xref: /linux/arch/powerpc/kernel/time.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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.
21  * - for astronomical applications: add a new function to get
22  * non ambiguous timestamps even around leap seconds. This needs
23  * a new timestamp format and a good name.
24  *
25  * 1997-09-10  Updated NTP code according to technical memorandum Jan '96
26  *             "A Kernel Model for Precision Timekeeping" by Dave Mills
27  *
28  *      This program is free software; you can redistribute it and/or
29  *      modify it under the terms of the GNU General Public License
30  *      as published by the Free Software Foundation; either version
31  *      2 of the License, or (at your option) any later version.
32  */
33 
34 #include <linux/errno.h>
35 #include <linux/export.h>
36 #include <linux/sched.h>
37 #include <linux/kernel.h>
38 #include <linux/param.h>
39 #include <linux/string.h>
40 #include <linux/mm.h>
41 #include <linux/interrupt.h>
42 #include <linux/timex.h>
43 #include <linux/kernel_stat.h>
44 #include <linux/time.h>
45 #include <linux/init.h>
46 #include <linux/profile.h>
47 #include <linux/cpu.h>
48 #include <linux/security.h>
49 #include <linux/percpu.h>
50 #include <linux/rtc.h>
51 #include <linux/jiffies.h>
52 #include <linux/posix-timers.h>
53 #include <linux/irq.h>
54 #include <linux/delay.h>
55 #include <linux/irq_work.h>
56 #include <asm/trace.h>
57 
58 #include <asm/io.h>
59 #include <asm/processor.h>
60 #include <asm/nvram.h>
61 #include <asm/cache.h>
62 #include <asm/machdep.h>
63 #include <asm/uaccess.h>
64 #include <asm/time.h>
65 #include <asm/prom.h>
66 #include <asm/irq.h>
67 #include <asm/div64.h>
68 #include <asm/smp.h>
69 #include <asm/vdso_datapage.h>
70 #include <asm/firmware.h>
71 #include <asm/cputime.h>
72 
73 /* powerpc clocksource/clockevent code */
74 
75 #include <linux/clockchips.h>
76 #include <linux/timekeeper_internal.h>
77 
78 static cycle_t rtc_read(struct clocksource *);
79 static struct clocksource clocksource_rtc = {
80 	.name         = "rtc",
81 	.rating       = 400,
82 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
83 	.mask         = CLOCKSOURCE_MASK(64),
84 	.read         = rtc_read,
85 };
86 
87 static cycle_t timebase_read(struct clocksource *);
88 static struct clocksource clocksource_timebase = {
89 	.name         = "timebase",
90 	.rating       = 400,
91 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
92 	.mask         = CLOCKSOURCE_MASK(64),
93 	.read         = timebase_read,
94 };
95 
96 #define DECREMENTER_MAX	0x7fffffff
97 
98 static int decrementer_set_next_event(unsigned long evt,
99 				      struct clock_event_device *dev);
100 static void decrementer_set_mode(enum clock_event_mode mode,
101 				 struct clock_event_device *dev);
102 
103 struct clock_event_device decrementer_clockevent = {
104 	.name           = "decrementer",
105 	.rating         = 200,
106 	.irq            = 0,
107 	.set_next_event = decrementer_set_next_event,
108 	.set_mode       = decrementer_set_mode,
109 	.features       = CLOCK_EVT_FEAT_ONESHOT,
110 };
111 EXPORT_SYMBOL(decrementer_clockevent);
112 
113 DEFINE_PER_CPU(u64, decrementers_next_tb);
114 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
115 
116 #define XSEC_PER_SEC (1024*1024)
117 
118 #ifdef CONFIG_PPC64
119 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
120 #else
121 /* compute ((xsec << 12) * max) >> 32 */
122 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
123 #endif
124 
125 unsigned long tb_ticks_per_jiffy;
126 unsigned long tb_ticks_per_usec = 100; /* sane default */
127 EXPORT_SYMBOL(tb_ticks_per_usec);
128 unsigned long tb_ticks_per_sec;
129 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
130 
131 DEFINE_SPINLOCK(rtc_lock);
132 EXPORT_SYMBOL_GPL(rtc_lock);
133 
134 static u64 tb_to_ns_scale __read_mostly;
135 static unsigned tb_to_ns_shift __read_mostly;
136 static u64 boot_tb __read_mostly;
137 
138 extern struct timezone sys_tz;
139 static long timezone_offset;
140 
141 unsigned long ppc_proc_freq;
142 EXPORT_SYMBOL_GPL(ppc_proc_freq);
143 unsigned long ppc_tb_freq;
144 EXPORT_SYMBOL_GPL(ppc_tb_freq);
145 
146 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
147 /*
148  * Factors for converting from cputime_t (timebase ticks) to
149  * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
150  * These are all stored as 0.64 fixed-point binary fractions.
151  */
152 u64 __cputime_jiffies_factor;
153 EXPORT_SYMBOL(__cputime_jiffies_factor);
154 u64 __cputime_usec_factor;
155 EXPORT_SYMBOL(__cputime_usec_factor);
156 u64 __cputime_sec_factor;
157 EXPORT_SYMBOL(__cputime_sec_factor);
158 u64 __cputime_clockt_factor;
159 EXPORT_SYMBOL(__cputime_clockt_factor);
160 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
161 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
162 
163 cputime_t cputime_one_jiffy;
164 
165 void (*dtl_consumer)(struct dtl_entry *, u64);
166 
167 static void calc_cputime_factors(void)
168 {
169 	struct div_result res;
170 
171 	div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
172 	__cputime_jiffies_factor = res.result_low;
173 	div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
174 	__cputime_usec_factor = res.result_low;
175 	div128_by_32(1, 0, tb_ticks_per_sec, &res);
176 	__cputime_sec_factor = res.result_low;
177 	div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
178 	__cputime_clockt_factor = res.result_low;
179 }
180 
181 /*
182  * Read the SPURR on systems that have it, otherwise the PURR,
183  * or if that doesn't exist return the timebase value passed in.
184  */
185 static u64 read_spurr(u64 tb)
186 {
187 	if (cpu_has_feature(CPU_FTR_SPURR))
188 		return mfspr(SPRN_SPURR);
189 	if (cpu_has_feature(CPU_FTR_PURR))
190 		return mfspr(SPRN_PURR);
191 	return tb;
192 }
193 
194 #ifdef CONFIG_PPC_SPLPAR
195 
196 /*
197  * Scan the dispatch trace log and count up the stolen time.
198  * Should be called with interrupts disabled.
199  */
200 static u64 scan_dispatch_log(u64 stop_tb)
201 {
202 	u64 i = local_paca->dtl_ridx;
203 	struct dtl_entry *dtl = local_paca->dtl_curr;
204 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
205 	struct lppaca *vpa = local_paca->lppaca_ptr;
206 	u64 tb_delta;
207 	u64 stolen = 0;
208 	u64 dtb;
209 
210 	if (!dtl)
211 		return 0;
212 
213 	if (i == vpa->dtl_idx)
214 		return 0;
215 	while (i < vpa->dtl_idx) {
216 		if (dtl_consumer)
217 			dtl_consumer(dtl, i);
218 		dtb = dtl->timebase;
219 		tb_delta = dtl->enqueue_to_dispatch_time +
220 			dtl->ready_to_enqueue_time;
221 		barrier();
222 		if (i + N_DISPATCH_LOG < vpa->dtl_idx) {
223 			/* buffer has overflowed */
224 			i = vpa->dtl_idx - N_DISPATCH_LOG;
225 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
226 			continue;
227 		}
228 		if (dtb > stop_tb)
229 			break;
230 		stolen += tb_delta;
231 		++i;
232 		++dtl;
233 		if (dtl == dtl_end)
234 			dtl = local_paca->dispatch_log;
235 	}
236 	local_paca->dtl_ridx = i;
237 	local_paca->dtl_curr = dtl;
238 	return stolen;
239 }
240 
241 /*
242  * Accumulate stolen time by scanning the dispatch trace log.
243  * Called on entry from user mode.
244  */
245 void accumulate_stolen_time(void)
246 {
247 	u64 sst, ust;
248 
249 	u8 save_soft_enabled = local_paca->soft_enabled;
250 
251 	/* We are called early in the exception entry, before
252 	 * soft/hard_enabled are sync'ed to the expected state
253 	 * for the exception. We are hard disabled but the PACA
254 	 * needs to reflect that so various debug stuff doesn't
255 	 * complain
256 	 */
257 	local_paca->soft_enabled = 0;
258 
259 	sst = scan_dispatch_log(local_paca->starttime_user);
260 	ust = scan_dispatch_log(local_paca->starttime);
261 	local_paca->system_time -= sst;
262 	local_paca->user_time -= ust;
263 	local_paca->stolen_time += ust + sst;
264 
265 	local_paca->soft_enabled = save_soft_enabled;
266 }
267 
268 static inline u64 calculate_stolen_time(u64 stop_tb)
269 {
270 	u64 stolen = 0;
271 
272 	if (get_paca()->dtl_ridx != get_paca()->lppaca_ptr->dtl_idx) {
273 		stolen = scan_dispatch_log(stop_tb);
274 		get_paca()->system_time -= stolen;
275 	}
276 
277 	stolen += get_paca()->stolen_time;
278 	get_paca()->stolen_time = 0;
279 	return stolen;
280 }
281 
282 #else /* CONFIG_PPC_SPLPAR */
283 static inline u64 calculate_stolen_time(u64 stop_tb)
284 {
285 	return 0;
286 }
287 
288 #endif /* CONFIG_PPC_SPLPAR */
289 
290 /*
291  * Account time for a transition between system, hard irq
292  * or soft irq state.
293  */
294 static u64 vtime_delta(struct task_struct *tsk,
295 			u64 *sys_scaled, u64 *stolen)
296 {
297 	u64 now, nowscaled, deltascaled;
298 	u64 udelta, delta, user_scaled;
299 
300 	WARN_ON_ONCE(!irqs_disabled());
301 
302 	now = mftb();
303 	nowscaled = read_spurr(now);
304 	get_paca()->system_time += now - get_paca()->starttime;
305 	get_paca()->starttime = now;
306 	deltascaled = nowscaled - get_paca()->startspurr;
307 	get_paca()->startspurr = nowscaled;
308 
309 	*stolen = calculate_stolen_time(now);
310 
311 	delta = get_paca()->system_time;
312 	get_paca()->system_time = 0;
313 	udelta = get_paca()->user_time - get_paca()->utime_sspurr;
314 	get_paca()->utime_sspurr = get_paca()->user_time;
315 
316 	/*
317 	 * Because we don't read the SPURR on every kernel entry/exit,
318 	 * deltascaled includes both user and system SPURR ticks.
319 	 * Apportion these ticks to system SPURR ticks and user
320 	 * SPURR ticks in the same ratio as the system time (delta)
321 	 * and user time (udelta) values obtained from the timebase
322 	 * over the same interval.  The system ticks get accounted here;
323 	 * the user ticks get saved up in paca->user_time_scaled to be
324 	 * used by account_process_tick.
325 	 */
326 	*sys_scaled = delta;
327 	user_scaled = udelta;
328 	if (deltascaled != delta + udelta) {
329 		if (udelta) {
330 			*sys_scaled = deltascaled * delta / (delta + udelta);
331 			user_scaled = deltascaled - *sys_scaled;
332 		} else {
333 			*sys_scaled = deltascaled;
334 		}
335 	}
336 	get_paca()->user_time_scaled += user_scaled;
337 
338 	return delta;
339 }
340 
341 void vtime_account_system(struct task_struct *tsk)
342 {
343 	u64 delta, sys_scaled, stolen;
344 
345 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
346 	account_system_time(tsk, 0, delta, sys_scaled);
347 	if (stolen)
348 		account_steal_time(stolen);
349 }
350 
351 void vtime_account_idle(struct task_struct *tsk)
352 {
353 	u64 delta, sys_scaled, stolen;
354 
355 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
356 	account_idle_time(delta + stolen);
357 }
358 
359 /*
360  * Transfer the user time accumulated in the paca
361  * by the exception entry and exit code to the generic
362  * process user time records.
363  * Must be called with interrupts disabled.
364  * Assumes that vtime_account_system/idle() has been called
365  * recently (i.e. since the last entry from usermode) so that
366  * get_paca()->user_time_scaled is up to date.
367  */
368 void vtime_account_user(struct task_struct *tsk)
369 {
370 	cputime_t utime, utimescaled;
371 
372 	utime = get_paca()->user_time;
373 	utimescaled = get_paca()->user_time_scaled;
374 	get_paca()->user_time = 0;
375 	get_paca()->user_time_scaled = 0;
376 	get_paca()->utime_sspurr = 0;
377 	account_user_time(tsk, utime, utimescaled);
378 }
379 
380 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
381 #define calc_cputime_factors()
382 #endif
383 
384 void __delay(unsigned long loops)
385 {
386 	unsigned long start;
387 	int diff;
388 
389 	if (__USE_RTC()) {
390 		start = get_rtcl();
391 		do {
392 			/* the RTCL register wraps at 1000000000 */
393 			diff = get_rtcl() - start;
394 			if (diff < 0)
395 				diff += 1000000000;
396 		} while (diff < loops);
397 	} else {
398 		start = get_tbl();
399 		while (get_tbl() - start < loops)
400 			HMT_low();
401 		HMT_medium();
402 	}
403 }
404 EXPORT_SYMBOL(__delay);
405 
406 void udelay(unsigned long usecs)
407 {
408 	__delay(tb_ticks_per_usec * usecs);
409 }
410 EXPORT_SYMBOL(udelay);
411 
412 #ifdef CONFIG_SMP
413 unsigned long profile_pc(struct pt_regs *regs)
414 {
415 	unsigned long pc = instruction_pointer(regs);
416 
417 	if (in_lock_functions(pc))
418 		return regs->link;
419 
420 	return pc;
421 }
422 EXPORT_SYMBOL(profile_pc);
423 #endif
424 
425 #ifdef CONFIG_IRQ_WORK
426 
427 /*
428  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
429  */
430 #ifdef CONFIG_PPC64
431 static inline unsigned long test_irq_work_pending(void)
432 {
433 	unsigned long x;
434 
435 	asm volatile("lbz %0,%1(13)"
436 		: "=r" (x)
437 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
438 	return x;
439 }
440 
441 static inline void set_irq_work_pending_flag(void)
442 {
443 	asm volatile("stb %0,%1(13)" : :
444 		"r" (1),
445 		"i" (offsetof(struct paca_struct, irq_work_pending)));
446 }
447 
448 static inline void clear_irq_work_pending(void)
449 {
450 	asm volatile("stb %0,%1(13)" : :
451 		"r" (0),
452 		"i" (offsetof(struct paca_struct, irq_work_pending)));
453 }
454 
455 #else /* 32-bit */
456 
457 DEFINE_PER_CPU(u8, irq_work_pending);
458 
459 #define set_irq_work_pending_flag()	__get_cpu_var(irq_work_pending) = 1
460 #define test_irq_work_pending()		__get_cpu_var(irq_work_pending)
461 #define clear_irq_work_pending()	__get_cpu_var(irq_work_pending) = 0
462 
463 #endif /* 32 vs 64 bit */
464 
465 void arch_irq_work_raise(void)
466 {
467 	preempt_disable();
468 	set_irq_work_pending_flag();
469 	set_dec(1);
470 	preempt_enable();
471 }
472 
473 #else  /* CONFIG_IRQ_WORK */
474 
475 #define test_irq_work_pending()	0
476 #define clear_irq_work_pending()
477 
478 #endif /* CONFIG_IRQ_WORK */
479 
480 /*
481  * timer_interrupt - gets called when the decrementer overflows,
482  * with interrupts disabled.
483  */
484 void timer_interrupt(struct pt_regs * regs)
485 {
486 	struct pt_regs *old_regs;
487 	u64 *next_tb = &__get_cpu_var(decrementers_next_tb);
488 	struct clock_event_device *evt = &__get_cpu_var(decrementers);
489 	u64 now;
490 
491 	/* Ensure a positive value is written to the decrementer, or else
492 	 * some CPUs will continue to take decrementer exceptions.
493 	 */
494 	set_dec(DECREMENTER_MAX);
495 
496 	/* Some implementations of hotplug will get timer interrupts while
497 	 * offline, just ignore these
498 	 */
499 	if (!cpu_online(smp_processor_id()))
500 		return;
501 
502 	/* Conditionally hard-enable interrupts now that the DEC has been
503 	 * bumped to its maximum value
504 	 */
505 	may_hard_irq_enable();
506 
507 	__get_cpu_var(irq_stat).timer_irqs++;
508 
509 #if defined(CONFIG_PPC32) && defined(CONFIG_PMAC)
510 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
511 		do_IRQ(regs);
512 #endif
513 
514 	old_regs = set_irq_regs(regs);
515 	irq_enter();
516 
517 	trace_timer_interrupt_entry(regs);
518 
519 	if (test_irq_work_pending()) {
520 		clear_irq_work_pending();
521 		irq_work_run();
522 	}
523 
524 	now = get_tb_or_rtc();
525 	if (now >= *next_tb) {
526 		*next_tb = ~(u64)0;
527 		if (evt->event_handler)
528 			evt->event_handler(evt);
529 	} else {
530 		now = *next_tb - now;
531 		if (now <= DECREMENTER_MAX)
532 			set_dec((int)now);
533 	}
534 
535 #ifdef CONFIG_PPC64
536 	/* collect purr register values often, for accurate calculations */
537 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
538 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
539 		cu->current_tb = mfspr(SPRN_PURR);
540 	}
541 #endif
542 
543 	trace_timer_interrupt_exit(regs);
544 
545 	irq_exit();
546 	set_irq_regs(old_regs);
547 }
548 
549 /*
550  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
551  * left pending on exit from a KVM guest.  We don't need to do anything
552  * to clear them, as they are edge-triggered.
553  */
554 void hdec_interrupt(struct pt_regs *regs)
555 {
556 }
557 
558 #ifdef CONFIG_SUSPEND
559 static void generic_suspend_disable_irqs(void)
560 {
561 	/* Disable the decrementer, so that it doesn't interfere
562 	 * with suspending.
563 	 */
564 
565 	set_dec(DECREMENTER_MAX);
566 	local_irq_disable();
567 	set_dec(DECREMENTER_MAX);
568 }
569 
570 static void generic_suspend_enable_irqs(void)
571 {
572 	local_irq_enable();
573 }
574 
575 /* Overrides the weak version in kernel/power/main.c */
576 void arch_suspend_disable_irqs(void)
577 {
578 	if (ppc_md.suspend_disable_irqs)
579 		ppc_md.suspend_disable_irqs();
580 	generic_suspend_disable_irqs();
581 }
582 
583 /* Overrides the weak version in kernel/power/main.c */
584 void arch_suspend_enable_irqs(void)
585 {
586 	generic_suspend_enable_irqs();
587 	if (ppc_md.suspend_enable_irqs)
588 		ppc_md.suspend_enable_irqs();
589 }
590 #endif
591 
592 /*
593  * Scheduler clock - returns current time in nanosec units.
594  *
595  * Note: mulhdu(a, b) (multiply high double unsigned) returns
596  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
597  * are 64-bit unsigned numbers.
598  */
599 unsigned long long sched_clock(void)
600 {
601 	if (__USE_RTC())
602 		return get_rtc();
603 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
604 }
605 
606 static int __init get_freq(char *name, int cells, unsigned long *val)
607 {
608 	struct device_node *cpu;
609 	const unsigned int *fp;
610 	int found = 0;
611 
612 	/* The cpu node should have timebase and clock frequency properties */
613 	cpu = of_find_node_by_type(NULL, "cpu");
614 
615 	if (cpu) {
616 		fp = of_get_property(cpu, name, NULL);
617 		if (fp) {
618 			found = 1;
619 			*val = of_read_ulong(fp, cells);
620 		}
621 
622 		of_node_put(cpu);
623 	}
624 
625 	return found;
626 }
627 
628 /* should become __cpuinit when secondary_cpu_time_init also is */
629 void start_cpu_decrementer(void)
630 {
631 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
632 	/* Clear any pending timer interrupts */
633 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
634 
635 	/* Enable decrementer interrupt */
636 	mtspr(SPRN_TCR, TCR_DIE);
637 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
638 }
639 
640 void __init generic_calibrate_decr(void)
641 {
642 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
643 
644 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
645 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
646 
647 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
648 				"(not found)\n");
649 	}
650 
651 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
652 
653 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
654 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
655 
656 		printk(KERN_ERR "WARNING: Estimating processor frequency "
657 				"(not found)\n");
658 	}
659 }
660 
661 int update_persistent_clock(struct timespec now)
662 {
663 	struct rtc_time tm;
664 
665 	if (!ppc_md.set_rtc_time)
666 		return 0;
667 
668 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
669 	tm.tm_year -= 1900;
670 	tm.tm_mon -= 1;
671 
672 	return ppc_md.set_rtc_time(&tm);
673 }
674 
675 static void __read_persistent_clock(struct timespec *ts)
676 {
677 	struct rtc_time tm;
678 	static int first = 1;
679 
680 	ts->tv_nsec = 0;
681 	/* XXX this is a litle fragile but will work okay in the short term */
682 	if (first) {
683 		first = 0;
684 		if (ppc_md.time_init)
685 			timezone_offset = ppc_md.time_init();
686 
687 		/* get_boot_time() isn't guaranteed to be safe to call late */
688 		if (ppc_md.get_boot_time) {
689 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
690 			return;
691 		}
692 	}
693 	if (!ppc_md.get_rtc_time) {
694 		ts->tv_sec = 0;
695 		return;
696 	}
697 	ppc_md.get_rtc_time(&tm);
698 
699 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
700 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
701 }
702 
703 void read_persistent_clock(struct timespec *ts)
704 {
705 	__read_persistent_clock(ts);
706 
707 	/* Sanitize it in case real time clock is set below EPOCH */
708 	if (ts->tv_sec < 0) {
709 		ts->tv_sec = 0;
710 		ts->tv_nsec = 0;
711 	}
712 
713 }
714 
715 /* clocksource code */
716 static cycle_t rtc_read(struct clocksource *cs)
717 {
718 	return (cycle_t)get_rtc();
719 }
720 
721 static cycle_t timebase_read(struct clocksource *cs)
722 {
723 	return (cycle_t)get_tb();
724 }
725 
726 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
727 			struct clocksource *clock, u32 mult)
728 {
729 	u64 new_tb_to_xs, new_stamp_xsec;
730 	u32 frac_sec;
731 
732 	if (clock != &clocksource_timebase)
733 		return;
734 
735 	/* Make userspace gettimeofday spin until we're done. */
736 	++vdso_data->tb_update_count;
737 	smp_mb();
738 
739 	/* 19342813113834067 ~= 2^(20+64) / 1e9 */
740 	new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
741 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
742 	do_div(new_stamp_xsec, 1000000000);
743 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
744 
745 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
746 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
747 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
748 
749 	/*
750 	 * tb_update_count is used to allow the userspace gettimeofday code
751 	 * to assure itself that it sees a consistent view of the tb_to_xs and
752 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
753 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
754 	 * the two values of tb_update_count match and are even then the
755 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
756 	 * loops back and reads them again until this criteria is met.
757 	 * We expect the caller to have done the first increment of
758 	 * vdso_data->tb_update_count already.
759 	 */
760 	vdso_data->tb_orig_stamp = clock->cycle_last;
761 	vdso_data->stamp_xsec = new_stamp_xsec;
762 	vdso_data->tb_to_xs = new_tb_to_xs;
763 	vdso_data->wtom_clock_sec = wtm->tv_sec;
764 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
765 	vdso_data->stamp_xtime = *wall_time;
766 	vdso_data->stamp_sec_fraction = frac_sec;
767 	smp_wmb();
768 	++(vdso_data->tb_update_count);
769 }
770 
771 void update_vsyscall_tz(void)
772 {
773 	/* Make userspace gettimeofday spin until we're done. */
774 	++vdso_data->tb_update_count;
775 	smp_mb();
776 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
777 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
778 	smp_mb();
779 	++vdso_data->tb_update_count;
780 }
781 
782 static void __init clocksource_init(void)
783 {
784 	struct clocksource *clock;
785 
786 	if (__USE_RTC())
787 		clock = &clocksource_rtc;
788 	else
789 		clock = &clocksource_timebase;
790 
791 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
792 		printk(KERN_ERR "clocksource: %s is already registered\n",
793 		       clock->name);
794 		return;
795 	}
796 
797 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
798 	       clock->name, clock->mult, clock->shift);
799 }
800 
801 static int decrementer_set_next_event(unsigned long evt,
802 				      struct clock_event_device *dev)
803 {
804 	__get_cpu_var(decrementers_next_tb) = get_tb_or_rtc() + evt;
805 	set_dec(evt);
806 	return 0;
807 }
808 
809 static void decrementer_set_mode(enum clock_event_mode mode,
810 				 struct clock_event_device *dev)
811 {
812 	if (mode != CLOCK_EVT_MODE_ONESHOT)
813 		decrementer_set_next_event(DECREMENTER_MAX, dev);
814 }
815 
816 static void register_decrementer_clockevent(int cpu)
817 {
818 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
819 
820 	*dec = decrementer_clockevent;
821 	dec->cpumask = cpumask_of(cpu);
822 
823 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
824 		    dec->name, dec->mult, dec->shift, cpu);
825 
826 	clockevents_register_device(dec);
827 }
828 
829 static void __init init_decrementer_clockevent(void)
830 {
831 	int cpu = smp_processor_id();
832 
833 	clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
834 
835 	decrementer_clockevent.max_delta_ns =
836 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
837 	decrementer_clockevent.min_delta_ns =
838 		clockevent_delta2ns(2, &decrementer_clockevent);
839 
840 	register_decrementer_clockevent(cpu);
841 }
842 
843 void secondary_cpu_time_init(void)
844 {
845 	/* Start the decrementer on CPUs that have manual control
846 	 * such as BookE
847 	 */
848 	start_cpu_decrementer();
849 
850 	/* FIME: Should make unrelatred change to move snapshot_timebase
851 	 * call here ! */
852 	register_decrementer_clockevent(smp_processor_id());
853 }
854 
855 /* This function is only called on the boot processor */
856 void __init time_init(void)
857 {
858 	struct div_result res;
859 	u64 scale;
860 	unsigned shift;
861 
862 	if (__USE_RTC()) {
863 		/* 601 processor: dec counts down by 128 every 128ns */
864 		ppc_tb_freq = 1000000000;
865 	} else {
866 		/* Normal PowerPC with timebase register */
867 		ppc_md.calibrate_decr();
868 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
869 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
870 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
871 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
872 	}
873 
874 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
875 	tb_ticks_per_sec = ppc_tb_freq;
876 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
877 	calc_cputime_factors();
878 	setup_cputime_one_jiffy();
879 
880 	/*
881 	 * Compute scale factor for sched_clock.
882 	 * The calibrate_decr() function has set tb_ticks_per_sec,
883 	 * which is the timebase frequency.
884 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
885 	 * the 128-bit result as a 64.64 fixed-point number.
886 	 * We then shift that number right until it is less than 1.0,
887 	 * giving us the scale factor and shift count to use in
888 	 * sched_clock().
889 	 */
890 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
891 	scale = res.result_low;
892 	for (shift = 0; res.result_high != 0; ++shift) {
893 		scale = (scale >> 1) | (res.result_high << 63);
894 		res.result_high >>= 1;
895 	}
896 	tb_to_ns_scale = scale;
897 	tb_to_ns_shift = shift;
898 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
899 	boot_tb = get_tb_or_rtc();
900 
901 	/* If platform provided a timezone (pmac), we correct the time */
902 	if (timezone_offset) {
903 		sys_tz.tz_minuteswest = -timezone_offset / 60;
904 		sys_tz.tz_dsttime = 0;
905 	}
906 
907 	vdso_data->tb_update_count = 0;
908 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
909 
910 	/* Start the decrementer on CPUs that have manual control
911 	 * such as BookE
912 	 */
913 	start_cpu_decrementer();
914 
915 	/* Register the clocksource */
916 	clocksource_init();
917 
918 	init_decrementer_clockevent();
919 }
920 
921 
922 #define FEBRUARY	2
923 #define	STARTOFTIME	1970
924 #define SECDAY		86400L
925 #define SECYR		(SECDAY * 365)
926 #define	leapyear(year)		((year) % 4 == 0 && \
927 				 ((year) % 100 != 0 || (year) % 400 == 0))
928 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
929 #define	days_in_month(a) 	(month_days[(a) - 1])
930 
931 static int month_days[12] = {
932 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
933 };
934 
935 /*
936  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
937  */
938 void GregorianDay(struct rtc_time * tm)
939 {
940 	int leapsToDate;
941 	int lastYear;
942 	int day;
943 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
944 
945 	lastYear = tm->tm_year - 1;
946 
947 	/*
948 	 * Number of leap corrections to apply up to end of last year
949 	 */
950 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
951 
952 	/*
953 	 * This year is a leap year if it is divisible by 4 except when it is
954 	 * divisible by 100 unless it is divisible by 400
955 	 *
956 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
957 	 */
958 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
959 
960 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
961 		   tm->tm_mday;
962 
963 	tm->tm_wday = day % 7;
964 }
965 
966 void to_tm(int tim, struct rtc_time * tm)
967 {
968 	register int    i;
969 	register long   hms, day;
970 
971 	day = tim / SECDAY;
972 	hms = tim % SECDAY;
973 
974 	/* Hours, minutes, seconds are easy */
975 	tm->tm_hour = hms / 3600;
976 	tm->tm_min = (hms % 3600) / 60;
977 	tm->tm_sec = (hms % 3600) % 60;
978 
979 	/* Number of years in days */
980 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
981 		day -= days_in_year(i);
982 	tm->tm_year = i;
983 
984 	/* Number of months in days left */
985 	if (leapyear(tm->tm_year))
986 		days_in_month(FEBRUARY) = 29;
987 	for (i = 1; day >= days_in_month(i); i++)
988 		day -= days_in_month(i);
989 	days_in_month(FEBRUARY) = 28;
990 	tm->tm_mon = i;
991 
992 	/* Days are what is left over (+1) from all that. */
993 	tm->tm_mday = day + 1;
994 
995 	/*
996 	 * Determine the day of week
997 	 */
998 	GregorianDay(tm);
999 }
1000 
1001 /*
1002  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1003  * result.
1004  */
1005 void div128_by_32(u64 dividend_high, u64 dividend_low,
1006 		  unsigned divisor, struct div_result *dr)
1007 {
1008 	unsigned long a, b, c, d;
1009 	unsigned long w, x, y, z;
1010 	u64 ra, rb, rc;
1011 
1012 	a = dividend_high >> 32;
1013 	b = dividend_high & 0xffffffff;
1014 	c = dividend_low >> 32;
1015 	d = dividend_low & 0xffffffff;
1016 
1017 	w = a / divisor;
1018 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1019 
1020 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1021 	x = ra;
1022 
1023 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1024 	y = rb;
1025 
1026 	do_div(rc, divisor);
1027 	z = rc;
1028 
1029 	dr->result_high = ((u64)w << 32) + x;
1030 	dr->result_low  = ((u64)y << 32) + z;
1031 
1032 }
1033 
1034 /* We don't need to calibrate delay, we use the CPU timebase for that */
1035 void calibrate_delay(void)
1036 {
1037 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1038 	 * as the number of __delay(1) in a jiffy, so make it so
1039 	 */
1040 	loops_per_jiffy = tb_ticks_per_jiffy;
1041 }
1042 
1043 static int __init rtc_init(void)
1044 {
1045 	struct platform_device *pdev;
1046 
1047 	if (!ppc_md.get_rtc_time)
1048 		return -ENODEV;
1049 
1050 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1051 	if (IS_ERR(pdev))
1052 		return PTR_ERR(pdev);
1053 
1054 	return 0;
1055 }
1056 
1057 module_init(rtc_init);
1058