xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision ce932d0c5589e9766e089c22c66890dfc48fbd94)
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/clocksource.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 static 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 
112 DEFINE_PER_CPU(u64, decrementers_next_tb);
113 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
114 
115 #define XSEC_PER_SEC (1024*1024)
116 
117 #ifdef CONFIG_PPC64
118 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
119 #else
120 /* compute ((xsec << 12) * max) >> 32 */
121 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
122 #endif
123 
124 unsigned long tb_ticks_per_jiffy;
125 unsigned long tb_ticks_per_usec = 100; /* sane default */
126 EXPORT_SYMBOL(tb_ticks_per_usec);
127 unsigned long tb_ticks_per_sec;
128 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
129 
130 DEFINE_SPINLOCK(rtc_lock);
131 EXPORT_SYMBOL_GPL(rtc_lock);
132 
133 static u64 tb_to_ns_scale __read_mostly;
134 static unsigned tb_to_ns_shift __read_mostly;
135 static u64 boot_tb __read_mostly;
136 
137 extern struct timezone sys_tz;
138 static long timezone_offset;
139 
140 unsigned long ppc_proc_freq;
141 EXPORT_SYMBOL_GPL(ppc_proc_freq);
142 unsigned long ppc_tb_freq;
143 EXPORT_SYMBOL_GPL(ppc_tb_freq);
144 
145 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
146 /*
147  * Factors for converting from cputime_t (timebase ticks) to
148  * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
149  * These are all stored as 0.64 fixed-point binary fractions.
150  */
151 u64 __cputime_jiffies_factor;
152 EXPORT_SYMBOL(__cputime_jiffies_factor);
153 u64 __cputime_usec_factor;
154 EXPORT_SYMBOL(__cputime_usec_factor);
155 u64 __cputime_sec_factor;
156 EXPORT_SYMBOL(__cputime_sec_factor);
157 u64 __cputime_clockt_factor;
158 EXPORT_SYMBOL(__cputime_clockt_factor);
159 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
160 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
161 
162 cputime_t cputime_one_jiffy;
163 
164 void (*dtl_consumer)(struct dtl_entry *, u64);
165 
166 static void calc_cputime_factors(void)
167 {
168 	struct div_result res;
169 
170 	div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
171 	__cputime_jiffies_factor = res.result_low;
172 	div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
173 	__cputime_usec_factor = res.result_low;
174 	div128_by_32(1, 0, tb_ticks_per_sec, &res);
175 	__cputime_sec_factor = res.result_low;
176 	div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
177 	__cputime_clockt_factor = res.result_low;
178 }
179 
180 /*
181  * Read the SPURR on systems that have it, otherwise the PURR,
182  * or if that doesn't exist return the timebase value passed in.
183  */
184 static u64 read_spurr(u64 tb)
185 {
186 	if (cpu_has_feature(CPU_FTR_SPURR))
187 		return mfspr(SPRN_SPURR);
188 	if (cpu_has_feature(CPU_FTR_PURR))
189 		return mfspr(SPRN_PURR);
190 	return tb;
191 }
192 
193 #ifdef CONFIG_PPC_SPLPAR
194 
195 /*
196  * Scan the dispatch trace log and count up the stolen time.
197  * Should be called with interrupts disabled.
198  */
199 static u64 scan_dispatch_log(u64 stop_tb)
200 {
201 	u64 i = local_paca->dtl_ridx;
202 	struct dtl_entry *dtl = local_paca->dtl_curr;
203 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
204 	struct lppaca *vpa = local_paca->lppaca_ptr;
205 	u64 tb_delta;
206 	u64 stolen = 0;
207 	u64 dtb;
208 
209 	if (!dtl)
210 		return 0;
211 
212 	if (i == vpa->dtl_idx)
213 		return 0;
214 	while (i < vpa->dtl_idx) {
215 		if (dtl_consumer)
216 			dtl_consumer(dtl, i);
217 		dtb = dtl->timebase;
218 		tb_delta = dtl->enqueue_to_dispatch_time +
219 			dtl->ready_to_enqueue_time;
220 		barrier();
221 		if (i + N_DISPATCH_LOG < vpa->dtl_idx) {
222 			/* buffer has overflowed */
223 			i = vpa->dtl_idx - N_DISPATCH_LOG;
224 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
225 			continue;
226 		}
227 		if (dtb > stop_tb)
228 			break;
229 		stolen += tb_delta;
230 		++i;
231 		++dtl;
232 		if (dtl == dtl_end)
233 			dtl = local_paca->dispatch_log;
234 	}
235 	local_paca->dtl_ridx = i;
236 	local_paca->dtl_curr = dtl;
237 	return stolen;
238 }
239 
240 /*
241  * Accumulate stolen time by scanning the dispatch trace log.
242  * Called on entry from user mode.
243  */
244 void accumulate_stolen_time(void)
245 {
246 	u64 sst, ust;
247 
248 	u8 save_soft_enabled = local_paca->soft_enabled;
249 
250 	/* We are called early in the exception entry, before
251 	 * soft/hard_enabled are sync'ed to the expected state
252 	 * for the exception. We are hard disabled but the PACA
253 	 * needs to reflect that so various debug stuff doesn't
254 	 * complain
255 	 */
256 	local_paca->soft_enabled = 0;
257 
258 	sst = scan_dispatch_log(local_paca->starttime_user);
259 	ust = scan_dispatch_log(local_paca->starttime);
260 	local_paca->system_time -= sst;
261 	local_paca->user_time -= ust;
262 	local_paca->stolen_time += ust + sst;
263 
264 	local_paca->soft_enabled = save_soft_enabled;
265 }
266 
267 static inline u64 calculate_stolen_time(u64 stop_tb)
268 {
269 	u64 stolen = 0;
270 
271 	if (get_paca()->dtl_ridx != get_paca()->lppaca_ptr->dtl_idx) {
272 		stolen = scan_dispatch_log(stop_tb);
273 		get_paca()->system_time -= stolen;
274 	}
275 
276 	stolen += get_paca()->stolen_time;
277 	get_paca()->stolen_time = 0;
278 	return stolen;
279 }
280 
281 #else /* CONFIG_PPC_SPLPAR */
282 static inline u64 calculate_stolen_time(u64 stop_tb)
283 {
284 	return 0;
285 }
286 
287 #endif /* CONFIG_PPC_SPLPAR */
288 
289 /*
290  * Account time for a transition between system, hard irq
291  * or soft irq state.
292  */
293 void account_system_vtime(struct task_struct *tsk)
294 {
295 	u64 now, nowscaled, delta, deltascaled;
296 	unsigned long flags;
297 	u64 stolen, udelta, sys_scaled, user_scaled;
298 
299 	local_irq_save(flags);
300 	now = mftb();
301 	nowscaled = read_spurr(now);
302 	get_paca()->system_time += now - get_paca()->starttime;
303 	get_paca()->starttime = now;
304 	deltascaled = nowscaled - get_paca()->startspurr;
305 	get_paca()->startspurr = nowscaled;
306 
307 	stolen = calculate_stolen_time(now);
308 
309 	delta = get_paca()->system_time;
310 	get_paca()->system_time = 0;
311 	udelta = get_paca()->user_time - get_paca()->utime_sspurr;
312 	get_paca()->utime_sspurr = get_paca()->user_time;
313 
314 	/*
315 	 * Because we don't read the SPURR on every kernel entry/exit,
316 	 * deltascaled includes both user and system SPURR ticks.
317 	 * Apportion these ticks to system SPURR ticks and user
318 	 * SPURR ticks in the same ratio as the system time (delta)
319 	 * and user time (udelta) values obtained from the timebase
320 	 * over the same interval.  The system ticks get accounted here;
321 	 * the user ticks get saved up in paca->user_time_scaled to be
322 	 * used by account_process_tick.
323 	 */
324 	sys_scaled = delta;
325 	user_scaled = udelta;
326 	if (deltascaled != delta + udelta) {
327 		if (udelta) {
328 			sys_scaled = deltascaled * delta / (delta + udelta);
329 			user_scaled = deltascaled - sys_scaled;
330 		} else {
331 			sys_scaled = deltascaled;
332 		}
333 	}
334 	get_paca()->user_time_scaled += user_scaled;
335 
336 	if (in_interrupt() || idle_task(smp_processor_id()) != tsk) {
337 		account_system_time(tsk, 0, delta, sys_scaled);
338 		if (stolen)
339 			account_steal_time(stolen);
340 	} else {
341 		account_idle_time(delta + stolen);
342 	}
343 	local_irq_restore(flags);
344 }
345 EXPORT_SYMBOL_GPL(account_system_vtime);
346 
347 /*
348  * Transfer the user and system times accumulated in the paca
349  * by the exception entry and exit code to the generic process
350  * user and system time records.
351  * Must be called with interrupts disabled.
352  * Assumes that account_system_vtime() has been called recently
353  * (i.e. since the last entry from usermode) so that
354  * get_paca()->user_time_scaled is up to date.
355  */
356 void account_process_tick(struct task_struct *tsk, int user_tick)
357 {
358 	cputime_t utime, utimescaled;
359 
360 	utime = get_paca()->user_time;
361 	utimescaled = get_paca()->user_time_scaled;
362 	get_paca()->user_time = 0;
363 	get_paca()->user_time_scaled = 0;
364 	get_paca()->utime_sspurr = 0;
365 	account_user_time(tsk, utime, utimescaled);
366 }
367 
368 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
369 #define calc_cputime_factors()
370 #endif
371 
372 void __delay(unsigned long loops)
373 {
374 	unsigned long start;
375 	int diff;
376 
377 	if (__USE_RTC()) {
378 		start = get_rtcl();
379 		do {
380 			/* the RTCL register wraps at 1000000000 */
381 			diff = get_rtcl() - start;
382 			if (diff < 0)
383 				diff += 1000000000;
384 		} while (diff < loops);
385 	} else {
386 		start = get_tbl();
387 		while (get_tbl() - start < loops)
388 			HMT_low();
389 		HMT_medium();
390 	}
391 }
392 EXPORT_SYMBOL(__delay);
393 
394 void udelay(unsigned long usecs)
395 {
396 	__delay(tb_ticks_per_usec * usecs);
397 }
398 EXPORT_SYMBOL(udelay);
399 
400 #ifdef CONFIG_SMP
401 unsigned long profile_pc(struct pt_regs *regs)
402 {
403 	unsigned long pc = instruction_pointer(regs);
404 
405 	if (in_lock_functions(pc))
406 		return regs->link;
407 
408 	return pc;
409 }
410 EXPORT_SYMBOL(profile_pc);
411 #endif
412 
413 #ifdef CONFIG_IRQ_WORK
414 
415 /*
416  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
417  */
418 #ifdef CONFIG_PPC64
419 static inline unsigned long test_irq_work_pending(void)
420 {
421 	unsigned long x;
422 
423 	asm volatile("lbz %0,%1(13)"
424 		: "=r" (x)
425 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
426 	return x;
427 }
428 
429 static inline void set_irq_work_pending_flag(void)
430 {
431 	asm volatile("stb %0,%1(13)" : :
432 		"r" (1),
433 		"i" (offsetof(struct paca_struct, irq_work_pending)));
434 }
435 
436 static inline void clear_irq_work_pending(void)
437 {
438 	asm volatile("stb %0,%1(13)" : :
439 		"r" (0),
440 		"i" (offsetof(struct paca_struct, irq_work_pending)));
441 }
442 
443 #else /* 32-bit */
444 
445 DEFINE_PER_CPU(u8, irq_work_pending);
446 
447 #define set_irq_work_pending_flag()	__get_cpu_var(irq_work_pending) = 1
448 #define test_irq_work_pending()		__get_cpu_var(irq_work_pending)
449 #define clear_irq_work_pending()	__get_cpu_var(irq_work_pending) = 0
450 
451 #endif /* 32 vs 64 bit */
452 
453 void arch_irq_work_raise(void)
454 {
455 	preempt_disable();
456 	set_irq_work_pending_flag();
457 	set_dec(1);
458 	preempt_enable();
459 }
460 
461 #else  /* CONFIG_IRQ_WORK */
462 
463 #define test_irq_work_pending()	0
464 #define clear_irq_work_pending()
465 
466 #endif /* CONFIG_IRQ_WORK */
467 
468 /*
469  * timer_interrupt - gets called when the decrementer overflows,
470  * with interrupts disabled.
471  */
472 void timer_interrupt(struct pt_regs * regs)
473 {
474 	struct pt_regs *old_regs;
475 	u64 *next_tb = &__get_cpu_var(decrementers_next_tb);
476 	struct clock_event_device *evt = &__get_cpu_var(decrementers);
477 
478 	/* Ensure a positive value is written to the decrementer, or else
479 	 * some CPUs will continue to take decrementer exceptions.
480 	 */
481 	set_dec(DECREMENTER_MAX);
482 
483 	/* Some implementations of hotplug will get timer interrupts while
484 	 * offline, just ignore these
485 	 */
486 	if (!cpu_online(smp_processor_id()))
487 		return;
488 
489 	/* Conditionally hard-enable interrupts now that the DEC has been
490 	 * bumped to its maximum value
491 	 */
492 	may_hard_irq_enable();
493 
494 	trace_timer_interrupt_entry(regs);
495 
496 	__get_cpu_var(irq_stat).timer_irqs++;
497 
498 #if defined(CONFIG_PPC32) && defined(CONFIG_PMAC)
499 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
500 		do_IRQ(regs);
501 #endif
502 
503 	old_regs = set_irq_regs(regs);
504 	irq_enter();
505 
506 	if (test_irq_work_pending()) {
507 		clear_irq_work_pending();
508 		irq_work_run();
509 	}
510 
511 	*next_tb = ~(u64)0;
512 	if (evt->event_handler)
513 		evt->event_handler(evt);
514 
515 #ifdef CONFIG_PPC64
516 	/* collect purr register values often, for accurate calculations */
517 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
518 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
519 		cu->current_tb = mfspr(SPRN_PURR);
520 	}
521 #endif
522 
523 	irq_exit();
524 	set_irq_regs(old_regs);
525 
526 	trace_timer_interrupt_exit(regs);
527 }
528 
529 #ifdef CONFIG_SUSPEND
530 static void generic_suspend_disable_irqs(void)
531 {
532 	/* Disable the decrementer, so that it doesn't interfere
533 	 * with suspending.
534 	 */
535 
536 	set_dec(DECREMENTER_MAX);
537 	local_irq_disable();
538 	set_dec(DECREMENTER_MAX);
539 }
540 
541 static void generic_suspend_enable_irqs(void)
542 {
543 	local_irq_enable();
544 }
545 
546 /* Overrides the weak version in kernel/power/main.c */
547 void arch_suspend_disable_irqs(void)
548 {
549 	if (ppc_md.suspend_disable_irqs)
550 		ppc_md.suspend_disable_irqs();
551 	generic_suspend_disable_irqs();
552 }
553 
554 /* Overrides the weak version in kernel/power/main.c */
555 void arch_suspend_enable_irqs(void)
556 {
557 	generic_suspend_enable_irqs();
558 	if (ppc_md.suspend_enable_irqs)
559 		ppc_md.suspend_enable_irqs();
560 }
561 #endif
562 
563 /*
564  * Scheduler clock - returns current time in nanosec units.
565  *
566  * Note: mulhdu(a, b) (multiply high double unsigned) returns
567  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
568  * are 64-bit unsigned numbers.
569  */
570 unsigned long long sched_clock(void)
571 {
572 	if (__USE_RTC())
573 		return get_rtc();
574 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
575 }
576 
577 static int __init get_freq(char *name, int cells, unsigned long *val)
578 {
579 	struct device_node *cpu;
580 	const unsigned int *fp;
581 	int found = 0;
582 
583 	/* The cpu node should have timebase and clock frequency properties */
584 	cpu = of_find_node_by_type(NULL, "cpu");
585 
586 	if (cpu) {
587 		fp = of_get_property(cpu, name, NULL);
588 		if (fp) {
589 			found = 1;
590 			*val = of_read_ulong(fp, cells);
591 		}
592 
593 		of_node_put(cpu);
594 	}
595 
596 	return found;
597 }
598 
599 /* should become __cpuinit when secondary_cpu_time_init also is */
600 void start_cpu_decrementer(void)
601 {
602 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
603 	/* Clear any pending timer interrupts */
604 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
605 
606 	/* Enable decrementer interrupt */
607 	mtspr(SPRN_TCR, TCR_DIE);
608 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
609 }
610 
611 void __init generic_calibrate_decr(void)
612 {
613 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
614 
615 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
616 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
617 
618 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
619 				"(not found)\n");
620 	}
621 
622 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
623 
624 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
625 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
626 
627 		printk(KERN_ERR "WARNING: Estimating processor frequency "
628 				"(not found)\n");
629 	}
630 }
631 
632 int update_persistent_clock(struct timespec now)
633 {
634 	struct rtc_time tm;
635 
636 	if (!ppc_md.set_rtc_time)
637 		return 0;
638 
639 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
640 	tm.tm_year -= 1900;
641 	tm.tm_mon -= 1;
642 
643 	return ppc_md.set_rtc_time(&tm);
644 }
645 
646 static void __read_persistent_clock(struct timespec *ts)
647 {
648 	struct rtc_time tm;
649 	static int first = 1;
650 
651 	ts->tv_nsec = 0;
652 	/* XXX this is a litle fragile but will work okay in the short term */
653 	if (first) {
654 		first = 0;
655 		if (ppc_md.time_init)
656 			timezone_offset = ppc_md.time_init();
657 
658 		/* get_boot_time() isn't guaranteed to be safe to call late */
659 		if (ppc_md.get_boot_time) {
660 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
661 			return;
662 		}
663 	}
664 	if (!ppc_md.get_rtc_time) {
665 		ts->tv_sec = 0;
666 		return;
667 	}
668 	ppc_md.get_rtc_time(&tm);
669 
670 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
671 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
672 }
673 
674 void read_persistent_clock(struct timespec *ts)
675 {
676 	__read_persistent_clock(ts);
677 
678 	/* Sanitize it in case real time clock is set below EPOCH */
679 	if (ts->tv_sec < 0) {
680 		ts->tv_sec = 0;
681 		ts->tv_nsec = 0;
682 	}
683 
684 }
685 
686 /* clocksource code */
687 static cycle_t rtc_read(struct clocksource *cs)
688 {
689 	return (cycle_t)get_rtc();
690 }
691 
692 static cycle_t timebase_read(struct clocksource *cs)
693 {
694 	return (cycle_t)get_tb();
695 }
696 
697 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
698 			struct clocksource *clock, u32 mult)
699 {
700 	u64 new_tb_to_xs, new_stamp_xsec;
701 	u32 frac_sec;
702 
703 	if (clock != &clocksource_timebase)
704 		return;
705 
706 	/* Make userspace gettimeofday spin until we're done. */
707 	++vdso_data->tb_update_count;
708 	smp_mb();
709 
710 	/* 19342813113834067 ~= 2^(20+64) / 1e9 */
711 	new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
712 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
713 	do_div(new_stamp_xsec, 1000000000);
714 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
715 
716 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
717 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
718 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
719 
720 	/*
721 	 * tb_update_count is used to allow the userspace gettimeofday code
722 	 * to assure itself that it sees a consistent view of the tb_to_xs and
723 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
724 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
725 	 * the two values of tb_update_count match and are even then the
726 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
727 	 * loops back and reads them again until this criteria is met.
728 	 * We expect the caller to have done the first increment of
729 	 * vdso_data->tb_update_count already.
730 	 */
731 	vdso_data->tb_orig_stamp = clock->cycle_last;
732 	vdso_data->stamp_xsec = new_stamp_xsec;
733 	vdso_data->tb_to_xs = new_tb_to_xs;
734 	vdso_data->wtom_clock_sec = wtm->tv_sec;
735 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
736 	vdso_data->stamp_xtime = *wall_time;
737 	vdso_data->stamp_sec_fraction = frac_sec;
738 	smp_wmb();
739 	++(vdso_data->tb_update_count);
740 }
741 
742 void update_vsyscall_tz(void)
743 {
744 	/* Make userspace gettimeofday spin until we're done. */
745 	++vdso_data->tb_update_count;
746 	smp_mb();
747 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
748 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
749 	smp_mb();
750 	++vdso_data->tb_update_count;
751 }
752 
753 static void __init clocksource_init(void)
754 {
755 	struct clocksource *clock;
756 
757 	if (__USE_RTC())
758 		clock = &clocksource_rtc;
759 	else
760 		clock = &clocksource_timebase;
761 
762 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
763 		printk(KERN_ERR "clocksource: %s is already registered\n",
764 		       clock->name);
765 		return;
766 	}
767 
768 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
769 	       clock->name, clock->mult, clock->shift);
770 }
771 
772 static int decrementer_set_next_event(unsigned long evt,
773 				      struct clock_event_device *dev)
774 {
775 	__get_cpu_var(decrementers_next_tb) = get_tb_or_rtc() + evt;
776 	set_dec(evt);
777 	return 0;
778 }
779 
780 static void decrementer_set_mode(enum clock_event_mode mode,
781 				 struct clock_event_device *dev)
782 {
783 	if (mode != CLOCK_EVT_MODE_ONESHOT)
784 		decrementer_set_next_event(DECREMENTER_MAX, dev);
785 }
786 
787 static void register_decrementer_clockevent(int cpu)
788 {
789 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
790 
791 	*dec = decrementer_clockevent;
792 	dec->cpumask = cpumask_of(cpu);
793 
794 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
795 		    dec->name, dec->mult, dec->shift, cpu);
796 
797 	clockevents_register_device(dec);
798 }
799 
800 static void __init init_decrementer_clockevent(void)
801 {
802 	int cpu = smp_processor_id();
803 
804 	clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
805 
806 	decrementer_clockevent.max_delta_ns =
807 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
808 	decrementer_clockevent.min_delta_ns =
809 		clockevent_delta2ns(2, &decrementer_clockevent);
810 
811 	register_decrementer_clockevent(cpu);
812 }
813 
814 void secondary_cpu_time_init(void)
815 {
816 	/* Start the decrementer on CPUs that have manual control
817 	 * such as BookE
818 	 */
819 	start_cpu_decrementer();
820 
821 	/* FIME: Should make unrelatred change to move snapshot_timebase
822 	 * call here ! */
823 	register_decrementer_clockevent(smp_processor_id());
824 }
825 
826 /* This function is only called on the boot processor */
827 void __init time_init(void)
828 {
829 	struct div_result res;
830 	u64 scale;
831 	unsigned shift;
832 
833 	if (__USE_RTC()) {
834 		/* 601 processor: dec counts down by 128 every 128ns */
835 		ppc_tb_freq = 1000000000;
836 	} else {
837 		/* Normal PowerPC with timebase register */
838 		ppc_md.calibrate_decr();
839 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
840 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
841 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
842 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
843 	}
844 
845 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
846 	tb_ticks_per_sec = ppc_tb_freq;
847 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
848 	calc_cputime_factors();
849 	setup_cputime_one_jiffy();
850 
851 	/*
852 	 * Compute scale factor for sched_clock.
853 	 * The calibrate_decr() function has set tb_ticks_per_sec,
854 	 * which is the timebase frequency.
855 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
856 	 * the 128-bit result as a 64.64 fixed-point number.
857 	 * We then shift that number right until it is less than 1.0,
858 	 * giving us the scale factor and shift count to use in
859 	 * sched_clock().
860 	 */
861 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
862 	scale = res.result_low;
863 	for (shift = 0; res.result_high != 0; ++shift) {
864 		scale = (scale >> 1) | (res.result_high << 63);
865 		res.result_high >>= 1;
866 	}
867 	tb_to_ns_scale = scale;
868 	tb_to_ns_shift = shift;
869 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
870 	boot_tb = get_tb_or_rtc();
871 
872 	/* If platform provided a timezone (pmac), we correct the time */
873 	if (timezone_offset) {
874 		sys_tz.tz_minuteswest = -timezone_offset / 60;
875 		sys_tz.tz_dsttime = 0;
876 	}
877 
878 	vdso_data->tb_update_count = 0;
879 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
880 
881 	/* Start the decrementer on CPUs that have manual control
882 	 * such as BookE
883 	 */
884 	start_cpu_decrementer();
885 
886 	/* Register the clocksource */
887 	clocksource_init();
888 
889 	init_decrementer_clockevent();
890 }
891 
892 
893 #define FEBRUARY	2
894 #define	STARTOFTIME	1970
895 #define SECDAY		86400L
896 #define SECYR		(SECDAY * 365)
897 #define	leapyear(year)		((year) % 4 == 0 && \
898 				 ((year) % 100 != 0 || (year) % 400 == 0))
899 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
900 #define	days_in_month(a) 	(month_days[(a) - 1])
901 
902 static int month_days[12] = {
903 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
904 };
905 
906 /*
907  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
908  */
909 void GregorianDay(struct rtc_time * tm)
910 {
911 	int leapsToDate;
912 	int lastYear;
913 	int day;
914 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
915 
916 	lastYear = tm->tm_year - 1;
917 
918 	/*
919 	 * Number of leap corrections to apply up to end of last year
920 	 */
921 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
922 
923 	/*
924 	 * This year is a leap year if it is divisible by 4 except when it is
925 	 * divisible by 100 unless it is divisible by 400
926 	 *
927 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
928 	 */
929 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
930 
931 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
932 		   tm->tm_mday;
933 
934 	tm->tm_wday = day % 7;
935 }
936 
937 void to_tm(int tim, struct rtc_time * tm)
938 {
939 	register int    i;
940 	register long   hms, day;
941 
942 	day = tim / SECDAY;
943 	hms = tim % SECDAY;
944 
945 	/* Hours, minutes, seconds are easy */
946 	tm->tm_hour = hms / 3600;
947 	tm->tm_min = (hms % 3600) / 60;
948 	tm->tm_sec = (hms % 3600) % 60;
949 
950 	/* Number of years in days */
951 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
952 		day -= days_in_year(i);
953 	tm->tm_year = i;
954 
955 	/* Number of months in days left */
956 	if (leapyear(tm->tm_year))
957 		days_in_month(FEBRUARY) = 29;
958 	for (i = 1; day >= days_in_month(i); i++)
959 		day -= days_in_month(i);
960 	days_in_month(FEBRUARY) = 28;
961 	tm->tm_mon = i;
962 
963 	/* Days are what is left over (+1) from all that. */
964 	tm->tm_mday = day + 1;
965 
966 	/*
967 	 * Determine the day of week
968 	 */
969 	GregorianDay(tm);
970 }
971 
972 /*
973  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
974  * result.
975  */
976 void div128_by_32(u64 dividend_high, u64 dividend_low,
977 		  unsigned divisor, struct div_result *dr)
978 {
979 	unsigned long a, b, c, d;
980 	unsigned long w, x, y, z;
981 	u64 ra, rb, rc;
982 
983 	a = dividend_high >> 32;
984 	b = dividend_high & 0xffffffff;
985 	c = dividend_low >> 32;
986 	d = dividend_low & 0xffffffff;
987 
988 	w = a / divisor;
989 	ra = ((u64)(a - (w * divisor)) << 32) + b;
990 
991 	rb = ((u64) do_div(ra, divisor) << 32) + c;
992 	x = ra;
993 
994 	rc = ((u64) do_div(rb, divisor) << 32) + d;
995 	y = rb;
996 
997 	do_div(rc, divisor);
998 	z = rc;
999 
1000 	dr->result_high = ((u64)w << 32) + x;
1001 	dr->result_low  = ((u64)y << 32) + z;
1002 
1003 }
1004 
1005 /* We don't need to calibrate delay, we use the CPU timebase for that */
1006 void calibrate_delay(void)
1007 {
1008 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1009 	 * as the number of __delay(1) in a jiffy, so make it so
1010 	 */
1011 	loops_per_jiffy = tb_ticks_per_jiffy;
1012 }
1013 
1014 static int __init rtc_init(void)
1015 {
1016 	struct platform_device *pdev;
1017 
1018 	if (!ppc_md.get_rtc_time)
1019 		return -ENODEV;
1020 
1021 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1022 	if (IS_ERR(pdev))
1023 		return PTR_ERR(pdev);
1024 
1025 	return 0;
1026 }
1027 
1028 module_init(rtc_init);
1029