xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision 7615856ebfee52b080c22d263ca4debbd0df0ac1)
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 	__get_cpu_var(cputime_last_delta) = delta;
269 	__get_cpu_var(cputime_scaled_last_delta) = 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 #ifdef CONFIG_SMP
427 unsigned long profile_pc(struct pt_regs *regs)
428 {
429 	unsigned long pc = instruction_pointer(regs);
430 
431 	if (in_lock_functions(pc))
432 		return regs->link;
433 
434 	return pc;
435 }
436 EXPORT_SYMBOL(profile_pc);
437 #endif
438 
439 #ifdef CONFIG_PPC_ISERIES
440 
441 /*
442  * This function recalibrates the timebase based on the 49-bit time-of-day
443  * value in the Titan chip.  The Titan is much more accurate than the value
444  * returned by the service processor for the timebase frequency.
445  */
446 
447 static int __init iSeries_tb_recal(void)
448 {
449 	struct div_result divres;
450 	unsigned long titan, tb;
451 
452 	/* Make sure we only run on iSeries */
453 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
454 		return -ENODEV;
455 
456 	tb = get_tb();
457 	titan = HvCallXm_loadTod();
458 	if ( iSeries_recal_titan ) {
459 		unsigned long tb_ticks = tb - iSeries_recal_tb;
460 		unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
461 		unsigned long new_tb_ticks_per_sec   = (tb_ticks * USEC_PER_SEC)/titan_usec;
462 		unsigned long new_tb_ticks_per_jiffy =
463 			DIV_ROUND_CLOSEST(new_tb_ticks_per_sec, HZ);
464 		long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
465 		char sign = '+';
466 		/* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
467 		new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;
468 
469 		if ( tick_diff < 0 ) {
470 			tick_diff = -tick_diff;
471 			sign = '-';
472 		}
473 		if ( tick_diff ) {
474 			if ( tick_diff < tb_ticks_per_jiffy/25 ) {
475 				printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
476 						new_tb_ticks_per_jiffy, sign, tick_diff );
477 				tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
478 				tb_ticks_per_sec   = new_tb_ticks_per_sec;
479 				calc_cputime_factors();
480 				div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres );
481 				tb_to_xs = divres.result_low;
482 				vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
483 				vdso_data->tb_to_xs = tb_to_xs;
484 				setup_cputime_one_jiffy();
485 			}
486 			else {
487 				printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
488 					"                   new tb_ticks_per_jiffy = %lu\n"
489 					"                   old tb_ticks_per_jiffy = %lu\n",
490 					new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
491 			}
492 		}
493 	}
494 	iSeries_recal_titan = titan;
495 	iSeries_recal_tb = tb;
496 
497 	/* Called here as now we know accurate values for the timebase */
498 	clocksource_init();
499 	return 0;
500 }
501 late_initcall(iSeries_tb_recal);
502 
503 /* Called from platform early init */
504 void __init iSeries_time_init_early(void)
505 {
506 	iSeries_recal_tb = get_tb();
507 	iSeries_recal_titan = HvCallXm_loadTod();
508 }
509 #endif /* CONFIG_PPC_ISERIES */
510 
511 #ifdef CONFIG_PERF_EVENTS
512 
513 /*
514  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
515  */
516 #ifdef CONFIG_PPC64
517 static inline unsigned long test_perf_event_pending(void)
518 {
519 	unsigned long x;
520 
521 	asm volatile("lbz %0,%1(13)"
522 		: "=r" (x)
523 		: "i" (offsetof(struct paca_struct, perf_event_pending)));
524 	return x;
525 }
526 
527 static inline void set_perf_event_pending_flag(void)
528 {
529 	asm volatile("stb %0,%1(13)" : :
530 		"r" (1),
531 		"i" (offsetof(struct paca_struct, perf_event_pending)));
532 }
533 
534 static inline void clear_perf_event_pending(void)
535 {
536 	asm volatile("stb %0,%1(13)" : :
537 		"r" (0),
538 		"i" (offsetof(struct paca_struct, perf_event_pending)));
539 }
540 
541 #else /* 32-bit */
542 
543 DEFINE_PER_CPU(u8, perf_event_pending);
544 
545 #define set_perf_event_pending_flag()	__get_cpu_var(perf_event_pending) = 1
546 #define test_perf_event_pending()	__get_cpu_var(perf_event_pending)
547 #define clear_perf_event_pending()	__get_cpu_var(perf_event_pending) = 0
548 
549 #endif /* 32 vs 64 bit */
550 
551 void set_perf_event_pending(void)
552 {
553 	preempt_disable();
554 	set_perf_event_pending_flag();
555 	set_dec(1);
556 	preempt_enable();
557 }
558 
559 #else  /* CONFIG_PERF_EVENTS */
560 
561 #define test_perf_event_pending()	0
562 #define clear_perf_event_pending()
563 
564 #endif /* CONFIG_PERF_EVENTS */
565 
566 /*
567  * For iSeries shared processors, we have to let the hypervisor
568  * set the hardware decrementer.  We set a virtual decrementer
569  * in the lppaca and call the hypervisor if the virtual
570  * decrementer is less than the current value in the hardware
571  * decrementer. (almost always the new decrementer value will
572  * be greater than the current hardware decementer so the hypervisor
573  * call will not be needed)
574  */
575 
576 /*
577  * timer_interrupt - gets called when the decrementer overflows,
578  * with interrupts disabled.
579  */
580 void timer_interrupt(struct pt_regs * regs)
581 {
582 	struct pt_regs *old_regs;
583 	struct decrementer_clock *decrementer =  &__get_cpu_var(decrementers);
584 	struct clock_event_device *evt = &decrementer->event;
585 	u64 now;
586 
587 	trace_timer_interrupt_entry(regs);
588 
589 	__get_cpu_var(irq_stat).timer_irqs++;
590 
591 	/* Ensure a positive value is written to the decrementer, or else
592 	 * some CPUs will continuue to take decrementer exceptions */
593 	set_dec(DECREMENTER_MAX);
594 
595 #ifdef CONFIG_PPC32
596 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
597 		do_IRQ(regs);
598 #endif
599 
600 	now = get_tb_or_rtc();
601 	if (now < decrementer->next_tb) {
602 		/* not time for this event yet */
603 		now = decrementer->next_tb - now;
604 		if (now <= DECREMENTER_MAX)
605 			set_dec((int)now);
606 		trace_timer_interrupt_exit(regs);
607 		return;
608 	}
609 	old_regs = set_irq_regs(regs);
610 	irq_enter();
611 
612 	calculate_steal_time();
613 
614 	if (test_perf_event_pending()) {
615 		clear_perf_event_pending();
616 		perf_event_do_pending();
617 	}
618 
619 #ifdef CONFIG_PPC_ISERIES
620 	if (firmware_has_feature(FW_FEATURE_ISERIES))
621 		get_lppaca()->int_dword.fields.decr_int = 0;
622 #endif
623 
624 	if (evt->event_handler)
625 		evt->event_handler(evt);
626 
627 #ifdef CONFIG_PPC_ISERIES
628 	if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending())
629 		process_hvlpevents();
630 #endif
631 
632 #ifdef CONFIG_PPC64
633 	/* collect purr register values often, for accurate calculations */
634 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
635 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
636 		cu->current_tb = mfspr(SPRN_PURR);
637 	}
638 #endif
639 
640 	irq_exit();
641 	set_irq_regs(old_regs);
642 
643 	trace_timer_interrupt_exit(regs);
644 }
645 
646 void wakeup_decrementer(void)
647 {
648 	unsigned long ticks;
649 
650 	/*
651 	 * The timebase gets saved on sleep and restored on wakeup,
652 	 * so all we need to do is to reset the decrementer.
653 	 */
654 	ticks = tb_ticks_since(__get_cpu_var(last_jiffy));
655 	if (ticks < tb_ticks_per_jiffy)
656 		ticks = tb_ticks_per_jiffy - ticks;
657 	else
658 		ticks = 1;
659 	set_dec(ticks);
660 }
661 
662 #ifdef CONFIG_SUSPEND
663 void generic_suspend_disable_irqs(void)
664 {
665 	preempt_disable();
666 
667 	/* Disable the decrementer, so that it doesn't interfere
668 	 * with suspending.
669 	 */
670 
671 	set_dec(0x7fffffff);
672 	local_irq_disable();
673 	set_dec(0x7fffffff);
674 }
675 
676 void generic_suspend_enable_irqs(void)
677 {
678 	wakeup_decrementer();
679 
680 	local_irq_enable();
681 	preempt_enable();
682 }
683 
684 /* Overrides the weak version in kernel/power/main.c */
685 void arch_suspend_disable_irqs(void)
686 {
687 	if (ppc_md.suspend_disable_irqs)
688 		ppc_md.suspend_disable_irqs();
689 	generic_suspend_disable_irqs();
690 }
691 
692 /* Overrides the weak version in kernel/power/main.c */
693 void arch_suspend_enable_irqs(void)
694 {
695 	generic_suspend_enable_irqs();
696 	if (ppc_md.suspend_enable_irqs)
697 		ppc_md.suspend_enable_irqs();
698 }
699 #endif
700 
701 #ifdef CONFIG_SMP
702 void __init smp_space_timers(unsigned int max_cpus)
703 {
704 	int i;
705 	u64 previous_tb = per_cpu(last_jiffy, boot_cpuid);
706 
707 	/* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */
708 	previous_tb -= tb_ticks_per_jiffy;
709 
710 	for_each_possible_cpu(i) {
711 		if (i == boot_cpuid)
712 			continue;
713 		per_cpu(last_jiffy, i) = previous_tb;
714 	}
715 }
716 #endif
717 
718 /*
719  * Scheduler clock - returns current time in nanosec units.
720  *
721  * Note: mulhdu(a, b) (multiply high double unsigned) returns
722  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
723  * are 64-bit unsigned numbers.
724  */
725 unsigned long long sched_clock(void)
726 {
727 	if (__USE_RTC())
728 		return get_rtc();
729 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
730 }
731 
732 static int __init get_freq(char *name, int cells, unsigned long *val)
733 {
734 	struct device_node *cpu;
735 	const unsigned int *fp;
736 	int found = 0;
737 
738 	/* The cpu node should have timebase and clock frequency properties */
739 	cpu = of_find_node_by_type(NULL, "cpu");
740 
741 	if (cpu) {
742 		fp = of_get_property(cpu, name, NULL);
743 		if (fp) {
744 			found = 1;
745 			*val = of_read_ulong(fp, cells);
746 		}
747 
748 		of_node_put(cpu);
749 	}
750 
751 	return found;
752 }
753 
754 /* should become __cpuinit when secondary_cpu_time_init also is */
755 void start_cpu_decrementer(void)
756 {
757 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
758 	/* Clear any pending timer interrupts */
759 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
760 
761 	/* Enable decrementer interrupt */
762 	mtspr(SPRN_TCR, TCR_DIE);
763 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
764 }
765 
766 void __init generic_calibrate_decr(void)
767 {
768 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
769 
770 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
771 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
772 
773 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
774 				"(not found)\n");
775 	}
776 
777 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
778 
779 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
780 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
781 
782 		printk(KERN_ERR "WARNING: Estimating processor frequency "
783 				"(not found)\n");
784 	}
785 }
786 
787 int update_persistent_clock(struct timespec now)
788 {
789 	struct rtc_time tm;
790 
791 	if (!ppc_md.set_rtc_time)
792 		return 0;
793 
794 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
795 	tm.tm_year -= 1900;
796 	tm.tm_mon -= 1;
797 
798 	return ppc_md.set_rtc_time(&tm);
799 }
800 
801 static void __read_persistent_clock(struct timespec *ts)
802 {
803 	struct rtc_time tm;
804 	static int first = 1;
805 
806 	ts->tv_nsec = 0;
807 	/* XXX this is a litle fragile but will work okay in the short term */
808 	if (first) {
809 		first = 0;
810 		if (ppc_md.time_init)
811 			timezone_offset = ppc_md.time_init();
812 
813 		/* get_boot_time() isn't guaranteed to be safe to call late */
814 		if (ppc_md.get_boot_time) {
815 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
816 			return;
817 		}
818 	}
819 	if (!ppc_md.get_rtc_time) {
820 		ts->tv_sec = 0;
821 		return;
822 	}
823 	ppc_md.get_rtc_time(&tm);
824 
825 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
826 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
827 }
828 
829 void read_persistent_clock(struct timespec *ts)
830 {
831 	__read_persistent_clock(ts);
832 
833 	/* Sanitize it in case real time clock is set below EPOCH */
834 	if (ts->tv_sec < 0) {
835 		ts->tv_sec = 0;
836 		ts->tv_nsec = 0;
837 	}
838 
839 }
840 
841 /* clocksource code */
842 static cycle_t rtc_read(struct clocksource *cs)
843 {
844 	return (cycle_t)get_rtc();
845 }
846 
847 static cycle_t timebase_read(struct clocksource *cs)
848 {
849 	return (cycle_t)get_tb();
850 }
851 
852 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
853 			struct clocksource *clock, u32 mult)
854 {
855 	u64 new_tb_to_xs, new_stamp_xsec;
856 
857 	if (clock != &clocksource_timebase)
858 		return;
859 
860 	/* Make userspace gettimeofday spin until we're done. */
861 	++vdso_data->tb_update_count;
862 	smp_mb();
863 
864 	/* XXX this assumes clock->shift == 22 */
865 	/* 4611686018 ~= 2^(20+64-22) / 1e9 */
866 	new_tb_to_xs = (u64) mult * 4611686018ULL;
867 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
868 	do_div(new_stamp_xsec, 1000000000);
869 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
870 
871 	/*
872 	 * tb_update_count is used to allow the userspace gettimeofday code
873 	 * to assure itself that it sees a consistent view of the tb_to_xs and
874 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
875 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
876 	 * the two values of tb_update_count match and are even then the
877 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
878 	 * loops back and reads them again until this criteria is met.
879 	 * We expect the caller to have done the first increment of
880 	 * vdso_data->tb_update_count already.
881 	 */
882 	vdso_data->tb_orig_stamp = clock->cycle_last;
883 	vdso_data->stamp_xsec = new_stamp_xsec;
884 	vdso_data->tb_to_xs = new_tb_to_xs;
885 	vdso_data->wtom_clock_sec = wtm->tv_sec;
886 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
887 	vdso_data->stamp_xtime = *wall_time;
888 	smp_wmb();
889 	++(vdso_data->tb_update_count);
890 }
891 
892 void update_vsyscall_tz(void)
893 {
894 	/* Make userspace gettimeofday spin until we're done. */
895 	++vdso_data->tb_update_count;
896 	smp_mb();
897 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
898 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
899 	smp_mb();
900 	++vdso_data->tb_update_count;
901 }
902 
903 static void __init clocksource_init(void)
904 {
905 	struct clocksource *clock;
906 
907 	if (__USE_RTC())
908 		clock = &clocksource_rtc;
909 	else
910 		clock = &clocksource_timebase;
911 
912 	clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
913 
914 	if (clocksource_register(clock)) {
915 		printk(KERN_ERR "clocksource: %s is already registered\n",
916 		       clock->name);
917 		return;
918 	}
919 
920 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
921 	       clock->name, clock->mult, clock->shift);
922 }
923 
924 static int decrementer_set_next_event(unsigned long evt,
925 				      struct clock_event_device *dev)
926 {
927 	__get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
928 	set_dec(evt);
929 	return 0;
930 }
931 
932 static void decrementer_set_mode(enum clock_event_mode mode,
933 				 struct clock_event_device *dev)
934 {
935 	if (mode != CLOCK_EVT_MODE_ONESHOT)
936 		decrementer_set_next_event(DECREMENTER_MAX, dev);
937 }
938 
939 static inline uint64_t div_sc64(unsigned long ticks, unsigned long nsec,
940 				int shift)
941 {
942 	uint64_t tmp = ((uint64_t)ticks) << shift;
943 
944 	do_div(tmp, nsec);
945 	return tmp;
946 }
947 
948 static void __init setup_clockevent_multiplier(unsigned long hz)
949 {
950 	u64 mult, shift = 32;
951 
952 	while (1) {
953 		mult = div_sc64(hz, NSEC_PER_SEC, shift);
954 		if (mult && (mult >> 32UL) == 0UL)
955 			break;
956 
957 		shift--;
958 	}
959 
960 	decrementer_clockevent.shift = shift;
961 	decrementer_clockevent.mult = mult;
962 }
963 
964 static void register_decrementer_clockevent(int cpu)
965 {
966 	struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
967 
968 	*dec = decrementer_clockevent;
969 	dec->cpumask = cpumask_of(cpu);
970 
971 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
972 		    dec->name, dec->mult, dec->shift, cpu);
973 
974 	clockevents_register_device(dec);
975 }
976 
977 static void __init init_decrementer_clockevent(void)
978 {
979 	int cpu = smp_processor_id();
980 
981 	setup_clockevent_multiplier(ppc_tb_freq);
982 	decrementer_clockevent.max_delta_ns =
983 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
984 	decrementer_clockevent.min_delta_ns =
985 		clockevent_delta2ns(2, &decrementer_clockevent);
986 
987 	register_decrementer_clockevent(cpu);
988 }
989 
990 void secondary_cpu_time_init(void)
991 {
992 	/* Start the decrementer on CPUs that have manual control
993 	 * such as BookE
994 	 */
995 	start_cpu_decrementer();
996 
997 	/* FIME: Should make unrelatred change to move snapshot_timebase
998 	 * call here ! */
999 	register_decrementer_clockevent(smp_processor_id());
1000 }
1001 
1002 /* This function is only called on the boot processor */
1003 void __init time_init(void)
1004 {
1005 	unsigned long flags;
1006 	struct div_result res;
1007 	u64 scale, x;
1008 	unsigned shift;
1009 
1010 	if (__USE_RTC()) {
1011 		/* 601 processor: dec counts down by 128 every 128ns */
1012 		ppc_tb_freq = 1000000000;
1013 		tb_last_jiffy = get_rtcl();
1014 	} else {
1015 		/* Normal PowerPC with timebase register */
1016 		ppc_md.calibrate_decr();
1017 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1018 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1019 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
1020 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1021 		tb_last_jiffy = get_tb();
1022 	}
1023 
1024 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1025 	tb_ticks_per_sec = ppc_tb_freq;
1026 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
1027 	tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000);
1028 	calc_cputime_factors();
1029 	setup_cputime_one_jiffy();
1030 
1031 	/*
1032 	 * Calculate the length of each tick in ns.  It will not be
1033 	 * exactly 1e9/HZ unless ppc_tb_freq is divisible by HZ.
1034 	 * We compute 1e9 * tb_ticks_per_jiffy / ppc_tb_freq,
1035 	 * rounded up.
1036 	 */
1037 	x = (u64) NSEC_PER_SEC * tb_ticks_per_jiffy + ppc_tb_freq - 1;
1038 	do_div(x, ppc_tb_freq);
1039 	tick_nsec = x;
1040 	last_tick_len = x << TICKLEN_SCALE;
1041 
1042 	/*
1043 	 * Compute ticklen_to_xs, which is a factor which gets multiplied
1044 	 * by (last_tick_len << TICKLEN_SHIFT) to get a tb_to_xs value.
1045 	 * It is computed as:
1046 	 * ticklen_to_xs = 2^N / (tb_ticks_per_jiffy * 1e9)
1047 	 * where N = 64 + 20 - TICKLEN_SCALE - TICKLEN_SHIFT
1048 	 * which turns out to be N = 51 - SHIFT_HZ.
1049 	 * This gives the result as a 0.64 fixed-point fraction.
1050 	 * That value is reduced by an offset amounting to 1 xsec per
1051 	 * 2^31 timebase ticks to avoid problems with time going backwards
1052 	 * by 1 xsec when we do timer_recalc_offset due to losing the
1053 	 * fractional xsec.  That offset is equal to ppc_tb_freq/2^51
1054 	 * since there are 2^20 xsec in a second.
1055 	 */
1056 	div128_by_32((1ULL << 51) - ppc_tb_freq, 0,
1057 		     tb_ticks_per_jiffy << SHIFT_HZ, &res);
1058 	div128_by_32(res.result_high, res.result_low, NSEC_PER_SEC, &res);
1059 	ticklen_to_xs = res.result_low;
1060 
1061 	/* Compute tb_to_xs from tick_nsec */
1062 	tb_to_xs = mulhdu(last_tick_len << TICKLEN_SHIFT, ticklen_to_xs);
1063 
1064 	/*
1065 	 * Compute scale factor for sched_clock.
1066 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1067 	 * which is the timebase frequency.
1068 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1069 	 * the 128-bit result as a 64.64 fixed-point number.
1070 	 * We then shift that number right until it is less than 1.0,
1071 	 * giving us the scale factor and shift count to use in
1072 	 * sched_clock().
1073 	 */
1074 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1075 	scale = res.result_low;
1076 	for (shift = 0; res.result_high != 0; ++shift) {
1077 		scale = (scale >> 1) | (res.result_high << 63);
1078 		res.result_high >>= 1;
1079 	}
1080 	tb_to_ns_scale = scale;
1081 	tb_to_ns_shift = shift;
1082 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1083 	boot_tb = get_tb_or_rtc();
1084 
1085 	write_seqlock_irqsave(&xtime_lock, flags);
1086 
1087 	/* If platform provided a timezone (pmac), we correct the time */
1088         if (timezone_offset) {
1089 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1090 		sys_tz.tz_dsttime = 0;
1091         }
1092 
1093 	vdso_data->tb_orig_stamp = tb_last_jiffy;
1094 	vdso_data->tb_update_count = 0;
1095 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1096 	vdso_data->stamp_xsec = (u64) get_seconds() * XSEC_PER_SEC;
1097 	vdso_data->tb_to_xs = tb_to_xs;
1098 
1099 	write_sequnlock_irqrestore(&xtime_lock, flags);
1100 
1101 	/* Start the decrementer on CPUs that have manual control
1102 	 * such as BookE
1103 	 */
1104 	start_cpu_decrementer();
1105 
1106 	/* Register the clocksource, if we're not running on iSeries */
1107 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
1108 		clocksource_init();
1109 
1110 	init_decrementer_clockevent();
1111 }
1112 
1113 
1114 #define FEBRUARY	2
1115 #define	STARTOFTIME	1970
1116 #define SECDAY		86400L
1117 #define SECYR		(SECDAY * 365)
1118 #define	leapyear(year)		((year) % 4 == 0 && \
1119 				 ((year) % 100 != 0 || (year) % 400 == 0))
1120 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1121 #define	days_in_month(a) 	(month_days[(a) - 1])
1122 
1123 static int month_days[12] = {
1124 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1125 };
1126 
1127 /*
1128  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1129  */
1130 void GregorianDay(struct rtc_time * tm)
1131 {
1132 	int leapsToDate;
1133 	int lastYear;
1134 	int day;
1135 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1136 
1137 	lastYear = tm->tm_year - 1;
1138 
1139 	/*
1140 	 * Number of leap corrections to apply up to end of last year
1141 	 */
1142 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
1143 
1144 	/*
1145 	 * This year is a leap year if it is divisible by 4 except when it is
1146 	 * divisible by 100 unless it is divisible by 400
1147 	 *
1148 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1149 	 */
1150 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
1151 
1152 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
1153 		   tm->tm_mday;
1154 
1155 	tm->tm_wday = day % 7;
1156 }
1157 
1158 void to_tm(int tim, struct rtc_time * tm)
1159 {
1160 	register int    i;
1161 	register long   hms, day;
1162 
1163 	day = tim / SECDAY;
1164 	hms = tim % SECDAY;
1165 
1166 	/* Hours, minutes, seconds are easy */
1167 	tm->tm_hour = hms / 3600;
1168 	tm->tm_min = (hms % 3600) / 60;
1169 	tm->tm_sec = (hms % 3600) % 60;
1170 
1171 	/* Number of years in days */
1172 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1173 		day -= days_in_year(i);
1174 	tm->tm_year = i;
1175 
1176 	/* Number of months in days left */
1177 	if (leapyear(tm->tm_year))
1178 		days_in_month(FEBRUARY) = 29;
1179 	for (i = 1; day >= days_in_month(i); i++)
1180 		day -= days_in_month(i);
1181 	days_in_month(FEBRUARY) = 28;
1182 	tm->tm_mon = i;
1183 
1184 	/* Days are what is left over (+1) from all that. */
1185 	tm->tm_mday = day + 1;
1186 
1187 	/*
1188 	 * Determine the day of week
1189 	 */
1190 	GregorianDay(tm);
1191 }
1192 
1193 /* Auxiliary function to compute scaling factors */
1194 /* Actually the choice of a timebase running at 1/4 the of the bus
1195  * frequency giving resolution of a few tens of nanoseconds is quite nice.
1196  * It makes this computation very precise (27-28 bits typically) which
1197  * is optimistic considering the stability of most processor clock
1198  * oscillators and the precision with which the timebase frequency
1199  * is measured but does not harm.
1200  */
1201 unsigned mulhwu_scale_factor(unsigned inscale, unsigned outscale)
1202 {
1203         unsigned mlt=0, tmp, err;
1204         /* No concern for performance, it's done once: use a stupid
1205          * but safe and compact method to find the multiplier.
1206          */
1207 
1208         for (tmp = 1U<<31; tmp != 0; tmp >>= 1) {
1209                 if (mulhwu(inscale, mlt|tmp) < outscale)
1210 			mlt |= tmp;
1211         }
1212 
1213         /* We might still be off by 1 for the best approximation.
1214          * A side effect of this is that if outscale is too large
1215          * the returned value will be zero.
1216          * Many corner cases have been checked and seem to work,
1217          * some might have been forgotten in the test however.
1218          */
1219 
1220         err = inscale * (mlt+1);
1221         if (err <= inscale/2)
1222 		mlt++;
1223         return mlt;
1224 }
1225 
1226 /*
1227  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1228  * result.
1229  */
1230 void div128_by_32(u64 dividend_high, u64 dividend_low,
1231 		  unsigned divisor, struct div_result *dr)
1232 {
1233 	unsigned long a, b, c, d;
1234 	unsigned long w, x, y, z;
1235 	u64 ra, rb, rc;
1236 
1237 	a = dividend_high >> 32;
1238 	b = dividend_high & 0xffffffff;
1239 	c = dividend_low >> 32;
1240 	d = dividend_low & 0xffffffff;
1241 
1242 	w = a / divisor;
1243 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1244 
1245 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1246 	x = ra;
1247 
1248 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1249 	y = rb;
1250 
1251 	do_div(rc, divisor);
1252 	z = rc;
1253 
1254 	dr->result_high = ((u64)w << 32) + x;
1255 	dr->result_low  = ((u64)y << 32) + z;
1256 
1257 }
1258 
1259 /* We don't need to calibrate delay, we use the CPU timebase for that */
1260 void calibrate_delay(void)
1261 {
1262 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1263 	 * as the number of __delay(1) in a jiffy, so make it so
1264 	 */
1265 	loops_per_jiffy = tb_ticks_per_jiffy;
1266 }
1267 
1268 static int __init rtc_init(void)
1269 {
1270 	struct platform_device *pdev;
1271 
1272 	if (!ppc_md.get_rtc_time)
1273 		return -ENODEV;
1274 
1275 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1276 	if (IS_ERR(pdev))
1277 		return PTR_ERR(pdev);
1278 
1279 	return 0;
1280 }
1281 
1282 module_init(rtc_init);
1283