xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision 84d69848)
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 <linux/rtc.h>
60 #include <asm/trace.h>
61 
62 #include <asm/io.h>
63 #include <asm/processor.h>
64 #include <asm/nvram.h>
65 #include <asm/cache.h>
66 #include <asm/machdep.h>
67 #include <asm/uaccess.h>
68 #include <asm/time.h>
69 #include <asm/prom.h>
70 #include <asm/irq.h>
71 #include <asm/div64.h>
72 #include <asm/smp.h>
73 #include <asm/vdso_datapage.h>
74 #include <asm/firmware.h>
75 #include <asm/cputime.h>
76 #include <asm/asm-prototypes.h>
77 
78 /* powerpc clocksource/clockevent code */
79 
80 #include <linux/clockchips.h>
81 #include <linux/timekeeper_internal.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 	.read         = rtc_read,
90 };
91 
92 static cycle_t timebase_read(struct clocksource *);
93 static struct clocksource clocksource_timebase = {
94 	.name         = "timebase",
95 	.rating       = 400,
96 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
97 	.mask         = CLOCKSOURCE_MASK(64),
98 	.read         = timebase_read,
99 };
100 
101 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
102 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
103 
104 static int decrementer_set_next_event(unsigned long evt,
105 				      struct clock_event_device *dev);
106 static int decrementer_shutdown(struct clock_event_device *evt);
107 
108 struct clock_event_device decrementer_clockevent = {
109 	.name			= "decrementer",
110 	.rating			= 200,
111 	.irq			= 0,
112 	.set_next_event		= decrementer_set_next_event,
113 	.set_state_shutdown	= decrementer_shutdown,
114 	.tick_resume		= decrementer_shutdown,
115 	.features		= CLOCK_EVT_FEAT_ONESHOT |
116 				  CLOCK_EVT_FEAT_C3STOP,
117 };
118 EXPORT_SYMBOL(decrementer_clockevent);
119 
120 DEFINE_PER_CPU(u64, decrementers_next_tb);
121 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
122 
123 #define XSEC_PER_SEC (1024*1024)
124 
125 #ifdef CONFIG_PPC64
126 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
127 #else
128 /* compute ((xsec << 12) * max) >> 32 */
129 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
130 #endif
131 
132 unsigned long tb_ticks_per_jiffy;
133 unsigned long tb_ticks_per_usec = 100; /* sane default */
134 EXPORT_SYMBOL(tb_ticks_per_usec);
135 unsigned long tb_ticks_per_sec;
136 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
137 
138 DEFINE_SPINLOCK(rtc_lock);
139 EXPORT_SYMBOL_GPL(rtc_lock);
140 
141 static u64 tb_to_ns_scale __read_mostly;
142 static unsigned tb_to_ns_shift __read_mostly;
143 static u64 boot_tb __read_mostly;
144 
145 extern struct timezone sys_tz;
146 static long timezone_offset;
147 
148 unsigned long ppc_proc_freq;
149 EXPORT_SYMBOL_GPL(ppc_proc_freq);
150 unsigned long ppc_tb_freq;
151 EXPORT_SYMBOL_GPL(ppc_tb_freq);
152 
153 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
154 /*
155  * Factors for converting from cputime_t (timebase ticks) to
156  * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
157  * These are all stored as 0.64 fixed-point binary fractions.
158  */
159 u64 __cputime_jiffies_factor;
160 EXPORT_SYMBOL(__cputime_jiffies_factor);
161 u64 __cputime_usec_factor;
162 EXPORT_SYMBOL(__cputime_usec_factor);
163 u64 __cputime_sec_factor;
164 EXPORT_SYMBOL(__cputime_sec_factor);
165 u64 __cputime_clockt_factor;
166 EXPORT_SYMBOL(__cputime_clockt_factor);
167 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
168 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
169 
170 cputime_t cputime_one_jiffy;
171 
172 #ifdef CONFIG_PPC_SPLPAR
173 void (*dtl_consumer)(struct dtl_entry *, u64);
174 #endif
175 
176 #ifdef CONFIG_PPC64
177 #define get_accounting(tsk)	(&get_paca()->accounting)
178 #else
179 #define get_accounting(tsk)	(&task_thread_info(tsk)->accounting)
180 #endif
181 
182 static void calc_cputime_factors(void)
183 {
184 	struct div_result res;
185 
186 	div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
187 	__cputime_jiffies_factor = res.result_low;
188 	div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
189 	__cputime_usec_factor = res.result_low;
190 	div128_by_32(1, 0, tb_ticks_per_sec, &res);
191 	__cputime_sec_factor = res.result_low;
192 	div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
193 	__cputime_clockt_factor = res.result_low;
194 }
195 
196 /*
197  * Read the SPURR on systems that have it, otherwise the PURR,
198  * or if that doesn't exist return the timebase value passed in.
199  */
200 static unsigned long read_spurr(unsigned long tb)
201 {
202 	if (cpu_has_feature(CPU_FTR_SPURR))
203 		return mfspr(SPRN_SPURR);
204 	if (cpu_has_feature(CPU_FTR_PURR))
205 		return mfspr(SPRN_PURR);
206 	return tb;
207 }
208 
209 #ifdef CONFIG_PPC_SPLPAR
210 
211 /*
212  * Scan the dispatch trace log and count up the stolen time.
213  * Should be called with interrupts disabled.
214  */
215 static u64 scan_dispatch_log(u64 stop_tb)
216 {
217 	u64 i = local_paca->dtl_ridx;
218 	struct dtl_entry *dtl = local_paca->dtl_curr;
219 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
220 	struct lppaca *vpa = local_paca->lppaca_ptr;
221 	u64 tb_delta;
222 	u64 stolen = 0;
223 	u64 dtb;
224 
225 	if (!dtl)
226 		return 0;
227 
228 	if (i == be64_to_cpu(vpa->dtl_idx))
229 		return 0;
230 	while (i < be64_to_cpu(vpa->dtl_idx)) {
231 		dtb = be64_to_cpu(dtl->timebase);
232 		tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
233 			be32_to_cpu(dtl->ready_to_enqueue_time);
234 		barrier();
235 		if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
236 			/* buffer has overflowed */
237 			i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
238 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
239 			continue;
240 		}
241 		if (dtb > stop_tb)
242 			break;
243 		if (dtl_consumer)
244 			dtl_consumer(dtl, i);
245 		stolen += tb_delta;
246 		++i;
247 		++dtl;
248 		if (dtl == dtl_end)
249 			dtl = local_paca->dispatch_log;
250 	}
251 	local_paca->dtl_ridx = i;
252 	local_paca->dtl_curr = dtl;
253 	return stolen;
254 }
255 
256 /*
257  * Accumulate stolen time by scanning the dispatch trace log.
258  * Called on entry from user mode.
259  */
260 void accumulate_stolen_time(void)
261 {
262 	u64 sst, ust;
263 	u8 save_soft_enabled = local_paca->soft_enabled;
264 	struct cpu_accounting_data *acct = &local_paca->accounting;
265 
266 	/* We are called early in the exception entry, before
267 	 * soft/hard_enabled are sync'ed to the expected state
268 	 * for the exception. We are hard disabled but the PACA
269 	 * needs to reflect that so various debug stuff doesn't
270 	 * complain
271 	 */
272 	local_paca->soft_enabled = 0;
273 
274 	sst = scan_dispatch_log(acct->starttime_user);
275 	ust = scan_dispatch_log(acct->starttime);
276 	acct->system_time -= sst;
277 	acct->user_time -= ust;
278 	local_paca->stolen_time += ust + sst;
279 
280 	local_paca->soft_enabled = save_soft_enabled;
281 }
282 
283 static inline u64 calculate_stolen_time(u64 stop_tb)
284 {
285 	u64 stolen = 0;
286 
287 	if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
288 		stolen = scan_dispatch_log(stop_tb);
289 		get_paca()->accounting.system_time -= stolen;
290 	}
291 
292 	stolen += get_paca()->stolen_time;
293 	get_paca()->stolen_time = 0;
294 	return stolen;
295 }
296 
297 #else /* CONFIG_PPC_SPLPAR */
298 static inline u64 calculate_stolen_time(u64 stop_tb)
299 {
300 	return 0;
301 }
302 
303 #endif /* CONFIG_PPC_SPLPAR */
304 
305 /*
306  * Account time for a transition between system, hard irq
307  * or soft irq state.
308  */
309 static unsigned long vtime_delta(struct task_struct *tsk,
310 				 unsigned long *sys_scaled,
311 				 unsigned long *stolen)
312 {
313 	unsigned long now, nowscaled, deltascaled;
314 	unsigned long udelta, delta, user_scaled;
315 	struct cpu_accounting_data *acct = get_accounting(tsk);
316 
317 	WARN_ON_ONCE(!irqs_disabled());
318 
319 	now = mftb();
320 	nowscaled = read_spurr(now);
321 	acct->system_time += now - acct->starttime;
322 	acct->starttime = now;
323 	deltascaled = nowscaled - acct->startspurr;
324 	acct->startspurr = nowscaled;
325 
326 	*stolen = calculate_stolen_time(now);
327 
328 	delta = acct->system_time;
329 	acct->system_time = 0;
330 	udelta = acct->user_time - acct->utime_sspurr;
331 	acct->utime_sspurr = acct->user_time;
332 
333 	/*
334 	 * Because we don't read the SPURR on every kernel entry/exit,
335 	 * deltascaled includes both user and system SPURR ticks.
336 	 * Apportion these ticks to system SPURR ticks and user
337 	 * SPURR ticks in the same ratio as the system time (delta)
338 	 * and user time (udelta) values obtained from the timebase
339 	 * over the same interval.  The system ticks get accounted here;
340 	 * the user ticks get saved up in paca->user_time_scaled to be
341 	 * used by account_process_tick.
342 	 */
343 	*sys_scaled = delta;
344 	user_scaled = udelta;
345 	if (deltascaled != delta + udelta) {
346 		if (udelta) {
347 			*sys_scaled = deltascaled * delta / (delta + udelta);
348 			user_scaled = deltascaled - *sys_scaled;
349 		} else {
350 			*sys_scaled = deltascaled;
351 		}
352 	}
353 	acct->user_time_scaled += user_scaled;
354 
355 	return delta;
356 }
357 
358 void vtime_account_system(struct task_struct *tsk)
359 {
360 	unsigned long delta, sys_scaled, stolen;
361 
362 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
363 	account_system_time(tsk, 0, delta, sys_scaled);
364 	if (stolen)
365 		account_steal_time(stolen);
366 }
367 EXPORT_SYMBOL_GPL(vtime_account_system);
368 
369 void vtime_account_idle(struct task_struct *tsk)
370 {
371 	unsigned long delta, sys_scaled, stolen;
372 
373 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
374 	account_idle_time(delta + stolen);
375 }
376 
377 /*
378  * Transfer the user time accumulated in the paca
379  * by the exception entry and exit code to the generic
380  * process user time records.
381  * Must be called with interrupts disabled.
382  * Assumes that vtime_account_system/idle() has been called
383  * recently (i.e. since the last entry from usermode) so that
384  * get_paca()->user_time_scaled is up to date.
385  */
386 void vtime_account_user(struct task_struct *tsk)
387 {
388 	cputime_t utime, utimescaled;
389 	struct cpu_accounting_data *acct = get_accounting(tsk);
390 
391 	utime = acct->user_time;
392 	utimescaled = acct->user_time_scaled;
393 	acct->user_time = 0;
394 	acct->user_time_scaled = 0;
395 	acct->utime_sspurr = 0;
396 	account_user_time(tsk, utime, utimescaled);
397 }
398 
399 #ifdef CONFIG_PPC32
400 /*
401  * Called from the context switch with interrupts disabled, to charge all
402  * accumulated times to the current process, and to prepare accounting on
403  * the next process.
404  */
405 void arch_vtime_task_switch(struct task_struct *prev)
406 {
407 	struct cpu_accounting_data *acct = get_accounting(current);
408 
409 	acct->starttime = get_accounting(prev)->starttime;
410 	acct->system_time = 0;
411 	acct->user_time = 0;
412 }
413 #endif /* CONFIG_PPC32 */
414 
415 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
416 #define calc_cputime_factors()
417 #endif
418 
419 void __delay(unsigned long loops)
420 {
421 	unsigned long start;
422 	int diff;
423 
424 	if (__USE_RTC()) {
425 		start = get_rtcl();
426 		do {
427 			/* the RTCL register wraps at 1000000000 */
428 			diff = get_rtcl() - start;
429 			if (diff < 0)
430 				diff += 1000000000;
431 		} while (diff < loops);
432 	} else {
433 		start = get_tbl();
434 		while (get_tbl() - start < loops)
435 			HMT_low();
436 		HMT_medium();
437 	}
438 }
439 EXPORT_SYMBOL(__delay);
440 
441 void udelay(unsigned long usecs)
442 {
443 	__delay(tb_ticks_per_usec * usecs);
444 }
445 EXPORT_SYMBOL(udelay);
446 
447 #ifdef CONFIG_SMP
448 unsigned long profile_pc(struct pt_regs *regs)
449 {
450 	unsigned long pc = instruction_pointer(regs);
451 
452 	if (in_lock_functions(pc))
453 		return regs->link;
454 
455 	return pc;
456 }
457 EXPORT_SYMBOL(profile_pc);
458 #endif
459 
460 #ifdef CONFIG_IRQ_WORK
461 
462 /*
463  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
464  */
465 #ifdef CONFIG_PPC64
466 static inline unsigned long test_irq_work_pending(void)
467 {
468 	unsigned long x;
469 
470 	asm volatile("lbz %0,%1(13)"
471 		: "=r" (x)
472 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
473 	return x;
474 }
475 
476 static inline void set_irq_work_pending_flag(void)
477 {
478 	asm volatile("stb %0,%1(13)" : :
479 		"r" (1),
480 		"i" (offsetof(struct paca_struct, irq_work_pending)));
481 }
482 
483 static inline void clear_irq_work_pending(void)
484 {
485 	asm volatile("stb %0,%1(13)" : :
486 		"r" (0),
487 		"i" (offsetof(struct paca_struct, irq_work_pending)));
488 }
489 
490 #else /* 32-bit */
491 
492 DEFINE_PER_CPU(u8, irq_work_pending);
493 
494 #define set_irq_work_pending_flag()	__this_cpu_write(irq_work_pending, 1)
495 #define test_irq_work_pending()		__this_cpu_read(irq_work_pending)
496 #define clear_irq_work_pending()	__this_cpu_write(irq_work_pending, 0)
497 
498 #endif /* 32 vs 64 bit */
499 
500 void arch_irq_work_raise(void)
501 {
502 	preempt_disable();
503 	set_irq_work_pending_flag();
504 	set_dec(1);
505 	preempt_enable();
506 }
507 
508 #else  /* CONFIG_IRQ_WORK */
509 
510 #define test_irq_work_pending()	0
511 #define clear_irq_work_pending()
512 
513 #endif /* CONFIG_IRQ_WORK */
514 
515 static void __timer_interrupt(void)
516 {
517 	struct pt_regs *regs = get_irq_regs();
518 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
519 	struct clock_event_device *evt = this_cpu_ptr(&decrementers);
520 	u64 now;
521 
522 	trace_timer_interrupt_entry(regs);
523 
524 	if (test_irq_work_pending()) {
525 		clear_irq_work_pending();
526 		irq_work_run();
527 	}
528 
529 	now = get_tb_or_rtc();
530 	if (now >= *next_tb) {
531 		*next_tb = ~(u64)0;
532 		if (evt->event_handler)
533 			evt->event_handler(evt);
534 		__this_cpu_inc(irq_stat.timer_irqs_event);
535 	} else {
536 		now = *next_tb - now;
537 		if (now <= decrementer_max)
538 			set_dec(now);
539 		/* We may have raced with new irq work */
540 		if (test_irq_work_pending())
541 			set_dec(1);
542 		__this_cpu_inc(irq_stat.timer_irqs_others);
543 	}
544 
545 #ifdef CONFIG_PPC64
546 	/* collect purr register values often, for accurate calculations */
547 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
548 		struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
549 		cu->current_tb = mfspr(SPRN_PURR);
550 	}
551 #endif
552 
553 	trace_timer_interrupt_exit(regs);
554 }
555 
556 /*
557  * timer_interrupt - gets called when the decrementer overflows,
558  * with interrupts disabled.
559  */
560 void timer_interrupt(struct pt_regs * regs)
561 {
562 	struct pt_regs *old_regs;
563 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
564 
565 	/* Ensure a positive value is written to the decrementer, or else
566 	 * some CPUs will continue to take decrementer exceptions.
567 	 */
568 	set_dec(decrementer_max);
569 
570 	/* Some implementations of hotplug will get timer interrupts while
571 	 * offline, just ignore these and we also need to set
572 	 * decrementers_next_tb as MAX to make sure __check_irq_replay
573 	 * don't replay timer interrupt when return, otherwise we'll trap
574 	 * here infinitely :(
575 	 */
576 	if (!cpu_online(smp_processor_id())) {
577 		*next_tb = ~(u64)0;
578 		return;
579 	}
580 
581 	/* Conditionally hard-enable interrupts now that the DEC has been
582 	 * bumped to its maximum value
583 	 */
584 	may_hard_irq_enable();
585 
586 
587 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
588 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
589 		do_IRQ(regs);
590 #endif
591 
592 	old_regs = set_irq_regs(regs);
593 	irq_enter();
594 
595 	__timer_interrupt();
596 	irq_exit();
597 	set_irq_regs(old_regs);
598 }
599 EXPORT_SYMBOL(timer_interrupt);
600 
601 /*
602  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
603  * left pending on exit from a KVM guest.  We don't need to do anything
604  * to clear them, as they are edge-triggered.
605  */
606 void hdec_interrupt(struct pt_regs *regs)
607 {
608 }
609 
610 #ifdef CONFIG_SUSPEND
611 static void generic_suspend_disable_irqs(void)
612 {
613 	/* Disable the decrementer, so that it doesn't interfere
614 	 * with suspending.
615 	 */
616 
617 	set_dec(decrementer_max);
618 	local_irq_disable();
619 	set_dec(decrementer_max);
620 }
621 
622 static void generic_suspend_enable_irqs(void)
623 {
624 	local_irq_enable();
625 }
626 
627 /* Overrides the weak version in kernel/power/main.c */
628 void arch_suspend_disable_irqs(void)
629 {
630 	if (ppc_md.suspend_disable_irqs)
631 		ppc_md.suspend_disable_irqs();
632 	generic_suspend_disable_irqs();
633 }
634 
635 /* Overrides the weak version in kernel/power/main.c */
636 void arch_suspend_enable_irqs(void)
637 {
638 	generic_suspend_enable_irqs();
639 	if (ppc_md.suspend_enable_irqs)
640 		ppc_md.suspend_enable_irqs();
641 }
642 #endif
643 
644 unsigned long long tb_to_ns(unsigned long long ticks)
645 {
646 	return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
647 }
648 EXPORT_SYMBOL_GPL(tb_to_ns);
649 
650 /*
651  * Scheduler clock - returns current time in nanosec units.
652  *
653  * Note: mulhdu(a, b) (multiply high double unsigned) returns
654  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
655  * are 64-bit unsigned numbers.
656  */
657 unsigned long long sched_clock(void)
658 {
659 	if (__USE_RTC())
660 		return get_rtc();
661 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
662 }
663 
664 
665 #ifdef CONFIG_PPC_PSERIES
666 
667 /*
668  * Running clock - attempts to give a view of time passing for a virtualised
669  * kernels.
670  * Uses the VTB register if available otherwise a next best guess.
671  */
672 unsigned long long running_clock(void)
673 {
674 	/*
675 	 * Don't read the VTB as a host since KVM does not switch in host
676 	 * timebase into the VTB when it takes a guest off the CPU, reading the
677 	 * VTB would result in reading 'last switched out' guest VTB.
678 	 *
679 	 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
680 	 * would be unsafe to rely only on the #ifdef above.
681 	 */
682 	if (firmware_has_feature(FW_FEATURE_LPAR) &&
683 	    cpu_has_feature(CPU_FTR_ARCH_207S))
684 		return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
685 
686 	/*
687 	 * This is a next best approximation without a VTB.
688 	 * On a host which is running bare metal there should never be any stolen
689 	 * time and on a host which doesn't do any virtualisation TB *should* equal
690 	 * VTB so it makes no difference anyway.
691 	 */
692 	return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
693 }
694 #endif
695 
696 static int __init get_freq(char *name, int cells, unsigned long *val)
697 {
698 	struct device_node *cpu;
699 	const __be32 *fp;
700 	int found = 0;
701 
702 	/* The cpu node should have timebase and clock frequency properties */
703 	cpu = of_find_node_by_type(NULL, "cpu");
704 
705 	if (cpu) {
706 		fp = of_get_property(cpu, name, NULL);
707 		if (fp) {
708 			found = 1;
709 			*val = of_read_ulong(fp, cells);
710 		}
711 
712 		of_node_put(cpu);
713 	}
714 
715 	return found;
716 }
717 
718 static void start_cpu_decrementer(void)
719 {
720 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
721 	/* Clear any pending timer interrupts */
722 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
723 
724 	/* Enable decrementer interrupt */
725 	mtspr(SPRN_TCR, TCR_DIE);
726 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
727 }
728 
729 void __init generic_calibrate_decr(void)
730 {
731 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
732 
733 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
734 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
735 
736 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
737 				"(not found)\n");
738 	}
739 
740 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
741 
742 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
743 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
744 
745 		printk(KERN_ERR "WARNING: Estimating processor frequency "
746 				"(not found)\n");
747 	}
748 }
749 
750 int update_persistent_clock(struct timespec now)
751 {
752 	struct rtc_time tm;
753 
754 	if (!ppc_md.set_rtc_time)
755 		return -ENODEV;
756 
757 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
758 	tm.tm_year -= 1900;
759 	tm.tm_mon -= 1;
760 
761 	return ppc_md.set_rtc_time(&tm);
762 }
763 
764 static void __read_persistent_clock(struct timespec *ts)
765 {
766 	struct rtc_time tm;
767 	static int first = 1;
768 
769 	ts->tv_nsec = 0;
770 	/* XXX this is a litle fragile but will work okay in the short term */
771 	if (first) {
772 		first = 0;
773 		if (ppc_md.time_init)
774 			timezone_offset = ppc_md.time_init();
775 
776 		/* get_boot_time() isn't guaranteed to be safe to call late */
777 		if (ppc_md.get_boot_time) {
778 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
779 			return;
780 		}
781 	}
782 	if (!ppc_md.get_rtc_time) {
783 		ts->tv_sec = 0;
784 		return;
785 	}
786 	ppc_md.get_rtc_time(&tm);
787 
788 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
789 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
790 }
791 
792 void read_persistent_clock(struct timespec *ts)
793 {
794 	__read_persistent_clock(ts);
795 
796 	/* Sanitize it in case real time clock is set below EPOCH */
797 	if (ts->tv_sec < 0) {
798 		ts->tv_sec = 0;
799 		ts->tv_nsec = 0;
800 	}
801 
802 }
803 
804 /* clocksource code */
805 static cycle_t rtc_read(struct clocksource *cs)
806 {
807 	return (cycle_t)get_rtc();
808 }
809 
810 static cycle_t timebase_read(struct clocksource *cs)
811 {
812 	return (cycle_t)get_tb();
813 }
814 
815 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
816 			 struct clocksource *clock, u32 mult, cycle_t cycle_last)
817 {
818 	u64 new_tb_to_xs, new_stamp_xsec;
819 	u32 frac_sec;
820 
821 	if (clock != &clocksource_timebase)
822 		return;
823 
824 	/* Make userspace gettimeofday spin until we're done. */
825 	++vdso_data->tb_update_count;
826 	smp_mb();
827 
828 	/* 19342813113834067 ~= 2^(20+64) / 1e9 */
829 	new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
830 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
831 	do_div(new_stamp_xsec, 1000000000);
832 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
833 
834 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
835 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
836 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
837 
838 	/*
839 	 * tb_update_count is used to allow the userspace gettimeofday code
840 	 * to assure itself that it sees a consistent view of the tb_to_xs and
841 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
842 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
843 	 * the two values of tb_update_count match and are even then the
844 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
845 	 * loops back and reads them again until this criteria is met.
846 	 * We expect the caller to have done the first increment of
847 	 * vdso_data->tb_update_count already.
848 	 */
849 	vdso_data->tb_orig_stamp = cycle_last;
850 	vdso_data->stamp_xsec = new_stamp_xsec;
851 	vdso_data->tb_to_xs = new_tb_to_xs;
852 	vdso_data->wtom_clock_sec = wtm->tv_sec;
853 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
854 	vdso_data->stamp_xtime = *wall_time;
855 	vdso_data->stamp_sec_fraction = frac_sec;
856 	smp_wmb();
857 	++(vdso_data->tb_update_count);
858 }
859 
860 void update_vsyscall_tz(void)
861 {
862 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
863 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
864 }
865 
866 static void __init clocksource_init(void)
867 {
868 	struct clocksource *clock;
869 
870 	if (__USE_RTC())
871 		clock = &clocksource_rtc;
872 	else
873 		clock = &clocksource_timebase;
874 
875 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
876 		printk(KERN_ERR "clocksource: %s is already registered\n",
877 		       clock->name);
878 		return;
879 	}
880 
881 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
882 	       clock->name, clock->mult, clock->shift);
883 }
884 
885 static int decrementer_set_next_event(unsigned long evt,
886 				      struct clock_event_device *dev)
887 {
888 	__this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
889 	set_dec(evt);
890 
891 	/* We may have raced with new irq work */
892 	if (test_irq_work_pending())
893 		set_dec(1);
894 
895 	return 0;
896 }
897 
898 static int decrementer_shutdown(struct clock_event_device *dev)
899 {
900 	decrementer_set_next_event(decrementer_max, dev);
901 	return 0;
902 }
903 
904 /* Interrupt handler for the timer broadcast IPI */
905 void tick_broadcast_ipi_handler(void)
906 {
907 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
908 
909 	*next_tb = get_tb_or_rtc();
910 	__timer_interrupt();
911 }
912 
913 static void register_decrementer_clockevent(int cpu)
914 {
915 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
916 
917 	*dec = decrementer_clockevent;
918 	dec->cpumask = cpumask_of(cpu);
919 
920 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
921 		    dec->name, dec->mult, dec->shift, cpu);
922 
923 	clockevents_register_device(dec);
924 }
925 
926 static void enable_large_decrementer(void)
927 {
928 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
929 		return;
930 
931 	if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
932 		return;
933 
934 	/*
935 	 * If we're running as the hypervisor we need to enable the LD manually
936 	 * otherwise firmware should have done it for us.
937 	 */
938 	if (cpu_has_feature(CPU_FTR_HVMODE))
939 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
940 }
941 
942 static void __init set_decrementer_max(void)
943 {
944 	struct device_node *cpu;
945 	u32 bits = 32;
946 
947 	/* Prior to ISAv3 the decrementer is always 32 bit */
948 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
949 		return;
950 
951 	cpu = of_find_node_by_type(NULL, "cpu");
952 
953 	if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
954 		if (bits > 64 || bits < 32) {
955 			pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
956 			bits = 32;
957 		}
958 
959 		/* calculate the signed maximum given this many bits */
960 		decrementer_max = (1ul << (bits - 1)) - 1;
961 	}
962 
963 	of_node_put(cpu);
964 
965 	pr_info("time_init: %u bit decrementer (max: %llx)\n",
966 		bits, decrementer_max);
967 }
968 
969 static void __init init_decrementer_clockevent(void)
970 {
971 	int cpu = smp_processor_id();
972 
973 	clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
974 
975 	decrementer_clockevent.max_delta_ns =
976 		clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
977 	decrementer_clockevent.min_delta_ns =
978 		clockevent_delta2ns(2, &decrementer_clockevent);
979 
980 	register_decrementer_clockevent(cpu);
981 }
982 
983 void secondary_cpu_time_init(void)
984 {
985 	/* Enable and test the large decrementer for this cpu */
986 	enable_large_decrementer();
987 
988 	/* Start the decrementer on CPUs that have manual control
989 	 * such as BookE
990 	 */
991 	start_cpu_decrementer();
992 
993 	/* FIME: Should make unrelatred change to move snapshot_timebase
994 	 * call here ! */
995 	register_decrementer_clockevent(smp_processor_id());
996 }
997 
998 /* This function is only called on the boot processor */
999 void __init time_init(void)
1000 {
1001 	struct div_result res;
1002 	u64 scale;
1003 	unsigned shift;
1004 
1005 	if (__USE_RTC()) {
1006 		/* 601 processor: dec counts down by 128 every 128ns */
1007 		ppc_tb_freq = 1000000000;
1008 	} else {
1009 		/* Normal PowerPC with timebase register */
1010 		ppc_md.calibrate_decr();
1011 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1012 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1013 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
1014 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1015 	}
1016 
1017 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1018 	tb_ticks_per_sec = ppc_tb_freq;
1019 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
1020 	calc_cputime_factors();
1021 	setup_cputime_one_jiffy();
1022 
1023 	/*
1024 	 * Compute scale factor for sched_clock.
1025 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1026 	 * which is the timebase frequency.
1027 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1028 	 * the 128-bit result as a 64.64 fixed-point number.
1029 	 * We then shift that number right until it is less than 1.0,
1030 	 * giving us the scale factor and shift count to use in
1031 	 * sched_clock().
1032 	 */
1033 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1034 	scale = res.result_low;
1035 	for (shift = 0; res.result_high != 0; ++shift) {
1036 		scale = (scale >> 1) | (res.result_high << 63);
1037 		res.result_high >>= 1;
1038 	}
1039 	tb_to_ns_scale = scale;
1040 	tb_to_ns_shift = shift;
1041 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1042 	boot_tb = get_tb_or_rtc();
1043 
1044 	/* If platform provided a timezone (pmac), we correct the time */
1045 	if (timezone_offset) {
1046 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1047 		sys_tz.tz_dsttime = 0;
1048 	}
1049 
1050 	vdso_data->tb_update_count = 0;
1051 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1052 
1053 	/* initialise and enable the large decrementer (if we have one) */
1054 	set_decrementer_max();
1055 	enable_large_decrementer();
1056 
1057 	/* Start the decrementer on CPUs that have manual control
1058 	 * such as BookE
1059 	 */
1060 	start_cpu_decrementer();
1061 
1062 	/* Register the clocksource */
1063 	clocksource_init();
1064 
1065 	init_decrementer_clockevent();
1066 	tick_setup_hrtimer_broadcast();
1067 
1068 #ifdef CONFIG_COMMON_CLK
1069 	of_clk_init(NULL);
1070 #endif
1071 }
1072 
1073 
1074 #define FEBRUARY	2
1075 #define	STARTOFTIME	1970
1076 #define SECDAY		86400L
1077 #define SECYR		(SECDAY * 365)
1078 #define	leapyear(year)		((year) % 4 == 0 && \
1079 				 ((year) % 100 != 0 || (year) % 400 == 0))
1080 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1081 #define	days_in_month(a) 	(month_days[(a) - 1])
1082 
1083 static int month_days[12] = {
1084 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1085 };
1086 
1087 void to_tm(int tim, struct rtc_time * tm)
1088 {
1089 	register int    i;
1090 	register long   hms, day;
1091 
1092 	day = tim / SECDAY;
1093 	hms = tim % SECDAY;
1094 
1095 	/* Hours, minutes, seconds are easy */
1096 	tm->tm_hour = hms / 3600;
1097 	tm->tm_min = (hms % 3600) / 60;
1098 	tm->tm_sec = (hms % 3600) % 60;
1099 
1100 	/* Number of years in days */
1101 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1102 		day -= days_in_year(i);
1103 	tm->tm_year = i;
1104 
1105 	/* Number of months in days left */
1106 	if (leapyear(tm->tm_year))
1107 		days_in_month(FEBRUARY) = 29;
1108 	for (i = 1; day >= days_in_month(i); i++)
1109 		day -= days_in_month(i);
1110 	days_in_month(FEBRUARY) = 28;
1111 	tm->tm_mon = i;
1112 
1113 	/* Days are what is left over (+1) from all that. */
1114 	tm->tm_mday = day + 1;
1115 
1116 	/*
1117 	 * No-one uses the day of the week.
1118 	 */
1119 	tm->tm_wday = -1;
1120 }
1121 EXPORT_SYMBOL(to_tm);
1122 
1123 /*
1124  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1125  * result.
1126  */
1127 void div128_by_32(u64 dividend_high, u64 dividend_low,
1128 		  unsigned divisor, struct div_result *dr)
1129 {
1130 	unsigned long a, b, c, d;
1131 	unsigned long w, x, y, z;
1132 	u64 ra, rb, rc;
1133 
1134 	a = dividend_high >> 32;
1135 	b = dividend_high & 0xffffffff;
1136 	c = dividend_low >> 32;
1137 	d = dividend_low & 0xffffffff;
1138 
1139 	w = a / divisor;
1140 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1141 
1142 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1143 	x = ra;
1144 
1145 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1146 	y = rb;
1147 
1148 	do_div(rc, divisor);
1149 	z = rc;
1150 
1151 	dr->result_high = ((u64)w << 32) + x;
1152 	dr->result_low  = ((u64)y << 32) + z;
1153 
1154 }
1155 
1156 /* We don't need to calibrate delay, we use the CPU timebase for that */
1157 void calibrate_delay(void)
1158 {
1159 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1160 	 * as the number of __delay(1) in a jiffy, so make it so
1161 	 */
1162 	loops_per_jiffy = tb_ticks_per_jiffy;
1163 }
1164 
1165 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1166 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1167 {
1168 	ppc_md.get_rtc_time(tm);
1169 	return rtc_valid_tm(tm);
1170 }
1171 
1172 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1173 {
1174 	if (!ppc_md.set_rtc_time)
1175 		return -EOPNOTSUPP;
1176 
1177 	if (ppc_md.set_rtc_time(tm) < 0)
1178 		return -EOPNOTSUPP;
1179 
1180 	return 0;
1181 }
1182 
1183 static const struct rtc_class_ops rtc_generic_ops = {
1184 	.read_time = rtc_generic_get_time,
1185 	.set_time = rtc_generic_set_time,
1186 };
1187 
1188 static int __init rtc_init(void)
1189 {
1190 	struct platform_device *pdev;
1191 
1192 	if (!ppc_md.get_rtc_time)
1193 		return -ENODEV;
1194 
1195 	pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1196 					     &rtc_generic_ops,
1197 					     sizeof(rtc_generic_ops));
1198 
1199 	return PTR_ERR_OR_ZERO(pdev);
1200 }
1201 
1202 device_initcall(rtc_init);
1203 #endif
1204