xref: /linux/arch/powerpc/kernel/time.c (revision 273b281fa22c293963ee3e6eec418f5dda2dbc83)
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 
273 /*
274  * Transfer the user and system times accumulated in the paca
275  * by the exception entry and exit code to the generic process
276  * user and system time records.
277  * Must be called with interrupts disabled.
278  */
279 void account_process_tick(struct task_struct *tsk, int user_tick)
280 {
281 	cputime_t utime, utimescaled;
282 
283 	utime = get_paca()->user_time;
284 	get_paca()->user_time = 0;
285 	utimescaled = cputime_to_scaled(utime);
286 	account_user_time(tsk, utime, utimescaled);
287 }
288 
289 /*
290  * Stuff for accounting stolen time.
291  */
292 struct cpu_purr_data {
293 	int	initialized;			/* thread is running */
294 	u64	tb;			/* last TB value read */
295 	u64	purr;			/* last PURR value read */
296 	u64	spurr;			/* last SPURR value read */
297 };
298 
299 /*
300  * Each entry in the cpu_purr_data array is manipulated only by its
301  * "owner" cpu -- usually in the timer interrupt but also occasionally
302  * in process context for cpu online.  As long as cpus do not touch
303  * each others' cpu_purr_data, disabling local interrupts is
304  * sufficient to serialize accesses.
305  */
306 static DEFINE_PER_CPU(struct cpu_purr_data, cpu_purr_data);
307 
308 static void snapshot_tb_and_purr(void *data)
309 {
310 	unsigned long flags;
311 	struct cpu_purr_data *p = &__get_cpu_var(cpu_purr_data);
312 
313 	local_irq_save(flags);
314 	p->tb = get_tb_or_rtc();
315 	p->purr = mfspr(SPRN_PURR);
316 	wmb();
317 	p->initialized = 1;
318 	local_irq_restore(flags);
319 }
320 
321 /*
322  * Called during boot when all cpus have come up.
323  */
324 void snapshot_timebases(void)
325 {
326 	if (!cpu_has_feature(CPU_FTR_PURR))
327 		return;
328 	on_each_cpu(snapshot_tb_and_purr, NULL, 1);
329 }
330 
331 /*
332  * Must be called with interrupts disabled.
333  */
334 void calculate_steal_time(void)
335 {
336 	u64 tb, purr;
337 	s64 stolen;
338 	struct cpu_purr_data *pme;
339 
340 	pme = &__get_cpu_var(cpu_purr_data);
341 	if (!pme->initialized)
342 		return;		/* !CPU_FTR_PURR or early in early boot */
343 	tb = mftb();
344 	purr = mfspr(SPRN_PURR);
345 	stolen = (tb - pme->tb) - (purr - pme->purr);
346 	if (stolen > 0) {
347 		if (idle_task(smp_processor_id()) != current)
348 			account_steal_time(stolen);
349 		else
350 			account_idle_time(stolen);
351 	}
352 	pme->tb = tb;
353 	pme->purr = purr;
354 }
355 
356 #ifdef CONFIG_PPC_SPLPAR
357 /*
358  * Must be called before the cpu is added to the online map when
359  * a cpu is being brought up at runtime.
360  */
361 static void snapshot_purr(void)
362 {
363 	struct cpu_purr_data *pme;
364 	unsigned long flags;
365 
366 	if (!cpu_has_feature(CPU_FTR_PURR))
367 		return;
368 	local_irq_save(flags);
369 	pme = &__get_cpu_var(cpu_purr_data);
370 	pme->tb = mftb();
371 	pme->purr = mfspr(SPRN_PURR);
372 	pme->initialized = 1;
373 	local_irq_restore(flags);
374 }
375 
376 #endif /* CONFIG_PPC_SPLPAR */
377 
378 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
379 #define calc_cputime_factors()
380 #define calculate_steal_time()		do { } while (0)
381 #endif
382 
383 #if !(defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR))
384 #define snapshot_purr()			do { } while (0)
385 #endif
386 
387 /*
388  * Called when a cpu comes up after the system has finished booting,
389  * i.e. as a result of a hotplug cpu action.
390  */
391 void snapshot_timebase(void)
392 {
393 	__get_cpu_var(last_jiffy) = get_tb_or_rtc();
394 	snapshot_purr();
395 }
396 
397 void __delay(unsigned long loops)
398 {
399 	unsigned long start;
400 	int diff;
401 
402 	if (__USE_RTC()) {
403 		start = get_rtcl();
404 		do {
405 			/* the RTCL register wraps at 1000000000 */
406 			diff = get_rtcl() - start;
407 			if (diff < 0)
408 				diff += 1000000000;
409 		} while (diff < loops);
410 	} else {
411 		start = get_tbl();
412 		while (get_tbl() - start < loops)
413 			HMT_low();
414 		HMT_medium();
415 	}
416 }
417 EXPORT_SYMBOL(__delay);
418 
419 void udelay(unsigned long usecs)
420 {
421 	__delay(tb_ticks_per_usec * usecs);
422 }
423 EXPORT_SYMBOL(udelay);
424 
425 static inline void update_gtod(u64 new_tb_stamp, u64 new_stamp_xsec,
426 			       u64 new_tb_to_xs)
427 {
428 	/*
429 	 * tb_update_count is used to allow the userspace gettimeofday code
430 	 * to assure itself that it sees a consistent view of the tb_to_xs and
431 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
432 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
433 	 * the two values of tb_update_count match and are even then the
434 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
435 	 * loops back and reads them again until this criteria is met.
436 	 * We expect the caller to have done the first increment of
437 	 * vdso_data->tb_update_count already.
438 	 */
439 	vdso_data->tb_orig_stamp = new_tb_stamp;
440 	vdso_data->stamp_xsec = new_stamp_xsec;
441 	vdso_data->tb_to_xs = new_tb_to_xs;
442 	vdso_data->wtom_clock_sec = wall_to_monotonic.tv_sec;
443 	vdso_data->wtom_clock_nsec = wall_to_monotonic.tv_nsec;
444 	vdso_data->stamp_xtime = xtime;
445 	smp_wmb();
446 	++(vdso_data->tb_update_count);
447 }
448 
449 #ifdef CONFIG_SMP
450 unsigned long profile_pc(struct pt_regs *regs)
451 {
452 	unsigned long pc = instruction_pointer(regs);
453 
454 	if (in_lock_functions(pc))
455 		return regs->link;
456 
457 	return pc;
458 }
459 EXPORT_SYMBOL(profile_pc);
460 #endif
461 
462 #ifdef CONFIG_PPC_ISERIES
463 
464 /*
465  * This function recalibrates the timebase based on the 49-bit time-of-day
466  * value in the Titan chip.  The Titan is much more accurate than the value
467  * returned by the service processor for the timebase frequency.
468  */
469 
470 static int __init iSeries_tb_recal(void)
471 {
472 	struct div_result divres;
473 	unsigned long titan, tb;
474 
475 	/* Make sure we only run on iSeries */
476 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
477 		return -ENODEV;
478 
479 	tb = get_tb();
480 	titan = HvCallXm_loadTod();
481 	if ( iSeries_recal_titan ) {
482 		unsigned long tb_ticks = tb - iSeries_recal_tb;
483 		unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
484 		unsigned long new_tb_ticks_per_sec   = (tb_ticks * USEC_PER_SEC)/titan_usec;
485 		unsigned long new_tb_ticks_per_jiffy =
486 			DIV_ROUND_CLOSEST(new_tb_ticks_per_sec, HZ);
487 		long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
488 		char sign = '+';
489 		/* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
490 		new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;
491 
492 		if ( tick_diff < 0 ) {
493 			tick_diff = -tick_diff;
494 			sign = '-';
495 		}
496 		if ( tick_diff ) {
497 			if ( tick_diff < tb_ticks_per_jiffy/25 ) {
498 				printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
499 						new_tb_ticks_per_jiffy, sign, tick_diff );
500 				tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
501 				tb_ticks_per_sec   = new_tb_ticks_per_sec;
502 				calc_cputime_factors();
503 				div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres );
504 				tb_to_xs = divres.result_low;
505 				vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
506 				vdso_data->tb_to_xs = tb_to_xs;
507 				setup_cputime_one_jiffy();
508 			}
509 			else {
510 				printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
511 					"                   new tb_ticks_per_jiffy = %lu\n"
512 					"                   old tb_ticks_per_jiffy = %lu\n",
513 					new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
514 			}
515 		}
516 	}
517 	iSeries_recal_titan = titan;
518 	iSeries_recal_tb = tb;
519 
520 	/* Called here as now we know accurate values for the timebase */
521 	clocksource_init();
522 	return 0;
523 }
524 late_initcall(iSeries_tb_recal);
525 
526 /* Called from platform early init */
527 void __init iSeries_time_init_early(void)
528 {
529 	iSeries_recal_tb = get_tb();
530 	iSeries_recal_titan = HvCallXm_loadTod();
531 }
532 #endif /* CONFIG_PPC_ISERIES */
533 
534 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_PPC32)
535 DEFINE_PER_CPU(u8, perf_event_pending);
536 
537 void set_perf_event_pending(void)
538 {
539 	get_cpu_var(perf_event_pending) = 1;
540 	set_dec(1);
541 	put_cpu_var(perf_event_pending);
542 }
543 
544 #define test_perf_event_pending()	__get_cpu_var(perf_event_pending)
545 #define clear_perf_event_pending()	__get_cpu_var(perf_event_pending) = 0
546 
547 #else  /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */
548 
549 #define test_perf_event_pending()	0
550 #define clear_perf_event_pending()
551 
552 #endif /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */
553 
554 /*
555  * For iSeries shared processors, we have to let the hypervisor
556  * set the hardware decrementer.  We set a virtual decrementer
557  * in the lppaca and call the hypervisor if the virtual
558  * decrementer is less than the current value in the hardware
559  * decrementer. (almost always the new decrementer value will
560  * be greater than the current hardware decementer so the hypervisor
561  * call will not be needed)
562  */
563 
564 /*
565  * timer_interrupt - gets called when the decrementer overflows,
566  * with interrupts disabled.
567  */
568 void timer_interrupt(struct pt_regs * regs)
569 {
570 	struct pt_regs *old_regs;
571 	struct decrementer_clock *decrementer =  &__get_cpu_var(decrementers);
572 	struct clock_event_device *evt = &decrementer->event;
573 	u64 now;
574 
575 	trace_timer_interrupt_entry(regs);
576 
577 	/* Ensure a positive value is written to the decrementer, or else
578 	 * some CPUs will continuue to take decrementer exceptions */
579 	set_dec(DECREMENTER_MAX);
580 
581 #ifdef CONFIG_PPC32
582 	if (test_perf_event_pending()) {
583 		clear_perf_event_pending();
584 		perf_event_do_pending();
585 	}
586 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
587 		do_IRQ(regs);
588 #endif
589 
590 	now = get_tb_or_rtc();
591 	if (now < decrementer->next_tb) {
592 		/* not time for this event yet */
593 		now = decrementer->next_tb - now;
594 		if (now <= DECREMENTER_MAX)
595 			set_dec((int)now);
596 		trace_timer_interrupt_exit(regs);
597 		return;
598 	}
599 	old_regs = set_irq_regs(regs);
600 	irq_enter();
601 
602 	calculate_steal_time();
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 void wakeup_decrementer(void)
632 {
633 	unsigned long ticks;
634 
635 	/*
636 	 * The timebase gets saved on sleep and restored on wakeup,
637 	 * so all we need to do is to reset the decrementer.
638 	 */
639 	ticks = tb_ticks_since(__get_cpu_var(last_jiffy));
640 	if (ticks < tb_ticks_per_jiffy)
641 		ticks = tb_ticks_per_jiffy - ticks;
642 	else
643 		ticks = 1;
644 	set_dec(ticks);
645 }
646 
647 #ifdef CONFIG_SUSPEND
648 void generic_suspend_disable_irqs(void)
649 {
650 	preempt_disable();
651 
652 	/* Disable the decrementer, so that it doesn't interfere
653 	 * with suspending.
654 	 */
655 
656 	set_dec(0x7fffffff);
657 	local_irq_disable();
658 	set_dec(0x7fffffff);
659 }
660 
661 void generic_suspend_enable_irqs(void)
662 {
663 	wakeup_decrementer();
664 
665 	local_irq_enable();
666 	preempt_enable();
667 }
668 
669 /* Overrides the weak version in kernel/power/main.c */
670 void arch_suspend_disable_irqs(void)
671 {
672 	if (ppc_md.suspend_disable_irqs)
673 		ppc_md.suspend_disable_irqs();
674 	generic_suspend_disable_irqs();
675 }
676 
677 /* Overrides the weak version in kernel/power/main.c */
678 void arch_suspend_enable_irqs(void)
679 {
680 	generic_suspend_enable_irqs();
681 	if (ppc_md.suspend_enable_irqs)
682 		ppc_md.suspend_enable_irqs();
683 }
684 #endif
685 
686 #ifdef CONFIG_SMP
687 void __init smp_space_timers(unsigned int max_cpus)
688 {
689 	int i;
690 	u64 previous_tb = per_cpu(last_jiffy, boot_cpuid);
691 
692 	/* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */
693 	previous_tb -= tb_ticks_per_jiffy;
694 
695 	for_each_possible_cpu(i) {
696 		if (i == boot_cpuid)
697 			continue;
698 		per_cpu(last_jiffy, i) = previous_tb;
699 	}
700 }
701 #endif
702 
703 /*
704  * Scheduler clock - returns current time in nanosec units.
705  *
706  * Note: mulhdu(a, b) (multiply high double unsigned) returns
707  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
708  * are 64-bit unsigned numbers.
709  */
710 unsigned long long sched_clock(void)
711 {
712 	if (__USE_RTC())
713 		return get_rtc();
714 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
715 }
716 
717 static int __init get_freq(char *name, int cells, unsigned long *val)
718 {
719 	struct device_node *cpu;
720 	const unsigned int *fp;
721 	int found = 0;
722 
723 	/* The cpu node should have timebase and clock frequency properties */
724 	cpu = of_find_node_by_type(NULL, "cpu");
725 
726 	if (cpu) {
727 		fp = of_get_property(cpu, name, NULL);
728 		if (fp) {
729 			found = 1;
730 			*val = of_read_ulong(fp, cells);
731 		}
732 
733 		of_node_put(cpu);
734 	}
735 
736 	return found;
737 }
738 
739 /* should become __cpuinit when secondary_cpu_time_init also is */
740 void start_cpu_decrementer(void)
741 {
742 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
743 	/* Clear any pending timer interrupts */
744 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
745 
746 	/* Enable decrementer interrupt */
747 	mtspr(SPRN_TCR, TCR_DIE);
748 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
749 }
750 
751 void __init generic_calibrate_decr(void)
752 {
753 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
754 
755 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
756 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
757 
758 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
759 				"(not found)\n");
760 	}
761 
762 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
763 
764 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
765 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
766 
767 		printk(KERN_ERR "WARNING: Estimating processor frequency "
768 				"(not found)\n");
769 	}
770 }
771 
772 int update_persistent_clock(struct timespec now)
773 {
774 	struct rtc_time tm;
775 
776 	if (!ppc_md.set_rtc_time)
777 		return 0;
778 
779 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
780 	tm.tm_year -= 1900;
781 	tm.tm_mon -= 1;
782 
783 	return ppc_md.set_rtc_time(&tm);
784 }
785 
786 static void __read_persistent_clock(struct timespec *ts)
787 {
788 	struct rtc_time tm;
789 	static int first = 1;
790 
791 	ts->tv_nsec = 0;
792 	/* XXX this is a litle fragile but will work okay in the short term */
793 	if (first) {
794 		first = 0;
795 		if (ppc_md.time_init)
796 			timezone_offset = ppc_md.time_init();
797 
798 		/* get_boot_time() isn't guaranteed to be safe to call late */
799 		if (ppc_md.get_boot_time) {
800 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
801 			return;
802 		}
803 	}
804 	if (!ppc_md.get_rtc_time) {
805 		ts->tv_sec = 0;
806 		return;
807 	}
808 	ppc_md.get_rtc_time(&tm);
809 
810 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
811 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
812 }
813 
814 void read_persistent_clock(struct timespec *ts)
815 {
816 	__read_persistent_clock(ts);
817 
818 	/* Sanitize it in case real time clock is set below EPOCH */
819 	if (ts->tv_sec < 0) {
820 		ts->tv_sec = 0;
821 		ts->tv_nsec = 0;
822 	}
823 
824 }
825 
826 /* clocksource code */
827 static cycle_t rtc_read(struct clocksource *cs)
828 {
829 	return (cycle_t)get_rtc();
830 }
831 
832 static cycle_t timebase_read(struct clocksource *cs)
833 {
834 	return (cycle_t)get_tb();
835 }
836 
837 void update_vsyscall(struct timespec *wall_time, struct clocksource *clock,
838 		     u32 mult)
839 {
840 	u64 t2x, stamp_xsec;
841 
842 	if (clock != &clocksource_timebase)
843 		return;
844 
845 	/* Make userspace gettimeofday spin until we're done. */
846 	++vdso_data->tb_update_count;
847 	smp_mb();
848 
849 	/* XXX this assumes clock->shift == 22 */
850 	/* 4611686018 ~= 2^(20+64-22) / 1e9 */
851 	t2x = (u64) mult * 4611686018ULL;
852 	stamp_xsec = (u64) xtime.tv_nsec * XSEC_PER_SEC;
853 	do_div(stamp_xsec, 1000000000);
854 	stamp_xsec += (u64) xtime.tv_sec * XSEC_PER_SEC;
855 	update_gtod(clock->cycle_last, stamp_xsec, t2x);
856 }
857 
858 void update_vsyscall_tz(void)
859 {
860 	/* Make userspace gettimeofday spin until we're done. */
861 	++vdso_data->tb_update_count;
862 	smp_mb();
863 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
864 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
865 	smp_mb();
866 	++vdso_data->tb_update_count;
867 }
868 
869 static void __init clocksource_init(void)
870 {
871 	struct clocksource *clock;
872 
873 	if (__USE_RTC())
874 		clock = &clocksource_rtc;
875 	else
876 		clock = &clocksource_timebase;
877 
878 	clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
879 
880 	if (clocksource_register(clock)) {
881 		printk(KERN_ERR "clocksource: %s is already registered\n",
882 		       clock->name);
883 		return;
884 	}
885 
886 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
887 	       clock->name, clock->mult, clock->shift);
888 }
889 
890 static int decrementer_set_next_event(unsigned long evt,
891 				      struct clock_event_device *dev)
892 {
893 	__get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
894 	set_dec(evt);
895 	return 0;
896 }
897 
898 static void decrementer_set_mode(enum clock_event_mode mode,
899 				 struct clock_event_device *dev)
900 {
901 	if (mode != CLOCK_EVT_MODE_ONESHOT)
902 		decrementer_set_next_event(DECREMENTER_MAX, dev);
903 }
904 
905 static void __init setup_clockevent_multiplier(unsigned long hz)
906 {
907 	u64 mult, shift = 32;
908 
909 	while (1) {
910 		mult = div_sc(hz, NSEC_PER_SEC, shift);
911 		if (mult && (mult >> 32UL) == 0UL)
912 			break;
913 
914 		shift--;
915 	}
916 
917 	decrementer_clockevent.shift = shift;
918 	decrementer_clockevent.mult = mult;
919 }
920 
921 static void register_decrementer_clockevent(int cpu)
922 {
923 	struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
924 
925 	*dec = decrementer_clockevent;
926 	dec->cpumask = cpumask_of(cpu);
927 
928 	printk(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
929 	       dec->name, dec->mult, dec->shift, cpu);
930 
931 	clockevents_register_device(dec);
932 }
933 
934 static void __init init_decrementer_clockevent(void)
935 {
936 	int cpu = smp_processor_id();
937 
938 	setup_clockevent_multiplier(ppc_tb_freq);
939 	decrementer_clockevent.max_delta_ns =
940 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
941 	decrementer_clockevent.min_delta_ns =
942 		clockevent_delta2ns(2, &decrementer_clockevent);
943 
944 	register_decrementer_clockevent(cpu);
945 }
946 
947 void secondary_cpu_time_init(void)
948 {
949 	/* Start the decrementer on CPUs that have manual control
950 	 * such as BookE
951 	 */
952 	start_cpu_decrementer();
953 
954 	/* FIME: Should make unrelatred change to move snapshot_timebase
955 	 * call here ! */
956 	register_decrementer_clockevent(smp_processor_id());
957 }
958 
959 /* This function is only called on the boot processor */
960 void __init time_init(void)
961 {
962 	unsigned long flags;
963 	struct div_result res;
964 	u64 scale, x;
965 	unsigned shift;
966 
967 	if (__USE_RTC()) {
968 		/* 601 processor: dec counts down by 128 every 128ns */
969 		ppc_tb_freq = 1000000000;
970 		tb_last_jiffy = get_rtcl();
971 	} else {
972 		/* Normal PowerPC with timebase register */
973 		ppc_md.calibrate_decr();
974 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
975 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
976 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
977 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
978 		tb_last_jiffy = get_tb();
979 	}
980 
981 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
982 	tb_ticks_per_sec = ppc_tb_freq;
983 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
984 	tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000);
985 	calc_cputime_factors();
986 	setup_cputime_one_jiffy();
987 
988 	/*
989 	 * Calculate the length of each tick in ns.  It will not be
990 	 * exactly 1e9/HZ unless ppc_tb_freq is divisible by HZ.
991 	 * We compute 1e9 * tb_ticks_per_jiffy / ppc_tb_freq,
992 	 * rounded up.
993 	 */
994 	x = (u64) NSEC_PER_SEC * tb_ticks_per_jiffy + ppc_tb_freq - 1;
995 	do_div(x, ppc_tb_freq);
996 	tick_nsec = x;
997 	last_tick_len = x << TICKLEN_SCALE;
998 
999 	/*
1000 	 * Compute ticklen_to_xs, which is a factor which gets multiplied
1001 	 * by (last_tick_len << TICKLEN_SHIFT) to get a tb_to_xs value.
1002 	 * It is computed as:
1003 	 * ticklen_to_xs = 2^N / (tb_ticks_per_jiffy * 1e9)
1004 	 * where N = 64 + 20 - TICKLEN_SCALE - TICKLEN_SHIFT
1005 	 * which turns out to be N = 51 - SHIFT_HZ.
1006 	 * This gives the result as a 0.64 fixed-point fraction.
1007 	 * That value is reduced by an offset amounting to 1 xsec per
1008 	 * 2^31 timebase ticks to avoid problems with time going backwards
1009 	 * by 1 xsec when we do timer_recalc_offset due to losing the
1010 	 * fractional xsec.  That offset is equal to ppc_tb_freq/2^51
1011 	 * since there are 2^20 xsec in a second.
1012 	 */
1013 	div128_by_32((1ULL << 51) - ppc_tb_freq, 0,
1014 		     tb_ticks_per_jiffy << SHIFT_HZ, &res);
1015 	div128_by_32(res.result_high, res.result_low, NSEC_PER_SEC, &res);
1016 	ticklen_to_xs = res.result_low;
1017 
1018 	/* Compute tb_to_xs from tick_nsec */
1019 	tb_to_xs = mulhdu(last_tick_len << TICKLEN_SHIFT, ticklen_to_xs);
1020 
1021 	/*
1022 	 * Compute scale factor for sched_clock.
1023 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1024 	 * which is the timebase frequency.
1025 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1026 	 * the 128-bit result as a 64.64 fixed-point number.
1027 	 * We then shift that number right until it is less than 1.0,
1028 	 * giving us the scale factor and shift count to use in
1029 	 * sched_clock().
1030 	 */
1031 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1032 	scale = res.result_low;
1033 	for (shift = 0; res.result_high != 0; ++shift) {
1034 		scale = (scale >> 1) | (res.result_high << 63);
1035 		res.result_high >>= 1;
1036 	}
1037 	tb_to_ns_scale = scale;
1038 	tb_to_ns_shift = shift;
1039 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1040 	boot_tb = get_tb_or_rtc();
1041 
1042 	write_seqlock_irqsave(&xtime_lock, flags);
1043 
1044 	/* If platform provided a timezone (pmac), we correct the time */
1045         if (timezone_offset) {
1046 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1047 		sys_tz.tz_dsttime = 0;
1048         }
1049 
1050 	vdso_data->tb_orig_stamp = tb_last_jiffy;
1051 	vdso_data->tb_update_count = 0;
1052 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1053 	vdso_data->stamp_xsec = (u64) xtime.tv_sec * XSEC_PER_SEC;
1054 	vdso_data->tb_to_xs = tb_to_xs;
1055 
1056 	write_sequnlock_irqrestore(&xtime_lock, flags);
1057 
1058 	/* Start the decrementer on CPUs that have manual control
1059 	 * such as BookE
1060 	 */
1061 	start_cpu_decrementer();
1062 
1063 	/* Register the clocksource, if we're not running on iSeries */
1064 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
1065 		clocksource_init();
1066 
1067 	init_decrementer_clockevent();
1068 }
1069 
1070 
1071 #define FEBRUARY	2
1072 #define	STARTOFTIME	1970
1073 #define SECDAY		86400L
1074 #define SECYR		(SECDAY * 365)
1075 #define	leapyear(year)		((year) % 4 == 0 && \
1076 				 ((year) % 100 != 0 || (year) % 400 == 0))
1077 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1078 #define	days_in_month(a) 	(month_days[(a) - 1])
1079 
1080 static int month_days[12] = {
1081 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1082 };
1083 
1084 /*
1085  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1086  */
1087 void GregorianDay(struct rtc_time * tm)
1088 {
1089 	int leapsToDate;
1090 	int lastYear;
1091 	int day;
1092 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1093 
1094 	lastYear = tm->tm_year - 1;
1095 
1096 	/*
1097 	 * Number of leap corrections to apply up to end of last year
1098 	 */
1099 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
1100 
1101 	/*
1102 	 * This year is a leap year if it is divisible by 4 except when it is
1103 	 * divisible by 100 unless it is divisible by 400
1104 	 *
1105 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1106 	 */
1107 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
1108 
1109 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
1110 		   tm->tm_mday;
1111 
1112 	tm->tm_wday = day % 7;
1113 }
1114 
1115 void to_tm(int tim, struct rtc_time * tm)
1116 {
1117 	register int    i;
1118 	register long   hms, day;
1119 
1120 	day = tim / SECDAY;
1121 	hms = tim % SECDAY;
1122 
1123 	/* Hours, minutes, seconds are easy */
1124 	tm->tm_hour = hms / 3600;
1125 	tm->tm_min = (hms % 3600) / 60;
1126 	tm->tm_sec = (hms % 3600) % 60;
1127 
1128 	/* Number of years in days */
1129 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1130 		day -= days_in_year(i);
1131 	tm->tm_year = i;
1132 
1133 	/* Number of months in days left */
1134 	if (leapyear(tm->tm_year))
1135 		days_in_month(FEBRUARY) = 29;
1136 	for (i = 1; day >= days_in_month(i); i++)
1137 		day -= days_in_month(i);
1138 	days_in_month(FEBRUARY) = 28;
1139 	tm->tm_mon = i;
1140 
1141 	/* Days are what is left over (+1) from all that. */
1142 	tm->tm_mday = day + 1;
1143 
1144 	/*
1145 	 * Determine the day of week
1146 	 */
1147 	GregorianDay(tm);
1148 }
1149 
1150 /* Auxiliary function to compute scaling factors */
1151 /* Actually the choice of a timebase running at 1/4 the of the bus
1152  * frequency giving resolution of a few tens of nanoseconds is quite nice.
1153  * It makes this computation very precise (27-28 bits typically) which
1154  * is optimistic considering the stability of most processor clock
1155  * oscillators and the precision with which the timebase frequency
1156  * is measured but does not harm.
1157  */
1158 unsigned mulhwu_scale_factor(unsigned inscale, unsigned outscale)
1159 {
1160         unsigned mlt=0, tmp, err;
1161         /* No concern for performance, it's done once: use a stupid
1162          * but safe and compact method to find the multiplier.
1163          */
1164 
1165         for (tmp = 1U<<31; tmp != 0; tmp >>= 1) {
1166                 if (mulhwu(inscale, mlt|tmp) < outscale)
1167 			mlt |= tmp;
1168         }
1169 
1170         /* We might still be off by 1 for the best approximation.
1171          * A side effect of this is that if outscale is too large
1172          * the returned value will be zero.
1173          * Many corner cases have been checked and seem to work,
1174          * some might have been forgotten in the test however.
1175          */
1176 
1177         err = inscale * (mlt+1);
1178         if (err <= inscale/2)
1179 		mlt++;
1180         return mlt;
1181 }
1182 
1183 /*
1184  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1185  * result.
1186  */
1187 void div128_by_32(u64 dividend_high, u64 dividend_low,
1188 		  unsigned divisor, struct div_result *dr)
1189 {
1190 	unsigned long a, b, c, d;
1191 	unsigned long w, x, y, z;
1192 	u64 ra, rb, rc;
1193 
1194 	a = dividend_high >> 32;
1195 	b = dividend_high & 0xffffffff;
1196 	c = dividend_low >> 32;
1197 	d = dividend_low & 0xffffffff;
1198 
1199 	w = a / divisor;
1200 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1201 
1202 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1203 	x = ra;
1204 
1205 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1206 	y = rb;
1207 
1208 	do_div(rc, divisor);
1209 	z = rc;
1210 
1211 	dr->result_high = ((u64)w << 32) + x;
1212 	dr->result_low  = ((u64)y << 32) + z;
1213 
1214 }
1215 
1216 /* We don't need to calibrate delay, we use the CPU timebase for that */
1217 void calibrate_delay(void)
1218 {
1219 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1220 	 * as the number of __delay(1) in a jiffy, so make it so
1221 	 */
1222 	loops_per_jiffy = tb_ticks_per_jiffy;
1223 }
1224 
1225 static int __init rtc_init(void)
1226 {
1227 	struct platform_device *pdev;
1228 
1229 	if (!ppc_md.get_rtc_time)
1230 		return -ENODEV;
1231 
1232 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1233 	if (IS_ERR(pdev))
1234 		return PTR_ERR(pdev);
1235 
1236 	return 0;
1237 }
1238 
1239 module_init(rtc_init);
1240