xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision 77a87824)
1 /*
2  * Common time routines among all ppc machines.
3  *
4  * Written by Cort Dougan (cort@cs.nmt.edu) to merge
5  * Paul Mackerras' version and mine for PReP and Pmac.
6  * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net).
7  * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com)
8  *
9  * First round of bugfixes by Gabriel Paubert (paubert@iram.es)
10  * to make clock more stable (2.4.0-test5). The only thing
11  * that this code assumes is that the timebases have been synchronized
12  * by firmware on SMP and are never stopped (never do sleep
13  * on SMP then, nap and doze are OK).
14  *
15  * Speeded up do_gettimeofday by getting rid of references to
16  * xtime (which required locks for consistency). (mikejc@us.ibm.com)
17  *
18  * TODO (not necessarily in this file):
19  * - improve precision and reproducibility of timebase frequency
20  * measurement at boot time.
21  * - for astronomical applications: add a new function to get
22  * non ambiguous timestamps even around leap seconds. This needs
23  * a new timestamp format and a good name.
24  *
25  * 1997-09-10  Updated NTP code according to technical memorandum Jan '96
26  *             "A Kernel Model for Precision Timekeeping" by Dave Mills
27  *
28  *      This program is free software; you can redistribute it and/or
29  *      modify it under the terms of the GNU General Public License
30  *      as published by the Free Software Foundation; either version
31  *      2 of the License, or (at your option) any later version.
32  */
33 
34 #include <linux/errno.h>
35 #include <linux/export.h>
36 #include <linux/sched.h>
37 #include <linux/kernel.h>
38 #include <linux/param.h>
39 #include <linux/string.h>
40 #include <linux/mm.h>
41 #include <linux/interrupt.h>
42 #include <linux/timex.h>
43 #include <linux/kernel_stat.h>
44 #include <linux/time.h>
45 #include <linux/clockchips.h>
46 #include <linux/init.h>
47 #include <linux/profile.h>
48 #include <linux/cpu.h>
49 #include <linux/security.h>
50 #include <linux/percpu.h>
51 #include <linux/rtc.h>
52 #include <linux/jiffies.h>
53 #include <linux/posix-timers.h>
54 #include <linux/irq.h>
55 #include <linux/delay.h>
56 #include <linux/irq_work.h>
57 #include <linux/clk-provider.h>
58 #include <linux/suspend.h>
59 #include <asm/trace.h>
60 
61 #include <asm/io.h>
62 #include <asm/processor.h>
63 #include <asm/nvram.h>
64 #include <asm/cache.h>
65 #include <asm/machdep.h>
66 #include <asm/uaccess.h>
67 #include <asm/time.h>
68 #include <asm/prom.h>
69 #include <asm/irq.h>
70 #include <asm/div64.h>
71 #include <asm/smp.h>
72 #include <asm/vdso_datapage.h>
73 #include <asm/firmware.h>
74 #include <asm/cputime.h>
75 
76 /* powerpc clocksource/clockevent code */
77 
78 #include <linux/clockchips.h>
79 #include <linux/timekeeper_internal.h>
80 
81 static cycle_t rtc_read(struct clocksource *);
82 static struct clocksource clocksource_rtc = {
83 	.name         = "rtc",
84 	.rating       = 400,
85 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
86 	.mask         = CLOCKSOURCE_MASK(64),
87 	.read         = rtc_read,
88 };
89 
90 static cycle_t timebase_read(struct clocksource *);
91 static struct clocksource clocksource_timebase = {
92 	.name         = "timebase",
93 	.rating       = 400,
94 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
95 	.mask         = CLOCKSOURCE_MASK(64),
96 	.read         = timebase_read,
97 };
98 
99 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
100 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
101 
102 static int decrementer_set_next_event(unsigned long evt,
103 				      struct clock_event_device *dev);
104 static int decrementer_shutdown(struct clock_event_device *evt);
105 
106 struct clock_event_device decrementer_clockevent = {
107 	.name			= "decrementer",
108 	.rating			= 200,
109 	.irq			= 0,
110 	.set_next_event		= decrementer_set_next_event,
111 	.set_state_shutdown	= decrementer_shutdown,
112 	.tick_resume		= decrementer_shutdown,
113 	.features		= CLOCK_EVT_FEAT_ONESHOT |
114 				  CLOCK_EVT_FEAT_C3STOP,
115 };
116 EXPORT_SYMBOL(decrementer_clockevent);
117 
118 DEFINE_PER_CPU(u64, decrementers_next_tb);
119 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
120 
121 #define XSEC_PER_SEC (1024*1024)
122 
123 #ifdef CONFIG_PPC64
124 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
125 #else
126 /* compute ((xsec << 12) * max) >> 32 */
127 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
128 #endif
129 
130 unsigned long tb_ticks_per_jiffy;
131 unsigned long tb_ticks_per_usec = 100; /* sane default */
132 EXPORT_SYMBOL(tb_ticks_per_usec);
133 unsigned long tb_ticks_per_sec;
134 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
135 
136 DEFINE_SPINLOCK(rtc_lock);
137 EXPORT_SYMBOL_GPL(rtc_lock);
138 
139 static u64 tb_to_ns_scale __read_mostly;
140 static unsigned tb_to_ns_shift __read_mostly;
141 static u64 boot_tb __read_mostly;
142 
143 extern struct timezone sys_tz;
144 static long timezone_offset;
145 
146 unsigned long ppc_proc_freq;
147 EXPORT_SYMBOL_GPL(ppc_proc_freq);
148 unsigned long ppc_tb_freq;
149 EXPORT_SYMBOL_GPL(ppc_tb_freq);
150 
151 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
152 /*
153  * Factors for converting from cputime_t (timebase ticks) to
154  * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
155  * These are all stored as 0.64 fixed-point binary fractions.
156  */
157 u64 __cputime_jiffies_factor;
158 EXPORT_SYMBOL(__cputime_jiffies_factor);
159 u64 __cputime_usec_factor;
160 EXPORT_SYMBOL(__cputime_usec_factor);
161 u64 __cputime_sec_factor;
162 EXPORT_SYMBOL(__cputime_sec_factor);
163 u64 __cputime_clockt_factor;
164 EXPORT_SYMBOL(__cputime_clockt_factor);
165 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
166 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
167 
168 cputime_t cputime_one_jiffy;
169 
170 #ifdef CONFIG_PPC_SPLPAR
171 void (*dtl_consumer)(struct dtl_entry *, u64);
172 #endif
173 
174 #ifdef CONFIG_PPC64
175 #define get_accounting(tsk)	(&get_paca()->accounting)
176 #else
177 #define get_accounting(tsk)	(&task_thread_info(tsk)->accounting)
178 #endif
179 
180 static void calc_cputime_factors(void)
181 {
182 	struct div_result res;
183 
184 	div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
185 	__cputime_jiffies_factor = res.result_low;
186 	div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
187 	__cputime_usec_factor = res.result_low;
188 	div128_by_32(1, 0, tb_ticks_per_sec, &res);
189 	__cputime_sec_factor = res.result_low;
190 	div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
191 	__cputime_clockt_factor = res.result_low;
192 }
193 
194 /*
195  * Read the SPURR on systems that have it, otherwise the PURR,
196  * or if that doesn't exist return the timebase value passed in.
197  */
198 static unsigned long read_spurr(unsigned long tb)
199 {
200 	if (cpu_has_feature(CPU_FTR_SPURR))
201 		return mfspr(SPRN_SPURR);
202 	if (cpu_has_feature(CPU_FTR_PURR))
203 		return mfspr(SPRN_PURR);
204 	return tb;
205 }
206 
207 #ifdef CONFIG_PPC_SPLPAR
208 
209 /*
210  * Scan the dispatch trace log and count up the stolen time.
211  * Should be called with interrupts disabled.
212  */
213 static u64 scan_dispatch_log(u64 stop_tb)
214 {
215 	u64 i = local_paca->dtl_ridx;
216 	struct dtl_entry *dtl = local_paca->dtl_curr;
217 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
218 	struct lppaca *vpa = local_paca->lppaca_ptr;
219 	u64 tb_delta;
220 	u64 stolen = 0;
221 	u64 dtb;
222 
223 	if (!dtl)
224 		return 0;
225 
226 	if (i == be64_to_cpu(vpa->dtl_idx))
227 		return 0;
228 	while (i < be64_to_cpu(vpa->dtl_idx)) {
229 		dtb = be64_to_cpu(dtl->timebase);
230 		tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
231 			be32_to_cpu(dtl->ready_to_enqueue_time);
232 		barrier();
233 		if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
234 			/* buffer has overflowed */
235 			i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
236 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
237 			continue;
238 		}
239 		if (dtb > stop_tb)
240 			break;
241 		if (dtl_consumer)
242 			dtl_consumer(dtl, i);
243 		stolen += tb_delta;
244 		++i;
245 		++dtl;
246 		if (dtl == dtl_end)
247 			dtl = local_paca->dispatch_log;
248 	}
249 	local_paca->dtl_ridx = i;
250 	local_paca->dtl_curr = dtl;
251 	return stolen;
252 }
253 
254 /*
255  * Accumulate stolen time by scanning the dispatch trace log.
256  * Called on entry from user mode.
257  */
258 void accumulate_stolen_time(void)
259 {
260 	u64 sst, ust;
261 	u8 save_soft_enabled = local_paca->soft_enabled;
262 	struct cpu_accounting_data *acct = &local_paca->accounting;
263 
264 	/* We are called early in the exception entry, before
265 	 * soft/hard_enabled are sync'ed to the expected state
266 	 * for the exception. We are hard disabled but the PACA
267 	 * needs to reflect that so various debug stuff doesn't
268 	 * complain
269 	 */
270 	local_paca->soft_enabled = 0;
271 
272 	sst = scan_dispatch_log(acct->starttime_user);
273 	ust = scan_dispatch_log(acct->starttime);
274 	acct->system_time -= sst;
275 	acct->user_time -= ust;
276 	local_paca->stolen_time += ust + sst;
277 
278 	local_paca->soft_enabled = save_soft_enabled;
279 }
280 
281 static inline u64 calculate_stolen_time(u64 stop_tb)
282 {
283 	u64 stolen = 0;
284 
285 	if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
286 		stolen = scan_dispatch_log(stop_tb);
287 		get_paca()->accounting.system_time -= stolen;
288 	}
289 
290 	stolen += get_paca()->stolen_time;
291 	get_paca()->stolen_time = 0;
292 	return stolen;
293 }
294 
295 #else /* CONFIG_PPC_SPLPAR */
296 static inline u64 calculate_stolen_time(u64 stop_tb)
297 {
298 	return 0;
299 }
300 
301 #endif /* CONFIG_PPC_SPLPAR */
302 
303 /*
304  * Account time for a transition between system, hard irq
305  * or soft irq state.
306  */
307 static unsigned long vtime_delta(struct task_struct *tsk,
308 				 unsigned long *sys_scaled,
309 				 unsigned long *stolen)
310 {
311 	unsigned long now, nowscaled, deltascaled;
312 	unsigned long udelta, delta, user_scaled;
313 	struct cpu_accounting_data *acct = get_accounting(tsk);
314 
315 	WARN_ON_ONCE(!irqs_disabled());
316 
317 	now = mftb();
318 	nowscaled = read_spurr(now);
319 	acct->system_time += now - acct->starttime;
320 	acct->starttime = now;
321 	deltascaled = nowscaled - acct->startspurr;
322 	acct->startspurr = nowscaled;
323 
324 	*stolen = calculate_stolen_time(now);
325 
326 	delta = acct->system_time;
327 	acct->system_time = 0;
328 	udelta = acct->user_time - acct->utime_sspurr;
329 	acct->utime_sspurr = acct->user_time;
330 
331 	/*
332 	 * Because we don't read the SPURR on every kernel entry/exit,
333 	 * deltascaled includes both user and system SPURR ticks.
334 	 * Apportion these ticks to system SPURR ticks and user
335 	 * SPURR ticks in the same ratio as the system time (delta)
336 	 * and user time (udelta) values obtained from the timebase
337 	 * over the same interval.  The system ticks get accounted here;
338 	 * the user ticks get saved up in paca->user_time_scaled to be
339 	 * used by account_process_tick.
340 	 */
341 	*sys_scaled = delta;
342 	user_scaled = udelta;
343 	if (deltascaled != delta + udelta) {
344 		if (udelta) {
345 			*sys_scaled = deltascaled * delta / (delta + udelta);
346 			user_scaled = deltascaled - *sys_scaled;
347 		} else {
348 			*sys_scaled = deltascaled;
349 		}
350 	}
351 	acct->user_time_scaled += user_scaled;
352 
353 	return delta;
354 }
355 
356 void vtime_account_system(struct task_struct *tsk)
357 {
358 	unsigned long delta, sys_scaled, stolen;
359 
360 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
361 	account_system_time(tsk, 0, delta, sys_scaled);
362 	if (stolen)
363 		account_steal_time(stolen);
364 }
365 EXPORT_SYMBOL_GPL(vtime_account_system);
366 
367 void vtime_account_idle(struct task_struct *tsk)
368 {
369 	unsigned long delta, sys_scaled, stolen;
370 
371 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
372 	account_idle_time(delta + stolen);
373 }
374 
375 /*
376  * Transfer the user time accumulated in the paca
377  * by the exception entry and exit code to the generic
378  * process user time records.
379  * Must be called with interrupts disabled.
380  * Assumes that vtime_account_system/idle() has been called
381  * recently (i.e. since the last entry from usermode) so that
382  * get_paca()->user_time_scaled is up to date.
383  */
384 void vtime_account_user(struct task_struct *tsk)
385 {
386 	cputime_t utime, utimescaled;
387 	struct cpu_accounting_data *acct = get_accounting(tsk);
388 
389 	utime = acct->user_time;
390 	utimescaled = acct->user_time_scaled;
391 	acct->user_time = 0;
392 	acct->user_time_scaled = 0;
393 	acct->utime_sspurr = 0;
394 	account_user_time(tsk, utime, utimescaled);
395 }
396 
397 #ifdef CONFIG_PPC32
398 /*
399  * Called from the context switch with interrupts disabled, to charge all
400  * accumulated times to the current process, and to prepare accounting on
401  * the next process.
402  */
403 void arch_vtime_task_switch(struct task_struct *prev)
404 {
405 	struct cpu_accounting_data *acct = get_accounting(current);
406 
407 	acct->starttime = get_accounting(prev)->starttime;
408 	acct->system_time = 0;
409 	acct->user_time = 0;
410 }
411 #endif /* CONFIG_PPC32 */
412 
413 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
414 #define calc_cputime_factors()
415 #endif
416 
417 void __delay(unsigned long loops)
418 {
419 	unsigned long start;
420 	int diff;
421 
422 	if (__USE_RTC()) {
423 		start = get_rtcl();
424 		do {
425 			/* the RTCL register wraps at 1000000000 */
426 			diff = get_rtcl() - start;
427 			if (diff < 0)
428 				diff += 1000000000;
429 		} while (diff < loops);
430 	} else {
431 		start = get_tbl();
432 		while (get_tbl() - start < loops)
433 			HMT_low();
434 		HMT_medium();
435 	}
436 }
437 EXPORT_SYMBOL(__delay);
438 
439 void udelay(unsigned long usecs)
440 {
441 	__delay(tb_ticks_per_usec * usecs);
442 }
443 EXPORT_SYMBOL(udelay);
444 
445 #ifdef CONFIG_SMP
446 unsigned long profile_pc(struct pt_regs *regs)
447 {
448 	unsigned long pc = instruction_pointer(regs);
449 
450 	if (in_lock_functions(pc))
451 		return regs->link;
452 
453 	return pc;
454 }
455 EXPORT_SYMBOL(profile_pc);
456 #endif
457 
458 #ifdef CONFIG_IRQ_WORK
459 
460 /*
461  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
462  */
463 #ifdef CONFIG_PPC64
464 static inline unsigned long test_irq_work_pending(void)
465 {
466 	unsigned long x;
467 
468 	asm volatile("lbz %0,%1(13)"
469 		: "=r" (x)
470 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
471 	return x;
472 }
473 
474 static inline void set_irq_work_pending_flag(void)
475 {
476 	asm volatile("stb %0,%1(13)" : :
477 		"r" (1),
478 		"i" (offsetof(struct paca_struct, irq_work_pending)));
479 }
480 
481 static inline void clear_irq_work_pending(void)
482 {
483 	asm volatile("stb %0,%1(13)" : :
484 		"r" (0),
485 		"i" (offsetof(struct paca_struct, irq_work_pending)));
486 }
487 
488 #else /* 32-bit */
489 
490 DEFINE_PER_CPU(u8, irq_work_pending);
491 
492 #define set_irq_work_pending_flag()	__this_cpu_write(irq_work_pending, 1)
493 #define test_irq_work_pending()		__this_cpu_read(irq_work_pending)
494 #define clear_irq_work_pending()	__this_cpu_write(irq_work_pending, 0)
495 
496 #endif /* 32 vs 64 bit */
497 
498 void arch_irq_work_raise(void)
499 {
500 	preempt_disable();
501 	set_irq_work_pending_flag();
502 	set_dec(1);
503 	preempt_enable();
504 }
505 
506 #else  /* CONFIG_IRQ_WORK */
507 
508 #define test_irq_work_pending()	0
509 #define clear_irq_work_pending()
510 
511 #endif /* CONFIG_IRQ_WORK */
512 
513 static void __timer_interrupt(void)
514 {
515 	struct pt_regs *regs = get_irq_regs();
516 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
517 	struct clock_event_device *evt = this_cpu_ptr(&decrementers);
518 	u64 now;
519 
520 	trace_timer_interrupt_entry(regs);
521 
522 	if (test_irq_work_pending()) {
523 		clear_irq_work_pending();
524 		irq_work_run();
525 	}
526 
527 	now = get_tb_or_rtc();
528 	if (now >= *next_tb) {
529 		*next_tb = ~(u64)0;
530 		if (evt->event_handler)
531 			evt->event_handler(evt);
532 		__this_cpu_inc(irq_stat.timer_irqs_event);
533 	} else {
534 		now = *next_tb - now;
535 		if (now <= decrementer_max)
536 			set_dec(now);
537 		/* We may have raced with new irq work */
538 		if (test_irq_work_pending())
539 			set_dec(1);
540 		__this_cpu_inc(irq_stat.timer_irqs_others);
541 	}
542 
543 #ifdef CONFIG_PPC64
544 	/* collect purr register values often, for accurate calculations */
545 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
546 		struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
547 		cu->current_tb = mfspr(SPRN_PURR);
548 	}
549 #endif
550 
551 	trace_timer_interrupt_exit(regs);
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 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
562 
563 	/* Ensure a positive value is written to the decrementer, or else
564 	 * some CPUs will continue to take decrementer exceptions.
565 	 */
566 	set_dec(decrementer_max);
567 
568 	/* Some implementations of hotplug will get timer interrupts while
569 	 * offline, just ignore these and we also need to set
570 	 * decrementers_next_tb as MAX to make sure __check_irq_replay
571 	 * don't replay timer interrupt when return, otherwise we'll trap
572 	 * here infinitely :(
573 	 */
574 	if (!cpu_online(smp_processor_id())) {
575 		*next_tb = ~(u64)0;
576 		return;
577 	}
578 
579 	/* Conditionally hard-enable interrupts now that the DEC has been
580 	 * bumped to its maximum value
581 	 */
582 	may_hard_irq_enable();
583 
584 
585 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
586 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
587 		do_IRQ(regs);
588 #endif
589 
590 	old_regs = set_irq_regs(regs);
591 	irq_enter();
592 
593 	__timer_interrupt();
594 	irq_exit();
595 	set_irq_regs(old_regs);
596 }
597 
598 /*
599  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
600  * left pending on exit from a KVM guest.  We don't need to do anything
601  * to clear them, as they are edge-triggered.
602  */
603 void hdec_interrupt(struct pt_regs *regs)
604 {
605 }
606 
607 #ifdef CONFIG_SUSPEND
608 static void generic_suspend_disable_irqs(void)
609 {
610 	/* Disable the decrementer, so that it doesn't interfere
611 	 * with suspending.
612 	 */
613 
614 	set_dec(decrementer_max);
615 	local_irq_disable();
616 	set_dec(decrementer_max);
617 }
618 
619 static void generic_suspend_enable_irqs(void)
620 {
621 	local_irq_enable();
622 }
623 
624 /* Overrides the weak version in kernel/power/main.c */
625 void arch_suspend_disable_irqs(void)
626 {
627 	if (ppc_md.suspend_disable_irqs)
628 		ppc_md.suspend_disable_irqs();
629 	generic_suspend_disable_irqs();
630 }
631 
632 /* Overrides the weak version in kernel/power/main.c */
633 void arch_suspend_enable_irqs(void)
634 {
635 	generic_suspend_enable_irqs();
636 	if (ppc_md.suspend_enable_irqs)
637 		ppc_md.suspend_enable_irqs();
638 }
639 #endif
640 
641 unsigned long long tb_to_ns(unsigned long long ticks)
642 {
643 	return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
644 }
645 EXPORT_SYMBOL_GPL(tb_to_ns);
646 
647 /*
648  * Scheduler clock - returns current time in nanosec units.
649  *
650  * Note: mulhdu(a, b) (multiply high double unsigned) returns
651  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
652  * are 64-bit unsigned numbers.
653  */
654 unsigned long long sched_clock(void)
655 {
656 	if (__USE_RTC())
657 		return get_rtc();
658 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
659 }
660 
661 
662 #ifdef CONFIG_PPC_PSERIES
663 
664 /*
665  * Running clock - attempts to give a view of time passing for a virtualised
666  * kernels.
667  * Uses the VTB register if available otherwise a next best guess.
668  */
669 unsigned long long running_clock(void)
670 {
671 	/*
672 	 * Don't read the VTB as a host since KVM does not switch in host
673 	 * timebase into the VTB when it takes a guest off the CPU, reading the
674 	 * VTB would result in reading 'last switched out' guest VTB.
675 	 *
676 	 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
677 	 * would be unsafe to rely only on the #ifdef above.
678 	 */
679 	if (firmware_has_feature(FW_FEATURE_LPAR) &&
680 	    cpu_has_feature(CPU_FTR_ARCH_207S))
681 		return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
682 
683 	/*
684 	 * This is a next best approximation without a VTB.
685 	 * On a host which is running bare metal there should never be any stolen
686 	 * time and on a host which doesn't do any virtualisation TB *should* equal
687 	 * VTB so it makes no difference anyway.
688 	 */
689 	return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
690 }
691 #endif
692 
693 static int __init get_freq(char *name, int cells, unsigned long *val)
694 {
695 	struct device_node *cpu;
696 	const __be32 *fp;
697 	int found = 0;
698 
699 	/* The cpu node should have timebase and clock frequency properties */
700 	cpu = of_find_node_by_type(NULL, "cpu");
701 
702 	if (cpu) {
703 		fp = of_get_property(cpu, name, NULL);
704 		if (fp) {
705 			found = 1;
706 			*val = of_read_ulong(fp, cells);
707 		}
708 
709 		of_node_put(cpu);
710 	}
711 
712 	return found;
713 }
714 
715 static void start_cpu_decrementer(void)
716 {
717 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
718 	/* Clear any pending timer interrupts */
719 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
720 
721 	/* Enable decrementer interrupt */
722 	mtspr(SPRN_TCR, TCR_DIE);
723 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
724 }
725 
726 void __init generic_calibrate_decr(void)
727 {
728 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
729 
730 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
731 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
732 
733 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
734 				"(not found)\n");
735 	}
736 
737 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
738 
739 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
740 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
741 
742 		printk(KERN_ERR "WARNING: Estimating processor frequency "
743 				"(not found)\n");
744 	}
745 }
746 
747 int update_persistent_clock(struct timespec now)
748 {
749 	struct rtc_time tm;
750 
751 	if (!ppc_md.set_rtc_time)
752 		return -ENODEV;
753 
754 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
755 	tm.tm_year -= 1900;
756 	tm.tm_mon -= 1;
757 
758 	return ppc_md.set_rtc_time(&tm);
759 }
760 
761 static void __read_persistent_clock(struct timespec *ts)
762 {
763 	struct rtc_time tm;
764 	static int first = 1;
765 
766 	ts->tv_nsec = 0;
767 	/* XXX this is a litle fragile but will work okay in the short term */
768 	if (first) {
769 		first = 0;
770 		if (ppc_md.time_init)
771 			timezone_offset = ppc_md.time_init();
772 
773 		/* get_boot_time() isn't guaranteed to be safe to call late */
774 		if (ppc_md.get_boot_time) {
775 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
776 			return;
777 		}
778 	}
779 	if (!ppc_md.get_rtc_time) {
780 		ts->tv_sec = 0;
781 		return;
782 	}
783 	ppc_md.get_rtc_time(&tm);
784 
785 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
786 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
787 }
788 
789 void read_persistent_clock(struct timespec *ts)
790 {
791 	__read_persistent_clock(ts);
792 
793 	/* Sanitize it in case real time clock is set below EPOCH */
794 	if (ts->tv_sec < 0) {
795 		ts->tv_sec = 0;
796 		ts->tv_nsec = 0;
797 	}
798 
799 }
800 
801 /* clocksource code */
802 static cycle_t rtc_read(struct clocksource *cs)
803 {
804 	return (cycle_t)get_rtc();
805 }
806 
807 static cycle_t timebase_read(struct clocksource *cs)
808 {
809 	return (cycle_t)get_tb();
810 }
811 
812 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
813 			 struct clocksource *clock, u32 mult, cycle_t cycle_last)
814 {
815 	u64 new_tb_to_xs, new_stamp_xsec;
816 	u32 frac_sec;
817 
818 	if (clock != &clocksource_timebase)
819 		return;
820 
821 	/* Make userspace gettimeofday spin until we're done. */
822 	++vdso_data->tb_update_count;
823 	smp_mb();
824 
825 	/* 19342813113834067 ~= 2^(20+64) / 1e9 */
826 	new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
827 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
828 	do_div(new_stamp_xsec, 1000000000);
829 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
830 
831 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
832 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
833 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
834 
835 	/*
836 	 * tb_update_count is used to allow the userspace gettimeofday code
837 	 * to assure itself that it sees a consistent view of the tb_to_xs and
838 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
839 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
840 	 * the two values of tb_update_count match and are even then the
841 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
842 	 * loops back and reads them again until this criteria is met.
843 	 * We expect the caller to have done the first increment of
844 	 * vdso_data->tb_update_count already.
845 	 */
846 	vdso_data->tb_orig_stamp = cycle_last;
847 	vdso_data->stamp_xsec = new_stamp_xsec;
848 	vdso_data->tb_to_xs = new_tb_to_xs;
849 	vdso_data->wtom_clock_sec = wtm->tv_sec;
850 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
851 	vdso_data->stamp_xtime = *wall_time;
852 	vdso_data->stamp_sec_fraction = frac_sec;
853 	smp_wmb();
854 	++(vdso_data->tb_update_count);
855 }
856 
857 void update_vsyscall_tz(void)
858 {
859 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
860 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
861 }
862 
863 static void __init clocksource_init(void)
864 {
865 	struct clocksource *clock;
866 
867 	if (__USE_RTC())
868 		clock = &clocksource_rtc;
869 	else
870 		clock = &clocksource_timebase;
871 
872 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
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 	__this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
886 	set_dec(evt);
887 
888 	/* We may have raced with new irq work */
889 	if (test_irq_work_pending())
890 		set_dec(1);
891 
892 	return 0;
893 }
894 
895 static int decrementer_shutdown(struct clock_event_device *dev)
896 {
897 	decrementer_set_next_event(decrementer_max, dev);
898 	return 0;
899 }
900 
901 /* Interrupt handler for the timer broadcast IPI */
902 void tick_broadcast_ipi_handler(void)
903 {
904 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
905 
906 	*next_tb = get_tb_or_rtc();
907 	__timer_interrupt();
908 }
909 
910 static void register_decrementer_clockevent(int cpu)
911 {
912 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
913 
914 	*dec = decrementer_clockevent;
915 	dec->cpumask = cpumask_of(cpu);
916 
917 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
918 		    dec->name, dec->mult, dec->shift, cpu);
919 
920 	clockevents_register_device(dec);
921 }
922 
923 static void enable_large_decrementer(void)
924 {
925 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
926 		return;
927 
928 	if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
929 		return;
930 
931 	/*
932 	 * If we're running as the hypervisor we need to enable the LD manually
933 	 * otherwise firmware should have done it for us.
934 	 */
935 	if (cpu_has_feature(CPU_FTR_HVMODE))
936 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
937 }
938 
939 static void __init set_decrementer_max(void)
940 {
941 	struct device_node *cpu;
942 	u32 bits = 32;
943 
944 	/* Prior to ISAv3 the decrementer is always 32 bit */
945 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
946 		return;
947 
948 	cpu = of_find_node_by_type(NULL, "cpu");
949 
950 	if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
951 		if (bits > 64 || bits < 32) {
952 			pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
953 			bits = 32;
954 		}
955 
956 		/* calculate the signed maximum given this many bits */
957 		decrementer_max = (1ul << (bits - 1)) - 1;
958 	}
959 
960 	of_node_put(cpu);
961 
962 	pr_info("time_init: %u bit decrementer (max: %llx)\n",
963 		bits, decrementer_max);
964 }
965 
966 static void __init init_decrementer_clockevent(void)
967 {
968 	int cpu = smp_processor_id();
969 
970 	clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
971 
972 	decrementer_clockevent.max_delta_ns =
973 		clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
974 	decrementer_clockevent.min_delta_ns =
975 		clockevent_delta2ns(2, &decrementer_clockevent);
976 
977 	register_decrementer_clockevent(cpu);
978 }
979 
980 void secondary_cpu_time_init(void)
981 {
982 	/* Enable and test the large decrementer for this cpu */
983 	enable_large_decrementer();
984 
985 	/* Start the decrementer on CPUs that have manual control
986 	 * such as BookE
987 	 */
988 	start_cpu_decrementer();
989 
990 	/* FIME: Should make unrelatred change to move snapshot_timebase
991 	 * call here ! */
992 	register_decrementer_clockevent(smp_processor_id());
993 }
994 
995 /* This function is only called on the boot processor */
996 void __init time_init(void)
997 {
998 	struct div_result res;
999 	u64 scale;
1000 	unsigned shift;
1001 
1002 	if (__USE_RTC()) {
1003 		/* 601 processor: dec counts down by 128 every 128ns */
1004 		ppc_tb_freq = 1000000000;
1005 	} else {
1006 		/* Normal PowerPC with timebase register */
1007 		ppc_md.calibrate_decr();
1008 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1009 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1010 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
1011 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1012 	}
1013 
1014 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1015 	tb_ticks_per_sec = ppc_tb_freq;
1016 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
1017 	calc_cputime_factors();
1018 	setup_cputime_one_jiffy();
1019 
1020 	/*
1021 	 * Compute scale factor for sched_clock.
1022 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1023 	 * which is the timebase frequency.
1024 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1025 	 * the 128-bit result as a 64.64 fixed-point number.
1026 	 * We then shift that number right until it is less than 1.0,
1027 	 * giving us the scale factor and shift count to use in
1028 	 * sched_clock().
1029 	 */
1030 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1031 	scale = res.result_low;
1032 	for (shift = 0; res.result_high != 0; ++shift) {
1033 		scale = (scale >> 1) | (res.result_high << 63);
1034 		res.result_high >>= 1;
1035 	}
1036 	tb_to_ns_scale = scale;
1037 	tb_to_ns_shift = shift;
1038 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1039 	boot_tb = get_tb_or_rtc();
1040 
1041 	/* If platform provided a timezone (pmac), we correct the time */
1042 	if (timezone_offset) {
1043 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1044 		sys_tz.tz_dsttime = 0;
1045 	}
1046 
1047 	vdso_data->tb_update_count = 0;
1048 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1049 
1050 	/* initialise and enable the large decrementer (if we have one) */
1051 	set_decrementer_max();
1052 	enable_large_decrementer();
1053 
1054 	/* Start the decrementer on CPUs that have manual control
1055 	 * such as BookE
1056 	 */
1057 	start_cpu_decrementer();
1058 
1059 	/* Register the clocksource */
1060 	clocksource_init();
1061 
1062 	init_decrementer_clockevent();
1063 	tick_setup_hrtimer_broadcast();
1064 
1065 #ifdef CONFIG_COMMON_CLK
1066 	of_clk_init(NULL);
1067 #endif
1068 }
1069 
1070 
1071 #define FEBRUARY	2
1072 #define	STARTOFTIME	1970
1073 #define SECDAY		86400L
1074 #define SECYR		(SECDAY * 365)
1075 #define	leapyear(year)		((year) % 4 == 0 && \
1076 				 ((year) % 100 != 0 || (year) % 400 == 0))
1077 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1078 #define	days_in_month(a) 	(month_days[(a) - 1])
1079 
1080 static int month_days[12] = {
1081 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1082 };
1083 
1084 void to_tm(int tim, struct rtc_time * tm)
1085 {
1086 	register int    i;
1087 	register long   hms, day;
1088 
1089 	day = tim / SECDAY;
1090 	hms = tim % SECDAY;
1091 
1092 	/* Hours, minutes, seconds are easy */
1093 	tm->tm_hour = hms / 3600;
1094 	tm->tm_min = (hms % 3600) / 60;
1095 	tm->tm_sec = (hms % 3600) % 60;
1096 
1097 	/* Number of years in days */
1098 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1099 		day -= days_in_year(i);
1100 	tm->tm_year = i;
1101 
1102 	/* Number of months in days left */
1103 	if (leapyear(tm->tm_year))
1104 		days_in_month(FEBRUARY) = 29;
1105 	for (i = 1; day >= days_in_month(i); i++)
1106 		day -= days_in_month(i);
1107 	days_in_month(FEBRUARY) = 28;
1108 	tm->tm_mon = i;
1109 
1110 	/* Days are what is left over (+1) from all that. */
1111 	tm->tm_mday = day + 1;
1112 
1113 	/*
1114 	 * No-one uses the day of the week.
1115 	 */
1116 	tm->tm_wday = -1;
1117 }
1118 EXPORT_SYMBOL(to_tm);
1119 
1120 /*
1121  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1122  * result.
1123  */
1124 void div128_by_32(u64 dividend_high, u64 dividend_low,
1125 		  unsigned divisor, struct div_result *dr)
1126 {
1127 	unsigned long a, b, c, d;
1128 	unsigned long w, x, y, z;
1129 	u64 ra, rb, rc;
1130 
1131 	a = dividend_high >> 32;
1132 	b = dividend_high & 0xffffffff;
1133 	c = dividend_low >> 32;
1134 	d = dividend_low & 0xffffffff;
1135 
1136 	w = a / divisor;
1137 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1138 
1139 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1140 	x = ra;
1141 
1142 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1143 	y = rb;
1144 
1145 	do_div(rc, divisor);
1146 	z = rc;
1147 
1148 	dr->result_high = ((u64)w << 32) + x;
1149 	dr->result_low  = ((u64)y << 32) + z;
1150 
1151 }
1152 
1153 /* We don't need to calibrate delay, we use the CPU timebase for that */
1154 void calibrate_delay(void)
1155 {
1156 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1157 	 * as the number of __delay(1) in a jiffy, so make it so
1158 	 */
1159 	loops_per_jiffy = tb_ticks_per_jiffy;
1160 }
1161 
1162 static int __init rtc_init(void)
1163 {
1164 	struct platform_device *pdev;
1165 
1166 	if (!ppc_md.get_rtc_time)
1167 		return -ENODEV;
1168 
1169 	pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1170 
1171 	return PTR_ERR_OR_ZERO(pdev);
1172 }
1173 
1174 device_initcall(rtc_init);
1175