xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision d2999e1b)
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/clockchips.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/irq_work.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 
74 /* powerpc clocksource/clockevent code */
75 
76 #include <linux/clockchips.h>
77 #include <linux/timekeeper_internal.h>
78 
79 static cycle_t rtc_read(struct clocksource *);
80 static struct clocksource clocksource_rtc = {
81 	.name         = "rtc",
82 	.rating       = 400,
83 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
84 	.mask         = CLOCKSOURCE_MASK(64),
85 	.read         = rtc_read,
86 };
87 
88 static cycle_t timebase_read(struct clocksource *);
89 static struct clocksource clocksource_timebase = {
90 	.name         = "timebase",
91 	.rating       = 400,
92 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
93 	.mask         = CLOCKSOURCE_MASK(64),
94 	.read         = timebase_read,
95 };
96 
97 #define DECREMENTER_MAX	0x7fffffff
98 
99 static int decrementer_set_next_event(unsigned long evt,
100 				      struct clock_event_device *dev);
101 static void decrementer_set_mode(enum clock_event_mode mode,
102 				 struct clock_event_device *dev);
103 
104 struct clock_event_device decrementer_clockevent = {
105 	.name           = "decrementer",
106 	.rating         = 200,
107 	.irq            = 0,
108 	.set_next_event = decrementer_set_next_event,
109 	.set_mode       = decrementer_set_mode,
110 	.features       = CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_C3STOP,
111 };
112 EXPORT_SYMBOL(decrementer_clockevent);
113 
114 DEFINE_PER_CPU(u64, decrementers_next_tb);
115 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
116 
117 #define XSEC_PER_SEC (1024*1024)
118 
119 #ifdef CONFIG_PPC64
120 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
121 #else
122 /* compute ((xsec << 12) * max) >> 32 */
123 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
124 #endif
125 
126 unsigned long tb_ticks_per_jiffy;
127 unsigned long tb_ticks_per_usec = 100; /* sane default */
128 EXPORT_SYMBOL(tb_ticks_per_usec);
129 unsigned long tb_ticks_per_sec;
130 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
131 
132 DEFINE_SPINLOCK(rtc_lock);
133 EXPORT_SYMBOL_GPL(rtc_lock);
134 
135 static u64 tb_to_ns_scale __read_mostly;
136 static unsigned tb_to_ns_shift __read_mostly;
137 static u64 boot_tb __read_mostly;
138 
139 extern struct timezone sys_tz;
140 static long timezone_offset;
141 
142 unsigned long ppc_proc_freq;
143 EXPORT_SYMBOL_GPL(ppc_proc_freq);
144 unsigned long ppc_tb_freq;
145 EXPORT_SYMBOL_GPL(ppc_tb_freq);
146 
147 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
148 /*
149  * Factors for converting from cputime_t (timebase ticks) to
150  * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
151  * These are all stored as 0.64 fixed-point binary fractions.
152  */
153 u64 __cputime_jiffies_factor;
154 EXPORT_SYMBOL(__cputime_jiffies_factor);
155 u64 __cputime_usec_factor;
156 EXPORT_SYMBOL(__cputime_usec_factor);
157 u64 __cputime_sec_factor;
158 EXPORT_SYMBOL(__cputime_sec_factor);
159 u64 __cputime_clockt_factor;
160 EXPORT_SYMBOL(__cputime_clockt_factor);
161 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
162 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
163 
164 cputime_t cputime_one_jiffy;
165 
166 void (*dtl_consumer)(struct dtl_entry *, u64);
167 
168 static void calc_cputime_factors(void)
169 {
170 	struct div_result res;
171 
172 	div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
173 	__cputime_jiffies_factor = res.result_low;
174 	div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
175 	__cputime_usec_factor = res.result_low;
176 	div128_by_32(1, 0, tb_ticks_per_sec, &res);
177 	__cputime_sec_factor = res.result_low;
178 	div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
179 	__cputime_clockt_factor = res.result_low;
180 }
181 
182 /*
183  * Read the SPURR on systems that have it, otherwise the PURR,
184  * or if that doesn't exist return the timebase value passed in.
185  */
186 static u64 read_spurr(u64 tb)
187 {
188 	if (cpu_has_feature(CPU_FTR_SPURR))
189 		return mfspr(SPRN_SPURR);
190 	if (cpu_has_feature(CPU_FTR_PURR))
191 		return mfspr(SPRN_PURR);
192 	return tb;
193 }
194 
195 #ifdef CONFIG_PPC_SPLPAR
196 
197 /*
198  * Scan the dispatch trace log and count up the stolen time.
199  * Should be called with interrupts disabled.
200  */
201 static u64 scan_dispatch_log(u64 stop_tb)
202 {
203 	u64 i = local_paca->dtl_ridx;
204 	struct dtl_entry *dtl = local_paca->dtl_curr;
205 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
206 	struct lppaca *vpa = local_paca->lppaca_ptr;
207 	u64 tb_delta;
208 	u64 stolen = 0;
209 	u64 dtb;
210 
211 	if (!dtl)
212 		return 0;
213 
214 	if (i == be64_to_cpu(vpa->dtl_idx))
215 		return 0;
216 	while (i < be64_to_cpu(vpa->dtl_idx)) {
217 		dtb = be64_to_cpu(dtl->timebase);
218 		tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
219 			be32_to_cpu(dtl->ready_to_enqueue_time);
220 		barrier();
221 		if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
222 			/* buffer has overflowed */
223 			i = be64_to_cpu(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 		if (dtl_consumer)
230 			dtl_consumer(dtl, i);
231 		stolen += tb_delta;
232 		++i;
233 		++dtl;
234 		if (dtl == dtl_end)
235 			dtl = local_paca->dispatch_log;
236 	}
237 	local_paca->dtl_ridx = i;
238 	local_paca->dtl_curr = dtl;
239 	return stolen;
240 }
241 
242 /*
243  * Accumulate stolen time by scanning the dispatch trace log.
244  * Called on entry from user mode.
245  */
246 void accumulate_stolen_time(void)
247 {
248 	u64 sst, ust;
249 
250 	u8 save_soft_enabled = local_paca->soft_enabled;
251 
252 	/* We are called early in the exception entry, before
253 	 * soft/hard_enabled are sync'ed to the expected state
254 	 * for the exception. We are hard disabled but the PACA
255 	 * needs to reflect that so various debug stuff doesn't
256 	 * complain
257 	 */
258 	local_paca->soft_enabled = 0;
259 
260 	sst = scan_dispatch_log(local_paca->starttime_user);
261 	ust = scan_dispatch_log(local_paca->starttime);
262 	local_paca->system_time -= sst;
263 	local_paca->user_time -= ust;
264 	local_paca->stolen_time += ust + sst;
265 
266 	local_paca->soft_enabled = save_soft_enabled;
267 }
268 
269 static inline u64 calculate_stolen_time(u64 stop_tb)
270 {
271 	u64 stolen = 0;
272 
273 	if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
274 		stolen = scan_dispatch_log(stop_tb);
275 		get_paca()->system_time -= stolen;
276 	}
277 
278 	stolen += get_paca()->stolen_time;
279 	get_paca()->stolen_time = 0;
280 	return stolen;
281 }
282 
283 #else /* CONFIG_PPC_SPLPAR */
284 static inline u64 calculate_stolen_time(u64 stop_tb)
285 {
286 	return 0;
287 }
288 
289 #endif /* CONFIG_PPC_SPLPAR */
290 
291 /*
292  * Account time for a transition between system, hard irq
293  * or soft irq state.
294  */
295 static u64 vtime_delta(struct task_struct *tsk,
296 			u64 *sys_scaled, u64 *stolen)
297 {
298 	u64 now, nowscaled, deltascaled;
299 	u64 udelta, delta, user_scaled;
300 
301 	WARN_ON_ONCE(!irqs_disabled());
302 
303 	now = mftb();
304 	nowscaled = read_spurr(now);
305 	get_paca()->system_time += now - get_paca()->starttime;
306 	get_paca()->starttime = now;
307 	deltascaled = nowscaled - get_paca()->startspurr;
308 	get_paca()->startspurr = nowscaled;
309 
310 	*stolen = calculate_stolen_time(now);
311 
312 	delta = get_paca()->system_time;
313 	get_paca()->system_time = 0;
314 	udelta = get_paca()->user_time - get_paca()->utime_sspurr;
315 	get_paca()->utime_sspurr = get_paca()->user_time;
316 
317 	/*
318 	 * Because we don't read the SPURR on every kernel entry/exit,
319 	 * deltascaled includes both user and system SPURR ticks.
320 	 * Apportion these ticks to system SPURR ticks and user
321 	 * SPURR ticks in the same ratio as the system time (delta)
322 	 * and user time (udelta) values obtained from the timebase
323 	 * over the same interval.  The system ticks get accounted here;
324 	 * the user ticks get saved up in paca->user_time_scaled to be
325 	 * used by account_process_tick.
326 	 */
327 	*sys_scaled = delta;
328 	user_scaled = udelta;
329 	if (deltascaled != delta + udelta) {
330 		if (udelta) {
331 			*sys_scaled = deltascaled * delta / (delta + udelta);
332 			user_scaled = deltascaled - *sys_scaled;
333 		} else {
334 			*sys_scaled = deltascaled;
335 		}
336 	}
337 	get_paca()->user_time_scaled += user_scaled;
338 
339 	return delta;
340 }
341 
342 void vtime_account_system(struct task_struct *tsk)
343 {
344 	u64 delta, sys_scaled, stolen;
345 
346 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
347 	account_system_time(tsk, 0, delta, sys_scaled);
348 	if (stolen)
349 		account_steal_time(stolen);
350 }
351 EXPORT_SYMBOL_GPL(vtime_account_system);
352 
353 void vtime_account_idle(struct task_struct *tsk)
354 {
355 	u64 delta, sys_scaled, stolen;
356 
357 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
358 	account_idle_time(delta + stolen);
359 }
360 
361 /*
362  * Transfer the user time accumulated in the paca
363  * by the exception entry and exit code to the generic
364  * process user time records.
365  * Must be called with interrupts disabled.
366  * Assumes that vtime_account_system/idle() has been called
367  * recently (i.e. since the last entry from usermode) so that
368  * get_paca()->user_time_scaled is up to date.
369  */
370 void vtime_account_user(struct task_struct *tsk)
371 {
372 	cputime_t utime, utimescaled;
373 
374 	utime = get_paca()->user_time;
375 	utimescaled = get_paca()->user_time_scaled;
376 	get_paca()->user_time = 0;
377 	get_paca()->user_time_scaled = 0;
378 	get_paca()->utime_sspurr = 0;
379 	account_user_time(tsk, utime, utimescaled);
380 }
381 
382 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
383 #define calc_cputime_factors()
384 #endif
385 
386 void __delay(unsigned long loops)
387 {
388 	unsigned long start;
389 	int diff;
390 
391 	if (__USE_RTC()) {
392 		start = get_rtcl();
393 		do {
394 			/* the RTCL register wraps at 1000000000 */
395 			diff = get_rtcl() - start;
396 			if (diff < 0)
397 				diff += 1000000000;
398 		} while (diff < loops);
399 	} else {
400 		start = get_tbl();
401 		while (get_tbl() - start < loops)
402 			HMT_low();
403 		HMT_medium();
404 	}
405 }
406 EXPORT_SYMBOL(__delay);
407 
408 void udelay(unsigned long usecs)
409 {
410 	__delay(tb_ticks_per_usec * usecs);
411 }
412 EXPORT_SYMBOL(udelay);
413 
414 #ifdef CONFIG_SMP
415 unsigned long profile_pc(struct pt_regs *regs)
416 {
417 	unsigned long pc = instruction_pointer(regs);
418 
419 	if (in_lock_functions(pc))
420 		return regs->link;
421 
422 	return pc;
423 }
424 EXPORT_SYMBOL(profile_pc);
425 #endif
426 
427 #ifdef CONFIG_IRQ_WORK
428 
429 /*
430  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
431  */
432 #ifdef CONFIG_PPC64
433 static inline unsigned long test_irq_work_pending(void)
434 {
435 	unsigned long x;
436 
437 	asm volatile("lbz %0,%1(13)"
438 		: "=r" (x)
439 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
440 	return x;
441 }
442 
443 static inline void set_irq_work_pending_flag(void)
444 {
445 	asm volatile("stb %0,%1(13)" : :
446 		"r" (1),
447 		"i" (offsetof(struct paca_struct, irq_work_pending)));
448 }
449 
450 static inline void clear_irq_work_pending(void)
451 {
452 	asm volatile("stb %0,%1(13)" : :
453 		"r" (0),
454 		"i" (offsetof(struct paca_struct, irq_work_pending)));
455 }
456 
457 #else /* 32-bit */
458 
459 DEFINE_PER_CPU(u8, irq_work_pending);
460 
461 #define set_irq_work_pending_flag()	__get_cpu_var(irq_work_pending) = 1
462 #define test_irq_work_pending()		__get_cpu_var(irq_work_pending)
463 #define clear_irq_work_pending()	__get_cpu_var(irq_work_pending) = 0
464 
465 #endif /* 32 vs 64 bit */
466 
467 void arch_irq_work_raise(void)
468 {
469 	preempt_disable();
470 	set_irq_work_pending_flag();
471 	set_dec(1);
472 	preempt_enable();
473 }
474 
475 #else  /* CONFIG_IRQ_WORK */
476 
477 #define test_irq_work_pending()	0
478 #define clear_irq_work_pending()
479 
480 #endif /* CONFIG_IRQ_WORK */
481 
482 void __timer_interrupt(void)
483 {
484 	struct pt_regs *regs = get_irq_regs();
485 	u64 *next_tb = &__get_cpu_var(decrementers_next_tb);
486 	struct clock_event_device *evt = &__get_cpu_var(decrementers);
487 	u64 now;
488 
489 	trace_timer_interrupt_entry(regs);
490 
491 	if (test_irq_work_pending()) {
492 		clear_irq_work_pending();
493 		irq_work_run();
494 	}
495 
496 	now = get_tb_or_rtc();
497 	if (now >= *next_tb) {
498 		*next_tb = ~(u64)0;
499 		if (evt->event_handler)
500 			evt->event_handler(evt);
501 		__get_cpu_var(irq_stat).timer_irqs_event++;
502 	} else {
503 		now = *next_tb - now;
504 		if (now <= DECREMENTER_MAX)
505 			set_dec((int)now);
506 		/* We may have raced with new irq work */
507 		if (test_irq_work_pending())
508 			set_dec(1);
509 		__get_cpu_var(irq_stat).timer_irqs_others++;
510 	}
511 
512 #ifdef CONFIG_PPC64
513 	/* collect purr register values often, for accurate calculations */
514 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
515 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
516 		cu->current_tb = mfspr(SPRN_PURR);
517 	}
518 #endif
519 
520 	trace_timer_interrupt_exit(regs);
521 }
522 
523 /*
524  * timer_interrupt - gets called when the decrementer overflows,
525  * with interrupts disabled.
526  */
527 void timer_interrupt(struct pt_regs * regs)
528 {
529 	struct pt_regs *old_regs;
530 	u64 *next_tb = &__get_cpu_var(decrementers_next_tb);
531 
532 	/* Ensure a positive value is written to the decrementer, or else
533 	 * some CPUs will continue to take decrementer exceptions.
534 	 */
535 	set_dec(DECREMENTER_MAX);
536 
537 	/* Some implementations of hotplug will get timer interrupts while
538 	 * offline, just ignore these and we also need to set
539 	 * decrementers_next_tb as MAX to make sure __check_irq_replay
540 	 * don't replay timer interrupt when return, otherwise we'll trap
541 	 * here infinitely :(
542 	 */
543 	if (!cpu_online(smp_processor_id())) {
544 		*next_tb = ~(u64)0;
545 		return;
546 	}
547 
548 	/* Conditionally hard-enable interrupts now that the DEC has been
549 	 * bumped to its maximum value
550 	 */
551 	may_hard_irq_enable();
552 
553 
554 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
555 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
556 		do_IRQ(regs);
557 #endif
558 
559 	old_regs = set_irq_regs(regs);
560 	irq_enter();
561 
562 	__timer_interrupt();
563 	irq_exit();
564 	set_irq_regs(old_regs);
565 }
566 
567 /*
568  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
569  * left pending on exit from a KVM guest.  We don't need to do anything
570  * to clear them, as they are edge-triggered.
571  */
572 void hdec_interrupt(struct pt_regs *regs)
573 {
574 }
575 
576 #ifdef CONFIG_SUSPEND
577 static void generic_suspend_disable_irqs(void)
578 {
579 	/* Disable the decrementer, so that it doesn't interfere
580 	 * with suspending.
581 	 */
582 
583 	set_dec(DECREMENTER_MAX);
584 	local_irq_disable();
585 	set_dec(DECREMENTER_MAX);
586 }
587 
588 static void generic_suspend_enable_irqs(void)
589 {
590 	local_irq_enable();
591 }
592 
593 /* Overrides the weak version in kernel/power/main.c */
594 void arch_suspend_disable_irqs(void)
595 {
596 	if (ppc_md.suspend_disable_irqs)
597 		ppc_md.suspend_disable_irqs();
598 	generic_suspend_disable_irqs();
599 }
600 
601 /* Overrides the weak version in kernel/power/main.c */
602 void arch_suspend_enable_irqs(void)
603 {
604 	generic_suspend_enable_irqs();
605 	if (ppc_md.suspend_enable_irqs)
606 		ppc_md.suspend_enable_irqs();
607 }
608 #endif
609 
610 /*
611  * Scheduler clock - returns current time in nanosec units.
612  *
613  * Note: mulhdu(a, b) (multiply high double unsigned) returns
614  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
615  * are 64-bit unsigned numbers.
616  */
617 unsigned long long sched_clock(void)
618 {
619 	if (__USE_RTC())
620 		return get_rtc();
621 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
622 }
623 
624 static int __init get_freq(char *name, int cells, unsigned long *val)
625 {
626 	struct device_node *cpu;
627 	const __be32 *fp;
628 	int found = 0;
629 
630 	/* The cpu node should have timebase and clock frequency properties */
631 	cpu = of_find_node_by_type(NULL, "cpu");
632 
633 	if (cpu) {
634 		fp = of_get_property(cpu, name, NULL);
635 		if (fp) {
636 			found = 1;
637 			*val = of_read_ulong(fp, cells);
638 		}
639 
640 		of_node_put(cpu);
641 	}
642 
643 	return found;
644 }
645 
646 void start_cpu_decrementer(void)
647 {
648 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
649 	/* Clear any pending timer interrupts */
650 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
651 
652 	/* Enable decrementer interrupt */
653 	mtspr(SPRN_TCR, TCR_DIE);
654 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
655 }
656 
657 void __init generic_calibrate_decr(void)
658 {
659 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
660 
661 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
662 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
663 
664 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
665 				"(not found)\n");
666 	}
667 
668 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
669 
670 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
671 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
672 
673 		printk(KERN_ERR "WARNING: Estimating processor frequency "
674 				"(not found)\n");
675 	}
676 }
677 
678 int update_persistent_clock(struct timespec now)
679 {
680 	struct rtc_time tm;
681 
682 	if (!ppc_md.set_rtc_time)
683 		return -ENODEV;
684 
685 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
686 	tm.tm_year -= 1900;
687 	tm.tm_mon -= 1;
688 
689 	return ppc_md.set_rtc_time(&tm);
690 }
691 
692 static void __read_persistent_clock(struct timespec *ts)
693 {
694 	struct rtc_time tm;
695 	static int first = 1;
696 
697 	ts->tv_nsec = 0;
698 	/* XXX this is a litle fragile but will work okay in the short term */
699 	if (first) {
700 		first = 0;
701 		if (ppc_md.time_init)
702 			timezone_offset = ppc_md.time_init();
703 
704 		/* get_boot_time() isn't guaranteed to be safe to call late */
705 		if (ppc_md.get_boot_time) {
706 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
707 			return;
708 		}
709 	}
710 	if (!ppc_md.get_rtc_time) {
711 		ts->tv_sec = 0;
712 		return;
713 	}
714 	ppc_md.get_rtc_time(&tm);
715 
716 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
717 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
718 }
719 
720 void read_persistent_clock(struct timespec *ts)
721 {
722 	__read_persistent_clock(ts);
723 
724 	/* Sanitize it in case real time clock is set below EPOCH */
725 	if (ts->tv_sec < 0) {
726 		ts->tv_sec = 0;
727 		ts->tv_nsec = 0;
728 	}
729 
730 }
731 
732 /* clocksource code */
733 static cycle_t rtc_read(struct clocksource *cs)
734 {
735 	return (cycle_t)get_rtc();
736 }
737 
738 static cycle_t timebase_read(struct clocksource *cs)
739 {
740 	return (cycle_t)get_tb();
741 }
742 
743 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
744 			struct clocksource *clock, u32 mult)
745 {
746 	u64 new_tb_to_xs, new_stamp_xsec;
747 	u32 frac_sec;
748 
749 	if (clock != &clocksource_timebase)
750 		return;
751 
752 	/* Make userspace gettimeofday spin until we're done. */
753 	++vdso_data->tb_update_count;
754 	smp_mb();
755 
756 	/* 19342813113834067 ~= 2^(20+64) / 1e9 */
757 	new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
758 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
759 	do_div(new_stamp_xsec, 1000000000);
760 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
761 
762 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
763 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
764 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
765 
766 	/*
767 	 * tb_update_count is used to allow the userspace gettimeofday code
768 	 * to assure itself that it sees a consistent view of the tb_to_xs and
769 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
770 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
771 	 * the two values of tb_update_count match and are even then the
772 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
773 	 * loops back and reads them again until this criteria is met.
774 	 * We expect the caller to have done the first increment of
775 	 * vdso_data->tb_update_count already.
776 	 */
777 	vdso_data->tb_orig_stamp = clock->cycle_last;
778 	vdso_data->stamp_xsec = new_stamp_xsec;
779 	vdso_data->tb_to_xs = new_tb_to_xs;
780 	vdso_data->wtom_clock_sec = wtm->tv_sec;
781 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
782 	vdso_data->stamp_xtime = *wall_time;
783 	vdso_data->stamp_sec_fraction = frac_sec;
784 	smp_wmb();
785 	++(vdso_data->tb_update_count);
786 }
787 
788 void update_vsyscall_tz(void)
789 {
790 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
791 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
792 }
793 
794 static void __init clocksource_init(void)
795 {
796 	struct clocksource *clock;
797 
798 	if (__USE_RTC())
799 		clock = &clocksource_rtc;
800 	else
801 		clock = &clocksource_timebase;
802 
803 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
804 		printk(KERN_ERR "clocksource: %s is already registered\n",
805 		       clock->name);
806 		return;
807 	}
808 
809 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
810 	       clock->name, clock->mult, clock->shift);
811 }
812 
813 static int decrementer_set_next_event(unsigned long evt,
814 				      struct clock_event_device *dev)
815 {
816 	__get_cpu_var(decrementers_next_tb) = get_tb_or_rtc() + evt;
817 	set_dec(evt);
818 
819 	/* We may have raced with new irq work */
820 	if (test_irq_work_pending())
821 		set_dec(1);
822 
823 	return 0;
824 }
825 
826 static void decrementer_set_mode(enum clock_event_mode mode,
827 				 struct clock_event_device *dev)
828 {
829 	if (mode != CLOCK_EVT_MODE_ONESHOT)
830 		decrementer_set_next_event(DECREMENTER_MAX, dev);
831 }
832 
833 /* Interrupt handler for the timer broadcast IPI */
834 void tick_broadcast_ipi_handler(void)
835 {
836 	u64 *next_tb = &__get_cpu_var(decrementers_next_tb);
837 
838 	*next_tb = get_tb_or_rtc();
839 	__timer_interrupt();
840 }
841 
842 static void register_decrementer_clockevent(int cpu)
843 {
844 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
845 
846 	*dec = decrementer_clockevent;
847 	dec->cpumask = cpumask_of(cpu);
848 
849 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
850 		    dec->name, dec->mult, dec->shift, cpu);
851 
852 	clockevents_register_device(dec);
853 }
854 
855 static void __init init_decrementer_clockevent(void)
856 {
857 	int cpu = smp_processor_id();
858 
859 	clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
860 
861 	decrementer_clockevent.max_delta_ns =
862 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
863 	decrementer_clockevent.min_delta_ns =
864 		clockevent_delta2ns(2, &decrementer_clockevent);
865 
866 	register_decrementer_clockevent(cpu);
867 }
868 
869 void secondary_cpu_time_init(void)
870 {
871 	/* Start the decrementer on CPUs that have manual control
872 	 * such as BookE
873 	 */
874 	start_cpu_decrementer();
875 
876 	/* FIME: Should make unrelatred change to move snapshot_timebase
877 	 * call here ! */
878 	register_decrementer_clockevent(smp_processor_id());
879 }
880 
881 /* This function is only called on the boot processor */
882 void __init time_init(void)
883 {
884 	struct div_result res;
885 	u64 scale;
886 	unsigned shift;
887 
888 	if (__USE_RTC()) {
889 		/* 601 processor: dec counts down by 128 every 128ns */
890 		ppc_tb_freq = 1000000000;
891 	} else {
892 		/* Normal PowerPC with timebase register */
893 		ppc_md.calibrate_decr();
894 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
895 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
896 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
897 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
898 	}
899 
900 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
901 	tb_ticks_per_sec = ppc_tb_freq;
902 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
903 	calc_cputime_factors();
904 	setup_cputime_one_jiffy();
905 
906 	/*
907 	 * Compute scale factor for sched_clock.
908 	 * The calibrate_decr() function has set tb_ticks_per_sec,
909 	 * which is the timebase frequency.
910 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
911 	 * the 128-bit result as a 64.64 fixed-point number.
912 	 * We then shift that number right until it is less than 1.0,
913 	 * giving us the scale factor and shift count to use in
914 	 * sched_clock().
915 	 */
916 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
917 	scale = res.result_low;
918 	for (shift = 0; res.result_high != 0; ++shift) {
919 		scale = (scale >> 1) | (res.result_high << 63);
920 		res.result_high >>= 1;
921 	}
922 	tb_to_ns_scale = scale;
923 	tb_to_ns_shift = shift;
924 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
925 	boot_tb = get_tb_or_rtc();
926 
927 	/* If platform provided a timezone (pmac), we correct the time */
928 	if (timezone_offset) {
929 		sys_tz.tz_minuteswest = -timezone_offset / 60;
930 		sys_tz.tz_dsttime = 0;
931 	}
932 
933 	vdso_data->tb_update_count = 0;
934 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
935 
936 	/* Start the decrementer on CPUs that have manual control
937 	 * such as BookE
938 	 */
939 	start_cpu_decrementer();
940 
941 	/* Register the clocksource */
942 	clocksource_init();
943 
944 	init_decrementer_clockevent();
945 	tick_setup_hrtimer_broadcast();
946 }
947 
948 
949 #define FEBRUARY	2
950 #define	STARTOFTIME	1970
951 #define SECDAY		86400L
952 #define SECYR		(SECDAY * 365)
953 #define	leapyear(year)		((year) % 4 == 0 && \
954 				 ((year) % 100 != 0 || (year) % 400 == 0))
955 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
956 #define	days_in_month(a) 	(month_days[(a) - 1])
957 
958 static int month_days[12] = {
959 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
960 };
961 
962 /*
963  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
964  */
965 void GregorianDay(struct rtc_time * tm)
966 {
967 	int leapsToDate;
968 	int lastYear;
969 	int day;
970 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
971 
972 	lastYear = tm->tm_year - 1;
973 
974 	/*
975 	 * Number of leap corrections to apply up to end of last year
976 	 */
977 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
978 
979 	/*
980 	 * This year is a leap year if it is divisible by 4 except when it is
981 	 * divisible by 100 unless it is divisible by 400
982 	 *
983 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
984 	 */
985 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
986 
987 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
988 		   tm->tm_mday;
989 
990 	tm->tm_wday = day % 7;
991 }
992 
993 void to_tm(int tim, struct rtc_time * tm)
994 {
995 	register int    i;
996 	register long   hms, day;
997 
998 	day = tim / SECDAY;
999 	hms = tim % SECDAY;
1000 
1001 	/* Hours, minutes, seconds are easy */
1002 	tm->tm_hour = hms / 3600;
1003 	tm->tm_min = (hms % 3600) / 60;
1004 	tm->tm_sec = (hms % 3600) % 60;
1005 
1006 	/* Number of years in days */
1007 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1008 		day -= days_in_year(i);
1009 	tm->tm_year = i;
1010 
1011 	/* Number of months in days left */
1012 	if (leapyear(tm->tm_year))
1013 		days_in_month(FEBRUARY) = 29;
1014 	for (i = 1; day >= days_in_month(i); i++)
1015 		day -= days_in_month(i);
1016 	days_in_month(FEBRUARY) = 28;
1017 	tm->tm_mon = i;
1018 
1019 	/* Days are what is left over (+1) from all that. */
1020 	tm->tm_mday = day + 1;
1021 
1022 	/*
1023 	 * Determine the day of week
1024 	 */
1025 	GregorianDay(tm);
1026 }
1027 
1028 /*
1029  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1030  * result.
1031  */
1032 void div128_by_32(u64 dividend_high, u64 dividend_low,
1033 		  unsigned divisor, struct div_result *dr)
1034 {
1035 	unsigned long a, b, c, d;
1036 	unsigned long w, x, y, z;
1037 	u64 ra, rb, rc;
1038 
1039 	a = dividend_high >> 32;
1040 	b = dividend_high & 0xffffffff;
1041 	c = dividend_low >> 32;
1042 	d = dividend_low & 0xffffffff;
1043 
1044 	w = a / divisor;
1045 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1046 
1047 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1048 	x = ra;
1049 
1050 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1051 	y = rb;
1052 
1053 	do_div(rc, divisor);
1054 	z = rc;
1055 
1056 	dr->result_high = ((u64)w << 32) + x;
1057 	dr->result_low  = ((u64)y << 32) + z;
1058 
1059 }
1060 
1061 /* We don't need to calibrate delay, we use the CPU timebase for that */
1062 void calibrate_delay(void)
1063 {
1064 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1065 	 * as the number of __delay(1) in a jiffy, so make it so
1066 	 */
1067 	loops_per_jiffy = tb_ticks_per_jiffy;
1068 }
1069 
1070 static int __init rtc_init(void)
1071 {
1072 	struct platform_device *pdev;
1073 
1074 	if (!ppc_md.get_rtc_time)
1075 		return -ENODEV;
1076 
1077 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1078 
1079 	return PTR_ERR_OR_ZERO(pdev);
1080 }
1081 
1082 module_init(rtc_init);
1083