (!ppc_md.get_rtc_time)
1018 return -ENODEV;
1019
1020 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1021 &rtc_generic_ops,
1022 sizeof(rtc_generic_ops));
1023
1024 return PTR_ERR_OR_ZERO(pdev);
1025}
1026
1027device_initcall(rtc_init);
1028#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