xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision 81d67439)
1 /*
2  * Common time routines among all ppc machines.
3  *
4  * Written by Cort Dougan (cort@cs.nmt.edu) to merge
5  * Paul Mackerras' version and mine for PReP and Pmac.
6  * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net).
7  * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com)
8  *
9  * First round of bugfixes by Gabriel Paubert (paubert@iram.es)
10  * to make clock more stable (2.4.0-test5). The only thing
11  * that this code assumes is that the timebases have been synchronized
12  * by firmware on SMP and are never stopped (never do sleep
13  * on SMP then, nap and doze are OK).
14  *
15  * Speeded up do_gettimeofday by getting rid of references to
16  * xtime (which required locks for consistency). (mikejc@us.ibm.com)
17  *
18  * TODO (not necessarily in this file):
19  * - improve precision and reproducibility of timebase frequency
20  * measurement at boot time. (for iSeries, we calibrate the timebase
21  * against the Titan chip's clock.)
22  * - for astronomical applications: add a new function to get
23  * non ambiguous timestamps even around leap seconds. This needs
24  * a new timestamp format and a good name.
25  *
26  * 1997-09-10  Updated NTP code according to technical memorandum Jan '96
27  *             "A Kernel Model for Precision Timekeeping" by Dave Mills
28  *
29  *      This program is free software; you can redistribute it and/or
30  *      modify it under the terms of the GNU General Public License
31  *      as published by the Free Software Foundation; either version
32  *      2 of the License, or (at your option) any later version.
33  */
34 
35 #include <linux/errno.h>
36 #include <linux/module.h>
37 #include <linux/sched.h>
38 #include <linux/kernel.h>
39 #include <linux/param.h>
40 #include <linux/string.h>
41 #include <linux/mm.h>
42 #include <linux/interrupt.h>
43 #include <linux/timex.h>
44 #include <linux/kernel_stat.h>
45 #include <linux/time.h>
46 #include <linux/init.h>
47 #include <linux/profile.h>
48 #include <linux/cpu.h>
49 #include <linux/security.h>
50 #include <linux/percpu.h>
51 #include <linux/rtc.h>
52 #include <linux/jiffies.h>
53 #include <linux/posix-timers.h>
54 #include <linux/irq.h>
55 #include <linux/delay.h>
56 #include <linux/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 #ifdef CONFIG_PPC_ISERIES
74 #include <asm/iseries/it_lp_queue.h>
75 #include <asm/iseries/hv_call_xm.h>
76 #endif
77 
78 /* powerpc clocksource/clockevent code */
79 
80 #include <linux/clockchips.h>
81 #include <linux/clocksource.h>
82 
83 static cycle_t rtc_read(struct clocksource *);
84 static struct clocksource clocksource_rtc = {
85 	.name         = "rtc",
86 	.rating       = 400,
87 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
88 	.mask         = CLOCKSOURCE_MASK(64),
89 	.shift        = 22,
90 	.mult         = 0,	/* To be filled in */
91 	.read         = rtc_read,
92 };
93 
94 static cycle_t timebase_read(struct clocksource *);
95 static struct clocksource clocksource_timebase = {
96 	.name         = "timebase",
97 	.rating       = 400,
98 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
99 	.mask         = CLOCKSOURCE_MASK(64),
100 	.shift        = 22,
101 	.mult         = 0,	/* To be filled in */
102 	.read         = timebase_read,
103 };
104 
105 #define DECREMENTER_MAX	0x7fffffff
106 
107 static int decrementer_set_next_event(unsigned long evt,
108 				      struct clock_event_device *dev);
109 static void decrementer_set_mode(enum clock_event_mode mode,
110 				 struct clock_event_device *dev);
111 
112 static struct clock_event_device decrementer_clockevent = {
113        .name           = "decrementer",
114        .rating         = 200,
115        .shift          = 0,	/* To be filled in */
116        .mult           = 0,	/* To be filled in */
117        .irq            = 0,
118        .set_next_event = decrementer_set_next_event,
119        .set_mode       = decrementer_set_mode,
120        .features       = CLOCK_EVT_FEAT_ONESHOT,
121 };
122 
123 struct decrementer_clock {
124 	struct clock_event_device event;
125 	u64 next_tb;
126 };
127 
128 static DEFINE_PER_CPU(struct decrementer_clock, decrementers);
129 
130 #ifdef CONFIG_PPC_ISERIES
131 static unsigned long __initdata iSeries_recal_titan;
132 static signed long __initdata iSeries_recal_tb;
133 
134 /* Forward declaration is only needed for iSereis compiles */
135 static void __init clocksource_init(void);
136 #endif
137 
138 #define XSEC_PER_SEC (1024*1024)
139 
140 #ifdef CONFIG_PPC64
141 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
142 #else
143 /* compute ((xsec << 12) * max) >> 32 */
144 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
145 #endif
146 
147 unsigned long tb_ticks_per_jiffy;
148 unsigned long tb_ticks_per_usec = 100; /* sane default */
149 EXPORT_SYMBOL(tb_ticks_per_usec);
150 unsigned long tb_ticks_per_sec;
151 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
152 
153 DEFINE_SPINLOCK(rtc_lock);
154 EXPORT_SYMBOL_GPL(rtc_lock);
155 
156 static u64 tb_to_ns_scale __read_mostly;
157 static unsigned tb_to_ns_shift __read_mostly;
158 static u64 boot_tb __read_mostly;
159 
160 extern struct timezone sys_tz;
161 static long timezone_offset;
162 
163 unsigned long ppc_proc_freq;
164 EXPORT_SYMBOL_GPL(ppc_proc_freq);
165 unsigned long ppc_tb_freq;
166 EXPORT_SYMBOL_GPL(ppc_tb_freq);
167 
168 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
169 /*
170  * Factors for converting from cputime_t (timebase ticks) to
171  * jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds).
172  * These are all stored as 0.64 fixed-point binary fractions.
173  */
174 u64 __cputime_jiffies_factor;
175 EXPORT_SYMBOL(__cputime_jiffies_factor);
176 u64 __cputime_msec_factor;
177 EXPORT_SYMBOL(__cputime_msec_factor);
178 u64 __cputime_sec_factor;
179 EXPORT_SYMBOL(__cputime_sec_factor);
180 u64 __cputime_clockt_factor;
181 EXPORT_SYMBOL(__cputime_clockt_factor);
182 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
183 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
184 
185 cputime_t cputime_one_jiffy;
186 
187 void (*dtl_consumer)(struct dtl_entry *, u64);
188 
189 static void calc_cputime_factors(void)
190 {
191 	struct div_result res;
192 
193 	div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
194 	__cputime_jiffies_factor = res.result_low;
195 	div128_by_32(1000, 0, tb_ticks_per_sec, &res);
196 	__cputime_msec_factor = res.result_low;
197 	div128_by_32(1, 0, tb_ticks_per_sec, &res);
198 	__cputime_sec_factor = res.result_low;
199 	div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
200 	__cputime_clockt_factor = res.result_low;
201 }
202 
203 /*
204  * Read the SPURR on systems that have it, otherwise the PURR,
205  * or if that doesn't exist return the timebase value passed in.
206  */
207 static u64 read_spurr(u64 tb)
208 {
209 	if (cpu_has_feature(CPU_FTR_SPURR))
210 		return mfspr(SPRN_SPURR);
211 	if (cpu_has_feature(CPU_FTR_PURR))
212 		return mfspr(SPRN_PURR);
213 	return tb;
214 }
215 
216 #ifdef CONFIG_PPC_SPLPAR
217 
218 /*
219  * Scan the dispatch trace log and count up the stolen time.
220  * Should be called with interrupts disabled.
221  */
222 static u64 scan_dispatch_log(u64 stop_tb)
223 {
224 	u64 i = local_paca->dtl_ridx;
225 	struct dtl_entry *dtl = local_paca->dtl_curr;
226 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
227 	struct lppaca *vpa = local_paca->lppaca_ptr;
228 	u64 tb_delta;
229 	u64 stolen = 0;
230 	u64 dtb;
231 
232 	if (!dtl)
233 		return 0;
234 
235 	if (i == vpa->dtl_idx)
236 		return 0;
237 	while (i < vpa->dtl_idx) {
238 		if (dtl_consumer)
239 			dtl_consumer(dtl, i);
240 		dtb = dtl->timebase;
241 		tb_delta = dtl->enqueue_to_dispatch_time +
242 			dtl->ready_to_enqueue_time;
243 		barrier();
244 		if (i + N_DISPATCH_LOG < vpa->dtl_idx) {
245 			/* buffer has overflowed */
246 			i = vpa->dtl_idx - N_DISPATCH_LOG;
247 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
248 			continue;
249 		}
250 		if (dtb > stop_tb)
251 			break;
252 		stolen += tb_delta;
253 		++i;
254 		++dtl;
255 		if (dtl == dtl_end)
256 			dtl = local_paca->dispatch_log;
257 	}
258 	local_paca->dtl_ridx = i;
259 	local_paca->dtl_curr = dtl;
260 	return stolen;
261 }
262 
263 /*
264  * Accumulate stolen time by scanning the dispatch trace log.
265  * Called on entry from user mode.
266  */
267 void accumulate_stolen_time(void)
268 {
269 	u64 sst, ust;
270 
271 	u8 save_soft_enabled = local_paca->soft_enabled;
272 	u8 save_hard_enabled = local_paca->hard_enabled;
273 
274 	/* We are called early in the exception entry, before
275 	 * soft/hard_enabled are sync'ed to the expected state
276 	 * for the exception. We are hard disabled but the PACA
277 	 * needs to reflect that so various debug stuff doesn't
278 	 * complain
279 	 */
280 	local_paca->soft_enabled = 0;
281 	local_paca->hard_enabled = 0;
282 
283 	sst = scan_dispatch_log(local_paca->starttime_user);
284 	ust = scan_dispatch_log(local_paca->starttime);
285 	local_paca->system_time -= sst;
286 	local_paca->user_time -= ust;
287 	local_paca->stolen_time += ust + sst;
288 
289 	local_paca->soft_enabled = save_soft_enabled;
290 	local_paca->hard_enabled = save_hard_enabled;
291 }
292 
293 static inline u64 calculate_stolen_time(u64 stop_tb)
294 {
295 	u64 stolen = 0;
296 
297 	if (get_paca()->dtl_ridx != get_paca()->lppaca_ptr->dtl_idx) {
298 		stolen = scan_dispatch_log(stop_tb);
299 		get_paca()->system_time -= stolen;
300 	}
301 
302 	stolen += get_paca()->stolen_time;
303 	get_paca()->stolen_time = 0;
304 	return stolen;
305 }
306 
307 #else /* CONFIG_PPC_SPLPAR */
308 static inline u64 calculate_stolen_time(u64 stop_tb)
309 {
310 	return 0;
311 }
312 
313 #endif /* CONFIG_PPC_SPLPAR */
314 
315 /*
316  * Account time for a transition between system, hard irq
317  * or soft irq state.
318  */
319 void account_system_vtime(struct task_struct *tsk)
320 {
321 	u64 now, nowscaled, delta, deltascaled;
322 	unsigned long flags;
323 	u64 stolen, udelta, sys_scaled, user_scaled;
324 
325 	local_irq_save(flags);
326 	now = mftb();
327 	nowscaled = read_spurr(now);
328 	get_paca()->system_time += now - get_paca()->starttime;
329 	get_paca()->starttime = now;
330 	deltascaled = nowscaled - get_paca()->startspurr;
331 	get_paca()->startspurr = nowscaled;
332 
333 	stolen = calculate_stolen_time(now);
334 
335 	delta = get_paca()->system_time;
336 	get_paca()->system_time = 0;
337 	udelta = get_paca()->user_time - get_paca()->utime_sspurr;
338 	get_paca()->utime_sspurr = get_paca()->user_time;
339 
340 	/*
341 	 * Because we don't read the SPURR on every kernel entry/exit,
342 	 * deltascaled includes both user and system SPURR ticks.
343 	 * Apportion these ticks to system SPURR ticks and user
344 	 * SPURR ticks in the same ratio as the system time (delta)
345 	 * and user time (udelta) values obtained from the timebase
346 	 * over the same interval.  The system ticks get accounted here;
347 	 * the user ticks get saved up in paca->user_time_scaled to be
348 	 * used by account_process_tick.
349 	 */
350 	sys_scaled = delta;
351 	user_scaled = udelta;
352 	if (deltascaled != delta + udelta) {
353 		if (udelta) {
354 			sys_scaled = deltascaled * delta / (delta + udelta);
355 			user_scaled = deltascaled - sys_scaled;
356 		} else {
357 			sys_scaled = deltascaled;
358 		}
359 	}
360 	get_paca()->user_time_scaled += user_scaled;
361 
362 	if (in_interrupt() || idle_task(smp_processor_id()) != tsk) {
363 		account_system_time(tsk, 0, delta, sys_scaled);
364 		if (stolen)
365 			account_steal_time(stolen);
366 	} else {
367 		account_idle_time(delta + stolen);
368 	}
369 	local_irq_restore(flags);
370 }
371 EXPORT_SYMBOL_GPL(account_system_vtime);
372 
373 /*
374  * Transfer the user and system times accumulated in the paca
375  * by the exception entry and exit code to the generic process
376  * user and system time records.
377  * Must be called with interrupts disabled.
378  * Assumes that account_system_vtime() has been called recently
379  * (i.e. since the last entry from usermode) so that
380  * get_paca()->user_time_scaled is up to date.
381  */
382 void account_process_tick(struct task_struct *tsk, int user_tick)
383 {
384 	cputime_t utime, utimescaled;
385 
386 	utime = get_paca()->user_time;
387 	utimescaled = get_paca()->user_time_scaled;
388 	get_paca()->user_time = 0;
389 	get_paca()->user_time_scaled = 0;
390 	get_paca()->utime_sspurr = 0;
391 	account_user_time(tsk, utime, utimescaled);
392 }
393 
394 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
395 #define calc_cputime_factors()
396 #endif
397 
398 void __delay(unsigned long loops)
399 {
400 	unsigned long start;
401 	int diff;
402 
403 	if (__USE_RTC()) {
404 		start = get_rtcl();
405 		do {
406 			/* the RTCL register wraps at 1000000000 */
407 			diff = get_rtcl() - start;
408 			if (diff < 0)
409 				diff += 1000000000;
410 		} while (diff < loops);
411 	} else {
412 		start = get_tbl();
413 		while (get_tbl() - start < loops)
414 			HMT_low();
415 		HMT_medium();
416 	}
417 }
418 EXPORT_SYMBOL(__delay);
419 
420 void udelay(unsigned long usecs)
421 {
422 	__delay(tb_ticks_per_usec * usecs);
423 }
424 EXPORT_SYMBOL(udelay);
425 
426 #ifdef CONFIG_SMP
427 unsigned long profile_pc(struct pt_regs *regs)
428 {
429 	unsigned long pc = instruction_pointer(regs);
430 
431 	if (in_lock_functions(pc))
432 		return regs->link;
433 
434 	return pc;
435 }
436 EXPORT_SYMBOL(profile_pc);
437 #endif
438 
439 #ifdef CONFIG_PPC_ISERIES
440 
441 /*
442  * This function recalibrates the timebase based on the 49-bit time-of-day
443  * value in the Titan chip.  The Titan is much more accurate than the value
444  * returned by the service processor for the timebase frequency.
445  */
446 
447 static int __init iSeries_tb_recal(void)
448 {
449 	unsigned long titan, tb;
450 
451 	/* Make sure we only run on iSeries */
452 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
453 		return -ENODEV;
454 
455 	tb = get_tb();
456 	titan = HvCallXm_loadTod();
457 	if ( iSeries_recal_titan ) {
458 		unsigned long tb_ticks = tb - iSeries_recal_tb;
459 		unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
460 		unsigned long new_tb_ticks_per_sec   = (tb_ticks * USEC_PER_SEC)/titan_usec;
461 		unsigned long new_tb_ticks_per_jiffy =
462 			DIV_ROUND_CLOSEST(new_tb_ticks_per_sec, HZ);
463 		long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
464 		char sign = '+';
465 		/* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
466 		new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;
467 
468 		if ( tick_diff < 0 ) {
469 			tick_diff = -tick_diff;
470 			sign = '-';
471 		}
472 		if ( tick_diff ) {
473 			if ( tick_diff < tb_ticks_per_jiffy/25 ) {
474 				printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
475 						new_tb_ticks_per_jiffy, sign, tick_diff );
476 				tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
477 				tb_ticks_per_sec   = new_tb_ticks_per_sec;
478 				calc_cputime_factors();
479 				vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
480 				setup_cputime_one_jiffy();
481 			}
482 			else {
483 				printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
484 					"                   new tb_ticks_per_jiffy = %lu\n"
485 					"                   old tb_ticks_per_jiffy = %lu\n",
486 					new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
487 			}
488 		}
489 	}
490 	iSeries_recal_titan = titan;
491 	iSeries_recal_tb = tb;
492 
493 	/* Called here as now we know accurate values for the timebase */
494 	clocksource_init();
495 	return 0;
496 }
497 late_initcall(iSeries_tb_recal);
498 
499 /* Called from platform early init */
500 void __init iSeries_time_init_early(void)
501 {
502 	iSeries_recal_tb = get_tb();
503 	iSeries_recal_titan = HvCallXm_loadTod();
504 }
505 #endif /* CONFIG_PPC_ISERIES */
506 
507 #ifdef CONFIG_IRQ_WORK
508 
509 /*
510  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
511  */
512 #ifdef CONFIG_PPC64
513 static inline unsigned long test_irq_work_pending(void)
514 {
515 	unsigned long x;
516 
517 	asm volatile("lbz %0,%1(13)"
518 		: "=r" (x)
519 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
520 	return x;
521 }
522 
523 static inline void set_irq_work_pending_flag(void)
524 {
525 	asm volatile("stb %0,%1(13)" : :
526 		"r" (1),
527 		"i" (offsetof(struct paca_struct, irq_work_pending)));
528 }
529 
530 static inline void clear_irq_work_pending(void)
531 {
532 	asm volatile("stb %0,%1(13)" : :
533 		"r" (0),
534 		"i" (offsetof(struct paca_struct, irq_work_pending)));
535 }
536 
537 #else /* 32-bit */
538 
539 DEFINE_PER_CPU(u8, irq_work_pending);
540 
541 #define set_irq_work_pending_flag()	__get_cpu_var(irq_work_pending) = 1
542 #define test_irq_work_pending()		__get_cpu_var(irq_work_pending)
543 #define clear_irq_work_pending()	__get_cpu_var(irq_work_pending) = 0
544 
545 #endif /* 32 vs 64 bit */
546 
547 void arch_irq_work_raise(void)
548 {
549 	preempt_disable();
550 	set_irq_work_pending_flag();
551 	set_dec(1);
552 	preempt_enable();
553 }
554 
555 #else  /* CONFIG_IRQ_WORK */
556 
557 #define test_irq_work_pending()	0
558 #define clear_irq_work_pending()
559 
560 #endif /* CONFIG_IRQ_WORK */
561 
562 /*
563  * For iSeries shared processors, we have to let the hypervisor
564  * set the hardware decrementer.  We set a virtual decrementer
565  * in the lppaca and call the hypervisor if the virtual
566  * decrementer is less than the current value in the hardware
567  * decrementer. (almost always the new decrementer value will
568  * be greater than the current hardware decementer so the hypervisor
569  * call will not be needed)
570  */
571 
572 /*
573  * timer_interrupt - gets called when the decrementer overflows,
574  * with interrupts disabled.
575  */
576 void timer_interrupt(struct pt_regs * regs)
577 {
578 	struct pt_regs *old_regs;
579 	struct decrementer_clock *decrementer =  &__get_cpu_var(decrementers);
580 	struct clock_event_device *evt = &decrementer->event;
581 	u64 now;
582 
583 	/* Ensure a positive value is written to the decrementer, or else
584 	 * some CPUs will continue to take decrementer exceptions.
585 	 */
586 	set_dec(DECREMENTER_MAX);
587 
588 	/* Some implementations of hotplug will get timer interrupts while
589 	 * offline, just ignore these
590 	 */
591 	if (!cpu_online(smp_processor_id()))
592 		return;
593 
594 	trace_timer_interrupt_entry(regs);
595 
596 	__get_cpu_var(irq_stat).timer_irqs++;
597 
598 #if defined(CONFIG_PPC32) && defined(CONFIG_PMAC)
599 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
600 		do_IRQ(regs);
601 #endif
602 
603 	old_regs = set_irq_regs(regs);
604 	irq_enter();
605 
606 	if (test_irq_work_pending()) {
607 		clear_irq_work_pending();
608 		irq_work_run();
609 	}
610 
611 #ifdef CONFIG_PPC_ISERIES
612 	if (firmware_has_feature(FW_FEATURE_ISERIES))
613 		get_lppaca()->int_dword.fields.decr_int = 0;
614 #endif
615 
616 	now = get_tb_or_rtc();
617 	if (now >= decrementer->next_tb) {
618 		decrementer->next_tb = ~(u64)0;
619 		if (evt->event_handler)
620 			evt->event_handler(evt);
621 	} else {
622 		now = decrementer->next_tb - now;
623 		if (now <= DECREMENTER_MAX)
624 			set_dec((int)now);
625 	}
626 
627 #ifdef CONFIG_PPC_ISERIES
628 	if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending())
629 		process_hvlpevents();
630 #endif
631 
632 #ifdef CONFIG_PPC64
633 	/* collect purr register values often, for accurate calculations */
634 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
635 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
636 		cu->current_tb = mfspr(SPRN_PURR);
637 	}
638 #endif
639 
640 	irq_exit();
641 	set_irq_regs(old_regs);
642 
643 	trace_timer_interrupt_exit(regs);
644 }
645 
646 #ifdef CONFIG_SUSPEND
647 static void generic_suspend_disable_irqs(void)
648 {
649 	/* Disable the decrementer, so that it doesn't interfere
650 	 * with suspending.
651 	 */
652 
653 	set_dec(0x7fffffff);
654 	local_irq_disable();
655 	set_dec(0x7fffffff);
656 }
657 
658 static void generic_suspend_enable_irqs(void)
659 {
660 	local_irq_enable();
661 }
662 
663 /* Overrides the weak version in kernel/power/main.c */
664 void arch_suspend_disable_irqs(void)
665 {
666 	if (ppc_md.suspend_disable_irqs)
667 		ppc_md.suspend_disable_irqs();
668 	generic_suspend_disable_irqs();
669 }
670 
671 /* Overrides the weak version in kernel/power/main.c */
672 void arch_suspend_enable_irqs(void)
673 {
674 	generic_suspend_enable_irqs();
675 	if (ppc_md.suspend_enable_irqs)
676 		ppc_md.suspend_enable_irqs();
677 }
678 #endif
679 
680 /*
681  * Scheduler clock - returns current time in nanosec units.
682  *
683  * Note: mulhdu(a, b) (multiply high double unsigned) returns
684  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
685  * are 64-bit unsigned numbers.
686  */
687 unsigned long long sched_clock(void)
688 {
689 	if (__USE_RTC())
690 		return get_rtc();
691 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
692 }
693 
694 static int __init get_freq(char *name, int cells, unsigned long *val)
695 {
696 	struct device_node *cpu;
697 	const unsigned int *fp;
698 	int found = 0;
699 
700 	/* The cpu node should have timebase and clock frequency properties */
701 	cpu = of_find_node_by_type(NULL, "cpu");
702 
703 	if (cpu) {
704 		fp = of_get_property(cpu, name, NULL);
705 		if (fp) {
706 			found = 1;
707 			*val = of_read_ulong(fp, cells);
708 		}
709 
710 		of_node_put(cpu);
711 	}
712 
713 	return found;
714 }
715 
716 /* should become __cpuinit when secondary_cpu_time_init also is */
717 void start_cpu_decrementer(void)
718 {
719 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
720 	/* Clear any pending timer interrupts */
721 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
722 
723 	/* Enable decrementer interrupt */
724 	mtspr(SPRN_TCR, TCR_DIE);
725 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
726 }
727 
728 void __init generic_calibrate_decr(void)
729 {
730 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
731 
732 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
733 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
734 
735 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
736 				"(not found)\n");
737 	}
738 
739 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
740 
741 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
742 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
743 
744 		printk(KERN_ERR "WARNING: Estimating processor frequency "
745 				"(not found)\n");
746 	}
747 }
748 
749 int update_persistent_clock(struct timespec now)
750 {
751 	struct rtc_time tm;
752 
753 	if (!ppc_md.set_rtc_time)
754 		return 0;
755 
756 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
757 	tm.tm_year -= 1900;
758 	tm.tm_mon -= 1;
759 
760 	return ppc_md.set_rtc_time(&tm);
761 }
762 
763 static void __read_persistent_clock(struct timespec *ts)
764 {
765 	struct rtc_time tm;
766 	static int first = 1;
767 
768 	ts->tv_nsec = 0;
769 	/* XXX this is a litle fragile but will work okay in the short term */
770 	if (first) {
771 		first = 0;
772 		if (ppc_md.time_init)
773 			timezone_offset = ppc_md.time_init();
774 
775 		/* get_boot_time() isn't guaranteed to be safe to call late */
776 		if (ppc_md.get_boot_time) {
777 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
778 			return;
779 		}
780 	}
781 	if (!ppc_md.get_rtc_time) {
782 		ts->tv_sec = 0;
783 		return;
784 	}
785 	ppc_md.get_rtc_time(&tm);
786 
787 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
788 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
789 }
790 
791 void read_persistent_clock(struct timespec *ts)
792 {
793 	__read_persistent_clock(ts);
794 
795 	/* Sanitize it in case real time clock is set below EPOCH */
796 	if (ts->tv_sec < 0) {
797 		ts->tv_sec = 0;
798 		ts->tv_nsec = 0;
799 	}
800 
801 }
802 
803 /* clocksource code */
804 static cycle_t rtc_read(struct clocksource *cs)
805 {
806 	return (cycle_t)get_rtc();
807 }
808 
809 static cycle_t timebase_read(struct clocksource *cs)
810 {
811 	return (cycle_t)get_tb();
812 }
813 
814 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
815 			struct clocksource *clock, u32 mult)
816 {
817 	u64 new_tb_to_xs, new_stamp_xsec;
818 	u32 frac_sec;
819 
820 	if (clock != &clocksource_timebase)
821 		return;
822 
823 	/* Make userspace gettimeofday spin until we're done. */
824 	++vdso_data->tb_update_count;
825 	smp_mb();
826 
827 	/* XXX this assumes clock->shift == 22 */
828 	/* 4611686018 ~= 2^(20+64-22) / 1e9 */
829 	new_tb_to_xs = (u64) mult * 4611686018ULL;
830 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
831 	do_div(new_stamp_xsec, 1000000000);
832 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
833 
834 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
835 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
836 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
837 
838 	/*
839 	 * tb_update_count is used to allow the userspace gettimeofday code
840 	 * to assure itself that it sees a consistent view of the tb_to_xs and
841 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
842 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
843 	 * the two values of tb_update_count match and are even then the
844 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
845 	 * loops back and reads them again until this criteria is met.
846 	 * We expect the caller to have done the first increment of
847 	 * vdso_data->tb_update_count already.
848 	 */
849 	vdso_data->tb_orig_stamp = clock->cycle_last;
850 	vdso_data->stamp_xsec = new_stamp_xsec;
851 	vdso_data->tb_to_xs = new_tb_to_xs;
852 	vdso_data->wtom_clock_sec = wtm->tv_sec;
853 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
854 	vdso_data->stamp_xtime = *wall_time;
855 	vdso_data->stamp_sec_fraction = frac_sec;
856 	smp_wmb();
857 	++(vdso_data->tb_update_count);
858 }
859 
860 void update_vsyscall_tz(void)
861 {
862 	/* Make userspace gettimeofday spin until we're done. */
863 	++vdso_data->tb_update_count;
864 	smp_mb();
865 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
866 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
867 	smp_mb();
868 	++vdso_data->tb_update_count;
869 }
870 
871 static void __init clocksource_init(void)
872 {
873 	struct clocksource *clock;
874 
875 	if (__USE_RTC())
876 		clock = &clocksource_rtc;
877 	else
878 		clock = &clocksource_timebase;
879 
880 	clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
881 
882 	if (clocksource_register(clock)) {
883 		printk(KERN_ERR "clocksource: %s is already registered\n",
884 		       clock->name);
885 		return;
886 	}
887 
888 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
889 	       clock->name, clock->mult, clock->shift);
890 }
891 
892 static int decrementer_set_next_event(unsigned long evt,
893 				      struct clock_event_device *dev)
894 {
895 	__get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
896 	set_dec(evt);
897 	return 0;
898 }
899 
900 static void decrementer_set_mode(enum clock_event_mode mode,
901 				 struct clock_event_device *dev)
902 {
903 	if (mode != CLOCK_EVT_MODE_ONESHOT)
904 		decrementer_set_next_event(DECREMENTER_MAX, dev);
905 }
906 
907 static inline uint64_t div_sc64(unsigned long ticks, unsigned long nsec,
908 				int shift)
909 {
910 	uint64_t tmp = ((uint64_t)ticks) << shift;
911 
912 	do_div(tmp, nsec);
913 	return tmp;
914 }
915 
916 static void __init setup_clockevent_multiplier(unsigned long hz)
917 {
918 	u64 mult, shift = 32;
919 
920 	while (1) {
921 		mult = div_sc64(hz, NSEC_PER_SEC, shift);
922 		if (mult && (mult >> 32UL) == 0UL)
923 			break;
924 
925 		shift--;
926 	}
927 
928 	decrementer_clockevent.shift = shift;
929 	decrementer_clockevent.mult = mult;
930 }
931 
932 static void register_decrementer_clockevent(int cpu)
933 {
934 	struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
935 
936 	*dec = decrementer_clockevent;
937 	dec->cpumask = cpumask_of(cpu);
938 
939 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
940 		    dec->name, dec->mult, dec->shift, cpu);
941 
942 	clockevents_register_device(dec);
943 }
944 
945 static void __init init_decrementer_clockevent(void)
946 {
947 	int cpu = smp_processor_id();
948 
949 	setup_clockevent_multiplier(ppc_tb_freq);
950 	decrementer_clockevent.max_delta_ns =
951 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
952 	decrementer_clockevent.min_delta_ns =
953 		clockevent_delta2ns(2, &decrementer_clockevent);
954 
955 	register_decrementer_clockevent(cpu);
956 }
957 
958 void secondary_cpu_time_init(void)
959 {
960 	/* Start the decrementer on CPUs that have manual control
961 	 * such as BookE
962 	 */
963 	start_cpu_decrementer();
964 
965 	/* FIME: Should make unrelatred change to move snapshot_timebase
966 	 * call here ! */
967 	register_decrementer_clockevent(smp_processor_id());
968 }
969 
970 /* This function is only called on the boot processor */
971 void __init time_init(void)
972 {
973 	struct div_result res;
974 	u64 scale;
975 	unsigned shift;
976 
977 	if (__USE_RTC()) {
978 		/* 601 processor: dec counts down by 128 every 128ns */
979 		ppc_tb_freq = 1000000000;
980 	} else {
981 		/* Normal PowerPC with timebase register */
982 		ppc_md.calibrate_decr();
983 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
984 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
985 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
986 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
987 	}
988 
989 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
990 	tb_ticks_per_sec = ppc_tb_freq;
991 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
992 	calc_cputime_factors();
993 	setup_cputime_one_jiffy();
994 
995 	/*
996 	 * Compute scale factor for sched_clock.
997 	 * The calibrate_decr() function has set tb_ticks_per_sec,
998 	 * which is the timebase frequency.
999 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1000 	 * the 128-bit result as a 64.64 fixed-point number.
1001 	 * We then shift that number right until it is less than 1.0,
1002 	 * giving us the scale factor and shift count to use in
1003 	 * sched_clock().
1004 	 */
1005 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1006 	scale = res.result_low;
1007 	for (shift = 0; res.result_high != 0; ++shift) {
1008 		scale = (scale >> 1) | (res.result_high << 63);
1009 		res.result_high >>= 1;
1010 	}
1011 	tb_to_ns_scale = scale;
1012 	tb_to_ns_shift = shift;
1013 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1014 	boot_tb = get_tb_or_rtc();
1015 
1016 	/* If platform provided a timezone (pmac), we correct the time */
1017         if (timezone_offset) {
1018 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1019 		sys_tz.tz_dsttime = 0;
1020         }
1021 
1022 	vdso_data->tb_update_count = 0;
1023 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1024 
1025 	/* Start the decrementer on CPUs that have manual control
1026 	 * such as BookE
1027 	 */
1028 	start_cpu_decrementer();
1029 
1030 	/* Register the clocksource, if we're not running on iSeries */
1031 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
1032 		clocksource_init();
1033 
1034 	init_decrementer_clockevent();
1035 }
1036 
1037 
1038 #define FEBRUARY	2
1039 #define	STARTOFTIME	1970
1040 #define SECDAY		86400L
1041 #define SECYR		(SECDAY * 365)
1042 #define	leapyear(year)		((year) % 4 == 0 && \
1043 				 ((year) % 100 != 0 || (year) % 400 == 0))
1044 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1045 #define	days_in_month(a) 	(month_days[(a) - 1])
1046 
1047 static int month_days[12] = {
1048 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1049 };
1050 
1051 /*
1052  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1053  */
1054 void GregorianDay(struct rtc_time * tm)
1055 {
1056 	int leapsToDate;
1057 	int lastYear;
1058 	int day;
1059 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1060 
1061 	lastYear = tm->tm_year - 1;
1062 
1063 	/*
1064 	 * Number of leap corrections to apply up to end of last year
1065 	 */
1066 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
1067 
1068 	/*
1069 	 * This year is a leap year if it is divisible by 4 except when it is
1070 	 * divisible by 100 unless it is divisible by 400
1071 	 *
1072 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1073 	 */
1074 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
1075 
1076 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
1077 		   tm->tm_mday;
1078 
1079 	tm->tm_wday = day % 7;
1080 }
1081 
1082 void to_tm(int tim, struct rtc_time * tm)
1083 {
1084 	register int    i;
1085 	register long   hms, day;
1086 
1087 	day = tim / SECDAY;
1088 	hms = tim % SECDAY;
1089 
1090 	/* Hours, minutes, seconds are easy */
1091 	tm->tm_hour = hms / 3600;
1092 	tm->tm_min = (hms % 3600) / 60;
1093 	tm->tm_sec = (hms % 3600) % 60;
1094 
1095 	/* Number of years in days */
1096 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1097 		day -= days_in_year(i);
1098 	tm->tm_year = i;
1099 
1100 	/* Number of months in days left */
1101 	if (leapyear(tm->tm_year))
1102 		days_in_month(FEBRUARY) = 29;
1103 	for (i = 1; day >= days_in_month(i); i++)
1104 		day -= days_in_month(i);
1105 	days_in_month(FEBRUARY) = 28;
1106 	tm->tm_mon = i;
1107 
1108 	/* Days are what is left over (+1) from all that. */
1109 	tm->tm_mday = day + 1;
1110 
1111 	/*
1112 	 * Determine the day of week
1113 	 */
1114 	GregorianDay(tm);
1115 }
1116 
1117 /*
1118  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1119  * result.
1120  */
1121 void div128_by_32(u64 dividend_high, u64 dividend_low,
1122 		  unsigned divisor, struct div_result *dr)
1123 {
1124 	unsigned long a, b, c, d;
1125 	unsigned long w, x, y, z;
1126 	u64 ra, rb, rc;
1127 
1128 	a = dividend_high >> 32;
1129 	b = dividend_high & 0xffffffff;
1130 	c = dividend_low >> 32;
1131 	d = dividend_low & 0xffffffff;
1132 
1133 	w = a / divisor;
1134 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1135 
1136 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1137 	x = ra;
1138 
1139 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1140 	y = rb;
1141 
1142 	do_div(rc, divisor);
1143 	z = rc;
1144 
1145 	dr->result_high = ((u64)w << 32) + x;
1146 	dr->result_low  = ((u64)y << 32) + z;
1147 
1148 }
1149 
1150 /* We don't need to calibrate delay, we use the CPU timebase for that */
1151 void calibrate_delay(void)
1152 {
1153 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1154 	 * as the number of __delay(1) in a jiffy, so make it so
1155 	 */
1156 	loops_per_jiffy = tb_ticks_per_jiffy;
1157 }
1158 
1159 static int __init rtc_init(void)
1160 {
1161 	struct platform_device *pdev;
1162 
1163 	if (!ppc_md.get_rtc_time)
1164 		return -ENODEV;
1165 
1166 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1167 	if (IS_ERR(pdev))
1168 		return PTR_ERR(pdev);
1169 
1170 	return 0;
1171 }
1172 
1173 module_init(rtc_init);
1174