xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision 55ec2fca3e99f83b5c674e9aba713d848392f6cc)
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/perf_event.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 unsigned long 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 	sst = scan_dispatch_log(get_paca()->starttime_user);
269 	ust = scan_dispatch_log(get_paca()->starttime);
270 	get_paca()->system_time -= sst;
271 	get_paca()->user_time -= ust;
272 	get_paca()->stolen_time += ust + sst;
273 }
274 
275 static inline u64 calculate_stolen_time(u64 stop_tb)
276 {
277 	u64 stolen = 0;
278 
279 	if (get_paca()->dtl_ridx != get_paca()->lppaca_ptr->dtl_idx) {
280 		stolen = scan_dispatch_log(stop_tb);
281 		get_paca()->system_time -= stolen;
282 	}
283 
284 	stolen += get_paca()->stolen_time;
285 	get_paca()->stolen_time = 0;
286 	return stolen;
287 }
288 
289 #else /* CONFIG_PPC_SPLPAR */
290 static inline u64 calculate_stolen_time(u64 stop_tb)
291 {
292 	return 0;
293 }
294 
295 #endif /* CONFIG_PPC_SPLPAR */
296 
297 /*
298  * Account time for a transition between system, hard irq
299  * or soft irq state.
300  */
301 void account_system_vtime(struct task_struct *tsk)
302 {
303 	u64 now, nowscaled, delta, deltascaled;
304 	unsigned long flags;
305 	u64 stolen, udelta, sys_scaled, user_scaled;
306 
307 	local_irq_save(flags);
308 	now = mftb();
309 	nowscaled = read_spurr(now);
310 	get_paca()->system_time += now - get_paca()->starttime;
311 	get_paca()->starttime = now;
312 	deltascaled = nowscaled - get_paca()->startspurr;
313 	get_paca()->startspurr = nowscaled;
314 
315 	stolen = calculate_stolen_time(now);
316 
317 	delta = get_paca()->system_time;
318 	get_paca()->system_time = 0;
319 	udelta = get_paca()->user_time - get_paca()->utime_sspurr;
320 	get_paca()->utime_sspurr = get_paca()->user_time;
321 
322 	/*
323 	 * Because we don't read the SPURR on every kernel entry/exit,
324 	 * deltascaled includes both user and system SPURR ticks.
325 	 * Apportion these ticks to system SPURR ticks and user
326 	 * SPURR ticks in the same ratio as the system time (delta)
327 	 * and user time (udelta) values obtained from the timebase
328 	 * over the same interval.  The system ticks get accounted here;
329 	 * the user ticks get saved up in paca->user_time_scaled to be
330 	 * used by account_process_tick.
331 	 */
332 	sys_scaled = delta;
333 	user_scaled = udelta;
334 	if (deltascaled != delta + udelta) {
335 		if (udelta) {
336 			sys_scaled = deltascaled * delta / (delta + udelta);
337 			user_scaled = deltascaled - sys_scaled;
338 		} else {
339 			sys_scaled = deltascaled;
340 		}
341 	}
342 	get_paca()->user_time_scaled += user_scaled;
343 
344 	if (in_irq() || idle_task(smp_processor_id()) != tsk) {
345 		account_system_time(tsk, 0, delta, sys_scaled);
346 		if (stolen)
347 			account_steal_time(stolen);
348 	} else {
349 		account_idle_time(delta + stolen);
350 	}
351 	local_irq_restore(flags);
352 }
353 EXPORT_SYMBOL_GPL(account_system_vtime);
354 
355 /*
356  * Transfer the user and system times accumulated in the paca
357  * by the exception entry and exit code to the generic process
358  * user and system time records.
359  * Must be called with interrupts disabled.
360  * Assumes that account_system_vtime() has been called recently
361  * (i.e. since the last entry from usermode) so that
362  * get_paca()->user_time_scaled is up to date.
363  */
364 void account_process_tick(struct task_struct *tsk, int user_tick)
365 {
366 	cputime_t utime, utimescaled;
367 
368 	utime = get_paca()->user_time;
369 	utimescaled = get_paca()->user_time_scaled;
370 	get_paca()->user_time = 0;
371 	get_paca()->user_time_scaled = 0;
372 	get_paca()->utime_sspurr = 0;
373 	account_user_time(tsk, utime, utimescaled);
374 }
375 
376 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
377 #define calc_cputime_factors()
378 #endif
379 
380 void __delay(unsigned long loops)
381 {
382 	unsigned long start;
383 	int diff;
384 
385 	if (__USE_RTC()) {
386 		start = get_rtcl();
387 		do {
388 			/* the RTCL register wraps at 1000000000 */
389 			diff = get_rtcl() - start;
390 			if (diff < 0)
391 				diff += 1000000000;
392 		} while (diff < loops);
393 	} else {
394 		start = get_tbl();
395 		while (get_tbl() - start < loops)
396 			HMT_low();
397 		HMT_medium();
398 	}
399 }
400 EXPORT_SYMBOL(__delay);
401 
402 void udelay(unsigned long usecs)
403 {
404 	__delay(tb_ticks_per_usec * usecs);
405 }
406 EXPORT_SYMBOL(udelay);
407 
408 #ifdef CONFIG_SMP
409 unsigned long profile_pc(struct pt_regs *regs)
410 {
411 	unsigned long pc = instruction_pointer(regs);
412 
413 	if (in_lock_functions(pc))
414 		return regs->link;
415 
416 	return pc;
417 }
418 EXPORT_SYMBOL(profile_pc);
419 #endif
420 
421 #ifdef CONFIG_PPC_ISERIES
422 
423 /*
424  * This function recalibrates the timebase based on the 49-bit time-of-day
425  * value in the Titan chip.  The Titan is much more accurate than the value
426  * returned by the service processor for the timebase frequency.
427  */
428 
429 static int __init iSeries_tb_recal(void)
430 {
431 	unsigned long titan, tb;
432 
433 	/* Make sure we only run on iSeries */
434 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
435 		return -ENODEV;
436 
437 	tb = get_tb();
438 	titan = HvCallXm_loadTod();
439 	if ( iSeries_recal_titan ) {
440 		unsigned long tb_ticks = tb - iSeries_recal_tb;
441 		unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12;
442 		unsigned long new_tb_ticks_per_sec   = (tb_ticks * USEC_PER_SEC)/titan_usec;
443 		unsigned long new_tb_ticks_per_jiffy =
444 			DIV_ROUND_CLOSEST(new_tb_ticks_per_sec, HZ);
445 		long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy;
446 		char sign = '+';
447 		/* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */
448 		new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ;
449 
450 		if ( tick_diff < 0 ) {
451 			tick_diff = -tick_diff;
452 			sign = '-';
453 		}
454 		if ( tick_diff ) {
455 			if ( tick_diff < tb_ticks_per_jiffy/25 ) {
456 				printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n",
457 						new_tb_ticks_per_jiffy, sign, tick_diff );
458 				tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
459 				tb_ticks_per_sec   = new_tb_ticks_per_sec;
460 				calc_cputime_factors();
461 				vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
462 				setup_cputime_one_jiffy();
463 			}
464 			else {
465 				printk( "Titan recalibrate: FAILED (difference > 4 percent)\n"
466 					"                   new tb_ticks_per_jiffy = %lu\n"
467 					"                   old tb_ticks_per_jiffy = %lu\n",
468 					new_tb_ticks_per_jiffy, tb_ticks_per_jiffy );
469 			}
470 		}
471 	}
472 	iSeries_recal_titan = titan;
473 	iSeries_recal_tb = tb;
474 
475 	/* Called here as now we know accurate values for the timebase */
476 	clocksource_init();
477 	return 0;
478 }
479 late_initcall(iSeries_tb_recal);
480 
481 /* Called from platform early init */
482 void __init iSeries_time_init_early(void)
483 {
484 	iSeries_recal_tb = get_tb();
485 	iSeries_recal_titan = HvCallXm_loadTod();
486 }
487 #endif /* CONFIG_PPC_ISERIES */
488 
489 #ifdef CONFIG_PERF_EVENTS
490 
491 /*
492  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
493  */
494 #ifdef CONFIG_PPC64
495 static inline unsigned long test_perf_event_pending(void)
496 {
497 	unsigned long x;
498 
499 	asm volatile("lbz %0,%1(13)"
500 		: "=r" (x)
501 		: "i" (offsetof(struct paca_struct, perf_event_pending)));
502 	return x;
503 }
504 
505 static inline void set_perf_event_pending_flag(void)
506 {
507 	asm volatile("stb %0,%1(13)" : :
508 		"r" (1),
509 		"i" (offsetof(struct paca_struct, perf_event_pending)));
510 }
511 
512 static inline void clear_perf_event_pending(void)
513 {
514 	asm volatile("stb %0,%1(13)" : :
515 		"r" (0),
516 		"i" (offsetof(struct paca_struct, perf_event_pending)));
517 }
518 
519 #else /* 32-bit */
520 
521 DEFINE_PER_CPU(u8, perf_event_pending);
522 
523 #define set_perf_event_pending_flag()	__get_cpu_var(perf_event_pending) = 1
524 #define test_perf_event_pending()	__get_cpu_var(perf_event_pending)
525 #define clear_perf_event_pending()	__get_cpu_var(perf_event_pending) = 0
526 
527 #endif /* 32 vs 64 bit */
528 
529 void set_perf_event_pending(void)
530 {
531 	preempt_disable();
532 	set_perf_event_pending_flag();
533 	set_dec(1);
534 	preempt_enable();
535 }
536 
537 #else  /* CONFIG_PERF_EVENTS */
538 
539 #define test_perf_event_pending()	0
540 #define clear_perf_event_pending()
541 
542 #endif /* CONFIG_PERF_EVENTS */
543 
544 /*
545  * For iSeries shared processors, we have to let the hypervisor
546  * set the hardware decrementer.  We set a virtual decrementer
547  * in the lppaca and call the hypervisor if the virtual
548  * decrementer is less than the current value in the hardware
549  * decrementer. (almost always the new decrementer value will
550  * be greater than the current hardware decementer so the hypervisor
551  * call will not be needed)
552  */
553 
554 /*
555  * timer_interrupt - gets called when the decrementer overflows,
556  * with interrupts disabled.
557  */
558 void timer_interrupt(struct pt_regs * regs)
559 {
560 	struct pt_regs *old_regs;
561 	struct decrementer_clock *decrementer =  &__get_cpu_var(decrementers);
562 	struct clock_event_device *evt = &decrementer->event;
563 	u64 now;
564 
565 	trace_timer_interrupt_entry(regs);
566 
567 	__get_cpu_var(irq_stat).timer_irqs++;
568 
569 	/* Ensure a positive value is written to the decrementer, or else
570 	 * some CPUs will continuue to take decrementer exceptions */
571 	set_dec(DECREMENTER_MAX);
572 
573 #if defined(CONFIG_PPC32) && defined(CONFIG_PMAC)
574 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
575 		do_IRQ(regs);
576 #endif
577 
578 	old_regs = set_irq_regs(regs);
579 	irq_enter();
580 
581 	if (test_perf_event_pending()) {
582 		clear_perf_event_pending();
583 		perf_event_do_pending();
584 	}
585 
586 #ifdef CONFIG_PPC_ISERIES
587 	if (firmware_has_feature(FW_FEATURE_ISERIES))
588 		get_lppaca()->int_dword.fields.decr_int = 0;
589 #endif
590 
591 	now = get_tb_or_rtc();
592 	if (now >= decrementer->next_tb) {
593 		decrementer->next_tb = ~(u64)0;
594 		if (evt->event_handler)
595 			evt->event_handler(evt);
596 	} else {
597 		now = decrementer->next_tb - now;
598 		if (now <= DECREMENTER_MAX)
599 			set_dec((int)now);
600 	}
601 
602 #ifdef CONFIG_PPC_ISERIES
603 	if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending())
604 		process_hvlpevents();
605 #endif
606 
607 #ifdef CONFIG_PPC64
608 	/* collect purr register values often, for accurate calculations */
609 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
610 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
611 		cu->current_tb = mfspr(SPRN_PURR);
612 	}
613 #endif
614 
615 	irq_exit();
616 	set_irq_regs(old_regs);
617 
618 	trace_timer_interrupt_exit(regs);
619 }
620 
621 #ifdef CONFIG_SUSPEND
622 static void generic_suspend_disable_irqs(void)
623 {
624 	/* Disable the decrementer, so that it doesn't interfere
625 	 * with suspending.
626 	 */
627 
628 	set_dec(0x7fffffff);
629 	local_irq_disable();
630 	set_dec(0x7fffffff);
631 }
632 
633 static void generic_suspend_enable_irqs(void)
634 {
635 	local_irq_enable();
636 }
637 
638 /* Overrides the weak version in kernel/power/main.c */
639 void arch_suspend_disable_irqs(void)
640 {
641 	if (ppc_md.suspend_disable_irqs)
642 		ppc_md.suspend_disable_irqs();
643 	generic_suspend_disable_irqs();
644 }
645 
646 /* Overrides the weak version in kernel/power/main.c */
647 void arch_suspend_enable_irqs(void)
648 {
649 	generic_suspend_enable_irqs();
650 	if (ppc_md.suspend_enable_irqs)
651 		ppc_md.suspend_enable_irqs();
652 }
653 #endif
654 
655 /*
656  * Scheduler clock - returns current time in nanosec units.
657  *
658  * Note: mulhdu(a, b) (multiply high double unsigned) returns
659  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
660  * are 64-bit unsigned numbers.
661  */
662 unsigned long long sched_clock(void)
663 {
664 	if (__USE_RTC())
665 		return get_rtc();
666 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
667 }
668 
669 static int __init get_freq(char *name, int cells, unsigned long *val)
670 {
671 	struct device_node *cpu;
672 	const unsigned int *fp;
673 	int found = 0;
674 
675 	/* The cpu node should have timebase and clock frequency properties */
676 	cpu = of_find_node_by_type(NULL, "cpu");
677 
678 	if (cpu) {
679 		fp = of_get_property(cpu, name, NULL);
680 		if (fp) {
681 			found = 1;
682 			*val = of_read_ulong(fp, cells);
683 		}
684 
685 		of_node_put(cpu);
686 	}
687 
688 	return found;
689 }
690 
691 /* should become __cpuinit when secondary_cpu_time_init also is */
692 void start_cpu_decrementer(void)
693 {
694 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
695 	/* Clear any pending timer interrupts */
696 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
697 
698 	/* Enable decrementer interrupt */
699 	mtspr(SPRN_TCR, TCR_DIE);
700 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
701 }
702 
703 void __init generic_calibrate_decr(void)
704 {
705 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
706 
707 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
708 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
709 
710 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
711 				"(not found)\n");
712 	}
713 
714 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
715 
716 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
717 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
718 
719 		printk(KERN_ERR "WARNING: Estimating processor frequency "
720 				"(not found)\n");
721 	}
722 }
723 
724 int update_persistent_clock(struct timespec now)
725 {
726 	struct rtc_time tm;
727 
728 	if (!ppc_md.set_rtc_time)
729 		return 0;
730 
731 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
732 	tm.tm_year -= 1900;
733 	tm.tm_mon -= 1;
734 
735 	return ppc_md.set_rtc_time(&tm);
736 }
737 
738 static void __read_persistent_clock(struct timespec *ts)
739 {
740 	struct rtc_time tm;
741 	static int first = 1;
742 
743 	ts->tv_nsec = 0;
744 	/* XXX this is a litle fragile but will work okay in the short term */
745 	if (first) {
746 		first = 0;
747 		if (ppc_md.time_init)
748 			timezone_offset = ppc_md.time_init();
749 
750 		/* get_boot_time() isn't guaranteed to be safe to call late */
751 		if (ppc_md.get_boot_time) {
752 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
753 			return;
754 		}
755 	}
756 	if (!ppc_md.get_rtc_time) {
757 		ts->tv_sec = 0;
758 		return;
759 	}
760 	ppc_md.get_rtc_time(&tm);
761 
762 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
763 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
764 }
765 
766 void read_persistent_clock(struct timespec *ts)
767 {
768 	__read_persistent_clock(ts);
769 
770 	/* Sanitize it in case real time clock is set below EPOCH */
771 	if (ts->tv_sec < 0) {
772 		ts->tv_sec = 0;
773 		ts->tv_nsec = 0;
774 	}
775 
776 }
777 
778 /* clocksource code */
779 static cycle_t rtc_read(struct clocksource *cs)
780 {
781 	return (cycle_t)get_rtc();
782 }
783 
784 static cycle_t timebase_read(struct clocksource *cs)
785 {
786 	return (cycle_t)get_tb();
787 }
788 
789 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
790 			struct clocksource *clock, u32 mult)
791 {
792 	u64 new_tb_to_xs, new_stamp_xsec;
793 	u32 frac_sec;
794 
795 	if (clock != &clocksource_timebase)
796 		return;
797 
798 	/* Make userspace gettimeofday spin until we're done. */
799 	++vdso_data->tb_update_count;
800 	smp_mb();
801 
802 	/* XXX this assumes clock->shift == 22 */
803 	/* 4611686018 ~= 2^(20+64-22) / 1e9 */
804 	new_tb_to_xs = (u64) mult * 4611686018ULL;
805 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
806 	do_div(new_stamp_xsec, 1000000000);
807 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
808 
809 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
810 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
811 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
812 
813 	/*
814 	 * tb_update_count is used to allow the userspace gettimeofday code
815 	 * to assure itself that it sees a consistent view of the tb_to_xs and
816 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
817 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
818 	 * the two values of tb_update_count match and are even then the
819 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
820 	 * loops back and reads them again until this criteria is met.
821 	 * We expect the caller to have done the first increment of
822 	 * vdso_data->tb_update_count already.
823 	 */
824 	vdso_data->tb_orig_stamp = clock->cycle_last;
825 	vdso_data->stamp_xsec = new_stamp_xsec;
826 	vdso_data->tb_to_xs = new_tb_to_xs;
827 	vdso_data->wtom_clock_sec = wtm->tv_sec;
828 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
829 	vdso_data->stamp_xtime = *wall_time;
830 	vdso_data->stamp_sec_fraction = frac_sec;
831 	smp_wmb();
832 	++(vdso_data->tb_update_count);
833 }
834 
835 void update_vsyscall_tz(void)
836 {
837 	/* Make userspace gettimeofday spin until we're done. */
838 	++vdso_data->tb_update_count;
839 	smp_mb();
840 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
841 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
842 	smp_mb();
843 	++vdso_data->tb_update_count;
844 }
845 
846 static void __init clocksource_init(void)
847 {
848 	struct clocksource *clock;
849 
850 	if (__USE_RTC())
851 		clock = &clocksource_rtc;
852 	else
853 		clock = &clocksource_timebase;
854 
855 	clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift);
856 
857 	if (clocksource_register(clock)) {
858 		printk(KERN_ERR "clocksource: %s is already registered\n",
859 		       clock->name);
860 		return;
861 	}
862 
863 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
864 	       clock->name, clock->mult, clock->shift);
865 }
866 
867 static int decrementer_set_next_event(unsigned long evt,
868 				      struct clock_event_device *dev)
869 {
870 	__get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt;
871 	set_dec(evt);
872 	return 0;
873 }
874 
875 static void decrementer_set_mode(enum clock_event_mode mode,
876 				 struct clock_event_device *dev)
877 {
878 	if (mode != CLOCK_EVT_MODE_ONESHOT)
879 		decrementer_set_next_event(DECREMENTER_MAX, dev);
880 }
881 
882 static inline uint64_t div_sc64(unsigned long ticks, unsigned long nsec,
883 				int shift)
884 {
885 	uint64_t tmp = ((uint64_t)ticks) << shift;
886 
887 	do_div(tmp, nsec);
888 	return tmp;
889 }
890 
891 static void __init setup_clockevent_multiplier(unsigned long hz)
892 {
893 	u64 mult, shift = 32;
894 
895 	while (1) {
896 		mult = div_sc64(hz, NSEC_PER_SEC, shift);
897 		if (mult && (mult >> 32UL) == 0UL)
898 			break;
899 
900 		shift--;
901 	}
902 
903 	decrementer_clockevent.shift = shift;
904 	decrementer_clockevent.mult = mult;
905 }
906 
907 static void register_decrementer_clockevent(int cpu)
908 {
909 	struct clock_event_device *dec = &per_cpu(decrementers, cpu).event;
910 
911 	*dec = decrementer_clockevent;
912 	dec->cpumask = cpumask_of(cpu);
913 
914 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
915 		    dec->name, dec->mult, dec->shift, cpu);
916 
917 	clockevents_register_device(dec);
918 }
919 
920 static void __init init_decrementer_clockevent(void)
921 {
922 	int cpu = smp_processor_id();
923 
924 	setup_clockevent_multiplier(ppc_tb_freq);
925 	decrementer_clockevent.max_delta_ns =
926 		clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
927 	decrementer_clockevent.min_delta_ns =
928 		clockevent_delta2ns(2, &decrementer_clockevent);
929 
930 	register_decrementer_clockevent(cpu);
931 }
932 
933 void secondary_cpu_time_init(void)
934 {
935 	/* Start the decrementer on CPUs that have manual control
936 	 * such as BookE
937 	 */
938 	start_cpu_decrementer();
939 
940 	/* FIME: Should make unrelatred change to move snapshot_timebase
941 	 * call here ! */
942 	register_decrementer_clockevent(smp_processor_id());
943 }
944 
945 /* This function is only called on the boot processor */
946 void __init time_init(void)
947 {
948 	struct div_result res;
949 	u64 scale;
950 	unsigned shift;
951 
952 	if (__USE_RTC()) {
953 		/* 601 processor: dec counts down by 128 every 128ns */
954 		ppc_tb_freq = 1000000000;
955 	} else {
956 		/* Normal PowerPC with timebase register */
957 		ppc_md.calibrate_decr();
958 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
959 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
960 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
961 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
962 	}
963 
964 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
965 	tb_ticks_per_sec = ppc_tb_freq;
966 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
967 	calc_cputime_factors();
968 	setup_cputime_one_jiffy();
969 
970 	/*
971 	 * Compute scale factor for sched_clock.
972 	 * The calibrate_decr() function has set tb_ticks_per_sec,
973 	 * which is the timebase frequency.
974 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
975 	 * the 128-bit result as a 64.64 fixed-point number.
976 	 * We then shift that number right until it is less than 1.0,
977 	 * giving us the scale factor and shift count to use in
978 	 * sched_clock().
979 	 */
980 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
981 	scale = res.result_low;
982 	for (shift = 0; res.result_high != 0; ++shift) {
983 		scale = (scale >> 1) | (res.result_high << 63);
984 		res.result_high >>= 1;
985 	}
986 	tb_to_ns_scale = scale;
987 	tb_to_ns_shift = shift;
988 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
989 	boot_tb = get_tb_or_rtc();
990 
991 	/* If platform provided a timezone (pmac), we correct the time */
992         if (timezone_offset) {
993 		sys_tz.tz_minuteswest = -timezone_offset / 60;
994 		sys_tz.tz_dsttime = 0;
995         }
996 
997 	vdso_data->tb_update_count = 0;
998 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
999 
1000 	/* Start the decrementer on CPUs that have manual control
1001 	 * such as BookE
1002 	 */
1003 	start_cpu_decrementer();
1004 
1005 	/* Register the clocksource, if we're not running on iSeries */
1006 	if (!firmware_has_feature(FW_FEATURE_ISERIES))
1007 		clocksource_init();
1008 
1009 	init_decrementer_clockevent();
1010 }
1011 
1012 
1013 #define FEBRUARY	2
1014 #define	STARTOFTIME	1970
1015 #define SECDAY		86400L
1016 #define SECYR		(SECDAY * 365)
1017 #define	leapyear(year)		((year) % 4 == 0 && \
1018 				 ((year) % 100 != 0 || (year) % 400 == 0))
1019 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1020 #define	days_in_month(a) 	(month_days[(a) - 1])
1021 
1022 static int month_days[12] = {
1023 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1024 };
1025 
1026 /*
1027  * This only works for the Gregorian calendar - i.e. after 1752 (in the UK)
1028  */
1029 void GregorianDay(struct rtc_time * tm)
1030 {
1031 	int leapsToDate;
1032 	int lastYear;
1033 	int day;
1034 	int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
1035 
1036 	lastYear = tm->tm_year - 1;
1037 
1038 	/*
1039 	 * Number of leap corrections to apply up to end of last year
1040 	 */
1041 	leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400;
1042 
1043 	/*
1044 	 * This year is a leap year if it is divisible by 4 except when it is
1045 	 * divisible by 100 unless it is divisible by 400
1046 	 *
1047 	 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was
1048 	 */
1049 	day = tm->tm_mon > 2 && leapyear(tm->tm_year);
1050 
1051 	day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] +
1052 		   tm->tm_mday;
1053 
1054 	tm->tm_wday = day % 7;
1055 }
1056 
1057 void to_tm(int tim, struct rtc_time * tm)
1058 {
1059 	register int    i;
1060 	register long   hms, day;
1061 
1062 	day = tim / SECDAY;
1063 	hms = tim % SECDAY;
1064 
1065 	/* Hours, minutes, seconds are easy */
1066 	tm->tm_hour = hms / 3600;
1067 	tm->tm_min = (hms % 3600) / 60;
1068 	tm->tm_sec = (hms % 3600) % 60;
1069 
1070 	/* Number of years in days */
1071 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1072 		day -= days_in_year(i);
1073 	tm->tm_year = i;
1074 
1075 	/* Number of months in days left */
1076 	if (leapyear(tm->tm_year))
1077 		days_in_month(FEBRUARY) = 29;
1078 	for (i = 1; day >= days_in_month(i); i++)
1079 		day -= days_in_month(i);
1080 	days_in_month(FEBRUARY) = 28;
1081 	tm->tm_mon = i;
1082 
1083 	/* Days are what is left over (+1) from all that. */
1084 	tm->tm_mday = day + 1;
1085 
1086 	/*
1087 	 * Determine the day of week
1088 	 */
1089 	GregorianDay(tm);
1090 }
1091 
1092 /*
1093  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1094  * result.
1095  */
1096 void div128_by_32(u64 dividend_high, u64 dividend_low,
1097 		  unsigned divisor, struct div_result *dr)
1098 {
1099 	unsigned long a, b, c, d;
1100 	unsigned long w, x, y, z;
1101 	u64 ra, rb, rc;
1102 
1103 	a = dividend_high >> 32;
1104 	b = dividend_high & 0xffffffff;
1105 	c = dividend_low >> 32;
1106 	d = dividend_low & 0xffffffff;
1107 
1108 	w = a / divisor;
1109 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1110 
1111 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1112 	x = ra;
1113 
1114 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1115 	y = rb;
1116 
1117 	do_div(rc, divisor);
1118 	z = rc;
1119 
1120 	dr->result_high = ((u64)w << 32) + x;
1121 	dr->result_low  = ((u64)y << 32) + z;
1122 
1123 }
1124 
1125 /* We don't need to calibrate delay, we use the CPU timebase for that */
1126 void calibrate_delay(void)
1127 {
1128 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1129 	 * as the number of __delay(1) in a jiffy, so make it so
1130 	 */
1131 	loops_per_jiffy = tb_ticks_per_jiffy;
1132 }
1133 
1134 static int __init rtc_init(void)
1135 {
1136 	struct platform_device *pdev;
1137 
1138 	if (!ppc_md.get_rtc_time)
1139 		return -ENODEV;
1140 
1141 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1142 	if (IS_ERR(pdev))
1143 		return PTR_ERR(pdev);
1144 
1145 	return 0;
1146 }
1147 
1148 module_init(rtc_init);
1149