xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision 7008eb99)
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 
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 EXPORT_SYMBOL(timer_interrupt);
598 
599 /*
600  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
601  * left pending on exit from a KVM guest.  We don't need to do anything
602  * to clear them, as they are edge-triggered.
603  */
604 void hdec_interrupt(struct pt_regs *regs)
605 {
606 }
607 
608 #ifdef CONFIG_SUSPEND
609 static void generic_suspend_disable_irqs(void)
610 {
611 	/* Disable the decrementer, so that it doesn't interfere
612 	 * with suspending.
613 	 */
614 
615 	set_dec(decrementer_max);
616 	local_irq_disable();
617 	set_dec(decrementer_max);
618 }
619 
620 static void generic_suspend_enable_irqs(void)
621 {
622 	local_irq_enable();
623 }
624 
625 /* Overrides the weak version in kernel/power/main.c */
626 void arch_suspend_disable_irqs(void)
627 {
628 	if (ppc_md.suspend_disable_irqs)
629 		ppc_md.suspend_disable_irqs();
630 	generic_suspend_disable_irqs();
631 }
632 
633 /* Overrides the weak version in kernel/power/main.c */
634 void arch_suspend_enable_irqs(void)
635 {
636 	generic_suspend_enable_irqs();
637 	if (ppc_md.suspend_enable_irqs)
638 		ppc_md.suspend_enable_irqs();
639 }
640 #endif
641 
642 unsigned long long tb_to_ns(unsigned long long ticks)
643 {
644 	return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
645 }
646 EXPORT_SYMBOL_GPL(tb_to_ns);
647 
648 /*
649  * Scheduler clock - returns current time in nanosec units.
650  *
651  * Note: mulhdu(a, b) (multiply high double unsigned) returns
652  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
653  * are 64-bit unsigned numbers.
654  */
655 unsigned long long sched_clock(void)
656 {
657 	if (__USE_RTC())
658 		return get_rtc();
659 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
660 }
661 
662 
663 #ifdef CONFIG_PPC_PSERIES
664 
665 /*
666  * Running clock - attempts to give a view of time passing for a virtualised
667  * kernels.
668  * Uses the VTB register if available otherwise a next best guess.
669  */
670 unsigned long long running_clock(void)
671 {
672 	/*
673 	 * Don't read the VTB as a host since KVM does not switch in host
674 	 * timebase into the VTB when it takes a guest off the CPU, reading the
675 	 * VTB would result in reading 'last switched out' guest VTB.
676 	 *
677 	 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
678 	 * would be unsafe to rely only on the #ifdef above.
679 	 */
680 	if (firmware_has_feature(FW_FEATURE_LPAR) &&
681 	    cpu_has_feature(CPU_FTR_ARCH_207S))
682 		return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
683 
684 	/*
685 	 * This is a next best approximation without a VTB.
686 	 * On a host which is running bare metal there should never be any stolen
687 	 * time and on a host which doesn't do any virtualisation TB *should* equal
688 	 * VTB so it makes no difference anyway.
689 	 */
690 	return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
691 }
692 #endif
693 
694 static int __init get_freq(char *name, int cells, unsigned long *val)
695 {
696 	struct device_node *cpu;
697 	const __be32 *fp;
698 	int found = 0;
699 
700 	/* The cpu node should have timebase and clock frequency properties */
701 	cpu = of_find_node_by_type(NULL, "cpu");
702 
703 	if (cpu) {
704 		fp = of_get_property(cpu, name, NULL);
705 		if (fp) {
706 			found = 1;
707 			*val = of_read_ulong(fp, cells);
708 		}
709 
710 		of_node_put(cpu);
711 	}
712 
713 	return found;
714 }
715 
716 static void start_cpu_decrementer(void)
717 {
718 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
719 	/* Clear any pending timer interrupts */
720 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
721 
722 	/* Enable decrementer interrupt */
723 	mtspr(SPRN_TCR, TCR_DIE);
724 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
725 }
726 
727 void __init generic_calibrate_decr(void)
728 {
729 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
730 
731 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
732 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
733 
734 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
735 				"(not found)\n");
736 	}
737 
738 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
739 
740 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
741 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
742 
743 		printk(KERN_ERR "WARNING: Estimating processor frequency "
744 				"(not found)\n");
745 	}
746 }
747 
748 int update_persistent_clock(struct timespec now)
749 {
750 	struct rtc_time tm;
751 
752 	if (!ppc_md.set_rtc_time)
753 		return -ENODEV;
754 
755 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
756 	tm.tm_year -= 1900;
757 	tm.tm_mon -= 1;
758 
759 	return ppc_md.set_rtc_time(&tm);
760 }
761 
762 static void __read_persistent_clock(struct timespec *ts)
763 {
764 	struct rtc_time tm;
765 	static int first = 1;
766 
767 	ts->tv_nsec = 0;
768 	/* XXX this is a litle fragile but will work okay in the short term */
769 	if (first) {
770 		first = 0;
771 		if (ppc_md.time_init)
772 			timezone_offset = ppc_md.time_init();
773 
774 		/* get_boot_time() isn't guaranteed to be safe to call late */
775 		if (ppc_md.get_boot_time) {
776 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
777 			return;
778 		}
779 	}
780 	if (!ppc_md.get_rtc_time) {
781 		ts->tv_sec = 0;
782 		return;
783 	}
784 	ppc_md.get_rtc_time(&tm);
785 
786 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
787 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
788 }
789 
790 void read_persistent_clock(struct timespec *ts)
791 {
792 	__read_persistent_clock(ts);
793 
794 	/* Sanitize it in case real time clock is set below EPOCH */
795 	if (ts->tv_sec < 0) {
796 		ts->tv_sec = 0;
797 		ts->tv_nsec = 0;
798 	}
799 
800 }
801 
802 /* clocksource code */
803 static cycle_t rtc_read(struct clocksource *cs)
804 {
805 	return (cycle_t)get_rtc();
806 }
807 
808 static cycle_t timebase_read(struct clocksource *cs)
809 {
810 	return (cycle_t)get_tb();
811 }
812 
813 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
814 			 struct clocksource *clock, u32 mult, cycle_t cycle_last)
815 {
816 	u64 new_tb_to_xs, new_stamp_xsec;
817 	u32 frac_sec;
818 
819 	if (clock != &clocksource_timebase)
820 		return;
821 
822 	/* Make userspace gettimeofday spin until we're done. */
823 	++vdso_data->tb_update_count;
824 	smp_mb();
825 
826 	/* 19342813113834067 ~= 2^(20+64) / 1e9 */
827 	new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
828 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
829 	do_div(new_stamp_xsec, 1000000000);
830 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
831 
832 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
833 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
834 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
835 
836 	/*
837 	 * tb_update_count is used to allow the userspace gettimeofday code
838 	 * to assure itself that it sees a consistent view of the tb_to_xs and
839 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
840 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
841 	 * the two values of tb_update_count match and are even then the
842 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
843 	 * loops back and reads them again until this criteria is met.
844 	 * We expect the caller to have done the first increment of
845 	 * vdso_data->tb_update_count already.
846 	 */
847 	vdso_data->tb_orig_stamp = cycle_last;
848 	vdso_data->stamp_xsec = new_stamp_xsec;
849 	vdso_data->tb_to_xs = new_tb_to_xs;
850 	vdso_data->wtom_clock_sec = wtm->tv_sec;
851 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
852 	vdso_data->stamp_xtime = *wall_time;
853 	vdso_data->stamp_sec_fraction = frac_sec;
854 	smp_wmb();
855 	++(vdso_data->tb_update_count);
856 }
857 
858 void update_vsyscall_tz(void)
859 {
860 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
861 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
862 }
863 
864 static void __init clocksource_init(void)
865 {
866 	struct clocksource *clock;
867 
868 	if (__USE_RTC())
869 		clock = &clocksource_rtc;
870 	else
871 		clock = &clocksource_timebase;
872 
873 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
874 		printk(KERN_ERR "clocksource: %s is already registered\n",
875 		       clock->name);
876 		return;
877 	}
878 
879 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
880 	       clock->name, clock->mult, clock->shift);
881 }
882 
883 static int decrementer_set_next_event(unsigned long evt,
884 				      struct clock_event_device *dev)
885 {
886 	__this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
887 	set_dec(evt);
888 
889 	/* We may have raced with new irq work */
890 	if (test_irq_work_pending())
891 		set_dec(1);
892 
893 	return 0;
894 }
895 
896 static int decrementer_shutdown(struct clock_event_device *dev)
897 {
898 	decrementer_set_next_event(decrementer_max, dev);
899 	return 0;
900 }
901 
902 /* Interrupt handler for the timer broadcast IPI */
903 void tick_broadcast_ipi_handler(void)
904 {
905 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
906 
907 	*next_tb = get_tb_or_rtc();
908 	__timer_interrupt();
909 }
910 
911 static void register_decrementer_clockevent(int cpu)
912 {
913 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
914 
915 	*dec = decrementer_clockevent;
916 	dec->cpumask = cpumask_of(cpu);
917 
918 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
919 		    dec->name, dec->mult, dec->shift, cpu);
920 
921 	clockevents_register_device(dec);
922 }
923 
924 static void enable_large_decrementer(void)
925 {
926 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
927 		return;
928 
929 	if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
930 		return;
931 
932 	/*
933 	 * If we're running as the hypervisor we need to enable the LD manually
934 	 * otherwise firmware should have done it for us.
935 	 */
936 	if (cpu_has_feature(CPU_FTR_HVMODE))
937 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
938 }
939 
940 static void __init set_decrementer_max(void)
941 {
942 	struct device_node *cpu;
943 	u32 bits = 32;
944 
945 	/* Prior to ISAv3 the decrementer is always 32 bit */
946 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
947 		return;
948 
949 	cpu = of_find_node_by_type(NULL, "cpu");
950 
951 	if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
952 		if (bits > 64 || bits < 32) {
953 			pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
954 			bits = 32;
955 		}
956 
957 		/* calculate the signed maximum given this many bits */
958 		decrementer_max = (1ul << (bits - 1)) - 1;
959 	}
960 
961 	of_node_put(cpu);
962 
963 	pr_info("time_init: %u bit decrementer (max: %llx)\n",
964 		bits, decrementer_max);
965 }
966 
967 static void __init init_decrementer_clockevent(void)
968 {
969 	int cpu = smp_processor_id();
970 
971 	clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
972 
973 	decrementer_clockevent.max_delta_ns =
974 		clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
975 	decrementer_clockevent.min_delta_ns =
976 		clockevent_delta2ns(2, &decrementer_clockevent);
977 
978 	register_decrementer_clockevent(cpu);
979 }
980 
981 void secondary_cpu_time_init(void)
982 {
983 	/* Enable and test the large decrementer for this cpu */
984 	enable_large_decrementer();
985 
986 	/* Start the decrementer on CPUs that have manual control
987 	 * such as BookE
988 	 */
989 	start_cpu_decrementer();
990 
991 	/* FIME: Should make unrelatred change to move snapshot_timebase
992 	 * call here ! */
993 	register_decrementer_clockevent(smp_processor_id());
994 }
995 
996 /* This function is only called on the boot processor */
997 void __init time_init(void)
998 {
999 	struct div_result res;
1000 	u64 scale;
1001 	unsigned shift;
1002 
1003 	if (__USE_RTC()) {
1004 		/* 601 processor: dec counts down by 128 every 128ns */
1005 		ppc_tb_freq = 1000000000;
1006 	} else {
1007 		/* Normal PowerPC with timebase register */
1008 		ppc_md.calibrate_decr();
1009 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1010 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1011 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
1012 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1013 	}
1014 
1015 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1016 	tb_ticks_per_sec = ppc_tb_freq;
1017 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
1018 	calc_cputime_factors();
1019 	setup_cputime_one_jiffy();
1020 
1021 	/*
1022 	 * Compute scale factor for sched_clock.
1023 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1024 	 * which is the timebase frequency.
1025 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1026 	 * the 128-bit result as a 64.64 fixed-point number.
1027 	 * We then shift that number right until it is less than 1.0,
1028 	 * giving us the scale factor and shift count to use in
1029 	 * sched_clock().
1030 	 */
1031 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1032 	scale = res.result_low;
1033 	for (shift = 0; res.result_high != 0; ++shift) {
1034 		scale = (scale >> 1) | (res.result_high << 63);
1035 		res.result_high >>= 1;
1036 	}
1037 	tb_to_ns_scale = scale;
1038 	tb_to_ns_shift = shift;
1039 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1040 	boot_tb = get_tb_or_rtc();
1041 
1042 	/* If platform provided a timezone (pmac), we correct the time */
1043 	if (timezone_offset) {
1044 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1045 		sys_tz.tz_dsttime = 0;
1046 	}
1047 
1048 	vdso_data->tb_update_count = 0;
1049 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1050 
1051 	/* initialise and enable the large decrementer (if we have one) */
1052 	set_decrementer_max();
1053 	enable_large_decrementer();
1054 
1055 	/* Start the decrementer on CPUs that have manual control
1056 	 * such as BookE
1057 	 */
1058 	start_cpu_decrementer();
1059 
1060 	/* Register the clocksource */
1061 	clocksource_init();
1062 
1063 	init_decrementer_clockevent();
1064 	tick_setup_hrtimer_broadcast();
1065 
1066 #ifdef CONFIG_COMMON_CLK
1067 	of_clk_init(NULL);
1068 #endif
1069 }
1070 
1071 
1072 #define FEBRUARY	2
1073 #define	STARTOFTIME	1970
1074 #define SECDAY		86400L
1075 #define SECYR		(SECDAY * 365)
1076 #define	leapyear(year)		((year) % 4 == 0 && \
1077 				 ((year) % 100 != 0 || (year) % 400 == 0))
1078 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1079 #define	days_in_month(a) 	(month_days[(a) - 1])
1080 
1081 static int month_days[12] = {
1082 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1083 };
1084 
1085 void to_tm(int tim, struct rtc_time * tm)
1086 {
1087 	register int    i;
1088 	register long   hms, day;
1089 
1090 	day = tim / SECDAY;
1091 	hms = tim % SECDAY;
1092 
1093 	/* Hours, minutes, seconds are easy */
1094 	tm->tm_hour = hms / 3600;
1095 	tm->tm_min = (hms % 3600) / 60;
1096 	tm->tm_sec = (hms % 3600) % 60;
1097 
1098 	/* Number of years in days */
1099 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1100 		day -= days_in_year(i);
1101 	tm->tm_year = i;
1102 
1103 	/* Number of months in days left */
1104 	if (leapyear(tm->tm_year))
1105 		days_in_month(FEBRUARY) = 29;
1106 	for (i = 1; day >= days_in_month(i); i++)
1107 		day -= days_in_month(i);
1108 	days_in_month(FEBRUARY) = 28;
1109 	tm->tm_mon = i;
1110 
1111 	/* Days are what is left over (+1) from all that. */
1112 	tm->tm_mday = day + 1;
1113 
1114 	/*
1115 	 * No-one uses the day of the week.
1116 	 */
1117 	tm->tm_wday = -1;
1118 }
1119 EXPORT_SYMBOL(to_tm);
1120 
1121 /*
1122  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1123  * result.
1124  */
1125 void div128_by_32(u64 dividend_high, u64 dividend_low,
1126 		  unsigned divisor, struct div_result *dr)
1127 {
1128 	unsigned long a, b, c, d;
1129 	unsigned long w, x, y, z;
1130 	u64 ra, rb, rc;
1131 
1132 	a = dividend_high >> 32;
1133 	b = dividend_high & 0xffffffff;
1134 	c = dividend_low >> 32;
1135 	d = dividend_low & 0xffffffff;
1136 
1137 	w = a / divisor;
1138 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1139 
1140 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1141 	x = ra;
1142 
1143 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1144 	y = rb;
1145 
1146 	do_div(rc, divisor);
1147 	z = rc;
1148 
1149 	dr->result_high = ((u64)w << 32) + x;
1150 	dr->result_low  = ((u64)y << 32) + z;
1151 
1152 }
1153 
1154 /* We don't need to calibrate delay, we use the CPU timebase for that */
1155 void calibrate_delay(void)
1156 {
1157 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1158 	 * as the number of __delay(1) in a jiffy, so make it so
1159 	 */
1160 	loops_per_jiffy = tb_ticks_per_jiffy;
1161 }
1162 
1163 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1164 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1165 {
1166 	ppc_md.get_rtc_time(tm);
1167 	return rtc_valid_tm(tm);
1168 }
1169 
1170 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1171 {
1172 	if (!ppc_md.set_rtc_time)
1173 		return -EOPNOTSUPP;
1174 
1175 	if (ppc_md.set_rtc_time(tm) < 0)
1176 		return -EOPNOTSUPP;
1177 
1178 	return 0;
1179 }
1180 
1181 static const struct rtc_class_ops rtc_generic_ops = {
1182 	.read_time = rtc_generic_get_time,
1183 	.set_time = rtc_generic_set_time,
1184 };
1185 
1186 static int __init rtc_init(void)
1187 {
1188 	struct platform_device *pdev;
1189 
1190 	if (!ppc_md.get_rtc_time)
1191 		return -ENODEV;
1192 
1193 	pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1194 					     &rtc_generic_ops,
1195 					     sizeof(rtc_generic_ops));
1196 
1197 	return PTR_ERR_OR_ZERO(pdev);
1198 }
1199 
1200 device_initcall(rtc_init);
1201 #endif
1202