xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
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 (i == vpa->dtl_idx)
233 		return 0;
234 	while (i < vpa->dtl_idx) {
235 		if (dtl_consumer)
236 			dtl_consumer(dtl, i);
237 		dtb = dtl->timebase;
238 		tb_delta = dtl->enqueue_to_dispatch_time +
239 			dtl->ready_to_enqueue_time;
240 		barrier();
241 		if (i + N_DISPATCH_LOG < vpa->dtl_idx) {
242 			/* buffer has overflowed */
243 			i = vpa->dtl_idx - N_DISPATCH_LOG;
244 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
245 			continue;
246 		}
247 		if (dtb > stop_tb)
248 			break;
249 		stolen += tb_delta;
250 		++i;
251 		++dtl;
252 		if (dtl == dtl_end)
253 			dtl = local_paca->dispatch_log;
254 	}
255 	local_paca->dtl_ridx = i;
256 	local_paca->dtl_curr = dtl;
257 	return stolen;
258 }
259 
260 /*
261  * Accumulate stolen time by scanning the dispatch trace log.
262  * Called on entry from user mode.
263  */
264 void accumulate_stolen_time(void)
265 {
266 	u64 sst, ust;
267 
268 	u8 save_soft_enabled = local_paca->soft_enabled;
269 	u8 save_hard_enabled = local_paca->hard_enabled;
270 
271 	/* We are called early in the exception entry, before
272 	 * soft/hard_enabled are sync'ed to the expected state
273 	 * for the exception. We are hard disabled but the PACA
274 	 * needs to reflect that so various debug stuff doesn't
275 	 * complain
276 	 */
277 	local_paca->soft_enabled = 0;
278 	local_paca->hard_enabled = 0;
279 
280 	sst = scan_dispatch_log(local_paca->starttime_user);
281 	ust = scan_dispatch_log(local_paca->starttime);
282 	local_paca->system_time -= sst;
283 	local_paca->user_time -= ust;
284 	local_paca->stolen_time += ust + sst;
285 
286 	local_paca->soft_enabled = save_soft_enabled;
287 	local_paca->hard_enabled = save_hard_enabled;
288 }
289 
290 static inline u64 calculate_stolen_time(u64 stop_tb)
291 {
292 	u64 stolen = 0;
293 
294 	if (get_paca()->dtl_ridx != get_paca()->lppaca_ptr->dtl_idx) {
295 		stolen = scan_dispatch_log(stop_tb);
296 		get_paca()->system_time -= stolen;
297 	}
298 
299 	stolen += get_paca()->stolen_time;
300 	get_paca()->stolen_time = 0;
301 	return stolen;
302 }
303 
304 #else /* CONFIG_PPC_SPLPAR */
305 static inline u64 calculate_stolen_time(u64 stop_tb)
306 {
307 	return 0;
308 }
309 
310 #endif /* CONFIG_PPC_SPLPAR */
311 
312 /*
313  * Account time for a transition between system, hard irq
314  * or soft irq state.
315  */
316 void account_system_vtime(struct task_struct *tsk)
317 {
318 	u64 now, nowscaled, delta, deltascaled;
319 	unsigned long flags;
320 	u64 stolen, udelta, sys_scaled, user_scaled;
321 
322 	local_irq_save(flags);
323 	now = mftb();
324 	nowscaled = read_spurr(now);
325 	get_paca()->system_time += now - get_paca()->starttime;
326 	get_paca()->starttime = now;
327 	deltascaled = nowscaled - get_paca()->startspurr;
328 	get_paca()->startspurr = nowscaled;
329 
330 	stolen = calculate_stolen_time(now);
331 
332 	delta = get_paca()->system_time;
333 	get_paca()->system_time = 0;
334 	udelta = get_paca()->user_time - get_paca()->utime_sspurr;
335 	get_paca()->utime_sspurr = get_paca()->user_time;
336 
337 	/*
338 	 * Because we don't read the SPURR on every kernel entry/exit,
339 	 * deltascaled includes both user and system SPURR ticks.
340 	 * Apportion these ticks to system SPURR ticks and user
341 	 * SPURR ticks in the same ratio as the system time (delta)
342 	 * and user time (udelta) values obtained from the timebase
343 	 * over the same interval.  The system ticks get accounted here;
344 	 * the user ticks get saved up in paca->user_time_scaled to be
345 	 * used by account_process_tick.
346 	 */
347 	sys_scaled = delta;
348 	user_scaled = udelta;
349 	if (deltascaled != delta + udelta) {
350 		if (udelta) {
351 			sys_scaled = deltascaled * delta / (delta + udelta);
352 			user_scaled = deltascaled - sys_scaled;
353 		} else {
354 			sys_scaled = deltascaled;
355 		}
356 	}
357 	get_paca()->user_time_scaled += user_scaled;
358 
359 	if (in_irq() || idle_task(smp_processor_id()) != tsk) {
360 		account_system_time(tsk, 0, delta, sys_scaled);
361 		if (stolen)
362 			account_steal_time(stolen);
363 	} else {
364 		account_idle_time(delta + stolen);
365 	}
366 	local_irq_restore(flags);
367 }
368 EXPORT_SYMBOL_GPL(account_system_vtime);
369 
370 /*
371  * Transfer the user and system times accumulated in the paca
372  * by the exception entry and exit code to the generic process
373  * user and system time records.
374  * Must be called with interrupts disabled.
375  * Assumes that account_system_vtime() has been called recently
376  * (i.e. since the last entry from usermode) so that
377  * get_paca()->user_time_scaled is up to date.
378  */
379 void account_process_tick(struct task_struct *tsk, int user_tick)
380 {
381 	cputime_t utime, utimescaled;
382 
383 	utime = get_paca()->user_time;
384 	utimescaled = get_paca()->user_time_scaled;
385 	get_paca()->user_time = 0;
386 	get_paca()->user_time_scaled = 0;
387 	get_paca()->utime_sspurr = 0;
388 	account_user_time(tsk, utime, utimescaled);
389 }
390 
391 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
392 #define calc_cputime_factors()
393 #endif
394 
395 void __delay(unsigned long loops)
396 {
397 	unsigned long start;
398 	int diff;
399 
400 	if (__USE_RTC()) {
401 		start = get_rtcl();
402 		do {
403 			/* the RTCL register wraps at 1000000000 */
404 			diff = get_rtcl() - start;
405 			if (diff < 0)
406 				diff += 1000000000;
407 		} while (diff < loops);
408 	} else {
409 		start = get_tbl();
410 		while (get_tbl() - start < loops)
411 			HMT_low();
412 		HMT_medium();
413 	}
414 }
415 EXPORT_SYMBOL(__delay);
416 
417 void udelay(unsigned long usecs)
418 {
419 	__delay(tb_ticks_per_usec * usecs);
420 }
421 EXPORT_SYMBOL(udelay);
422 
423 #ifdef CONFIG_SMP
424 unsigned long profile_pc(struct pt_regs *regs)
425 {
426 	unsigned long pc = instruction_pointer(regs);
427 
428 	if (in_lock_functions(pc))
429 		return regs->link;
430 
431 	return pc;
432 }
433 EXPORT_SYMBOL(profile_pc);
434 #endif
435 
436 #ifdef CONFIG_PPC_ISERIES
437 
438 /*
439  * This function recalibrates the timebase based on the 49-bit time-of-day
440  * value in the Titan chip.  The Titan is much more accurate than the value
441  * returned by the service processor for the timebase frequency.
442  */
443 
444 static int __init iSeries_tb_recal(void)
445 {
446 	unsigned long titan, tb;
447 
448 	/* Make sure we only run on iSeries */
449 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
450 		return -ENODEV;
451 
452 	tb = get_tb();
453 	titan = HvCallXm_loadTod();
454 	if ( iSeries_recal_titan ) {
455 		unsigned long tb_ticks = tb - iSeries_recal_tb;
456 		unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
457 		unsigned long new_tb_ticks_per_sec   = (tb_ticks * USEC_PER_SEC)/titan_usec;
458 		unsigned long new_tb_ticks_per_jiffy =
459 			DIV_ROUND_CLOSEST(new_tb_ticks_per_sec, HZ);
460 		long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
461 		char sign = '+';
462 		/* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
463 		new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;
464 
465 		if ( tick_diff < 0 ) {
466 			tick_diff = -tick_diff;
467 			sign = '-';
468 		}
469 		if ( tick_diff ) {
470 			if ( tick_diff < tb_ticks_per_jiffy/25 ) {
471 				printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
472 						new_tb_ticks_per_jiffy, sign, tick_diff );
473 				tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
474 				tb_ticks_per_sec   = new_tb_ticks_per_sec;
475 				calc_cputime_factors();
476 				vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
477 				setup_cputime_one_jiffy();
478 			}
479 			else {
480 				printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
481 					"                   new tb_ticks_per_jiffy = %lu\n"
482 					"                   old tb_ticks_per_jiffy = %lu\n",
483 					new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
484 			}
485 		}
486 	}
487 	iSeries_recal_titan = titan;
488 	iSeries_recal_tb = tb;
489 
490 	/* Called here as now we know accurate values for the timebase */
491 	clocksource_init();
492 	return 0;
493 }
494 late_initcall(iSeries_tb_recal);
495 
496 /* Called from platform early init */
497 void __init iSeries_time_init_early(void)
498 {
499 	iSeries_recal_tb = get_tb();
500 	iSeries_recal_titan = HvCallXm_loadTod();
501 }
502 #endif /* CONFIG_PPC_ISERIES */
503 
504 #ifdef CONFIG_IRQ_WORK
505 
506 /*
507  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
508  */
509 #ifdef CONFIG_PPC64
510 static inline unsigned long test_irq_work_pending(void)
511 {
512 	unsigned long x;
513 
514 	asm volatile("lbz %0,%1(13)"
515 		: "=r" (x)
516 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
517 	return x;
518 }
519 
520 static inline void set_irq_work_pending_flag(void)
521 {
522 	asm volatile("stb %0,%1(13)" : :
523 		"r" (1),
524 		"i" (offsetof(struct paca_struct, irq_work_pending)));
525 }
526 
527 static inline void clear_irq_work_pending(void)
528 {
529 	asm volatile("stb %0,%1(13)" : :
530 		"r" (0),
531 		"i" (offsetof(struct paca_struct, irq_work_pending)));
532 }
533 
534 #else /* 32-bit */
535 
536 DEFINE_PER_CPU(u8, irq_work_pending);
537 
538 #define set_irq_work_pending_flag()	__get_cpu_var(irq_work_pending) = 1
539 #define test_irq_work_pending()		__get_cpu_var(irq_work_pending)
540 #define clear_irq_work_pending()	__get_cpu_var(irq_work_pending) = 0
541 
542 #endif /* 32 vs 64 bit */
543 
544 void set_irq_work_pending(void)
545 {
546 	preempt_disable();
547 	set_irq_work_pending_flag();
548 	set_dec(1);
549 	preempt_enable();
550 }
551 
552 #else  /* CONFIG_IRQ_WORK */
553 
554 #define test_irq_work_pending()	0
555 #define clear_irq_work_pending()
556 
557 #endif /* CONFIG_IRQ_WORK */
558 
559 /*
560  * For iSeries shared processors, we have to let the hypervisor
561  * set the hardware decrementer.  We set a virtual decrementer
562  * in the lppaca and call the hypervisor if the virtual
563  * decrementer is less than the current value in the hardware
564  * decrementer. (almost always the new decrementer value will
565  * be greater than the current hardware decementer so the hypervisor
566  * call will not be needed)
567  */
568 
569 /*
570  * timer_interrupt - gets called when the decrementer overflows,
571  * with interrupts disabled.
572  */
573 void timer_interrupt(struct pt_regs * regs)
574 {
575 	struct pt_regs *old_regs;
576 	struct decrementer_clock *decrementer =  &__get_cpu_var(decrementers);
577 	struct clock_event_device *evt = &decrementer->event;
578 	u64 now;
579 
580 	trace_timer_interrupt_entry(regs);
581 
582 	__get_cpu_var(irq_stat).timer_irqs++;
583 
584 	/* Ensure a positive value is written to the decrementer, or else
585 	 * some CPUs will continuue to take decrementer exceptions */
586 	set_dec(DECREMENTER_MAX);
587 
588 #if defined(CONFIG_PPC32) && defined(CONFIG_PMAC)
589 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
590 		do_IRQ(regs);
591 #endif
592 
593 	old_regs = set_irq_regs(regs);
594 	irq_enter();
595 
596 	if (test_irq_work_pending()) {
597 		clear_irq_work_pending();
598 		irq_work_run();
599 	}
600 
601 #ifdef CONFIG_PPC_ISERIES
602 	if (firmware_has_feature(FW_FEATURE_ISERIES))
603 		get_lppaca()->int_dword.fields.decr_int = 0;
604 #endif
605 
606 	now = get_tb_or_rtc();
607 	if (now >= decrementer->next_tb) {
608 		decrementer->next_tb = ~(u64)0;
609 		if (evt->event_handler)
610 			evt->event_handler(evt);
611 	} else {
612 		now = decrementer->next_tb - now;
613 		if (now <= DECREMENTER_MAX)
614 			set_dec((int)now);
615 	}
616 
617 #ifdef CONFIG_PPC_ISERIES
618 	if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending())
619 		process_hvlpevents();
620 #endif
621 
622 #ifdef CONFIG_PPC64
623 	/* collect purr register values often, for accurate calculations */
624 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
625 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
626 		cu->current_tb = mfspr(SPRN_PURR);
627 	}
628 #endif
629 
630 	irq_exit();
631 	set_irq_regs(old_regs);
632 
633 	trace_timer_interrupt_exit(regs);
634 }
635 
636 #ifdef CONFIG_SUSPEND
637 static void generic_suspend_disable_irqs(void)
638 {
639 	/* Disable the decrementer, so that it doesn't interfere
640 	 * with suspending.
641 	 */
642 
643 	set_dec(0x7fffffff);
644 	local_irq_disable();
645 	set_dec(0x7fffffff);
646 }
647 
648 static void generic_suspend_enable_irqs(void)
649 {
650 	local_irq_enable();
651 }
652 
653 /* Overrides the weak version in kernel/power/main.c */
654 void arch_suspend_disable_irqs(void)
655 {
656 	if (ppc_md.suspend_disable_irqs)
657 		ppc_md.suspend_disable_irqs();
658 	generic_suspend_disable_irqs();
659 }
660 
661 /* Overrides the weak version in kernel/power/main.c */
662 void arch_suspend_enable_irqs(void)
663 {
664 	generic_suspend_enable_irqs();
665 	if (ppc_md.suspend_enable_irqs)
666 		ppc_md.suspend_enable_irqs();
667 }
668 #endif
669 
670 /*
671  * Scheduler clock - returns current time in nanosec units.
672  *
673  * Note: mulhdu(a, b) (multiply high double unsigned) returns
674  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
675  * are 64-bit unsigned numbers.
676  */
677 unsigned long long sched_clock(void)
678 {
679 	if (__USE_RTC())
680 		return get_rtc();
681 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
682 }
683 
684 static int __init get_freq(char *name, int cells, unsigned long *val)
685 {
686 	struct device_node *cpu;
687 	const unsigned int *fp;
688 	int found = 0;
689 
690 	/* The cpu node should have timebase and clock frequency properties */
691 	cpu = of_find_node_by_type(NULL, "cpu");
692 
693 	if (cpu) {
694 		fp = of_get_property(cpu, name, NULL);
695 		if (fp) {
696 			found = 1;
697 			*val = of_read_ulong(fp, cells);
698 		}
699 
700 		of_node_put(cpu);
701 	}
702 
703 	return found;
704 }
705 
706 /* should become __cpuinit when secondary_cpu_time_init also is */
707 void start_cpu_decrementer(void)
708 {
709 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
710 	/* Clear any pending timer interrupts */
711 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
712 
713 	/* Enable decrementer interrupt */
714 	mtspr(SPRN_TCR, TCR_DIE);
715 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
716 }
717 
718 void __init generic_calibrate_decr(void)
719 {
720 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
721 
722 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
723 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
724 
725 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
726 				"(not found)\n");
727 	}
728 
729 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
730 
731 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
732 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
733 
734 		printk(KERN_ERR "WARNING: Estimating processor frequency "
735 				"(not found)\n");
736 	}
737 }
738 
739 int update_persistent_clock(struct timespec now)
740 {
741 	struct rtc_time tm;
742 
743 	if (!ppc_md.set_rtc_time)
744 		return 0;
745 
746 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
747 	tm.tm_year -= 1900;
748 	tm.tm_mon -= 1;
749 
750 	return ppc_md.set_rtc_time(&tm);
751 }
752 
753 static void __read_persistent_clock(struct timespec *ts)
754 {
755 	struct rtc_time tm;
756 	static int first = 1;
757 
758 	ts->tv_nsec = 0;
759 	/* XXX this is a litle fragile but will work okay in the short term */
760 	if (first) {
761 		first = 0;
762 		if (ppc_md.time_init)
763 			timezone_offset = ppc_md.time_init();
764 
765 		/* get_boot_time() isn't guaranteed to be safe to call late */
766 		if (ppc_md.get_boot_time) {
767 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
768 			return;
769 		}
770 	}
771 	if (!ppc_md.get_rtc_time) {
772 		ts->tv_sec = 0;
773 		return;
774 	}
775 	ppc_md.get_rtc_time(&tm);
776 
777 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
778 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
779 }
780 
781 void read_persistent_clock(struct timespec *ts)
782 {
783 	__read_persistent_clock(ts);
784 
785 	/* Sanitize it in case real time clock is set below EPOCH */
786 	if (ts->tv_sec < 0) {
787 		ts->tv_sec = 0;
788 		ts->tv_nsec = 0;
789 	}
790 
791 }
792 
793 /* clocksource code */
794 static cycle_t rtc_read(struct clocksource *cs)
795 {
796 	return (cycle_t)get_rtc();
797 }
798 
799 static cycle_t timebase_read(struct clocksource *cs)
800 {
801 	return (cycle_t)get_tb();
802 }
803 
804 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
805 			struct clocksource *clock, u32 mult)
806 {
807 	u64 new_tb_to_xs, new_stamp_xsec;
808 	u32 frac_sec;
809 
810 	if (clock != &clocksource_timebase)
811 		return;
812 
813 	/* Make userspace gettimeofday spin until we're done. */
814 	++vdso_data->tb_update_count;
815 	smp_mb();
816 
817 	/* XXX this assumes clock->shift == 22 */
818 	/* 4611686018 ~= 2^(20+64-22) / 1e9 */
819 	new_tb_to_xs = (u64) mult * 4611686018ULL;
820 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
821 	do_div(new_stamp_xsec, 1000000000);
822 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
823 
824 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
825 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
826 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
827 
828 	/*
829 	 * tb_update_count is used to allow the userspace gettimeofday code
830 	 * to assure itself that it sees a consistent view of the tb_to_xs and
831 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
832 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
833 	 * the two values of tb_update_count match and are even then the
834 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
835 	 * loops back and reads them again until this criteria is met.
836 	 * We expect the caller to have done the first increment of
837 	 * vdso_data->tb_update_count already.
838 	 */
839 	vdso_data->tb_orig_stamp = clock->cycle_last;
840 	vdso_data->stamp_xsec = new_stamp_xsec;
841 	vdso_data->tb_to_xs = new_tb_to_xs;
842 	vdso_data->wtom_clock_sec = wtm->tv_sec;
843 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
844 	vdso_data->stamp_xtime = *wall_time;
845 	vdso_data->stamp_sec_fraction = frac_sec;
846 	smp_wmb();
847 	++(vdso_data->tb_update_count);
848 }
849 
850 void update_vsyscall_tz(void)
851 {
852 	/* Make userspace gettimeofday spin until we're done. */
853 	++vdso_data->tb_update_count;
854 	smp_mb();
855 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
856 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
857 	smp_mb();
858 	++vdso_data->tb_update_count;
859 }
860 
861 static void __init clocksource_init(void)
862 {
863 	struct clocksource *clock;
864 
865 	if (__USE_RTC())
866 		clock = &clocksource_rtc;
867 	else
868 		clock = &clocksource_timebase;
869 
870 	clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
871 
872 	if (clocksource_register(clock)) {
873 		printk(KERN_ERR "clocksource: %s is already registered\n",
874 		       clock->name);
875 		return;
876 	}
877 
878 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
879 	       clock->name, clock->mult, clock->shift);
880 }
881 
882 static int decrementer_set_next_event(unsigned long evt,
883 				      struct clock_event_device *dev)
884 {
885 	__get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
886 	set_dec(evt);
887 	return 0;
888 }
889 
890 static void decrementer_set_mode(enum clock_event_mode mode,
891 				 struct clock_event_device *dev)
892 {
893 	if (mode != CLOCK_EVT_MODE_ONESHOT)
894 		decrementer_set_next_event(DECREMENTER_MAX, dev);
895 }
896 
897 static inline uint64_t div_sc64(unsigned long ticks, unsigned long nsec,
898 				int shift)
899 {
900 	uint64_t tmp = ((uint64_t)ticks) << shift;
901 
902 	do_div(tmp, nsec);
903 	return tmp;
904 }
905 
906 static void __init setup_clockevent_multiplier(unsigned long hz)
907 {
908 	u64 mult, shift = 32;
909 
910 	while (1) {
911 		mult = div_sc64(hz, NSEC_PER_SEC, shift);
912 		if (mult && (mult >> 32UL) == 0UL)
913 			break;
914 
915 		shift--;
916 	}
917 
918 	decrementer_clockevent.shift = shift;
919 	decrementer_clockevent.mult = mult;
920 }
921 
922 static void register_decrementer_clockevent(int cpu)
923 {
924 	struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
925 
926 	*dec = decrementer_clockevent;
927 	dec->cpumask = cpumask_of(cpu);
928 
929 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
930 		    dec->name, dec->mult, dec->shift, cpu);
931 
932 	clockevents_register_device(dec);
933 }
934 
935 static void __init init_decrementer_clockevent(void)
936 {
937 	int cpu = smp_processor_id();
938 
939 	setup_clockevent_multiplier(ppc_tb_freq);
940 	decrementer_clockevent.max_delta_ns =
941 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
942 	decrementer_clockevent.min_delta_ns =
943 		clockevent_delta2ns(2, &decrementer_clockevent);
944 
945 	register_decrementer_clockevent(cpu);
946 }
947 
948 void secondary_cpu_time_init(void)
949 {
950 	/* Start the decrementer on CPUs that have manual control
951 	 * such as BookE
952 	 */
953 	start_cpu_decrementer();
954 
955 	/* FIME: Should make unrelatred change to move snapshot_timebase
956 	 * call here ! */
957 	register_decrementer_clockevent(smp_processor_id());
958 }
959 
960 /* This function is only called on the boot processor */
961 void __init time_init(void)
962 {
963 	struct div_result res;
964 	u64 scale;
965 	unsigned shift;
966 
967 	if (__USE_RTC()) {
968 		/* 601 processor: dec counts down by 128 every 128ns */
969 		ppc_tb_freq = 1000000000;
970 	} else {
971 		/* Normal PowerPC with timebase register */
972 		ppc_md.calibrate_decr();
973 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
974 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
975 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
976 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
977 	}
978 
979 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
980 	tb_ticks_per_sec = ppc_tb_freq;
981 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
982 	calc_cputime_factors();
983 	setup_cputime_one_jiffy();
984 
985 	/*
986 	 * Compute scale factor for sched_clock.
987 	 * The calibrate_decr() function has set tb_ticks_per_sec,
988 	 * which is the timebase frequency.
989 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
990 	 * the 128-bit result as a 64.64 fixed-point number.
991 	 * We then shift that number right until it is less than 1.0,
992 	 * giving us the scale factor and shift count to use in
993 	 * sched_clock().
994 	 */
995 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
996 	scale = res.result_low;
997 	for (shift = 0; res.result_high != 0; ++shift) {
998 		scale = (scale >> 1) | (res.result_high << 63);
999 		res.result_high >>= 1;
1000 	}
1001 	tb_to_ns_scale = scale;
1002 	tb_to_ns_shift = shift;
1003 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1004 	boot_tb = get_tb_or_rtc();
1005 
1006 	/* If platform provided a timezone (pmac), we correct the time */
1007         if (timezone_offset) {
1008 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1009 		sys_tz.tz_dsttime = 0;
1010         }
1011 
1012 	vdso_data->tb_update_count = 0;
1013 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1014 
1015 	/* Start the decrementer on CPUs that have manual control
1016 	 * such as BookE
1017 	 */
1018 	start_cpu_decrementer();
1019 
1020 	/* Register the clocksource, if we're not running on iSeries */
1021 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
1022 		clocksource_init();
1023 
1024 	init_decrementer_clockevent();
1025 }
1026 
1027 
1028 #define FEBRUARY	2
1029 #define	STARTOFTIME	1970
1030 #define SECDAY		86400L
1031 #define SECYR		(SECDAY * 365)
1032 #define	leapyear(year)		((year) % 4 == 0 && \
1033 				 ((year) % 100 != 0 || (year) % 400 == 0))
1034 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1035 #define	days_in_month(a) 	(month_days[(a) - 1])
1036 
1037 static int month_days[12] = {
1038 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1039 };
1040 
1041 /*
1042  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1043  */
1044 void GregorianDay(struct rtc_time * tm)
1045 {
1046 	int leapsToDate;
1047 	int lastYear;
1048 	int day;
1049 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1050 
1051 	lastYear = tm->tm_year - 1;
1052 
1053 	/*
1054 	 * Number of leap corrections to apply up to end of last year
1055 	 */
1056 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
1057 
1058 	/*
1059 	 * This year is a leap year if it is divisible by 4 except when it is
1060 	 * divisible by 100 unless it is divisible by 400
1061 	 *
1062 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1063 	 */
1064 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
1065 
1066 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
1067 		   tm->tm_mday;
1068 
1069 	tm->tm_wday = day % 7;
1070 }
1071 
1072 void to_tm(int tim, struct rtc_time * tm)
1073 {
1074 	register int    i;
1075 	register long   hms, day;
1076 
1077 	day = tim / SECDAY;
1078 	hms = tim % SECDAY;
1079 
1080 	/* Hours, minutes, seconds are easy */
1081 	tm->tm_hour = hms / 3600;
1082 	tm->tm_min = (hms % 3600) / 60;
1083 	tm->tm_sec = (hms % 3600) % 60;
1084 
1085 	/* Number of years in days */
1086 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1087 		day -= days_in_year(i);
1088 	tm->tm_year = i;
1089 
1090 	/* Number of months in days left */
1091 	if (leapyear(tm->tm_year))
1092 		days_in_month(FEBRUARY) = 29;
1093 	for (i = 1; day >= days_in_month(i); i++)
1094 		day -= days_in_month(i);
1095 	days_in_month(FEBRUARY) = 28;
1096 	tm->tm_mon = i;
1097 
1098 	/* Days are what is left over (+1) from all that. */
1099 	tm->tm_mday = day + 1;
1100 
1101 	/*
1102 	 * Determine the day of week
1103 	 */
1104 	GregorianDay(tm);
1105 }
1106 
1107 /*
1108  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1109  * result.
1110  */
1111 void div128_by_32(u64 dividend_high, u64 dividend_low,
1112 		  unsigned divisor, struct div_result *dr)
1113 {
1114 	unsigned long a, b, c, d;
1115 	unsigned long w, x, y, z;
1116 	u64 ra, rb, rc;
1117 
1118 	a = dividend_high >> 32;
1119 	b = dividend_high & 0xffffffff;
1120 	c = dividend_low >> 32;
1121 	d = dividend_low & 0xffffffff;
1122 
1123 	w = a / divisor;
1124 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1125 
1126 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1127 	x = ra;
1128 
1129 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1130 	y = rb;
1131 
1132 	do_div(rc, divisor);
1133 	z = rc;
1134 
1135 	dr->result_high = ((u64)w << 32) + x;
1136 	dr->result_low  = ((u64)y << 32) + z;
1137 
1138 }
1139 
1140 /* We don't need to calibrate delay, we use the CPU timebase for that */
1141 void calibrate_delay(void)
1142 {
1143 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1144 	 * as the number of __delay(1) in a jiffy, so make it so
1145 	 */
1146 	loops_per_jiffy = tb_ticks_per_jiffy;
1147 }
1148 
1149 static int __init rtc_init(void)
1150 {
1151 	struct platform_device *pdev;
1152 
1153 	if (!ppc_md.get_rtc_time)
1154 		return -ENODEV;
1155 
1156 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1157 	if (IS_ERR(pdev))
1158 		return PTR_ERR(pdev);
1159 
1160 	return 0;
1161 }
1162 
1163 module_init(rtc_init);
1164