xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision f828c3d0)
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 <linux/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 u64 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 u64 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->stime -= sst;
275 	acct->utime -= ust;
276 	acct->steal_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 	struct cpu_accounting_data *acct = &local_paca->accounting;
285 
286 	if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
287 		stolen = scan_dispatch_log(stop_tb);
288 		acct->stime -= stolen;
289 	}
290 
291 	stolen += acct->steal_time;
292 	acct->steal_time = 0;
293 	return stolen;
294 }
295 
296 #else /* CONFIG_PPC_SPLPAR */
297 static inline u64 calculate_stolen_time(u64 stop_tb)
298 {
299 	return 0;
300 }
301 
302 #endif /* CONFIG_PPC_SPLPAR */
303 
304 /*
305  * Account time for a transition between system, hard irq
306  * or soft irq state.
307  */
308 static unsigned long vtime_delta(struct task_struct *tsk,
309 				 unsigned long *sys_scaled,
310 				 unsigned long *stolen)
311 {
312 	unsigned long now, nowscaled, deltascaled;
313 	unsigned long udelta, delta, user_scaled;
314 	struct cpu_accounting_data *acct = get_accounting(tsk);
315 
316 	WARN_ON_ONCE(!irqs_disabled());
317 
318 	now = mftb();
319 	nowscaled = read_spurr(now);
320 	acct->stime += now - acct->starttime;
321 	acct->starttime = now;
322 	deltascaled = nowscaled - acct->startspurr;
323 	acct->startspurr = nowscaled;
324 
325 	*stolen = calculate_stolen_time(now);
326 
327 	delta = acct->stime;
328 	acct->stime = 0;
329 	udelta = acct->utime - acct->utime_sspurr;
330 	acct->utime_sspurr = acct->utime;
331 
332 	/*
333 	 * Because we don't read the SPURR on every kernel entry/exit,
334 	 * deltascaled includes both user and system SPURR ticks.
335 	 * Apportion these ticks to system SPURR ticks and user
336 	 * SPURR ticks in the same ratio as the system time (delta)
337 	 * and user time (udelta) values obtained from the timebase
338 	 * over the same interval.  The system ticks get accounted here;
339 	 * the user ticks get saved up in paca->user_time_scaled to be
340 	 * used by account_process_tick.
341 	 */
342 	*sys_scaled = delta;
343 	user_scaled = udelta;
344 	if (deltascaled != delta + udelta) {
345 		if (udelta) {
346 			*sys_scaled = deltascaled * delta / (delta + udelta);
347 			user_scaled = deltascaled - *sys_scaled;
348 		} else {
349 			*sys_scaled = deltascaled;
350 		}
351 	}
352 	acct->utime_scaled += user_scaled;
353 
354 	return delta;
355 }
356 
357 void vtime_account_system(struct task_struct *tsk)
358 {
359 	unsigned long delta, sys_scaled, stolen;
360 
361 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
362 	account_system_time(tsk, 0, delta);
363 	tsk->stimescaled += 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->utime;
392 	utimescaled = acct->utime_scaled;
393 	acct->utime = 0;
394 	acct->utime_scaled = 0;
395 	acct->utime_sspurr = 0;
396 	account_user_time(tsk, utime);
397 	tsk->utimescaled += utimescaled;
398 }
399 
400 #ifdef CONFIG_PPC32
401 /*
402  * Called from the context switch with interrupts disabled, to charge all
403  * accumulated times to the current process, and to prepare accounting on
404  * the next process.
405  */
406 void arch_vtime_task_switch(struct task_struct *prev)
407 {
408 	struct cpu_accounting_data *acct = get_accounting(current);
409 
410 	acct->starttime = get_accounting(prev)->starttime;
411 	acct->startspurr = get_accounting(prev)->startspurr;
412 	acct->stime = 0;
413 	acct->utime = 0;
414 }
415 #endif /* CONFIG_PPC32 */
416 
417 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
418 #define calc_cputime_factors()
419 #endif
420 
421 void __delay(unsigned long loops)
422 {
423 	unsigned long start;
424 	int diff;
425 
426 	if (__USE_RTC()) {
427 		start = get_rtcl();
428 		do {
429 			/* the RTCL register wraps at 1000000000 */
430 			diff = get_rtcl() - start;
431 			if (diff < 0)
432 				diff += 1000000000;
433 		} while (diff < loops);
434 	} else {
435 		start = get_tbl();
436 		while (get_tbl() - start < loops)
437 			HMT_low();
438 		HMT_medium();
439 	}
440 }
441 EXPORT_SYMBOL(__delay);
442 
443 void udelay(unsigned long usecs)
444 {
445 	__delay(tb_ticks_per_usec * usecs);
446 }
447 EXPORT_SYMBOL(udelay);
448 
449 #ifdef CONFIG_SMP
450 unsigned long profile_pc(struct pt_regs *regs)
451 {
452 	unsigned long pc = instruction_pointer(regs);
453 
454 	if (in_lock_functions(pc))
455 		return regs->link;
456 
457 	return pc;
458 }
459 EXPORT_SYMBOL(profile_pc);
460 #endif
461 
462 #ifdef CONFIG_IRQ_WORK
463 
464 /*
465  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
466  */
467 #ifdef CONFIG_PPC64
468 static inline unsigned long test_irq_work_pending(void)
469 {
470 	unsigned long x;
471 
472 	asm volatile("lbz %0,%1(13)"
473 		: "=r" (x)
474 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
475 	return x;
476 }
477 
478 static inline void set_irq_work_pending_flag(void)
479 {
480 	asm volatile("stb %0,%1(13)" : :
481 		"r" (1),
482 		"i" (offsetof(struct paca_struct, irq_work_pending)));
483 }
484 
485 static inline void clear_irq_work_pending(void)
486 {
487 	asm volatile("stb %0,%1(13)" : :
488 		"r" (0),
489 		"i" (offsetof(struct paca_struct, irq_work_pending)));
490 }
491 
492 #else /* 32-bit */
493 
494 DEFINE_PER_CPU(u8, irq_work_pending);
495 
496 #define set_irq_work_pending_flag()	__this_cpu_write(irq_work_pending, 1)
497 #define test_irq_work_pending()		__this_cpu_read(irq_work_pending)
498 #define clear_irq_work_pending()	__this_cpu_write(irq_work_pending, 0)
499 
500 #endif /* 32 vs 64 bit */
501 
502 void arch_irq_work_raise(void)
503 {
504 	preempt_disable();
505 	set_irq_work_pending_flag();
506 	set_dec(1);
507 	preempt_enable();
508 }
509 
510 #else  /* CONFIG_IRQ_WORK */
511 
512 #define test_irq_work_pending()	0
513 #define clear_irq_work_pending()
514 
515 #endif /* CONFIG_IRQ_WORK */
516 
517 static void __timer_interrupt(void)
518 {
519 	struct pt_regs *regs = get_irq_regs();
520 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
521 	struct clock_event_device *evt = this_cpu_ptr(&decrementers);
522 	u64 now;
523 
524 	trace_timer_interrupt_entry(regs);
525 
526 	if (test_irq_work_pending()) {
527 		clear_irq_work_pending();
528 		irq_work_run();
529 	}
530 
531 	now = get_tb_or_rtc();
532 	if (now >= *next_tb) {
533 		*next_tb = ~(u64)0;
534 		if (evt->event_handler)
535 			evt->event_handler(evt);
536 		__this_cpu_inc(irq_stat.timer_irqs_event);
537 	} else {
538 		now = *next_tb - now;
539 		if (now <= decrementer_max)
540 			set_dec(now);
541 		/* We may have raced with new irq work */
542 		if (test_irq_work_pending())
543 			set_dec(1);
544 		__this_cpu_inc(irq_stat.timer_irqs_others);
545 	}
546 
547 #ifdef CONFIG_PPC64
548 	/* collect purr register values often, for accurate calculations */
549 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
550 		struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
551 		cu->current_tb = mfspr(SPRN_PURR);
552 	}
553 #endif
554 
555 	trace_timer_interrupt_exit(regs);
556 }
557 
558 /*
559  * timer_interrupt - gets called when the decrementer overflows,
560  * with interrupts disabled.
561  */
562 void timer_interrupt(struct pt_regs * regs)
563 {
564 	struct pt_regs *old_regs;
565 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
566 
567 	/* Ensure a positive value is written to the decrementer, or else
568 	 * some CPUs will continue to take decrementer exceptions.
569 	 */
570 	set_dec(decrementer_max);
571 
572 	/* Some implementations of hotplug will get timer interrupts while
573 	 * offline, just ignore these and we also need to set
574 	 * decrementers_next_tb as MAX to make sure __check_irq_replay
575 	 * don't replay timer interrupt when return, otherwise we'll trap
576 	 * here infinitely :(
577 	 */
578 	if (!cpu_online(smp_processor_id())) {
579 		*next_tb = ~(u64)0;
580 		return;
581 	}
582 
583 	/* Conditionally hard-enable interrupts now that the DEC has been
584 	 * bumped to its maximum value
585 	 */
586 	may_hard_irq_enable();
587 
588 
589 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
590 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
591 		do_IRQ(regs);
592 #endif
593 
594 	old_regs = set_irq_regs(regs);
595 	irq_enter();
596 
597 	__timer_interrupt();
598 	irq_exit();
599 	set_irq_regs(old_regs);
600 }
601 EXPORT_SYMBOL(timer_interrupt);
602 
603 /*
604  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
605  * left pending on exit from a KVM guest.  We don't need to do anything
606  * to clear them, as they are edge-triggered.
607  */
608 void hdec_interrupt(struct pt_regs *regs)
609 {
610 }
611 
612 #ifdef CONFIG_SUSPEND
613 static void generic_suspend_disable_irqs(void)
614 {
615 	/* Disable the decrementer, so that it doesn't interfere
616 	 * with suspending.
617 	 */
618 
619 	set_dec(decrementer_max);
620 	local_irq_disable();
621 	set_dec(decrementer_max);
622 }
623 
624 static void generic_suspend_enable_irqs(void)
625 {
626 	local_irq_enable();
627 }
628 
629 /* Overrides the weak version in kernel/power/main.c */
630 void arch_suspend_disable_irqs(void)
631 {
632 	if (ppc_md.suspend_disable_irqs)
633 		ppc_md.suspend_disable_irqs();
634 	generic_suspend_disable_irqs();
635 }
636 
637 /* Overrides the weak version in kernel/power/main.c */
638 void arch_suspend_enable_irqs(void)
639 {
640 	generic_suspend_enable_irqs();
641 	if (ppc_md.suspend_enable_irqs)
642 		ppc_md.suspend_enable_irqs();
643 }
644 #endif
645 
646 unsigned long long tb_to_ns(unsigned long long ticks)
647 {
648 	return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
649 }
650 EXPORT_SYMBOL_GPL(tb_to_ns);
651 
652 /*
653  * Scheduler clock - returns current time in nanosec units.
654  *
655  * Note: mulhdu(a, b) (multiply high double unsigned) returns
656  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
657  * are 64-bit unsigned numbers.
658  */
659 unsigned long long sched_clock(void)
660 {
661 	if (__USE_RTC())
662 		return get_rtc();
663 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
664 }
665 
666 
667 #ifdef CONFIG_PPC_PSERIES
668 
669 /*
670  * Running clock - attempts to give a view of time passing for a virtualised
671  * kernels.
672  * Uses the VTB register if available otherwise a next best guess.
673  */
674 unsigned long long running_clock(void)
675 {
676 	/*
677 	 * Don't read the VTB as a host since KVM does not switch in host
678 	 * timebase into the VTB when it takes a guest off the CPU, reading the
679 	 * VTB would result in reading 'last switched out' guest VTB.
680 	 *
681 	 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
682 	 * would be unsafe to rely only on the #ifdef above.
683 	 */
684 	if (firmware_has_feature(FW_FEATURE_LPAR) &&
685 	    cpu_has_feature(CPU_FTR_ARCH_207S))
686 		return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
687 
688 	/*
689 	 * This is a next best approximation without a VTB.
690 	 * On a host which is running bare metal there should never be any stolen
691 	 * time and on a host which doesn't do any virtualisation TB *should* equal
692 	 * VTB so it makes no difference anyway.
693 	 */
694 	return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
695 }
696 #endif
697 
698 static int __init get_freq(char *name, int cells, unsigned long *val)
699 {
700 	struct device_node *cpu;
701 	const __be32 *fp;
702 	int found = 0;
703 
704 	/* The cpu node should have timebase and clock frequency properties */
705 	cpu = of_find_node_by_type(NULL, "cpu");
706 
707 	if (cpu) {
708 		fp = of_get_property(cpu, name, NULL);
709 		if (fp) {
710 			found = 1;
711 			*val = of_read_ulong(fp, cells);
712 		}
713 
714 		of_node_put(cpu);
715 	}
716 
717 	return found;
718 }
719 
720 static void start_cpu_decrementer(void)
721 {
722 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
723 	/* Clear any pending timer interrupts */
724 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
725 
726 	/* Enable decrementer interrupt */
727 	mtspr(SPRN_TCR, TCR_DIE);
728 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
729 }
730 
731 void __init generic_calibrate_decr(void)
732 {
733 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
734 
735 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
736 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
737 
738 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
739 				"(not found)\n");
740 	}
741 
742 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
743 
744 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
745 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
746 
747 		printk(KERN_ERR "WARNING: Estimating processor frequency "
748 				"(not found)\n");
749 	}
750 }
751 
752 int update_persistent_clock(struct timespec now)
753 {
754 	struct rtc_time tm;
755 
756 	if (!ppc_md.set_rtc_time)
757 		return -ENODEV;
758 
759 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
760 	tm.tm_year -= 1900;
761 	tm.tm_mon -= 1;
762 
763 	return ppc_md.set_rtc_time(&tm);
764 }
765 
766 static void __read_persistent_clock(struct timespec *ts)
767 {
768 	struct rtc_time tm;
769 	static int first = 1;
770 
771 	ts->tv_nsec = 0;
772 	/* XXX this is a litle fragile but will work okay in the short term */
773 	if (first) {
774 		first = 0;
775 		if (ppc_md.time_init)
776 			timezone_offset = ppc_md.time_init();
777 
778 		/* get_boot_time() isn't guaranteed to be safe to call late */
779 		if (ppc_md.get_boot_time) {
780 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
781 			return;
782 		}
783 	}
784 	if (!ppc_md.get_rtc_time) {
785 		ts->tv_sec = 0;
786 		return;
787 	}
788 	ppc_md.get_rtc_time(&tm);
789 
790 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
791 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
792 }
793 
794 void read_persistent_clock(struct timespec *ts)
795 {
796 	__read_persistent_clock(ts);
797 
798 	/* Sanitize it in case real time clock is set below EPOCH */
799 	if (ts->tv_sec < 0) {
800 		ts->tv_sec = 0;
801 		ts->tv_nsec = 0;
802 	}
803 
804 }
805 
806 /* clocksource code */
807 static u64 rtc_read(struct clocksource *cs)
808 {
809 	return (u64)get_rtc();
810 }
811 
812 static u64 timebase_read(struct clocksource *cs)
813 {
814 	return (u64)get_tb();
815 }
816 
817 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
818 			 struct clocksource *clock, u32 mult, u64 cycle_last)
819 {
820 	u64 new_tb_to_xs, new_stamp_xsec;
821 	u32 frac_sec;
822 
823 	if (clock != &clocksource_timebase)
824 		return;
825 
826 	/* Make userspace gettimeofday spin until we're done. */
827 	++vdso_data->tb_update_count;
828 	smp_mb();
829 
830 	/* 19342813113834067 ~= 2^(20+64) / 1e9 */
831 	new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
832 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
833 	do_div(new_stamp_xsec, 1000000000);
834 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
835 
836 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
837 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
838 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
839 
840 	/*
841 	 * tb_update_count is used to allow the userspace gettimeofday code
842 	 * to assure itself that it sees a consistent view of the tb_to_xs and
843 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
844 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
845 	 * the two values of tb_update_count match and are even then the
846 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
847 	 * loops back and reads them again until this criteria is met.
848 	 * We expect the caller to have done the first increment of
849 	 * vdso_data->tb_update_count already.
850 	 */
851 	vdso_data->tb_orig_stamp = cycle_last;
852 	vdso_data->stamp_xsec = new_stamp_xsec;
853 	vdso_data->tb_to_xs = new_tb_to_xs;
854 	vdso_data->wtom_clock_sec = wtm->tv_sec;
855 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
856 	vdso_data->stamp_xtime = *wall_time;
857 	vdso_data->stamp_sec_fraction = frac_sec;
858 	smp_wmb();
859 	++(vdso_data->tb_update_count);
860 }
861 
862 void update_vsyscall_tz(void)
863 {
864 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
865 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
866 }
867 
868 static void __init clocksource_init(void)
869 {
870 	struct clocksource *clock;
871 
872 	if (__USE_RTC())
873 		clock = &clocksource_rtc;
874 	else
875 		clock = &clocksource_timebase;
876 
877 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
878 		printk(KERN_ERR "clocksource: %s is already registered\n",
879 		       clock->name);
880 		return;
881 	}
882 
883 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
884 	       clock->name, clock->mult, clock->shift);
885 }
886 
887 static int decrementer_set_next_event(unsigned long evt,
888 				      struct clock_event_device *dev)
889 {
890 	__this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
891 	set_dec(evt);
892 
893 	/* We may have raced with new irq work */
894 	if (test_irq_work_pending())
895 		set_dec(1);
896 
897 	return 0;
898 }
899 
900 static int decrementer_shutdown(struct clock_event_device *dev)
901 {
902 	decrementer_set_next_event(decrementer_max, dev);
903 	return 0;
904 }
905 
906 /* Interrupt handler for the timer broadcast IPI */
907 void tick_broadcast_ipi_handler(void)
908 {
909 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
910 
911 	*next_tb = get_tb_or_rtc();
912 	__timer_interrupt();
913 }
914 
915 static void register_decrementer_clockevent(int cpu)
916 {
917 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
918 
919 	*dec = decrementer_clockevent;
920 	dec->cpumask = cpumask_of(cpu);
921 
922 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
923 		    dec->name, dec->mult, dec->shift, cpu);
924 
925 	clockevents_register_device(dec);
926 }
927 
928 static void enable_large_decrementer(void)
929 {
930 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
931 		return;
932 
933 	if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
934 		return;
935 
936 	/*
937 	 * If we're running as the hypervisor we need to enable the LD manually
938 	 * otherwise firmware should have done it for us.
939 	 */
940 	if (cpu_has_feature(CPU_FTR_HVMODE))
941 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
942 }
943 
944 static void __init set_decrementer_max(void)
945 {
946 	struct device_node *cpu;
947 	u32 bits = 32;
948 
949 	/* Prior to ISAv3 the decrementer is always 32 bit */
950 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
951 		return;
952 
953 	cpu = of_find_node_by_type(NULL, "cpu");
954 
955 	if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
956 		if (bits > 64 || bits < 32) {
957 			pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
958 			bits = 32;
959 		}
960 
961 		/* calculate the signed maximum given this many bits */
962 		decrementer_max = (1ul << (bits - 1)) - 1;
963 	}
964 
965 	of_node_put(cpu);
966 
967 	pr_info("time_init: %u bit decrementer (max: %llx)\n",
968 		bits, decrementer_max);
969 }
970 
971 static void __init init_decrementer_clockevent(void)
972 {
973 	int cpu = smp_processor_id();
974 
975 	clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
976 
977 	decrementer_clockevent.max_delta_ns =
978 		clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
979 	decrementer_clockevent.min_delta_ns =
980 		clockevent_delta2ns(2, &decrementer_clockevent);
981 
982 	register_decrementer_clockevent(cpu);
983 }
984 
985 void secondary_cpu_time_init(void)
986 {
987 	/* Enable and test the large decrementer for this cpu */
988 	enable_large_decrementer();
989 
990 	/* Start the decrementer on CPUs that have manual control
991 	 * such as BookE
992 	 */
993 	start_cpu_decrementer();
994 
995 	/* FIME: Should make unrelatred change to move snapshot_timebase
996 	 * call here ! */
997 	register_decrementer_clockevent(smp_processor_id());
998 }
999 
1000 /* This function is only called on the boot processor */
1001 void __init time_init(void)
1002 {
1003 	struct div_result res;
1004 	u64 scale;
1005 	unsigned shift;
1006 
1007 	if (__USE_RTC()) {
1008 		/* 601 processor: dec counts down by 128 every 128ns */
1009 		ppc_tb_freq = 1000000000;
1010 	} else {
1011 		/* Normal PowerPC with timebase register */
1012 		ppc_md.calibrate_decr();
1013 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1014 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1015 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
1016 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1017 	}
1018 
1019 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1020 	tb_ticks_per_sec = ppc_tb_freq;
1021 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
1022 	calc_cputime_factors();
1023 	setup_cputime_one_jiffy();
1024 
1025 	/*
1026 	 * Compute scale factor for sched_clock.
1027 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1028 	 * which is the timebase frequency.
1029 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1030 	 * the 128-bit result as a 64.64 fixed-point number.
1031 	 * We then shift that number right until it is less than 1.0,
1032 	 * giving us the scale factor and shift count to use in
1033 	 * sched_clock().
1034 	 */
1035 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1036 	scale = res.result_low;
1037 	for (shift = 0; res.result_high != 0; ++shift) {
1038 		scale = (scale >> 1) | (res.result_high << 63);
1039 		res.result_high >>= 1;
1040 	}
1041 	tb_to_ns_scale = scale;
1042 	tb_to_ns_shift = shift;
1043 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1044 	boot_tb = get_tb_or_rtc();
1045 
1046 	/* If platform provided a timezone (pmac), we correct the time */
1047 	if (timezone_offset) {
1048 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1049 		sys_tz.tz_dsttime = 0;
1050 	}
1051 
1052 	vdso_data->tb_update_count = 0;
1053 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1054 
1055 	/* initialise and enable the large decrementer (if we have one) */
1056 	set_decrementer_max();
1057 	enable_large_decrementer();
1058 
1059 	/* Start the decrementer on CPUs that have manual control
1060 	 * such as BookE
1061 	 */
1062 	start_cpu_decrementer();
1063 
1064 	/* Register the clocksource */
1065 	clocksource_init();
1066 
1067 	init_decrementer_clockevent();
1068 	tick_setup_hrtimer_broadcast();
1069 
1070 #ifdef CONFIG_COMMON_CLK
1071 	of_clk_init(NULL);
1072 #endif
1073 }
1074 
1075 
1076 #define FEBRUARY	2
1077 #define	STARTOFTIME	1970
1078 #define SECDAY		86400L
1079 #define SECYR		(SECDAY * 365)
1080 #define	leapyear(year)		((year) % 4 == 0 && \
1081 				 ((year) % 100 != 0 || (year) % 400 == 0))
1082 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1083 #define	days_in_month(a) 	(month_days[(a) - 1])
1084 
1085 static int month_days[12] = {
1086 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1087 };
1088 
1089 void to_tm(int tim, struct rtc_time * tm)
1090 {
1091 	register int    i;
1092 	register long   hms, day;
1093 
1094 	day = tim / SECDAY;
1095 	hms = tim % SECDAY;
1096 
1097 	/* Hours, minutes, seconds are easy */
1098 	tm->tm_hour = hms / 3600;
1099 	tm->tm_min = (hms % 3600) / 60;
1100 	tm->tm_sec = (hms % 3600) % 60;
1101 
1102 	/* Number of years in days */
1103 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1104 		day -= days_in_year(i);
1105 	tm->tm_year = i;
1106 
1107 	/* Number of months in days left */
1108 	if (leapyear(tm->tm_year))
1109 		days_in_month(FEBRUARY) = 29;
1110 	for (i = 1; day >= days_in_month(i); i++)
1111 		day -= days_in_month(i);
1112 	days_in_month(FEBRUARY) = 28;
1113 	tm->tm_mon = i;
1114 
1115 	/* Days are what is left over (+1) from all that. */
1116 	tm->tm_mday = day + 1;
1117 
1118 	/*
1119 	 * No-one uses the day of the week.
1120 	 */
1121 	tm->tm_wday = -1;
1122 }
1123 EXPORT_SYMBOL(to_tm);
1124 
1125 /*
1126  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1127  * result.
1128  */
1129 void div128_by_32(u64 dividend_high, u64 dividend_low,
1130 		  unsigned divisor, struct div_result *dr)
1131 {
1132 	unsigned long a, b, c, d;
1133 	unsigned long w, x, y, z;
1134 	u64 ra, rb, rc;
1135 
1136 	a = dividend_high >> 32;
1137 	b = dividend_high & 0xffffffff;
1138 	c = dividend_low >> 32;
1139 	d = dividend_low & 0xffffffff;
1140 
1141 	w = a / divisor;
1142 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1143 
1144 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1145 	x = ra;
1146 
1147 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1148 	y = rb;
1149 
1150 	do_div(rc, divisor);
1151 	z = rc;
1152 
1153 	dr->result_high = ((u64)w << 32) + x;
1154 	dr->result_low  = ((u64)y << 32) + z;
1155 
1156 }
1157 
1158 /* We don't need to calibrate delay, we use the CPU timebase for that */
1159 void calibrate_delay(void)
1160 {
1161 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1162 	 * as the number of __delay(1) in a jiffy, so make it so
1163 	 */
1164 	loops_per_jiffy = tb_ticks_per_jiffy;
1165 }
1166 
1167 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1168 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1169 {
1170 	ppc_md.get_rtc_time(tm);
1171 	return rtc_valid_tm(tm);
1172 }
1173 
1174 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1175 {
1176 	if (!ppc_md.set_rtc_time)
1177 		return -EOPNOTSUPP;
1178 
1179 	if (ppc_md.set_rtc_time(tm) < 0)
1180 		return -EOPNOTSUPP;
1181 
1182 	return 0;
1183 }
1184 
1185 static const struct rtc_class_ops rtc_generic_ops = {
1186 	.read_time = rtc_generic_get_time,
1187 	.set_time = rtc_generic_set_time,
1188 };
1189 
1190 static int __init rtc_init(void)
1191 {
1192 	struct platform_device *pdev;
1193 
1194 	if (!ppc_md.get_rtc_time)
1195 		return -ENODEV;
1196 
1197 	pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1198 					     &rtc_generic_ops,
1199 					     sizeof(rtc_generic_ops));
1200 
1201 	return PTR_ERR_OR_ZERO(pdev);
1202 }
1203 
1204 device_initcall(rtc_init);
1205 #endif
1206