xref: /linux/arch/powerpc/kernel/time.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
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. (for iSeries, we calibrate the timebase
21  * against the Titan chip's clock.)
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  *      This program is free software; you can redistribute it and/or
30  *      modify it under the terms of the GNU General Public License
31  *      as published by the Free Software Foundation; either version
32  *      2 of the License, or (at your option) any later version.
33  */
34 
35 #include <linux/errno.h>
36 #include <linux/module.h>
37 #include <linux/sched.h>
38 #include <linux/kernel.h>
39 #include <linux/param.h>
40 #include <linux/string.h>
41 #include <linux/mm.h>
42 #include <linux/interrupt.h>
43 #include <linux/timex.h>
44 #include <linux/kernel_stat.h>
45 #include <linux/time.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/perf_event.h>
57 #include <asm/trace.h>
58 
59 #include <asm/io.h>
60 #include <asm/processor.h>
61 #include <asm/nvram.h>
62 #include <asm/cache.h>
63 #include <asm/machdep.h>
64 #include <asm/uaccess.h>
65 #include <asm/time.h>
66 #include <asm/prom.h>
67 #include <asm/irq.h>
68 #include <asm/div64.h>
69 #include <asm/smp.h>
70 #include <asm/vdso_datapage.h>
71 #include <asm/firmware.h>
72 #include <asm/cputime.h>
73 #ifdef CONFIG_PPC_ISERIES
74 #include <asm/iseries/it_lp_queue.h>
75 #include <asm/iseries/hv_call_xm.h>
76 #endif
77 
78 /* powerpc clocksource/clockevent code */
79 
80 #include <linux/clockchips.h>
81 #include <linux/clocksource.h>
82 
83 static cycle_t rtc_read(struct clocksource *);
84 static struct clocksource clocksource_rtc = {
85 	.name         = "rtc",
86 	.rating       = 400,
87 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
88 	.mask         = CLOCKSOURCE_MASK(64),
89 	.shift        = 22,
90 	.mult         = 0,	/* To be filled in */
91 	.read         = rtc_read,
92 };
93 
94 static cycle_t timebase_read(struct clocksource *);
95 static struct clocksource clocksource_timebase = {
96 	.name         = "timebase",
97 	.rating       = 400,
98 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
99 	.mask         = CLOCKSOURCE_MASK(64),
100 	.shift        = 22,
101 	.mult         = 0,	/* To be filled in */
102 	.read         = timebase_read,
103 };
104 
105 #define DECREMENTER_MAX	0x7fffffff
106 
107 static int decrementer_set_next_event(unsigned long evt,
108 				      struct clock_event_device *dev);
109 static void decrementer_set_mode(enum clock_event_mode mode,
110 				 struct clock_event_device *dev);
111 
112 static struct clock_event_device decrementer_clockevent = {
113        .name           = "decrementer",
114        .rating         = 200,
115        .shift          = 0,	/* To be filled in */
116        .mult           = 0,	/* To be filled in */
117        .irq            = 0,
118        .set_next_event = decrementer_set_next_event,
119        .set_mode       = decrementer_set_mode,
120        .features       = CLOCK_EVT_FEAT_ONESHOT,
121 };
122 
123 struct decrementer_clock {
124 	struct clock_event_device event;
125 	u64 next_tb;
126 };
127 
128 static DEFINE_PER_CPU(struct decrementer_clock, decrementers);
129 
130 #ifdef CONFIG_PPC_ISERIES
131 static unsigned long __initdata iSeries_recal_titan;
132 static signed long __initdata iSeries_recal_tb;
133 
134 /* Forward declaration is only needed for iSereis compiles */
135 static void __init clocksource_init(void);
136 #endif
137 
138 #define XSEC_PER_SEC (1024*1024)
139 
140 #ifdef CONFIG_PPC64
141 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
142 #else
143 /* compute ((xsec << 12) * max) >> 32 */
144 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
145 #endif
146 
147 unsigned long tb_ticks_per_jiffy;
148 unsigned long tb_ticks_per_usec = 100; /* sane default */
149 EXPORT_SYMBOL(tb_ticks_per_usec);
150 unsigned long tb_ticks_per_sec;
151 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
152 
153 DEFINE_SPINLOCK(rtc_lock);
154 EXPORT_SYMBOL_GPL(rtc_lock);
155 
156 static u64 tb_to_ns_scale __read_mostly;
157 static unsigned tb_to_ns_shift __read_mostly;
158 static unsigned long boot_tb __read_mostly;
159 
160 extern struct timezone sys_tz;
161 static long timezone_offset;
162 
163 unsigned long ppc_proc_freq;
164 EXPORT_SYMBOL(ppc_proc_freq);
165 unsigned long ppc_tb_freq;
166 
167 static DEFINE_PER_CPU(u64, last_jiffy);
168 
169 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
170 /*
171  * Factors for converting from cputime_t (timebase ticks) to
172  * jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds).
173  * These are all stored as 0.64 fixed-point binary fractions.
174  */
175 u64 __cputime_jiffies_factor;
176 EXPORT_SYMBOL(__cputime_jiffies_factor);
177 u64 __cputime_msec_factor;
178 EXPORT_SYMBOL(__cputime_msec_factor);
179 u64 __cputime_sec_factor;
180 EXPORT_SYMBOL(__cputime_sec_factor);
181 u64 __cputime_clockt_factor;
182 EXPORT_SYMBOL(__cputime_clockt_factor);
183 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
184 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
185 
186 cputime_t cputime_one_jiffy;
187 
188 static void calc_cputime_factors(void)
189 {
190 	struct div_result res;
191 
192 	div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
193 	__cputime_jiffies_factor = res.result_low;
194 	div128_by_32(1000, 0, tb_ticks_per_sec, &res);
195 	__cputime_msec_factor = res.result_low;
196 	div128_by_32(1, 0, tb_ticks_per_sec, &res);
197 	__cputime_sec_factor = res.result_low;
198 	div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
199 	__cputime_clockt_factor = res.result_low;
200 }
201 
202 /*
203  * Read the PURR on systems that have it, otherwise the timebase.
204  */
205 static u64 read_purr(void)
206 {
207 	if (cpu_has_feature(CPU_FTR_PURR))
208 		return mfspr(SPRN_PURR);
209 	return mftb();
210 }
211 
212 /*
213  * Read the SPURR on systems that have it, otherwise the purr
214  */
215 static u64 read_spurr(u64 purr)
216 {
217 	/*
218 	 * cpus without PURR won't have a SPURR
219 	 * We already know the former when we use this, so tell gcc
220 	 */
221 	if (cpu_has_feature(CPU_FTR_PURR) && cpu_has_feature(CPU_FTR_SPURR))
222 		return mfspr(SPRN_SPURR);
223 	return purr;
224 }
225 
226 /*
227  * Account time for a transition between system, hard irq
228  * or soft irq state.
229  */
230 void account_system_vtime(struct task_struct *tsk)
231 {
232 	u64 now, nowscaled, delta, deltascaled, sys_time;
233 	unsigned long flags;
234 
235 	local_irq_save(flags);
236 	now = read_purr();
237 	nowscaled = read_spurr(now);
238 	delta = now - get_paca()->startpurr;
239 	deltascaled = nowscaled - get_paca()->startspurr;
240 	get_paca()->startpurr = now;
241 	get_paca()->startspurr = nowscaled;
242 	if (!in_interrupt()) {
243 		/* deltascaled includes both user and system time.
244 		 * Hence scale it based on the purr ratio to estimate
245 		 * the system time */
246 		sys_time = get_paca()->system_time;
247 		if (get_paca()->user_time)
248 			deltascaled = deltascaled * sys_time /
249 			     (sys_time + get_paca()->user_time);
250 		delta += sys_time;
251 		get_paca()->system_time = 0;
252 	}
253 	if (in_irq() || idle_task(smp_processor_id()) != tsk)
254 		account_system_time(tsk, 0, delta, deltascaled);
255 	else
256 		account_idle_time(delta);
257 	__get_cpu_var(cputime_last_delta) = delta;
258 	__get_cpu_var(cputime_scaled_last_delta) = deltascaled;
259 	local_irq_restore(flags);
260 }
261 EXPORT_SYMBOL_GPL(account_system_vtime);
262 
263 /*
264  * Transfer the user and system times accumulated in the paca
265  * by the exception entry and exit code to the generic process
266  * user and system time records.
267  * Must be called with interrupts disabled.
268  */
269 void account_process_tick(struct task_struct *tsk, int user_tick)
270 {
271 	cputime_t utime, utimescaled;
272 
273 	utime = get_paca()->user_time;
274 	get_paca()->user_time = 0;
275 	utimescaled = cputime_to_scaled(utime);
276 	account_user_time(tsk, utime, utimescaled);
277 }
278 
279 /*
280  * Stuff for accounting stolen time.
281  */
282 struct cpu_purr_data {
283 	int	initialized;			/* thread is running */
284 	u64	tb;			/* last TB value read */
285 	u64	purr;			/* last PURR value read */
286 	u64	spurr;			/* last SPURR value read */
287 };
288 
289 /*
290  * Each entry in the cpu_purr_data array is manipulated only by its
291  * "owner" cpu -- usually in the timer interrupt but also occasionally
292  * in process context for cpu online.  As long as cpus do not touch
293  * each others' cpu_purr_data, disabling local interrupts is
294  * sufficient to serialize accesses.
295  */
296 static DEFINE_PER_CPU(struct cpu_purr_data, cpu_purr_data);
297 
298 static void snapshot_tb_and_purr(void *data)
299 {
300 	unsigned long flags;
301 	struct cpu_purr_data *p = &__get_cpu_var(cpu_purr_data);
302 
303 	local_irq_save(flags);
304 	p->tb = get_tb_or_rtc();
305 	p->purr = mfspr(SPRN_PURR);
306 	wmb();
307 	p->initialized = 1;
308 	local_irq_restore(flags);
309 }
310 
311 /*
312  * Called during boot when all cpus have come up.
313  */
314 void snapshot_timebases(void)
315 {
316 	if (!cpu_has_feature(CPU_FTR_PURR))
317 		return;
318 	on_each_cpu(snapshot_tb_and_purr, NULL, 1);
319 }
320 
321 /*
322  * Must be called with interrupts disabled.
323  */
324 void calculate_steal_time(void)
325 {
326 	u64 tb, purr;
327 	s64 stolen;
328 	struct cpu_purr_data *pme;
329 
330 	pme = &__get_cpu_var(cpu_purr_data);
331 	if (!pme->initialized)
332 		return;		/* !CPU_FTR_PURR or early in early boot */
333 	tb = mftb();
334 	purr = mfspr(SPRN_PURR);
335 	stolen = (tb - pme->tb) - (purr - pme->purr);
336 	if (stolen > 0) {
337 		if (idle_task(smp_processor_id()) != current)
338 			account_steal_time(stolen);
339 		else
340 			account_idle_time(stolen);
341 	}
342 	pme->tb = tb;
343 	pme->purr = purr;
344 }
345 
346 #ifdef CONFIG_PPC_SPLPAR
347 /*
348  * Must be called before the cpu is added to the online map when
349  * a cpu is being brought up at runtime.
350  */
351 static void snapshot_purr(void)
352 {
353 	struct cpu_purr_data *pme;
354 	unsigned long flags;
355 
356 	if (!cpu_has_feature(CPU_FTR_PURR))
357 		return;
358 	local_irq_save(flags);
359 	pme = &__get_cpu_var(cpu_purr_data);
360 	pme->tb = mftb();
361 	pme->purr = mfspr(SPRN_PURR);
362 	pme->initialized = 1;
363 	local_irq_restore(flags);
364 }
365 
366 #endif /* CONFIG_PPC_SPLPAR */
367 
368 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
369 #define calc_cputime_factors()
370 #define calculate_steal_time()		do { } while (0)
371 #endif
372 
373 #if !(defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR))
374 #define snapshot_purr()			do { } while (0)
375 #endif
376 
377 /*
378  * Called when a cpu comes up after the system has finished booting,
379  * i.e. as a result of a hotplug cpu action.
380  */
381 void snapshot_timebase(void)
382 {
383 	__get_cpu_var(last_jiffy) = get_tb_or_rtc();
384 	snapshot_purr();
385 }
386 
387 void __delay(unsigned long loops)
388 {
389 	unsigned long start;
390 	int diff;
391 
392 	if (__USE_RTC()) {
393 		start = get_rtcl();
394 		do {
395 			/* the RTCL register wraps at 1000000000 */
396 			diff = get_rtcl() - start;
397 			if (diff < 0)
398 				diff += 1000000000;
399 		} while (diff < loops);
400 	} else {
401 		start = get_tbl();
402 		while (get_tbl() - start < loops)
403 			HMT_low();
404 		HMT_medium();
405 	}
406 }
407 EXPORT_SYMBOL(__delay);
408 
409 void udelay(unsigned long usecs)
410 {
411 	__delay(tb_ticks_per_usec * usecs);
412 }
413 EXPORT_SYMBOL(udelay);
414 
415 #ifdef CONFIG_SMP
416 unsigned long profile_pc(struct pt_regs *regs)
417 {
418 	unsigned long pc = instruction_pointer(regs);
419 
420 	if (in_lock_functions(pc))
421 		return regs->link;
422 
423 	return pc;
424 }
425 EXPORT_SYMBOL(profile_pc);
426 #endif
427 
428 #ifdef CONFIG_PPC_ISERIES
429 
430 /*
431  * This function recalibrates the timebase based on the 49-bit time-of-day
432  * value in the Titan chip.  The Titan is much more accurate than the value
433  * returned by the service processor for the timebase frequency.
434  */
435 
436 static int __init iSeries_tb_recal(void)
437 {
438 	unsigned long titan, tb;
439 
440 	/* Make sure we only run on iSeries */
441 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
442 		return -ENODEV;
443 
444 	tb = get_tb();
445 	titan = HvCallXm_loadTod();
446 	if ( iSeries_recal_titan ) {
447 		unsigned long tb_ticks = tb - iSeries_recal_tb;
448 		unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
449 		unsigned long new_tb_ticks_per_sec   = (tb_ticks * USEC_PER_SEC)/titan_usec;
450 		unsigned long new_tb_ticks_per_jiffy =
451 			DIV_ROUND_CLOSEST(new_tb_ticks_per_sec, HZ);
452 		long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
453 		char sign = '+';
454 		/* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
455 		new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;
456 
457 		if ( tick_diff < 0 ) {
458 			tick_diff = -tick_diff;
459 			sign = '-';
460 		}
461 		if ( tick_diff ) {
462 			if ( tick_diff < tb_ticks_per_jiffy/25 ) {
463 				printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
464 						new_tb_ticks_per_jiffy, sign, tick_diff );
465 				tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
466 				tb_ticks_per_sec   = new_tb_ticks_per_sec;
467 				calc_cputime_factors();
468 				vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
469 				setup_cputime_one_jiffy();
470 			}
471 			else {
472 				printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
473 					"                   new tb_ticks_per_jiffy = %lu\n"
474 					"                   old tb_ticks_per_jiffy = %lu\n",
475 					new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
476 			}
477 		}
478 	}
479 	iSeries_recal_titan = titan;
480 	iSeries_recal_tb = tb;
481 
482 	/* Called here as now we know accurate values for the timebase */
483 	clocksource_init();
484 	return 0;
485 }
486 late_initcall(iSeries_tb_recal);
487 
488 /* Called from platform early init */
489 void __init iSeries_time_init_early(void)
490 {
491 	iSeries_recal_tb = get_tb();
492 	iSeries_recal_titan = HvCallXm_loadTod();
493 }
494 #endif /* CONFIG_PPC_ISERIES */
495 
496 #ifdef CONFIG_PERF_EVENTS
497 
498 /*
499  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
500  */
501 #ifdef CONFIG_PPC64
502 static inline unsigned long test_perf_event_pending(void)
503 {
504 	unsigned long x;
505 
506 	asm volatile("lbz %0,%1(13)"
507 		: "=r" (x)
508 		: "i" (offsetof(struct paca_struct, perf_event_pending)));
509 	return x;
510 }
511 
512 static inline void set_perf_event_pending_flag(void)
513 {
514 	asm volatile("stb %0,%1(13)" : :
515 		"r" (1),
516 		"i" (offsetof(struct paca_struct, perf_event_pending)));
517 }
518 
519 static inline void clear_perf_event_pending(void)
520 {
521 	asm volatile("stb %0,%1(13)" : :
522 		"r" (0),
523 		"i" (offsetof(struct paca_struct, perf_event_pending)));
524 }
525 
526 #else /* 32-bit */
527 
528 DEFINE_PER_CPU(u8, perf_event_pending);
529 
530 #define set_perf_event_pending_flag()	__get_cpu_var(perf_event_pending) = 1
531 #define test_perf_event_pending()	__get_cpu_var(perf_event_pending)
532 #define clear_perf_event_pending()	__get_cpu_var(perf_event_pending) = 0
533 
534 #endif /* 32 vs 64 bit */
535 
536 void set_perf_event_pending(void)
537 {
538 	preempt_disable();
539 	set_perf_event_pending_flag();
540 	set_dec(1);
541 	preempt_enable();
542 }
543 
544 #else  /* CONFIG_PERF_EVENTS */
545 
546 #define test_perf_event_pending()	0
547 #define clear_perf_event_pending()
548 
549 #endif /* CONFIG_PERF_EVENTS */
550 
551 /*
552  * For iSeries shared processors, we have to let the hypervisor
553  * set the hardware decrementer.  We set a virtual decrementer
554  * in the lppaca and call the hypervisor if the virtual
555  * decrementer is less than the current value in the hardware
556  * decrementer. (almost always the new decrementer value will
557  * be greater than the current hardware decementer so the hypervisor
558  * call will not be needed)
559  */
560 
561 /*
562  * timer_interrupt - gets called when the decrementer overflows,
563  * with interrupts disabled.
564  */
565 void timer_interrupt(struct pt_regs * regs)
566 {
567 	struct pt_regs *old_regs;
568 	struct decrementer_clock *decrementer =  &__get_cpu_var(decrementers);
569 	struct clock_event_device *evt = &decrementer->event;
570 	u64 now;
571 
572 	trace_timer_interrupt_entry(regs);
573 
574 	__get_cpu_var(irq_stat).timer_irqs++;
575 
576 	/* Ensure a positive value is written to the decrementer, or else
577 	 * some CPUs will continuue to take decrementer exceptions */
578 	set_dec(DECREMENTER_MAX);
579 
580 #ifdef CONFIG_PPC32
581 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
582 		do_IRQ(regs);
583 #endif
584 
585 	now = get_tb_or_rtc();
586 	if (now < decrementer->next_tb) {
587 		/* not time for this event yet */
588 		now = decrementer->next_tb - now;
589 		if (now <= DECREMENTER_MAX)
590 			set_dec((int)now);
591 		trace_timer_interrupt_exit(regs);
592 		return;
593 	}
594 	old_regs = set_irq_regs(regs);
595 	irq_enter();
596 
597 	calculate_steal_time();
598 
599 	if (test_perf_event_pending()) {
600 		clear_perf_event_pending();
601 		perf_event_do_pending();
602 	}
603 
604 #ifdef CONFIG_PPC_ISERIES
605 	if (firmware_has_feature(FW_FEATURE_ISERIES))
606 		get_lppaca()->int_dword.fields.decr_int = 0;
607 #endif
608 
609 	if (evt->event_handler)
610 		evt->event_handler(evt);
611 
612 #ifdef CONFIG_PPC_ISERIES
613 	if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending())
614 		process_hvlpevents();
615 #endif
616 
617 #ifdef CONFIG_PPC64
618 	/* collect purr register values often, for accurate calculations */
619 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
620 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
621 		cu->current_tb = mfspr(SPRN_PURR);
622 	}
623 #endif
624 
625 	irq_exit();
626 	set_irq_regs(old_regs);
627 
628 	trace_timer_interrupt_exit(regs);
629 }
630 
631 #ifdef CONFIG_SUSPEND
632 static void generic_suspend_disable_irqs(void)
633 {
634 	/* Disable the decrementer, so that it doesn't interfere
635 	 * with suspending.
636 	 */
637 
638 	set_dec(0x7fffffff);
639 	local_irq_disable();
640 	set_dec(0x7fffffff);
641 }
642 
643 static void generic_suspend_enable_irqs(void)
644 {
645 	local_irq_enable();
646 }
647 
648 /* Overrides the weak version in kernel/power/main.c */
649 void arch_suspend_disable_irqs(void)
650 {
651 	if (ppc_md.suspend_disable_irqs)
652 		ppc_md.suspend_disable_irqs();
653 	generic_suspend_disable_irqs();
654 }
655 
656 /* Overrides the weak version in kernel/power/main.c */
657 void arch_suspend_enable_irqs(void)
658 {
659 	generic_suspend_enable_irqs();
660 	if (ppc_md.suspend_enable_irqs)
661 		ppc_md.suspend_enable_irqs();
662 }
663 #endif
664 
665 /*
666  * Scheduler clock - returns current time in nanosec units.
667  *
668  * Note: mulhdu(a, b) (multiply high double unsigned) returns
669  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
670  * are 64-bit unsigned numbers.
671  */
672 unsigned long long sched_clock(void)
673 {
674 	if (__USE_RTC())
675 		return get_rtc();
676 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
677 }
678 
679 static int __init get_freq(char *name, int cells, unsigned long *val)
680 {
681 	struct device_node *cpu;
682 	const unsigned int *fp;
683 	int found = 0;
684 
685 	/* The cpu node should have timebase and clock frequency properties */
686 	cpu = of_find_node_by_type(NULL, "cpu");
687 
688 	if (cpu) {
689 		fp = of_get_property(cpu, name, NULL);
690 		if (fp) {
691 			found = 1;
692 			*val = of_read_ulong(fp, cells);
693 		}
694 
695 		of_node_put(cpu);
696 	}
697 
698 	return found;
699 }
700 
701 /* should become __cpuinit when secondary_cpu_time_init also is */
702 void start_cpu_decrementer(void)
703 {
704 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
705 	/* Clear any pending timer interrupts */
706 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
707 
708 	/* Enable decrementer interrupt */
709 	mtspr(SPRN_TCR, TCR_DIE);
710 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
711 }
712 
713 void __init generic_calibrate_decr(void)
714 {
715 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
716 
717 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
718 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
719 
720 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
721 				"(not found)\n");
722 	}
723 
724 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
725 
726 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
727 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
728 
729 		printk(KERN_ERR "WARNING: Estimating processor frequency "
730 				"(not found)\n");
731 	}
732 }
733 
734 int update_persistent_clock(struct timespec now)
735 {
736 	struct rtc_time tm;
737 
738 	if (!ppc_md.set_rtc_time)
739 		return 0;
740 
741 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
742 	tm.tm_year -= 1900;
743 	tm.tm_mon -= 1;
744 
745 	return ppc_md.set_rtc_time(&tm);
746 }
747 
748 static void __read_persistent_clock(struct timespec *ts)
749 {
750 	struct rtc_time tm;
751 	static int first = 1;
752 
753 	ts->tv_nsec = 0;
754 	/* XXX this is a litle fragile but will work okay in the short term */
755 	if (first) {
756 		first = 0;
757 		if (ppc_md.time_init)
758 			timezone_offset = ppc_md.time_init();
759 
760 		/* get_boot_time() isn't guaranteed to be safe to call late */
761 		if (ppc_md.get_boot_time) {
762 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
763 			return;
764 		}
765 	}
766 	if (!ppc_md.get_rtc_time) {
767 		ts->tv_sec = 0;
768 		return;
769 	}
770 	ppc_md.get_rtc_time(&tm);
771 
772 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
773 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
774 }
775 
776 void read_persistent_clock(struct timespec *ts)
777 {
778 	__read_persistent_clock(ts);
779 
780 	/* Sanitize it in case real time clock is set below EPOCH */
781 	if (ts->tv_sec < 0) {
782 		ts->tv_sec = 0;
783 		ts->tv_nsec = 0;
784 	}
785 
786 }
787 
788 /* clocksource code */
789 static cycle_t rtc_read(struct clocksource *cs)
790 {
791 	return (cycle_t)get_rtc();
792 }
793 
794 static cycle_t timebase_read(struct clocksource *cs)
795 {
796 	return (cycle_t)get_tb();
797 }
798 
799 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
800 			struct clocksource *clock, u32 mult)
801 {
802 	u64 new_tb_to_xs, new_stamp_xsec;
803 	u32 frac_sec;
804 
805 	if (clock != &clocksource_timebase)
806 		return;
807 
808 	/* Make userspace gettimeofday spin until we're done. */
809 	++vdso_data->tb_update_count;
810 	smp_mb();
811 
812 	/* XXX this assumes clock->shift == 22 */
813 	/* 4611686018 ~= 2^(20+64-22) / 1e9 */
814 	new_tb_to_xs = (u64) mult * 4611686018ULL;
815 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
816 	do_div(new_stamp_xsec, 1000000000);
817 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
818 
819 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
820 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
821 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
822 
823 	/*
824 	 * tb_update_count is used to allow the userspace gettimeofday code
825 	 * to assure itself that it sees a consistent view of the tb_to_xs and
826 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
827 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
828 	 * the two values of tb_update_count match and are even then the
829 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
830 	 * loops back and reads them again until this criteria is met.
831 	 * We expect the caller to have done the first increment of
832 	 * vdso_data->tb_update_count already.
833 	 */
834 	vdso_data->tb_orig_stamp = clock->cycle_last;
835 	vdso_data->stamp_xsec = new_stamp_xsec;
836 	vdso_data->tb_to_xs = new_tb_to_xs;
837 	vdso_data->wtom_clock_sec = wtm->tv_sec;
838 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
839 	vdso_data->stamp_xtime = *wall_time;
840 	vdso_data->stamp_sec_fraction = frac_sec;
841 	smp_wmb();
842 	++(vdso_data->tb_update_count);
843 }
844 
845 void update_vsyscall_tz(void)
846 {
847 	/* Make userspace gettimeofday spin until we're done. */
848 	++vdso_data->tb_update_count;
849 	smp_mb();
850 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
851 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
852 	smp_mb();
853 	++vdso_data->tb_update_count;
854 }
855 
856 static void __init clocksource_init(void)
857 {
858 	struct clocksource *clock;
859 
860 	if (__USE_RTC())
861 		clock = &clocksource_rtc;
862 	else
863 		clock = &clocksource_timebase;
864 
865 	clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
866 
867 	if (clocksource_register(clock)) {
868 		printk(KERN_ERR "clocksource: %s is already registered\n",
869 		       clock->name);
870 		return;
871 	}
872 
873 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
874 	       clock->name, clock->mult, clock->shift);
875 }
876 
877 static int decrementer_set_next_event(unsigned long evt,
878 				      struct clock_event_device *dev)
879 {
880 	__get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
881 	set_dec(evt);
882 	return 0;
883 }
884 
885 static void decrementer_set_mode(enum clock_event_mode mode,
886 				 struct clock_event_device *dev)
887 {
888 	if (mode != CLOCK_EVT_MODE_ONESHOT)
889 		decrementer_set_next_event(DECREMENTER_MAX, dev);
890 }
891 
892 static inline uint64_t div_sc64(unsigned long ticks, unsigned long nsec,
893 				int shift)
894 {
895 	uint64_t tmp = ((uint64_t)ticks) << shift;
896 
897 	do_div(tmp, nsec);
898 	return tmp;
899 }
900 
901 static void __init setup_clockevent_multiplier(unsigned long hz)
902 {
903 	u64 mult, shift = 32;
904 
905 	while (1) {
906 		mult = div_sc64(hz, NSEC_PER_SEC, shift);
907 		if (mult && (mult >> 32UL) == 0UL)
908 			break;
909 
910 		shift--;
911 	}
912 
913 	decrementer_clockevent.shift = shift;
914 	decrementer_clockevent.mult = mult;
915 }
916 
917 static void register_decrementer_clockevent(int cpu)
918 {
919 	struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
920 
921 	*dec = decrementer_clockevent;
922 	dec->cpumask = cpumask_of(cpu);
923 
924 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
925 		    dec->name, dec->mult, dec->shift, cpu);
926 
927 	clockevents_register_device(dec);
928 }
929 
930 static void __init init_decrementer_clockevent(void)
931 {
932 	int cpu = smp_processor_id();
933 
934 	setup_clockevent_multiplier(ppc_tb_freq);
935 	decrementer_clockevent.max_delta_ns =
936 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
937 	decrementer_clockevent.min_delta_ns =
938 		clockevent_delta2ns(2, &decrementer_clockevent);
939 
940 	register_decrementer_clockevent(cpu);
941 }
942 
943 void secondary_cpu_time_init(void)
944 {
945 	/* Start the decrementer on CPUs that have manual control
946 	 * such as BookE
947 	 */
948 	start_cpu_decrementer();
949 
950 	/* FIME: Should make unrelatred change to move snapshot_timebase
951 	 * call here ! */
952 	register_decrementer_clockevent(smp_processor_id());
953 }
954 
955 /* This function is only called on the boot processor */
956 void __init time_init(void)
957 {
958 	struct div_result res;
959 	u64 scale;
960 	unsigned shift;
961 
962 	if (__USE_RTC()) {
963 		/* 601 processor: dec counts down by 128 every 128ns */
964 		ppc_tb_freq = 1000000000;
965 	} else {
966 		/* Normal PowerPC with timebase register */
967 		ppc_md.calibrate_decr();
968 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
969 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
970 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
971 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
972 	}
973 
974 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
975 	tb_ticks_per_sec = ppc_tb_freq;
976 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
977 	calc_cputime_factors();
978 	setup_cputime_one_jiffy();
979 
980 	/*
981 	 * Compute scale factor for sched_clock.
982 	 * The calibrate_decr() function has set tb_ticks_per_sec,
983 	 * which is the timebase frequency.
984 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
985 	 * the 128-bit result as a 64.64 fixed-point number.
986 	 * We then shift that number right until it is less than 1.0,
987 	 * giving us the scale factor and shift count to use in
988 	 * sched_clock().
989 	 */
990 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
991 	scale = res.result_low;
992 	for (shift = 0; res.result_high != 0; ++shift) {
993 		scale = (scale >> 1) | (res.result_high << 63);
994 		res.result_high >>= 1;
995 	}
996 	tb_to_ns_scale = scale;
997 	tb_to_ns_shift = shift;
998 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
999 	boot_tb = get_tb_or_rtc();
1000 
1001 	/* If platform provided a timezone (pmac), we correct the time */
1002         if (timezone_offset) {
1003 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1004 		sys_tz.tz_dsttime = 0;
1005         }
1006 
1007 	vdso_data->tb_update_count = 0;
1008 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1009 
1010 	/* Start the decrementer on CPUs that have manual control
1011 	 * such as BookE
1012 	 */
1013 	start_cpu_decrementer();
1014 
1015 	/* Register the clocksource, if we're not running on iSeries */
1016 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
1017 		clocksource_init();
1018 
1019 	init_decrementer_clockevent();
1020 }
1021 
1022 
1023 #define FEBRUARY	2
1024 #define	STARTOFTIME	1970
1025 #define SECDAY		86400L
1026 #define SECYR		(SECDAY * 365)
1027 #define	leapyear(year)		((year) % 4 == 0 && \
1028 				 ((year) % 100 != 0 || (year) % 400 == 0))
1029 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1030 #define	days_in_month(a) 	(month_days[(a) - 1])
1031 
1032 static int month_days[12] = {
1033 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1034 };
1035 
1036 /*
1037  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1038  */
1039 void GregorianDay(struct rtc_time * tm)
1040 {
1041 	int leapsToDate;
1042 	int lastYear;
1043 	int day;
1044 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1045 
1046 	lastYear = tm->tm_year - 1;
1047 
1048 	/*
1049 	 * Number of leap corrections to apply up to end of last year
1050 	 */
1051 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
1052 
1053 	/*
1054 	 * This year is a leap year if it is divisible by 4 except when it is
1055 	 * divisible by 100 unless it is divisible by 400
1056 	 *
1057 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1058 	 */
1059 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
1060 
1061 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
1062 		   tm->tm_mday;
1063 
1064 	tm->tm_wday = day % 7;
1065 }
1066 
1067 void to_tm(int tim, struct rtc_time * tm)
1068 {
1069 	register int    i;
1070 	register long   hms, day;
1071 
1072 	day = tim / SECDAY;
1073 	hms = tim % SECDAY;
1074 
1075 	/* Hours, minutes, seconds are easy */
1076 	tm->tm_hour = hms / 3600;
1077 	tm->tm_min = (hms % 3600) / 60;
1078 	tm->tm_sec = (hms % 3600) % 60;
1079 
1080 	/* Number of years in days */
1081 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1082 		day -= days_in_year(i);
1083 	tm->tm_year = i;
1084 
1085 	/* Number of months in days left */
1086 	if (leapyear(tm->tm_year))
1087 		days_in_month(FEBRUARY) = 29;
1088 	for (i = 1; day >= days_in_month(i); i++)
1089 		day -= days_in_month(i);
1090 	days_in_month(FEBRUARY) = 28;
1091 	tm->tm_mon = i;
1092 
1093 	/* Days are what is left over (+1) from all that. */
1094 	tm->tm_mday = day + 1;
1095 
1096 	/*
1097 	 * Determine the day of week
1098 	 */
1099 	GregorianDay(tm);
1100 }
1101 
1102 /*
1103  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1104  * result.
1105  */
1106 void div128_by_32(u64 dividend_high, u64 dividend_low,
1107 		  unsigned divisor, struct div_result *dr)
1108 {
1109 	unsigned long a, b, c, d;
1110 	unsigned long w, x, y, z;
1111 	u64 ra, rb, rc;
1112 
1113 	a = dividend_high >> 32;
1114 	b = dividend_high & 0xffffffff;
1115 	c = dividend_low >> 32;
1116 	d = dividend_low & 0xffffffff;
1117 
1118 	w = a / divisor;
1119 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1120 
1121 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1122 	x = ra;
1123 
1124 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1125 	y = rb;
1126 
1127 	do_div(rc, divisor);
1128 	z = rc;
1129 
1130 	dr->result_high = ((u64)w << 32) + x;
1131 	dr->result_low  = ((u64)y << 32) + z;
1132 
1133 }
1134 
1135 /* We don't need to calibrate delay, we use the CPU timebase for that */
1136 void calibrate_delay(void)
1137 {
1138 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1139 	 * as the number of __delay(1) in a jiffy, so make it so
1140 	 */
1141 	loops_per_jiffy = tb_ticks_per_jiffy;
1142 }
1143 
1144 static int __init rtc_init(void)
1145 {
1146 	struct platform_device *pdev;
1147 
1148 	if (!ppc_md.get_rtc_time)
1149 		return -ENODEV;
1150 
1151 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1152 	if (IS_ERR(pdev))
1153 		return PTR_ERR(pdev);
1154 
1155 	return 0;
1156 }
1157 
1158 module_init(rtc_init);
1159