xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision a09d2831)
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 u64 tb_to_xs;
153 unsigned tb_to_us;
154 
155 #define TICKLEN_SCALE	NTP_SCALE_SHIFT
156 static u64 last_tick_len;	/* units are ns / 2^TICKLEN_SCALE */
157 static u64 ticklen_to_xs;	/* 0.64 fraction */
158 
159 /* If last_tick_len corresponds to about 1/HZ seconds, then
160    last_tick_len << TICKLEN_SHIFT will be about 2^63. */
161 #define TICKLEN_SHIFT	(63 - 30 - TICKLEN_SCALE + SHIFT_HZ)
162 
163 DEFINE_SPINLOCK(rtc_lock);
164 EXPORT_SYMBOL_GPL(rtc_lock);
165 
166 static u64 tb_to_ns_scale __read_mostly;
167 static unsigned tb_to_ns_shift __read_mostly;
168 static unsigned long boot_tb __read_mostly;
169 
170 extern struct timezone sys_tz;
171 static long timezone_offset;
172 
173 unsigned long ppc_proc_freq;
174 EXPORT_SYMBOL(ppc_proc_freq);
175 unsigned long ppc_tb_freq;
176 
177 static u64 tb_last_jiffy __cacheline_aligned_in_smp;
178 static DEFINE_PER_CPU(u64, last_jiffy);
179 
180 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
181 /*
182  * Factors for converting from cputime_t (timebase ticks) to
183  * jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds).
184  * These are all stored as 0.64 fixed-point binary fractions.
185  */
186 u64 __cputime_jiffies_factor;
187 EXPORT_SYMBOL(__cputime_jiffies_factor);
188 u64 __cputime_msec_factor;
189 EXPORT_SYMBOL(__cputime_msec_factor);
190 u64 __cputime_sec_factor;
191 EXPORT_SYMBOL(__cputime_sec_factor);
192 u64 __cputime_clockt_factor;
193 EXPORT_SYMBOL(__cputime_clockt_factor);
194 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
195 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
196 
197 cputime_t cputime_one_jiffy;
198 
199 static void calc_cputime_factors(void)
200 {
201 	struct div_result res;
202 
203 	div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
204 	__cputime_jiffies_factor = res.result_low;
205 	div128_by_32(1000, 0, tb_ticks_per_sec, &res);
206 	__cputime_msec_factor = res.result_low;
207 	div128_by_32(1, 0, tb_ticks_per_sec, &res);
208 	__cputime_sec_factor = res.result_low;
209 	div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
210 	__cputime_clockt_factor = res.result_low;
211 }
212 
213 /*
214  * Read the PURR on systems that have it, otherwise the timebase.
215  */
216 static u64 read_purr(void)
217 {
218 	if (cpu_has_feature(CPU_FTR_PURR))
219 		return mfspr(SPRN_PURR);
220 	return mftb();
221 }
222 
223 /*
224  * Read the SPURR on systems that have it, otherwise the purr
225  */
226 static u64 read_spurr(u64 purr)
227 {
228 	/*
229 	 * cpus without PURR won't have a SPURR
230 	 * We already know the former when we use this, so tell gcc
231 	 */
232 	if (cpu_has_feature(CPU_FTR_PURR) && cpu_has_feature(CPU_FTR_SPURR))
233 		return mfspr(SPRN_SPURR);
234 	return purr;
235 }
236 
237 /*
238  * Account time for a transition between system, hard irq
239  * or soft irq state.
240  */
241 void account_system_vtime(struct task_struct *tsk)
242 {
243 	u64 now, nowscaled, delta, deltascaled, sys_time;
244 	unsigned long flags;
245 
246 	local_irq_save(flags);
247 	now = read_purr();
248 	nowscaled = read_spurr(now);
249 	delta = now - get_paca()->startpurr;
250 	deltascaled = nowscaled - get_paca()->startspurr;
251 	get_paca()->startpurr = now;
252 	get_paca()->startspurr = nowscaled;
253 	if (!in_interrupt()) {
254 		/* deltascaled includes both user and system time.
255 		 * Hence scale it based on the purr ratio to estimate
256 		 * the system time */
257 		sys_time = get_paca()->system_time;
258 		if (get_paca()->user_time)
259 			deltascaled = deltascaled * sys_time /
260 			     (sys_time + get_paca()->user_time);
261 		delta += sys_time;
262 		get_paca()->system_time = 0;
263 	}
264 	if (in_irq() || idle_task(smp_processor_id()) != tsk)
265 		account_system_time(tsk, 0, delta, deltascaled);
266 	else
267 		account_idle_time(delta);
268 	per_cpu(cputime_last_delta, smp_processor_id()) = delta;
269 	per_cpu(cputime_scaled_last_delta, smp_processor_id()) = deltascaled;
270 	local_irq_restore(flags);
271 }
272 EXPORT_SYMBOL_GPL(account_system_vtime);
273 
274 /*
275  * Transfer the user and system times accumulated in the paca
276  * by the exception entry and exit code to the generic process
277  * user and system time records.
278  * Must be called with interrupts disabled.
279  */
280 void account_process_tick(struct task_struct *tsk, int user_tick)
281 {
282 	cputime_t utime, utimescaled;
283 
284 	utime = get_paca()->user_time;
285 	get_paca()->user_time = 0;
286 	utimescaled = cputime_to_scaled(utime);
287 	account_user_time(tsk, utime, utimescaled);
288 }
289 
290 /*
291  * Stuff for accounting stolen time.
292  */
293 struct cpu_purr_data {
294 	int	initialized;			/* thread is running */
295 	u64	tb;			/* last TB value read */
296 	u64	purr;			/* last PURR value read */
297 	u64	spurr;			/* last SPURR value read */
298 };
299 
300 /*
301  * Each entry in the cpu_purr_data array is manipulated only by its
302  * "owner" cpu -- usually in the timer interrupt but also occasionally
303  * in process context for cpu online.  As long as cpus do not touch
304  * each others' cpu_purr_data, disabling local interrupts is
305  * sufficient to serialize accesses.
306  */
307 static DEFINE_PER_CPU(struct cpu_purr_data, cpu_purr_data);
308 
309 static void snapshot_tb_and_purr(void *data)
310 {
311 	unsigned long flags;
312 	struct cpu_purr_data *p = &__get_cpu_var(cpu_purr_data);
313 
314 	local_irq_save(flags);
315 	p->tb = get_tb_or_rtc();
316 	p->purr = mfspr(SPRN_PURR);
317 	wmb();
318 	p->initialized = 1;
319 	local_irq_restore(flags);
320 }
321 
322 /*
323  * Called during boot when all cpus have come up.
324  */
325 void snapshot_timebases(void)
326 {
327 	if (!cpu_has_feature(CPU_FTR_PURR))
328 		return;
329 	on_each_cpu(snapshot_tb_and_purr, NULL, 1);
330 }
331 
332 /*
333  * Must be called with interrupts disabled.
334  */
335 void calculate_steal_time(void)
336 {
337 	u64 tb, purr;
338 	s64 stolen;
339 	struct cpu_purr_data *pme;
340 
341 	pme = &__get_cpu_var(cpu_purr_data);
342 	if (!pme->initialized)
343 		return;		/* !CPU_FTR_PURR or early in early boot */
344 	tb = mftb();
345 	purr = mfspr(SPRN_PURR);
346 	stolen = (tb - pme->tb) - (purr - pme->purr);
347 	if (stolen > 0) {
348 		if (idle_task(smp_processor_id()) != current)
349 			account_steal_time(stolen);
350 		else
351 			account_idle_time(stolen);
352 	}
353 	pme->tb = tb;
354 	pme->purr = purr;
355 }
356 
357 #ifdef CONFIG_PPC_SPLPAR
358 /*
359  * Must be called before the cpu is added to the online map when
360  * a cpu is being brought up at runtime.
361  */
362 static void snapshot_purr(void)
363 {
364 	struct cpu_purr_data *pme;
365 	unsigned long flags;
366 
367 	if (!cpu_has_feature(CPU_FTR_PURR))
368 		return;
369 	local_irq_save(flags);
370 	pme = &__get_cpu_var(cpu_purr_data);
371 	pme->tb = mftb();
372 	pme->purr = mfspr(SPRN_PURR);
373 	pme->initialized = 1;
374 	local_irq_restore(flags);
375 }
376 
377 #endif /* CONFIG_PPC_SPLPAR */
378 
379 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
380 #define calc_cputime_factors()
381 #define calculate_steal_time()		do { } while (0)
382 #endif
383 
384 #if !(defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR))
385 #define snapshot_purr()			do { } while (0)
386 #endif
387 
388 /*
389  * Called when a cpu comes up after the system has finished booting,
390  * i.e. as a result of a hotplug cpu action.
391  */
392 void snapshot_timebase(void)
393 {
394 	__get_cpu_var(last_jiffy) = get_tb_or_rtc();
395 	snapshot_purr();
396 }
397 
398 void __delay(unsigned long loops)
399 {
400 	unsigned long start;
401 	int diff;
402 
403 	if (__USE_RTC()) {
404 		start = get_rtcl();
405 		do {
406 			/* the RTCL register wraps at 1000000000 */
407 			diff = get_rtcl() - start;
408 			if (diff < 0)
409 				diff += 1000000000;
410 		} while (diff < loops);
411 	} else {
412 		start = get_tbl();
413 		while (get_tbl() - start < loops)
414 			HMT_low();
415 		HMT_medium();
416 	}
417 }
418 EXPORT_SYMBOL(__delay);
419 
420 void udelay(unsigned long usecs)
421 {
422 	__delay(tb_ticks_per_usec * usecs);
423 }
424 EXPORT_SYMBOL(udelay);
425 
426 static inline void update_gtod(u64 new_tb_stamp, u64 new_stamp_xsec,
427 			       u64 new_tb_to_xs)
428 {
429 	/*
430 	 * tb_update_count is used to allow the userspace gettimeofday code
431 	 * to assure itself that it sees a consistent view of the tb_to_xs and
432 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
433 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
434 	 * the two values of tb_update_count match and are even then the
435 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
436 	 * loops back and reads them again until this criteria is met.
437 	 * We expect the caller to have done the first increment of
438 	 * vdso_data->tb_update_count already.
439 	 */
440 	vdso_data->tb_orig_stamp = new_tb_stamp;
441 	vdso_data->stamp_xsec = new_stamp_xsec;
442 	vdso_data->tb_to_xs = new_tb_to_xs;
443 	vdso_data->wtom_clock_sec = wall_to_monotonic.tv_sec;
444 	vdso_data->wtom_clock_nsec = wall_to_monotonic.tv_nsec;
445 	vdso_data->stamp_xtime = xtime;
446 	smp_wmb();
447 	++(vdso_data->tb_update_count);
448 }
449 
450 #ifdef CONFIG_SMP
451 unsigned long profile_pc(struct pt_regs *regs)
452 {
453 	unsigned long pc = instruction_pointer(regs);
454 
455 	if (in_lock_functions(pc))
456 		return regs->link;
457 
458 	return pc;
459 }
460 EXPORT_SYMBOL(profile_pc);
461 #endif
462 
463 #ifdef CONFIG_PPC_ISERIES
464 
465 /*
466  * This function recalibrates the timebase based on the 49-bit time-of-day
467  * value in the Titan chip.  The Titan is much more accurate than the value
468  * returned by the service processor for the timebase frequency.
469  */
470 
471 static int __init iSeries_tb_recal(void)
472 {
473 	struct div_result divres;
474 	unsigned long titan, tb;
475 
476 	/* Make sure we only run on iSeries */
477 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
478 		return -ENODEV;
479 
480 	tb = get_tb();
481 	titan = HvCallXm_loadTod();
482 	if ( iSeries_recal_titan ) {
483 		unsigned long tb_ticks = tb - iSeries_recal_tb;
484 		unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
485 		unsigned long new_tb_ticks_per_sec   = (tb_ticks * USEC_PER_SEC)/titan_usec;
486 		unsigned long new_tb_ticks_per_jiffy =
487 			DIV_ROUND_CLOSEST(new_tb_ticks_per_sec, HZ);
488 		long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
489 		char sign = '+';
490 		/* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
491 		new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;
492 
493 		if ( tick_diff < 0 ) {
494 			tick_diff = -tick_diff;
495 			sign = '-';
496 		}
497 		if ( tick_diff ) {
498 			if ( tick_diff < tb_ticks_per_jiffy/25 ) {
499 				printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
500 						new_tb_ticks_per_jiffy, sign, tick_diff );
501 				tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
502 				tb_ticks_per_sec   = new_tb_ticks_per_sec;
503 				calc_cputime_factors();
504 				div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres );
505 				tb_to_xs = divres.result_low;
506 				vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
507 				vdso_data->tb_to_xs = tb_to_xs;
508 				setup_cputime_one_jiffy();
509 			}
510 			else {
511 				printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
512 					"                   new tb_ticks_per_jiffy = %lu\n"
513 					"                   old tb_ticks_per_jiffy = %lu\n",
514 					new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
515 			}
516 		}
517 	}
518 	iSeries_recal_titan = titan;
519 	iSeries_recal_tb = tb;
520 
521 	/* Called here as now we know accurate values for the timebase */
522 	clocksource_init();
523 	return 0;
524 }
525 late_initcall(iSeries_tb_recal);
526 
527 /* Called from platform early init */
528 void __init iSeries_time_init_early(void)
529 {
530 	iSeries_recal_tb = get_tb();
531 	iSeries_recal_titan = HvCallXm_loadTod();
532 }
533 #endif /* CONFIG_PPC_ISERIES */
534 
535 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_PPC32)
536 DEFINE_PER_CPU(u8, perf_event_pending);
537 
538 void set_perf_event_pending(void)
539 {
540 	get_cpu_var(perf_event_pending) = 1;
541 	set_dec(1);
542 	put_cpu_var(perf_event_pending);
543 }
544 
545 #define test_perf_event_pending()	__get_cpu_var(perf_event_pending)
546 #define clear_perf_event_pending()	__get_cpu_var(perf_event_pending) = 0
547 
548 #else  /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */
549 
550 #define test_perf_event_pending()	0
551 #define clear_perf_event_pending()
552 
553 #endif /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */
554 
555 /*
556  * For iSeries shared processors, we have to let the hypervisor
557  * set the hardware decrementer.  We set a virtual decrementer
558  * in the lppaca and call the hypervisor if the virtual
559  * decrementer is less than the current value in the hardware
560  * decrementer. (almost always the new decrementer value will
561  * be greater than the current hardware decementer so the hypervisor
562  * call will not be needed)
563  */
564 
565 /*
566  * timer_interrupt - gets called when the decrementer overflows,
567  * with interrupts disabled.
568  */
569 void timer_interrupt(struct pt_regs * regs)
570 {
571 	struct pt_regs *old_regs;
572 	struct decrementer_clock *decrementer =  &__get_cpu_var(decrementers);
573 	struct clock_event_device *evt = &decrementer->event;
574 	u64 now;
575 
576 	trace_timer_interrupt_entry(regs);
577 
578 	/* Ensure a positive value is written to the decrementer, or else
579 	 * some CPUs will continuue to take decrementer exceptions */
580 	set_dec(DECREMENTER_MAX);
581 
582 #ifdef CONFIG_PPC32
583 	if (test_perf_event_pending()) {
584 		clear_perf_event_pending();
585 		perf_event_do_pending();
586 	}
587 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
588 		do_IRQ(regs);
589 #endif
590 
591 	now = get_tb_or_rtc();
592 	if (now < decrementer->next_tb) {
593 		/* not time for this event yet */
594 		now = decrementer->next_tb - now;
595 		if (now <= DECREMENTER_MAX)
596 			set_dec((int)now);
597 		trace_timer_interrupt_exit(regs);
598 		return;
599 	}
600 	old_regs = set_irq_regs(regs);
601 	irq_enter();
602 
603 	calculate_steal_time();
604 
605 #ifdef CONFIG_PPC_ISERIES
606 	if (firmware_has_feature(FW_FEATURE_ISERIES))
607 		get_lppaca()->int_dword.fields.decr_int = 0;
608 #endif
609 
610 	if (evt->event_handler)
611 		evt->event_handler(evt);
612 
613 #ifdef CONFIG_PPC_ISERIES
614 	if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending())
615 		process_hvlpevents();
616 #endif
617 
618 #ifdef CONFIG_PPC64
619 	/* collect purr register values often, for accurate calculations */
620 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
621 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
622 		cu->current_tb = mfspr(SPRN_PURR);
623 	}
624 #endif
625 
626 	irq_exit();
627 	set_irq_regs(old_regs);
628 
629 	trace_timer_interrupt_exit(regs);
630 }
631 
632 void wakeup_decrementer(void)
633 {
634 	unsigned long ticks;
635 
636 	/*
637 	 * The timebase gets saved on sleep and restored on wakeup,
638 	 * so all we need to do is to reset the decrementer.
639 	 */
640 	ticks = tb_ticks_since(__get_cpu_var(last_jiffy));
641 	if (ticks < tb_ticks_per_jiffy)
642 		ticks = tb_ticks_per_jiffy - ticks;
643 	else
644 		ticks = 1;
645 	set_dec(ticks);
646 }
647 
648 #ifdef CONFIG_SUSPEND
649 void generic_suspend_disable_irqs(void)
650 {
651 	preempt_disable();
652 
653 	/* Disable the decrementer, so that it doesn't interfere
654 	 * with suspending.
655 	 */
656 
657 	set_dec(0x7fffffff);
658 	local_irq_disable();
659 	set_dec(0x7fffffff);
660 }
661 
662 void generic_suspend_enable_irqs(void)
663 {
664 	wakeup_decrementer();
665 
666 	local_irq_enable();
667 	preempt_enable();
668 }
669 
670 /* Overrides the weak version in kernel/power/main.c */
671 void arch_suspend_disable_irqs(void)
672 {
673 	if (ppc_md.suspend_disable_irqs)
674 		ppc_md.suspend_disable_irqs();
675 	generic_suspend_disable_irqs();
676 }
677 
678 /* Overrides the weak version in kernel/power/main.c */
679 void arch_suspend_enable_irqs(void)
680 {
681 	generic_suspend_enable_irqs();
682 	if (ppc_md.suspend_enable_irqs)
683 		ppc_md.suspend_enable_irqs();
684 }
685 #endif
686 
687 #ifdef CONFIG_SMP
688 void __init smp_space_timers(unsigned int max_cpus)
689 {
690 	int i;
691 	u64 previous_tb = per_cpu(last_jiffy, boot_cpuid);
692 
693 	/* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */
694 	previous_tb -= tb_ticks_per_jiffy;
695 
696 	for_each_possible_cpu(i) {
697 		if (i == boot_cpuid)
698 			continue;
699 		per_cpu(last_jiffy, i) = previous_tb;
700 	}
701 }
702 #endif
703 
704 /*
705  * Scheduler clock - returns current time in nanosec units.
706  *
707  * Note: mulhdu(a, b) (multiply high double unsigned) returns
708  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
709  * are 64-bit unsigned numbers.
710  */
711 unsigned long long sched_clock(void)
712 {
713 	if (__USE_RTC())
714 		return get_rtc();
715 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
716 }
717 
718 static int __init get_freq(char *name, int cells, unsigned long *val)
719 {
720 	struct device_node *cpu;
721 	const unsigned int *fp;
722 	int found = 0;
723 
724 	/* The cpu node should have timebase and clock frequency properties */
725 	cpu = of_find_node_by_type(NULL, "cpu");
726 
727 	if (cpu) {
728 		fp = of_get_property(cpu, name, NULL);
729 		if (fp) {
730 			found = 1;
731 			*val = of_read_ulong(fp, cells);
732 		}
733 
734 		of_node_put(cpu);
735 	}
736 
737 	return found;
738 }
739 
740 /* should become __cpuinit when secondary_cpu_time_init also is */
741 void start_cpu_decrementer(void)
742 {
743 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
744 	/* Clear any pending timer interrupts */
745 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
746 
747 	/* Enable decrementer interrupt */
748 	mtspr(SPRN_TCR, TCR_DIE);
749 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
750 }
751 
752 void __init generic_calibrate_decr(void)
753 {
754 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
755 
756 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
757 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
758 
759 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
760 				"(not found)\n");
761 	}
762 
763 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
764 
765 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
766 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
767 
768 		printk(KERN_ERR "WARNING: Estimating processor frequency "
769 				"(not found)\n");
770 	}
771 }
772 
773 int update_persistent_clock(struct timespec now)
774 {
775 	struct rtc_time tm;
776 
777 	if (!ppc_md.set_rtc_time)
778 		return 0;
779 
780 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
781 	tm.tm_year -= 1900;
782 	tm.tm_mon -= 1;
783 
784 	return ppc_md.set_rtc_time(&tm);
785 }
786 
787 static void __read_persistent_clock(struct timespec *ts)
788 {
789 	struct rtc_time tm;
790 	static int first = 1;
791 
792 	ts->tv_nsec = 0;
793 	/* XXX this is a litle fragile but will work okay in the short term */
794 	if (first) {
795 		first = 0;
796 		if (ppc_md.time_init)
797 			timezone_offset = ppc_md.time_init();
798 
799 		/* get_boot_time() isn't guaranteed to be safe to call late */
800 		if (ppc_md.get_boot_time) {
801 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
802 			return;
803 		}
804 	}
805 	if (!ppc_md.get_rtc_time) {
806 		ts->tv_sec = 0;
807 		return;
808 	}
809 	ppc_md.get_rtc_time(&tm);
810 
811 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
812 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
813 }
814 
815 void read_persistent_clock(struct timespec *ts)
816 {
817 	__read_persistent_clock(ts);
818 
819 	/* Sanitize it in case real time clock is set below EPOCH */
820 	if (ts->tv_sec < 0) {
821 		ts->tv_sec = 0;
822 		ts->tv_nsec = 0;
823 	}
824 
825 }
826 
827 /* clocksource code */
828 static cycle_t rtc_read(struct clocksource *cs)
829 {
830 	return (cycle_t)get_rtc();
831 }
832 
833 static cycle_t timebase_read(struct clocksource *cs)
834 {
835 	return (cycle_t)get_tb();
836 }
837 
838 void update_vsyscall(struct timespec *wall_time, struct clocksource *clock,
839 		     u32 mult)
840 {
841 	u64 t2x, stamp_xsec;
842 
843 	if (clock != &clocksource_timebase)
844 		return;
845 
846 	/* Make userspace gettimeofday spin until we're done. */
847 	++vdso_data->tb_update_count;
848 	smp_mb();
849 
850 	/* XXX this assumes clock->shift == 22 */
851 	/* 4611686018 ~= 2^(20+64-22) / 1e9 */
852 	t2x = (u64) mult * 4611686018ULL;
853 	stamp_xsec = (u64) xtime.tv_nsec * XSEC_PER_SEC;
854 	do_div(stamp_xsec, 1000000000);
855 	stamp_xsec += (u64) xtime.tv_sec * XSEC_PER_SEC;
856 	update_gtod(clock->cycle_last, stamp_xsec, t2x);
857 }
858 
859 void update_vsyscall_tz(void)
860 {
861 	/* Make userspace gettimeofday spin until we're done. */
862 	++vdso_data->tb_update_count;
863 	smp_mb();
864 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
865 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
866 	smp_mb();
867 	++vdso_data->tb_update_count;
868 }
869 
870 static void __init clocksource_init(void)
871 {
872 	struct clocksource *clock;
873 
874 	if (__USE_RTC())
875 		clock = &clocksource_rtc;
876 	else
877 		clock = &clocksource_timebase;
878 
879 	clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
880 
881 	if (clocksource_register(clock)) {
882 		printk(KERN_ERR "clocksource: %s is already registered\n",
883 		       clock->name);
884 		return;
885 	}
886 
887 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
888 	       clock->name, clock->mult, clock->shift);
889 }
890 
891 static int decrementer_set_next_event(unsigned long evt,
892 				      struct clock_event_device *dev)
893 {
894 	__get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
895 	set_dec(evt);
896 	return 0;
897 }
898 
899 static void decrementer_set_mode(enum clock_event_mode mode,
900 				 struct clock_event_device *dev)
901 {
902 	if (mode != CLOCK_EVT_MODE_ONESHOT)
903 		decrementer_set_next_event(DECREMENTER_MAX, dev);
904 }
905 
906 static void __init setup_clockevent_multiplier(unsigned long hz)
907 {
908 	u64 mult, shift = 32;
909 
910 	while (1) {
911 		mult = div_sc(hz, NSEC_PER_SEC, shift);
912 		if (mult && (mult >> 32UL) == 0UL)
913 			break;
914 
915 		shift--;
916 	}
917 
918 	decrementer_clockevent.shift = shift;
919 	decrementer_clockevent.mult = mult;
920 }
921 
922 static void register_decrementer_clockevent(int cpu)
923 {
924 	struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
925 
926 	*dec = decrementer_clockevent;
927 	dec->cpumask = cpumask_of(cpu);
928 
929 	printk(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
930 	       dec->name, dec->mult, dec->shift, cpu);
931 
932 	clockevents_register_device(dec);
933 }
934 
935 static void __init init_decrementer_clockevent(void)
936 {
937 	int cpu = smp_processor_id();
938 
939 	setup_clockevent_multiplier(ppc_tb_freq);
940 	decrementer_clockevent.max_delta_ns =
941 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
942 	decrementer_clockevent.min_delta_ns =
943 		clockevent_delta2ns(2, &decrementer_clockevent);
944 
945 	register_decrementer_clockevent(cpu);
946 }
947 
948 void secondary_cpu_time_init(void)
949 {
950 	/* Start the decrementer on CPUs that have manual control
951 	 * such as BookE
952 	 */
953 	start_cpu_decrementer();
954 
955 	/* FIME: Should make unrelatred change to move snapshot_timebase
956 	 * call here ! */
957 	register_decrementer_clockevent(smp_processor_id());
958 }
959 
960 /* This function is only called on the boot processor */
961 void __init time_init(void)
962 {
963 	unsigned long flags;
964 	struct div_result res;
965 	u64 scale, x;
966 	unsigned shift;
967 
968 	if (__USE_RTC()) {
969 		/* 601 processor: dec counts down by 128 every 128ns */
970 		ppc_tb_freq = 1000000000;
971 		tb_last_jiffy = get_rtcl();
972 	} else {
973 		/* Normal PowerPC with timebase register */
974 		ppc_md.calibrate_decr();
975 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
976 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
977 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
978 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
979 		tb_last_jiffy = get_tb();
980 	}
981 
982 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
983 	tb_ticks_per_sec = ppc_tb_freq;
984 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
985 	tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000);
986 	calc_cputime_factors();
987 	setup_cputime_one_jiffy();
988 
989 	/*
990 	 * Calculate the length of each tick in ns.  It will not be
991 	 * exactly 1e9/HZ unless ppc_tb_freq is divisible by HZ.
992 	 * We compute 1e9 * tb_ticks_per_jiffy / ppc_tb_freq,
993 	 * rounded up.
994 	 */
995 	x = (u64) NSEC_PER_SEC * tb_ticks_per_jiffy + ppc_tb_freq - 1;
996 	do_div(x, ppc_tb_freq);
997 	tick_nsec = x;
998 	last_tick_len = x << TICKLEN_SCALE;
999 
1000 	/*
1001 	 * Compute ticklen_to_xs, which is a factor which gets multiplied
1002 	 * by (last_tick_len << TICKLEN_SHIFT) to get a tb_to_xs value.
1003 	 * It is computed as:
1004 	 * ticklen_to_xs = 2^N / (tb_ticks_per_jiffy * 1e9)
1005 	 * where N = 64 + 20 - TICKLEN_SCALE - TICKLEN_SHIFT
1006 	 * which turns out to be N = 51 - SHIFT_HZ.
1007 	 * This gives the result as a 0.64 fixed-point fraction.
1008 	 * That value is reduced by an offset amounting to 1 xsec per
1009 	 * 2^31 timebase ticks to avoid problems with time going backwards
1010 	 * by 1 xsec when we do timer_recalc_offset due to losing the
1011 	 * fractional xsec.  That offset is equal to ppc_tb_freq/2^51
1012 	 * since there are 2^20 xsec in a second.
1013 	 */
1014 	div128_by_32((1ULL << 51) - ppc_tb_freq, 0,
1015 		     tb_ticks_per_jiffy << SHIFT_HZ, &res);
1016 	div128_by_32(res.result_high, res.result_low, NSEC_PER_SEC, &res);
1017 	ticklen_to_xs = res.result_low;
1018 
1019 	/* Compute tb_to_xs from tick_nsec */
1020 	tb_to_xs = mulhdu(last_tick_len << TICKLEN_SHIFT, ticklen_to_xs);
1021 
1022 	/*
1023 	 * Compute scale factor for sched_clock.
1024 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1025 	 * which is the timebase frequency.
1026 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1027 	 * the 128-bit result as a 64.64 fixed-point number.
1028 	 * We then shift that number right until it is less than 1.0,
1029 	 * giving us the scale factor and shift count to use in
1030 	 * sched_clock().
1031 	 */
1032 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1033 	scale = res.result_low;
1034 	for (shift = 0; res.result_high != 0; ++shift) {
1035 		scale = (scale >> 1) | (res.result_high << 63);
1036 		res.result_high >>= 1;
1037 	}
1038 	tb_to_ns_scale = scale;
1039 	tb_to_ns_shift = shift;
1040 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1041 	boot_tb = get_tb_or_rtc();
1042 
1043 	write_seqlock_irqsave(&xtime_lock, flags);
1044 
1045 	/* If platform provided a timezone (pmac), we correct the time */
1046         if (timezone_offset) {
1047 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1048 		sys_tz.tz_dsttime = 0;
1049         }
1050 
1051 	vdso_data->tb_orig_stamp = tb_last_jiffy;
1052 	vdso_data->tb_update_count = 0;
1053 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1054 	vdso_data->stamp_xsec = (u64) xtime.tv_sec * XSEC_PER_SEC;
1055 	vdso_data->tb_to_xs = tb_to_xs;
1056 
1057 	write_sequnlock_irqrestore(&xtime_lock, flags);
1058 
1059 	/* Start the decrementer on CPUs that have manual control
1060 	 * such as BookE
1061 	 */
1062 	start_cpu_decrementer();
1063 
1064 	/* Register the clocksource, if we're not running on iSeries */
1065 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
1066 		clocksource_init();
1067 
1068 	init_decrementer_clockevent();
1069 }
1070 
1071 
1072 #define FEBRUARY	2
1073 #define	STARTOFTIME	1970
1074 #define SECDAY		86400L
1075 #define SECYR		(SECDAY * 365)
1076 #define	leapyear(year)		((year) % 4 == 0 && \
1077 				 ((year) % 100 != 0 || (year) % 400 == 0))
1078 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1079 #define	days_in_month(a) 	(month_days[(a) - 1])
1080 
1081 static int month_days[12] = {
1082 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1083 };
1084 
1085 /*
1086  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1087  */
1088 void GregorianDay(struct rtc_time * tm)
1089 {
1090 	int leapsToDate;
1091 	int lastYear;
1092 	int day;
1093 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1094 
1095 	lastYear = tm->tm_year - 1;
1096 
1097 	/*
1098 	 * Number of leap corrections to apply up to end of last year
1099 	 */
1100 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
1101 
1102 	/*
1103 	 * This year is a leap year if it is divisible by 4 except when it is
1104 	 * divisible by 100 unless it is divisible by 400
1105 	 *
1106 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1107 	 */
1108 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
1109 
1110 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
1111 		   tm->tm_mday;
1112 
1113 	tm->tm_wday = day % 7;
1114 }
1115 
1116 void to_tm(int tim, struct rtc_time * tm)
1117 {
1118 	register int    i;
1119 	register long   hms, day;
1120 
1121 	day = tim / SECDAY;
1122 	hms = tim % SECDAY;
1123 
1124 	/* Hours, minutes, seconds are easy */
1125 	tm->tm_hour = hms / 3600;
1126 	tm->tm_min = (hms % 3600) / 60;
1127 	tm->tm_sec = (hms % 3600) % 60;
1128 
1129 	/* Number of years in days */
1130 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1131 		day -= days_in_year(i);
1132 	tm->tm_year = i;
1133 
1134 	/* Number of months in days left */
1135 	if (leapyear(tm->tm_year))
1136 		days_in_month(FEBRUARY) = 29;
1137 	for (i = 1; day >= days_in_month(i); i++)
1138 		day -= days_in_month(i);
1139 	days_in_month(FEBRUARY) = 28;
1140 	tm->tm_mon = i;
1141 
1142 	/* Days are what is left over (+1) from all that. */
1143 	tm->tm_mday = day + 1;
1144 
1145 	/*
1146 	 * Determine the day of week
1147 	 */
1148 	GregorianDay(tm);
1149 }
1150 
1151 /* Auxiliary function to compute scaling factors */
1152 /* Actually the choice of a timebase running at 1/4 the of the bus
1153  * frequency giving resolution of a few tens of nanoseconds is quite nice.
1154  * It makes this computation very precise (27-28 bits typically) which
1155  * is optimistic considering the stability of most processor clock
1156  * oscillators and the precision with which the timebase frequency
1157  * is measured but does not harm.
1158  */
1159 unsigned mulhwu_scale_factor(unsigned inscale, unsigned outscale)
1160 {
1161         unsigned mlt=0, tmp, err;
1162         /* No concern for performance, it's done once: use a stupid
1163          * but safe and compact method to find the multiplier.
1164          */
1165 
1166         for (tmp = 1U<<31; tmp != 0; tmp >>= 1) {
1167                 if (mulhwu(inscale, mlt|tmp) < outscale)
1168 			mlt |= tmp;
1169         }
1170 
1171         /* We might still be off by 1 for the best approximation.
1172          * A side effect of this is that if outscale is too large
1173          * the returned value will be zero.
1174          * Many corner cases have been checked and seem to work,
1175          * some might have been forgotten in the test however.
1176          */
1177 
1178         err = inscale * (mlt+1);
1179         if (err <= inscale/2)
1180 		mlt++;
1181         return mlt;
1182 }
1183 
1184 /*
1185  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1186  * result.
1187  */
1188 void div128_by_32(u64 dividend_high, u64 dividend_low,
1189 		  unsigned divisor, struct div_result *dr)
1190 {
1191 	unsigned long a, b, c, d;
1192 	unsigned long w, x, y, z;
1193 	u64 ra, rb, rc;
1194 
1195 	a = dividend_high >> 32;
1196 	b = dividend_high & 0xffffffff;
1197 	c = dividend_low >> 32;
1198 	d = dividend_low & 0xffffffff;
1199 
1200 	w = a / divisor;
1201 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1202 
1203 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1204 	x = ra;
1205 
1206 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1207 	y = rb;
1208 
1209 	do_div(rc, divisor);
1210 	z = rc;
1211 
1212 	dr->result_high = ((u64)w << 32) + x;
1213 	dr->result_low  = ((u64)y << 32) + z;
1214 
1215 }
1216 
1217 /* We don't need to calibrate delay, we use the CPU timebase for that */
1218 void calibrate_delay(void)
1219 {
1220 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1221 	 * as the number of __delay(1) in a jiffy, so make it so
1222 	 */
1223 	loops_per_jiffy = tb_ticks_per_jiffy;
1224 }
1225 
1226 static int __init rtc_init(void)
1227 {
1228 	struct platform_device *pdev;
1229 
1230 	if (!ppc_md.get_rtc_time)
1231 		return -ENODEV;
1232 
1233 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1234 	if (IS_ERR(pdev))
1235 		return PTR_ERR(pdev);
1236 
1237 	return 0;
1238 }
1239 
1240 module_init(rtc_init);
1241