xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision 3d8a1a6a8af910cc2da566080d111e062a124ba6)
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 #if defined(CONFIG_PPC32) && defined(CONFIG_PMAC)
581 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
582 		do_IRQ(regs);
583 #endif
584 
585 	old_regs = set_irq_regs(regs);
586 	irq_enter();
587 
588 	calculate_steal_time();
589 
590 	if (test_perf_event_pending()) {
591 		clear_perf_event_pending();
592 		perf_event_do_pending();
593 	}
594 
595 #ifdef CONFIG_PPC_ISERIES
596 	if (firmware_has_feature(FW_FEATURE_ISERIES))
597 		get_lppaca()->int_dword.fields.decr_int = 0;
598 #endif
599 
600 	now = get_tb_or_rtc();
601 	if (now >= decrementer->next_tb) {
602 		decrementer->next_tb = ~(u64)0;
603 		if (evt->event_handler)
604 			evt->event_handler(evt);
605 	} else {
606 		now = decrementer->next_tb - now;
607 		if (now <= DECREMENTER_MAX)
608 			set_dec((int)now);
609 	}
610 
611 #ifdef CONFIG_PPC_ISERIES
612 	if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending())
613 		process_hvlpevents();
614 #endif
615 
616 #ifdef CONFIG_PPC64
617 	/* collect purr register values often, for accurate calculations */
618 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
619 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
620 		cu->current_tb = mfspr(SPRN_PURR);
621 	}
622 #endif
623 
624 	irq_exit();
625 	set_irq_regs(old_regs);
626 
627 	trace_timer_interrupt_exit(regs);
628 }
629 
630 #ifdef CONFIG_SUSPEND
631 static void generic_suspend_disable_irqs(void)
632 {
633 	/* Disable the decrementer, so that it doesn't interfere
634 	 * with suspending.
635 	 */
636 
637 	set_dec(0x7fffffff);
638 	local_irq_disable();
639 	set_dec(0x7fffffff);
640 }
641 
642 static void generic_suspend_enable_irqs(void)
643 {
644 	local_irq_enable();
645 }
646 
647 /* Overrides the weak version in kernel/power/main.c */
648 void arch_suspend_disable_irqs(void)
649 {
650 	if (ppc_md.suspend_disable_irqs)
651 		ppc_md.suspend_disable_irqs();
652 	generic_suspend_disable_irqs();
653 }
654 
655 /* Overrides the weak version in kernel/power/main.c */
656 void arch_suspend_enable_irqs(void)
657 {
658 	generic_suspend_enable_irqs();
659 	if (ppc_md.suspend_enable_irqs)
660 		ppc_md.suspend_enable_irqs();
661 }
662 #endif
663 
664 /*
665  * Scheduler clock - returns current time in nanosec units.
666  *
667  * Note: mulhdu(a, b) (multiply high double unsigned) returns
668  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
669  * are 64-bit unsigned numbers.
670  */
671 unsigned long long sched_clock(void)
672 {
673 	if (__USE_RTC())
674 		return get_rtc();
675 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
676 }
677 
678 static int __init get_freq(char *name, int cells, unsigned long *val)
679 {
680 	struct device_node *cpu;
681 	const unsigned int *fp;
682 	int found = 0;
683 
684 	/* The cpu node should have timebase and clock frequency properties */
685 	cpu = of_find_node_by_type(NULL, "cpu");
686 
687 	if (cpu) {
688 		fp = of_get_property(cpu, name, NULL);
689 		if (fp) {
690 			found = 1;
691 			*val = of_read_ulong(fp, cells);
692 		}
693 
694 		of_node_put(cpu);
695 	}
696 
697 	return found;
698 }
699 
700 /* should become __cpuinit when secondary_cpu_time_init also is */
701 void start_cpu_decrementer(void)
702 {
703 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
704 	/* Clear any pending timer interrupts */
705 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
706 
707 	/* Enable decrementer interrupt */
708 	mtspr(SPRN_TCR, TCR_DIE);
709 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
710 }
711 
712 void __init generic_calibrate_decr(void)
713 {
714 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
715 
716 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
717 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
718 
719 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
720 				"(not found)\n");
721 	}
722 
723 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
724 
725 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
726 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
727 
728 		printk(KERN_ERR "WARNING: Estimating processor frequency "
729 				"(not found)\n");
730 	}
731 }
732 
733 int update_persistent_clock(struct timespec now)
734 {
735 	struct rtc_time tm;
736 
737 	if (!ppc_md.set_rtc_time)
738 		return 0;
739 
740 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
741 	tm.tm_year -= 1900;
742 	tm.tm_mon -= 1;
743 
744 	return ppc_md.set_rtc_time(&tm);
745 }
746 
747 static void __read_persistent_clock(struct timespec *ts)
748 {
749 	struct rtc_time tm;
750 	static int first = 1;
751 
752 	ts->tv_nsec = 0;
753 	/* XXX this is a litle fragile but will work okay in the short term */
754 	if (first) {
755 		first = 0;
756 		if (ppc_md.time_init)
757 			timezone_offset = ppc_md.time_init();
758 
759 		/* get_boot_time() isn't guaranteed to be safe to call late */
760 		if (ppc_md.get_boot_time) {
761 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
762 			return;
763 		}
764 	}
765 	if (!ppc_md.get_rtc_time) {
766 		ts->tv_sec = 0;
767 		return;
768 	}
769 	ppc_md.get_rtc_time(&tm);
770 
771 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
772 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
773 }
774 
775 void read_persistent_clock(struct timespec *ts)
776 {
777 	__read_persistent_clock(ts);
778 
779 	/* Sanitize it in case real time clock is set below EPOCH */
780 	if (ts->tv_sec < 0) {
781 		ts->tv_sec = 0;
782 		ts->tv_nsec = 0;
783 	}
784 
785 }
786 
787 /* clocksource code */
788 static cycle_t rtc_read(struct clocksource *cs)
789 {
790 	return (cycle_t)get_rtc();
791 }
792 
793 static cycle_t timebase_read(struct clocksource *cs)
794 {
795 	return (cycle_t)get_tb();
796 }
797 
798 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
799 			struct clocksource *clock, u32 mult)
800 {
801 	u64 new_tb_to_xs, new_stamp_xsec;
802 	u32 frac_sec;
803 
804 	if (clock != &clocksource_timebase)
805 		return;
806 
807 	/* Make userspace gettimeofday spin until we're done. */
808 	++vdso_data->tb_update_count;
809 	smp_mb();
810 
811 	/* XXX this assumes clock->shift == 22 */
812 	/* 4611686018 ~= 2^(20+64-22) / 1e9 */
813 	new_tb_to_xs = (u64) mult * 4611686018ULL;
814 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
815 	do_div(new_stamp_xsec, 1000000000);
816 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
817 
818 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
819 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
820 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
821 
822 	/*
823 	 * tb_update_count is used to allow the userspace gettimeofday code
824 	 * to assure itself that it sees a consistent view of the tb_to_xs and
825 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
826 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
827 	 * the two values of tb_update_count match and are even then the
828 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
829 	 * loops back and reads them again until this criteria is met.
830 	 * We expect the caller to have done the first increment of
831 	 * vdso_data->tb_update_count already.
832 	 */
833 	vdso_data->tb_orig_stamp = clock->cycle_last;
834 	vdso_data->stamp_xsec = new_stamp_xsec;
835 	vdso_data->tb_to_xs = new_tb_to_xs;
836 	vdso_data->wtom_clock_sec = wtm->tv_sec;
837 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
838 	vdso_data->stamp_xtime = *wall_time;
839 	vdso_data->stamp_sec_fraction = frac_sec;
840 	smp_wmb();
841 	++(vdso_data->tb_update_count);
842 }
843 
844 void update_vsyscall_tz(void)
845 {
846 	/* Make userspace gettimeofday spin until we're done. */
847 	++vdso_data->tb_update_count;
848 	smp_mb();
849 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
850 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
851 	smp_mb();
852 	++vdso_data->tb_update_count;
853 }
854 
855 static void __init clocksource_init(void)
856 {
857 	struct clocksource *clock;
858 
859 	if (__USE_RTC())
860 		clock = &clocksource_rtc;
861 	else
862 		clock = &clocksource_timebase;
863 
864 	clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
865 
866 	if (clocksource_register(clock)) {
867 		printk(KERN_ERR "clocksource: %s is already registered\n",
868 		       clock->name);
869 		return;
870 	}
871 
872 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
873 	       clock->name, clock->mult, clock->shift);
874 }
875 
876 static int decrementer_set_next_event(unsigned long evt,
877 				      struct clock_event_device *dev)
878 {
879 	__get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
880 	set_dec(evt);
881 	return 0;
882 }
883 
884 static void decrementer_set_mode(enum clock_event_mode mode,
885 				 struct clock_event_device *dev)
886 {
887 	if (mode != CLOCK_EVT_MODE_ONESHOT)
888 		decrementer_set_next_event(DECREMENTER_MAX, dev);
889 }
890 
891 static inline uint64_t div_sc64(unsigned long ticks, unsigned long nsec,
892 				int shift)
893 {
894 	uint64_t tmp = ((uint64_t)ticks) << shift;
895 
896 	do_div(tmp, nsec);
897 	return tmp;
898 }
899 
900 static void __init setup_clockevent_multiplier(unsigned long hz)
901 {
902 	u64 mult, shift = 32;
903 
904 	while (1) {
905 		mult = div_sc64(hz, NSEC_PER_SEC, shift);
906 		if (mult && (mult >> 32UL) == 0UL)
907 			break;
908 
909 		shift--;
910 	}
911 
912 	decrementer_clockevent.shift = shift;
913 	decrementer_clockevent.mult = mult;
914 }
915 
916 static void register_decrementer_clockevent(int cpu)
917 {
918 	struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
919 
920 	*dec = decrementer_clockevent;
921 	dec->cpumask = cpumask_of(cpu);
922 
923 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
924 		    dec->name, dec->mult, dec->shift, cpu);
925 
926 	clockevents_register_device(dec);
927 }
928 
929 static void __init init_decrementer_clockevent(void)
930 {
931 	int cpu = smp_processor_id();
932 
933 	setup_clockevent_multiplier(ppc_tb_freq);
934 	decrementer_clockevent.max_delta_ns =
935 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
936 	decrementer_clockevent.min_delta_ns =
937 		clockevent_delta2ns(2, &decrementer_clockevent);
938 
939 	register_decrementer_clockevent(cpu);
940 }
941 
942 void secondary_cpu_time_init(void)
943 {
944 	/* Start the decrementer on CPUs that have manual control
945 	 * such as BookE
946 	 */
947 	start_cpu_decrementer();
948 
949 	/* FIME: Should make unrelatred change to move snapshot_timebase
950 	 * call here ! */
951 	register_decrementer_clockevent(smp_processor_id());
952 }
953 
954 /* This function is only called on the boot processor */
955 void __init time_init(void)
956 {
957 	struct div_result res;
958 	u64 scale;
959 	unsigned shift;
960 
961 	if (__USE_RTC()) {
962 		/* 601 processor: dec counts down by 128 every 128ns */
963 		ppc_tb_freq = 1000000000;
964 	} else {
965 		/* Normal PowerPC with timebase register */
966 		ppc_md.calibrate_decr();
967 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
968 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
969 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
970 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
971 	}
972 
973 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
974 	tb_ticks_per_sec = ppc_tb_freq;
975 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
976 	calc_cputime_factors();
977 	setup_cputime_one_jiffy();
978 
979 	/*
980 	 * Compute scale factor for sched_clock.
981 	 * The calibrate_decr() function has set tb_ticks_per_sec,
982 	 * which is the timebase frequency.
983 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
984 	 * the 128-bit result as a 64.64 fixed-point number.
985 	 * We then shift that number right until it is less than 1.0,
986 	 * giving us the scale factor and shift count to use in
987 	 * sched_clock().
988 	 */
989 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
990 	scale = res.result_low;
991 	for (shift = 0; res.result_high != 0; ++shift) {
992 		scale = (scale >> 1) | (res.result_high << 63);
993 		res.result_high >>= 1;
994 	}
995 	tb_to_ns_scale = scale;
996 	tb_to_ns_shift = shift;
997 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
998 	boot_tb = get_tb_or_rtc();
999 
1000 	/* If platform provided a timezone (pmac), we correct the time */
1001         if (timezone_offset) {
1002 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1003 		sys_tz.tz_dsttime = 0;
1004         }
1005 
1006 	vdso_data->tb_update_count = 0;
1007 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1008 
1009 	/* Start the decrementer on CPUs that have manual control
1010 	 * such as BookE
1011 	 */
1012 	start_cpu_decrementer();
1013 
1014 	/* Register the clocksource, if we're not running on iSeries */
1015 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
1016 		clocksource_init();
1017 
1018 	init_decrementer_clockevent();
1019 }
1020 
1021 
1022 #define FEBRUARY	2
1023 #define	STARTOFTIME	1970
1024 #define SECDAY		86400L
1025 #define SECYR		(SECDAY * 365)
1026 #define	leapyear(year)		((year) % 4 == 0 && \
1027 				 ((year) % 100 != 0 || (year) % 400 == 0))
1028 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1029 #define	days_in_month(a) 	(month_days[(a) - 1])
1030 
1031 static int month_days[12] = {
1032 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1033 };
1034 
1035 /*
1036  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1037  */
1038 void GregorianDay(struct rtc_time * tm)
1039 {
1040 	int leapsToDate;
1041 	int lastYear;
1042 	int day;
1043 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1044 
1045 	lastYear = tm->tm_year - 1;
1046 
1047 	/*
1048 	 * Number of leap corrections to apply up to end of last year
1049 	 */
1050 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
1051 
1052 	/*
1053 	 * This year is a leap year if it is divisible by 4 except when it is
1054 	 * divisible by 100 unless it is divisible by 400
1055 	 *
1056 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1057 	 */
1058 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
1059 
1060 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
1061 		   tm->tm_mday;
1062 
1063 	tm->tm_wday = day % 7;
1064 }
1065 
1066 void to_tm(int tim, struct rtc_time * tm)
1067 {
1068 	register int    i;
1069 	register long   hms, day;
1070 
1071 	day = tim / SECDAY;
1072 	hms = tim % SECDAY;
1073 
1074 	/* Hours, minutes, seconds are easy */
1075 	tm->tm_hour = hms / 3600;
1076 	tm->tm_min = (hms % 3600) / 60;
1077 	tm->tm_sec = (hms % 3600) % 60;
1078 
1079 	/* Number of years in days */
1080 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1081 		day -= days_in_year(i);
1082 	tm->tm_year = i;
1083 
1084 	/* Number of months in days left */
1085 	if (leapyear(tm->tm_year))
1086 		days_in_month(FEBRUARY) = 29;
1087 	for (i = 1; day >= days_in_month(i); i++)
1088 		day -= days_in_month(i);
1089 	days_in_month(FEBRUARY) = 28;
1090 	tm->tm_mon = i;
1091 
1092 	/* Days are what is left over (+1) from all that. */
1093 	tm->tm_mday = day + 1;
1094 
1095 	/*
1096 	 * Determine the day of week
1097 	 */
1098 	GregorianDay(tm);
1099 }
1100 
1101 /*
1102  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1103  * result.
1104  */
1105 void div128_by_32(u64 dividend_high, u64 dividend_low,
1106 		  unsigned divisor, struct div_result *dr)
1107 {
1108 	unsigned long a, b, c, d;
1109 	unsigned long w, x, y, z;
1110 	u64 ra, rb, rc;
1111 
1112 	a = dividend_high >> 32;
1113 	b = dividend_high & 0xffffffff;
1114 	c = dividend_low >> 32;
1115 	d = dividend_low & 0xffffffff;
1116 
1117 	w = a / divisor;
1118 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1119 
1120 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1121 	x = ra;
1122 
1123 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1124 	y = rb;
1125 
1126 	do_div(rc, divisor);
1127 	z = rc;
1128 
1129 	dr->result_high = ((u64)w << 32) + x;
1130 	dr->result_low  = ((u64)y << 32) + z;
1131 
1132 }
1133 
1134 /* We don't need to calibrate delay, we use the CPU timebase for that */
1135 void calibrate_delay(void)
1136 {
1137 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1138 	 * as the number of __delay(1) in a jiffy, so make it so
1139 	 */
1140 	loops_per_jiffy = tb_ticks_per_jiffy;
1141 }
1142 
1143 static int __init rtc_init(void)
1144 {
1145 	struct platform_device *pdev;
1146 
1147 	if (!ppc_md.get_rtc_time)
1148 		return -ENODEV;
1149 
1150 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1151 	if (IS_ERR(pdev))
1152 		return PTR_ERR(pdev);
1153 
1154 	return 0;
1155 }
1156 
1157 module_init(rtc_init);
1158