xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision 9445aa1a)
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 
77 /* powerpc clocksource/clockevent code */
78 
79 #include <linux/clockchips.h>
80 #include <linux/timekeeper_internal.h>
81 
82 static cycle_t rtc_read(struct clocksource *);
83 static struct clocksource clocksource_rtc = {
84 	.name         = "rtc",
85 	.rating       = 400,
86 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
87 	.mask         = CLOCKSOURCE_MASK(64),
88 	.read         = rtc_read,
89 };
90 
91 static cycle_t timebase_read(struct clocksource *);
92 static struct clocksource clocksource_timebase = {
93 	.name         = "timebase",
94 	.rating       = 400,
95 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
96 	.mask         = CLOCKSOURCE_MASK(64),
97 	.read         = timebase_read,
98 };
99 
100 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
101 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
102 
103 static int decrementer_set_next_event(unsigned long evt,
104 				      struct clock_event_device *dev);
105 static int decrementer_shutdown(struct clock_event_device *evt);
106 
107 struct clock_event_device decrementer_clockevent = {
108 	.name			= "decrementer",
109 	.rating			= 200,
110 	.irq			= 0,
111 	.set_next_event		= decrementer_set_next_event,
112 	.set_state_shutdown	= decrementer_shutdown,
113 	.tick_resume		= decrementer_shutdown,
114 	.features		= CLOCK_EVT_FEAT_ONESHOT |
115 				  CLOCK_EVT_FEAT_C3STOP,
116 };
117 EXPORT_SYMBOL(decrementer_clockevent);
118 
119 DEFINE_PER_CPU(u64, decrementers_next_tb);
120 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
121 
122 #define XSEC_PER_SEC (1024*1024)
123 
124 #ifdef CONFIG_PPC64
125 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
126 #else
127 /* compute ((xsec << 12) * max) >> 32 */
128 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
129 #endif
130 
131 unsigned long tb_ticks_per_jiffy;
132 unsigned long tb_ticks_per_usec = 100; /* sane default */
133 EXPORT_SYMBOL(tb_ticks_per_usec);
134 unsigned long tb_ticks_per_sec;
135 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
136 
137 DEFINE_SPINLOCK(rtc_lock);
138 EXPORT_SYMBOL_GPL(rtc_lock);
139 
140 static u64 tb_to_ns_scale __read_mostly;
141 static unsigned tb_to_ns_shift __read_mostly;
142 static u64 boot_tb __read_mostly;
143 
144 extern struct timezone sys_tz;
145 static long timezone_offset;
146 
147 unsigned long ppc_proc_freq;
148 EXPORT_SYMBOL_GPL(ppc_proc_freq);
149 unsigned long ppc_tb_freq;
150 EXPORT_SYMBOL_GPL(ppc_tb_freq);
151 
152 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
153 /*
154  * Factors for converting from cputime_t (timebase ticks) to
155  * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
156  * These are all stored as 0.64 fixed-point binary fractions.
157  */
158 u64 __cputime_jiffies_factor;
159 EXPORT_SYMBOL(__cputime_jiffies_factor);
160 u64 __cputime_usec_factor;
161 EXPORT_SYMBOL(__cputime_usec_factor);
162 u64 __cputime_sec_factor;
163 EXPORT_SYMBOL(__cputime_sec_factor);
164 u64 __cputime_clockt_factor;
165 EXPORT_SYMBOL(__cputime_clockt_factor);
166 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
167 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
168 
169 cputime_t cputime_one_jiffy;
170 
171 #ifdef CONFIG_PPC_SPLPAR
172 void (*dtl_consumer)(struct dtl_entry *, u64);
173 #endif
174 
175 #ifdef CONFIG_PPC64
176 #define get_accounting(tsk)	(&get_paca()->accounting)
177 #else
178 #define get_accounting(tsk)	(&task_thread_info(tsk)->accounting)
179 #endif
180 
181 static void calc_cputime_factors(void)
182 {
183 	struct div_result res;
184 
185 	div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
186 	__cputime_jiffies_factor = res.result_low;
187 	div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
188 	__cputime_usec_factor = res.result_low;
189 	div128_by_32(1, 0, tb_ticks_per_sec, &res);
190 	__cputime_sec_factor = res.result_low;
191 	div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
192 	__cputime_clockt_factor = res.result_low;
193 }
194 
195 /*
196  * Read the SPURR on systems that have it, otherwise the PURR,
197  * or if that doesn't exist return the timebase value passed in.
198  */
199 static unsigned long read_spurr(unsigned long tb)
200 {
201 	if (cpu_has_feature(CPU_FTR_SPURR))
202 		return mfspr(SPRN_SPURR);
203 	if (cpu_has_feature(CPU_FTR_PURR))
204 		return mfspr(SPRN_PURR);
205 	return tb;
206 }
207 
208 #ifdef CONFIG_PPC_SPLPAR
209 
210 /*
211  * Scan the dispatch trace log and count up the stolen time.
212  * Should be called with interrupts disabled.
213  */
214 static u64 scan_dispatch_log(u64 stop_tb)
215 {
216 	u64 i = local_paca->dtl_ridx;
217 	struct dtl_entry *dtl = local_paca->dtl_curr;
218 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
219 	struct lppaca *vpa = local_paca->lppaca_ptr;
220 	u64 tb_delta;
221 	u64 stolen = 0;
222 	u64 dtb;
223 
224 	if (!dtl)
225 		return 0;
226 
227 	if (i == be64_to_cpu(vpa->dtl_idx))
228 		return 0;
229 	while (i < be64_to_cpu(vpa->dtl_idx)) {
230 		dtb = be64_to_cpu(dtl->timebase);
231 		tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
232 			be32_to_cpu(dtl->ready_to_enqueue_time);
233 		barrier();
234 		if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
235 			/* buffer has overflowed */
236 			i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
237 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
238 			continue;
239 		}
240 		if (dtb > stop_tb)
241 			break;
242 		if (dtl_consumer)
243 			dtl_consumer(dtl, i);
244 		stolen += tb_delta;
245 		++i;
246 		++dtl;
247 		if (dtl == dtl_end)
248 			dtl = local_paca->dispatch_log;
249 	}
250 	local_paca->dtl_ridx = i;
251 	local_paca->dtl_curr = dtl;
252 	return stolen;
253 }
254 
255 /*
256  * Accumulate stolen time by scanning the dispatch trace log.
257  * Called on entry from user mode.
258  */
259 void accumulate_stolen_time(void)
260 {
261 	u64 sst, ust;
262 	u8 save_soft_enabled = local_paca->soft_enabled;
263 	struct cpu_accounting_data *acct = &local_paca->accounting;
264 
265 	/* We are called early in the exception entry, before
266 	 * soft/hard_enabled are sync'ed to the expected state
267 	 * for the exception. We are hard disabled but the PACA
268 	 * needs to reflect that so various debug stuff doesn't
269 	 * complain
270 	 */
271 	local_paca->soft_enabled = 0;
272 
273 	sst = scan_dispatch_log(acct->starttime_user);
274 	ust = scan_dispatch_log(acct->starttime);
275 	acct->system_time -= sst;
276 	acct->user_time -= ust;
277 	local_paca->stolen_time += ust + sst;
278 
279 	local_paca->soft_enabled = save_soft_enabled;
280 }
281 
282 static inline u64 calculate_stolen_time(u64 stop_tb)
283 {
284 	u64 stolen = 0;
285 
286 	if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
287 		stolen = scan_dispatch_log(stop_tb);
288 		get_paca()->accounting.system_time -= stolen;
289 	}
290 
291 	stolen += get_paca()->stolen_time;
292 	get_paca()->stolen_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->system_time += 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->system_time;
328 	acct->system_time = 0;
329 	udelta = acct->user_time - acct->utime_sspurr;
330 	acct->utime_sspurr = acct->user_time;
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->user_time_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, sys_scaled);
363 	if (stolen)
364 		account_steal_time(stolen);
365 }
366 EXPORT_SYMBOL_GPL(vtime_account_system);
367 
368 void vtime_account_idle(struct task_struct *tsk)
369 {
370 	unsigned long delta, sys_scaled, stolen;
371 
372 	delta = vtime_delta(tsk, &sys_scaled, &stolen);
373 	account_idle_time(delta + stolen);
374 }
375 
376 /*
377  * Transfer the user time accumulated in the paca
378  * by the exception entry and exit code to the generic
379  * process user time records.
380  * Must be called with interrupts disabled.
381  * Assumes that vtime_account_system/idle() has been called
382  * recently (i.e. since the last entry from usermode) so that
383  * get_paca()->user_time_scaled is up to date.
384  */
385 void vtime_account_user(struct task_struct *tsk)
386 {
387 	cputime_t utime, utimescaled;
388 	struct cpu_accounting_data *acct = get_accounting(tsk);
389 
390 	utime = acct->user_time;
391 	utimescaled = acct->user_time_scaled;
392 	acct->user_time = 0;
393 	acct->user_time_scaled = 0;
394 	acct->utime_sspurr = 0;
395 	account_user_time(tsk, utime, utimescaled);
396 }
397 
398 #ifdef CONFIG_PPC32
399 /*
400  * Called from the context switch with interrupts disabled, to charge all
401  * accumulated times to the current process, and to prepare accounting on
402  * the next process.
403  */
404 void arch_vtime_task_switch(struct task_struct *prev)
405 {
406 	struct cpu_accounting_data *acct = get_accounting(current);
407 
408 	acct->starttime = get_accounting(prev)->starttime;
409 	acct->system_time = 0;
410 	acct->user_time = 0;
411 }
412 #endif /* CONFIG_PPC32 */
413 
414 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
415 #define calc_cputime_factors()
416 #endif
417 
418 void __delay(unsigned long loops)
419 {
420 	unsigned long start;
421 	int diff;
422 
423 	if (__USE_RTC()) {
424 		start = get_rtcl();
425 		do {
426 			/* the RTCL register wraps at 1000000000 */
427 			diff = get_rtcl() - start;
428 			if (diff < 0)
429 				diff += 1000000000;
430 		} while (diff < loops);
431 	} else {
432 		start = get_tbl();
433 		while (get_tbl() - start < loops)
434 			HMT_low();
435 		HMT_medium();
436 	}
437 }
438 EXPORT_SYMBOL(__delay);
439 
440 void udelay(unsigned long usecs)
441 {
442 	__delay(tb_ticks_per_usec * usecs);
443 }
444 EXPORT_SYMBOL(udelay);
445 
446 #ifdef CONFIG_SMP
447 unsigned long profile_pc(struct pt_regs *regs)
448 {
449 	unsigned long pc = instruction_pointer(regs);
450 
451 	if (in_lock_functions(pc))
452 		return regs->link;
453 
454 	return pc;
455 }
456 EXPORT_SYMBOL(profile_pc);
457 #endif
458 
459 #ifdef CONFIG_IRQ_WORK
460 
461 /*
462  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
463  */
464 #ifdef CONFIG_PPC64
465 static inline unsigned long test_irq_work_pending(void)
466 {
467 	unsigned long x;
468 
469 	asm volatile("lbz %0,%1(13)"
470 		: "=r" (x)
471 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
472 	return x;
473 }
474 
475 static inline void set_irq_work_pending_flag(void)
476 {
477 	asm volatile("stb %0,%1(13)" : :
478 		"r" (1),
479 		"i" (offsetof(struct paca_struct, irq_work_pending)));
480 }
481 
482 static inline void clear_irq_work_pending(void)
483 {
484 	asm volatile("stb %0,%1(13)" : :
485 		"r" (0),
486 		"i" (offsetof(struct paca_struct, irq_work_pending)));
487 }
488 
489 #else /* 32-bit */
490 
491 DEFINE_PER_CPU(u8, irq_work_pending);
492 
493 #define set_irq_work_pending_flag()	__this_cpu_write(irq_work_pending, 1)
494 #define test_irq_work_pending()		__this_cpu_read(irq_work_pending)
495 #define clear_irq_work_pending()	__this_cpu_write(irq_work_pending, 0)
496 
497 #endif /* 32 vs 64 bit */
498 
499 void arch_irq_work_raise(void)
500 {
501 	preempt_disable();
502 	set_irq_work_pending_flag();
503 	set_dec(1);
504 	preempt_enable();
505 }
506 
507 #else  /* CONFIG_IRQ_WORK */
508 
509 #define test_irq_work_pending()	0
510 #define clear_irq_work_pending()
511 
512 #endif /* CONFIG_IRQ_WORK */
513 
514 static void __timer_interrupt(void)
515 {
516 	struct pt_regs *regs = get_irq_regs();
517 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
518 	struct clock_event_device *evt = this_cpu_ptr(&decrementers);
519 	u64 now;
520 
521 	trace_timer_interrupt_entry(regs);
522 
523 	if (test_irq_work_pending()) {
524 		clear_irq_work_pending();
525 		irq_work_run();
526 	}
527 
528 	now = get_tb_or_rtc();
529 	if (now >= *next_tb) {
530 		*next_tb = ~(u64)0;
531 		if (evt->event_handler)
532 			evt->event_handler(evt);
533 		__this_cpu_inc(irq_stat.timer_irqs_event);
534 	} else {
535 		now = *next_tb - now;
536 		if (now <= decrementer_max)
537 			set_dec(now);
538 		/* We may have raced with new irq work */
539 		if (test_irq_work_pending())
540 			set_dec(1);
541 		__this_cpu_inc(irq_stat.timer_irqs_others);
542 	}
543 
544 #ifdef CONFIG_PPC64
545 	/* collect purr register values often, for accurate calculations */
546 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
547 		struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
548 		cu->current_tb = mfspr(SPRN_PURR);
549 	}
550 #endif
551 
552 	trace_timer_interrupt_exit(regs);
553 }
554 
555 /*
556  * timer_interrupt - gets called when the decrementer overflows,
557  * with interrupts disabled.
558  */
559 void timer_interrupt(struct pt_regs * regs)
560 {
561 	struct pt_regs *old_regs;
562 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
563 
564 	/* Ensure a positive value is written to the decrementer, or else
565 	 * some CPUs will continue to take decrementer exceptions.
566 	 */
567 	set_dec(decrementer_max);
568 
569 	/* Some implementations of hotplug will get timer interrupts while
570 	 * offline, just ignore these and we also need to set
571 	 * decrementers_next_tb as MAX to make sure __check_irq_replay
572 	 * don't replay timer interrupt when return, otherwise we'll trap
573 	 * here infinitely :(
574 	 */
575 	if (!cpu_online(smp_processor_id())) {
576 		*next_tb = ~(u64)0;
577 		return;
578 	}
579 
580 	/* Conditionally hard-enable interrupts now that the DEC has been
581 	 * bumped to its maximum value
582 	 */
583 	may_hard_irq_enable();
584 
585 
586 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
587 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
588 		do_IRQ(regs);
589 #endif
590 
591 	old_regs = set_irq_regs(regs);
592 	irq_enter();
593 
594 	__timer_interrupt();
595 	irq_exit();
596 	set_irq_regs(old_regs);
597 }
598 EXPORT_SYMBOL(timer_interrupt);
599 
600 /*
601  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
602  * left pending on exit from a KVM guest.  We don't need to do anything
603  * to clear them, as they are edge-triggered.
604  */
605 void hdec_interrupt(struct pt_regs *regs)
606 {
607 }
608 
609 #ifdef CONFIG_SUSPEND
610 static void generic_suspend_disable_irqs(void)
611 {
612 	/* Disable the decrementer, so that it doesn't interfere
613 	 * with suspending.
614 	 */
615 
616 	set_dec(decrementer_max);
617 	local_irq_disable();
618 	set_dec(decrementer_max);
619 }
620 
621 static void generic_suspend_enable_irqs(void)
622 {
623 	local_irq_enable();
624 }
625 
626 /* Overrides the weak version in kernel/power/main.c */
627 void arch_suspend_disable_irqs(void)
628 {
629 	if (ppc_md.suspend_disable_irqs)
630 		ppc_md.suspend_disable_irqs();
631 	generic_suspend_disable_irqs();
632 }
633 
634 /* Overrides the weak version in kernel/power/main.c */
635 void arch_suspend_enable_irqs(void)
636 {
637 	generic_suspend_enable_irqs();
638 	if (ppc_md.suspend_enable_irqs)
639 		ppc_md.suspend_enable_irqs();
640 }
641 #endif
642 
643 unsigned long long tb_to_ns(unsigned long long ticks)
644 {
645 	return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
646 }
647 EXPORT_SYMBOL_GPL(tb_to_ns);
648 
649 /*
650  * Scheduler clock - returns current time in nanosec units.
651  *
652  * Note: mulhdu(a, b) (multiply high double unsigned) returns
653  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
654  * are 64-bit unsigned numbers.
655  */
656 unsigned long long sched_clock(void)
657 {
658 	if (__USE_RTC())
659 		return get_rtc();
660 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
661 }
662 
663 
664 #ifdef CONFIG_PPC_PSERIES
665 
666 /*
667  * Running clock - attempts to give a view of time passing for a virtualised
668  * kernels.
669  * Uses the VTB register if available otherwise a next best guess.
670  */
671 unsigned long long running_clock(void)
672 {
673 	/*
674 	 * Don't read the VTB as a host since KVM does not switch in host
675 	 * timebase into the VTB when it takes a guest off the CPU, reading the
676 	 * VTB would result in reading 'last switched out' guest VTB.
677 	 *
678 	 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
679 	 * would be unsafe to rely only on the #ifdef above.
680 	 */
681 	if (firmware_has_feature(FW_FEATURE_LPAR) &&
682 	    cpu_has_feature(CPU_FTR_ARCH_207S))
683 		return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
684 
685 	/*
686 	 * This is a next best approximation without a VTB.
687 	 * On a host which is running bare metal there should never be any stolen
688 	 * time and on a host which doesn't do any virtualisation TB *should* equal
689 	 * VTB so it makes no difference anyway.
690 	 */
691 	return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
692 }
693 #endif
694 
695 static int __init get_freq(char *name, int cells, unsigned long *val)
696 {
697 	struct device_node *cpu;
698 	const __be32 *fp;
699 	int found = 0;
700 
701 	/* The cpu node should have timebase and clock frequency properties */
702 	cpu = of_find_node_by_type(NULL, "cpu");
703 
704 	if (cpu) {
705 		fp = of_get_property(cpu, name, NULL);
706 		if (fp) {
707 			found = 1;
708 			*val = of_read_ulong(fp, cells);
709 		}
710 
711 		of_node_put(cpu);
712 	}
713 
714 	return found;
715 }
716 
717 static void start_cpu_decrementer(void)
718 {
719 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
720 	/* Clear any pending timer interrupts */
721 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
722 
723 	/* Enable decrementer interrupt */
724 	mtspr(SPRN_TCR, TCR_DIE);
725 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
726 }
727 
728 void __init generic_calibrate_decr(void)
729 {
730 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
731 
732 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
733 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
734 
735 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
736 				"(not found)\n");
737 	}
738 
739 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
740 
741 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
742 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
743 
744 		printk(KERN_ERR "WARNING: Estimating processor frequency "
745 				"(not found)\n");
746 	}
747 }
748 
749 int update_persistent_clock(struct timespec now)
750 {
751 	struct rtc_time tm;
752 
753 	if (!ppc_md.set_rtc_time)
754 		return -ENODEV;
755 
756 	to_tm(now.tv_sec + 1 + timezone_offset, &tm);
757 	tm.tm_year -= 1900;
758 	tm.tm_mon -= 1;
759 
760 	return ppc_md.set_rtc_time(&tm);
761 }
762 
763 static void __read_persistent_clock(struct timespec *ts)
764 {
765 	struct rtc_time tm;
766 	static int first = 1;
767 
768 	ts->tv_nsec = 0;
769 	/* XXX this is a litle fragile but will work okay in the short term */
770 	if (first) {
771 		first = 0;
772 		if (ppc_md.time_init)
773 			timezone_offset = ppc_md.time_init();
774 
775 		/* get_boot_time() isn't guaranteed to be safe to call late */
776 		if (ppc_md.get_boot_time) {
777 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
778 			return;
779 		}
780 	}
781 	if (!ppc_md.get_rtc_time) {
782 		ts->tv_sec = 0;
783 		return;
784 	}
785 	ppc_md.get_rtc_time(&tm);
786 
787 	ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
788 			    tm.tm_hour, tm.tm_min, tm.tm_sec);
789 }
790 
791 void read_persistent_clock(struct timespec *ts)
792 {
793 	__read_persistent_clock(ts);
794 
795 	/* Sanitize it in case real time clock is set below EPOCH */
796 	if (ts->tv_sec < 0) {
797 		ts->tv_sec = 0;
798 		ts->tv_nsec = 0;
799 	}
800 
801 }
802 
803 /* clocksource code */
804 static cycle_t rtc_read(struct clocksource *cs)
805 {
806 	return (cycle_t)get_rtc();
807 }
808 
809 static cycle_t timebase_read(struct clocksource *cs)
810 {
811 	return (cycle_t)get_tb();
812 }
813 
814 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
815 			 struct clocksource *clock, u32 mult, cycle_t cycle_last)
816 {
817 	u64 new_tb_to_xs, new_stamp_xsec;
818 	u32 frac_sec;
819 
820 	if (clock != &clocksource_timebase)
821 		return;
822 
823 	/* Make userspace gettimeofday spin until we're done. */
824 	++vdso_data->tb_update_count;
825 	smp_mb();
826 
827 	/* 19342813113834067 ~= 2^(20+64) / 1e9 */
828 	new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
829 	new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
830 	do_div(new_stamp_xsec, 1000000000);
831 	new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
832 
833 	BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
834 	/* this is tv_nsec / 1e9 as a 0.32 fraction */
835 	frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
836 
837 	/*
838 	 * tb_update_count is used to allow the userspace gettimeofday code
839 	 * to assure itself that it sees a consistent view of the tb_to_xs and
840 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
841 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
842 	 * the two values of tb_update_count match and are even then the
843 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
844 	 * loops back and reads them again until this criteria is met.
845 	 * We expect the caller to have done the first increment of
846 	 * vdso_data->tb_update_count already.
847 	 */
848 	vdso_data->tb_orig_stamp = cycle_last;
849 	vdso_data->stamp_xsec = new_stamp_xsec;
850 	vdso_data->tb_to_xs = new_tb_to_xs;
851 	vdso_data->wtom_clock_sec = wtm->tv_sec;
852 	vdso_data->wtom_clock_nsec = wtm->tv_nsec;
853 	vdso_data->stamp_xtime = *wall_time;
854 	vdso_data->stamp_sec_fraction = frac_sec;
855 	smp_wmb();
856 	++(vdso_data->tb_update_count);
857 }
858 
859 void update_vsyscall_tz(void)
860 {
861 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
862 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
863 }
864 
865 static void __init clocksource_init(void)
866 {
867 	struct clocksource *clock;
868 
869 	if (__USE_RTC())
870 		clock = &clocksource_rtc;
871 	else
872 		clock = &clocksource_timebase;
873 
874 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
875 		printk(KERN_ERR "clocksource: %s is already registered\n",
876 		       clock->name);
877 		return;
878 	}
879 
880 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
881 	       clock->name, clock->mult, clock->shift);
882 }
883 
884 static int decrementer_set_next_event(unsigned long evt,
885 				      struct clock_event_device *dev)
886 {
887 	__this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
888 	set_dec(evt);
889 
890 	/* We may have raced with new irq work */
891 	if (test_irq_work_pending())
892 		set_dec(1);
893 
894 	return 0;
895 }
896 
897 static int decrementer_shutdown(struct clock_event_device *dev)
898 {
899 	decrementer_set_next_event(decrementer_max, dev);
900 	return 0;
901 }
902 
903 /* Interrupt handler for the timer broadcast IPI */
904 void tick_broadcast_ipi_handler(void)
905 {
906 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
907 
908 	*next_tb = get_tb_or_rtc();
909 	__timer_interrupt();
910 }
911 
912 static void register_decrementer_clockevent(int cpu)
913 {
914 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
915 
916 	*dec = decrementer_clockevent;
917 	dec->cpumask = cpumask_of(cpu);
918 
919 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
920 		    dec->name, dec->mult, dec->shift, cpu);
921 
922 	clockevents_register_device(dec);
923 }
924 
925 static void enable_large_decrementer(void)
926 {
927 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
928 		return;
929 
930 	if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
931 		return;
932 
933 	/*
934 	 * If we're running as the hypervisor we need to enable the LD manually
935 	 * otherwise firmware should have done it for us.
936 	 */
937 	if (cpu_has_feature(CPU_FTR_HVMODE))
938 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
939 }
940 
941 static void __init set_decrementer_max(void)
942 {
943 	struct device_node *cpu;
944 	u32 bits = 32;
945 
946 	/* Prior to ISAv3 the decrementer is always 32 bit */
947 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
948 		return;
949 
950 	cpu = of_find_node_by_type(NULL, "cpu");
951 
952 	if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
953 		if (bits > 64 || bits < 32) {
954 			pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
955 			bits = 32;
956 		}
957 
958 		/* calculate the signed maximum given this many bits */
959 		decrementer_max = (1ul << (bits - 1)) - 1;
960 	}
961 
962 	of_node_put(cpu);
963 
964 	pr_info("time_init: %u bit decrementer (max: %llx)\n",
965 		bits, decrementer_max);
966 }
967 
968 static void __init init_decrementer_clockevent(void)
969 {
970 	int cpu = smp_processor_id();
971 
972 	clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
973 
974 	decrementer_clockevent.max_delta_ns =
975 		clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
976 	decrementer_clockevent.min_delta_ns =
977 		clockevent_delta2ns(2, &decrementer_clockevent);
978 
979 	register_decrementer_clockevent(cpu);
980 }
981 
982 void secondary_cpu_time_init(void)
983 {
984 	/* Enable and test the large decrementer for this cpu */
985 	enable_large_decrementer();
986 
987 	/* Start the decrementer on CPUs that have manual control
988 	 * such as BookE
989 	 */
990 	start_cpu_decrementer();
991 
992 	/* FIME: Should make unrelatred change to move snapshot_timebase
993 	 * call here ! */
994 	register_decrementer_clockevent(smp_processor_id());
995 }
996 
997 /* This function is only called on the boot processor */
998 void __init time_init(void)
999 {
1000 	struct div_result res;
1001 	u64 scale;
1002 	unsigned shift;
1003 
1004 	if (__USE_RTC()) {
1005 		/* 601 processor: dec counts down by 128 every 128ns */
1006 		ppc_tb_freq = 1000000000;
1007 	} else {
1008 		/* Normal PowerPC with timebase register */
1009 		ppc_md.calibrate_decr();
1010 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1011 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1012 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
1013 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1014 	}
1015 
1016 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1017 	tb_ticks_per_sec = ppc_tb_freq;
1018 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
1019 	calc_cputime_factors();
1020 	setup_cputime_one_jiffy();
1021 
1022 	/*
1023 	 * Compute scale factor for sched_clock.
1024 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1025 	 * which is the timebase frequency.
1026 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1027 	 * the 128-bit result as a 64.64 fixed-point number.
1028 	 * We then shift that number right until it is less than 1.0,
1029 	 * giving us the scale factor and shift count to use in
1030 	 * sched_clock().
1031 	 */
1032 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1033 	scale = res.result_low;
1034 	for (shift = 0; res.result_high != 0; ++shift) {
1035 		scale = (scale >> 1) | (res.result_high << 63);
1036 		res.result_high >>= 1;
1037 	}
1038 	tb_to_ns_scale = scale;
1039 	tb_to_ns_shift = shift;
1040 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1041 	boot_tb = get_tb_or_rtc();
1042 
1043 	/* If platform provided a timezone (pmac), we correct the time */
1044 	if (timezone_offset) {
1045 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1046 		sys_tz.tz_dsttime = 0;
1047 	}
1048 
1049 	vdso_data->tb_update_count = 0;
1050 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1051 
1052 	/* initialise and enable the large decrementer (if we have one) */
1053 	set_decrementer_max();
1054 	enable_large_decrementer();
1055 
1056 	/* Start the decrementer on CPUs that have manual control
1057 	 * such as BookE
1058 	 */
1059 	start_cpu_decrementer();
1060 
1061 	/* Register the clocksource */
1062 	clocksource_init();
1063 
1064 	init_decrementer_clockevent();
1065 	tick_setup_hrtimer_broadcast();
1066 
1067 #ifdef CONFIG_COMMON_CLK
1068 	of_clk_init(NULL);
1069 #endif
1070 }
1071 
1072 
1073 #define FEBRUARY	2
1074 #define	STARTOFTIME	1970
1075 #define SECDAY		86400L
1076 #define SECYR		(SECDAY * 365)
1077 #define	leapyear(year)		((year) % 4 == 0 && \
1078 				 ((year) % 100 != 0 || (year) % 400 == 0))
1079 #define	days_in_year(a) 	(leapyear(a) ? 366 : 365)
1080 #define	days_in_month(a) 	(month_days[(a) - 1])
1081 
1082 static int month_days[12] = {
1083 	31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1084 };
1085 
1086 void to_tm(int tim, struct rtc_time * tm)
1087 {
1088 	register int    i;
1089 	register long   hms, day;
1090 
1091 	day = tim / SECDAY;
1092 	hms = tim % SECDAY;
1093 
1094 	/* Hours, minutes, seconds are easy */
1095 	tm->tm_hour = hms / 3600;
1096 	tm->tm_min = (hms % 3600) / 60;
1097 	tm->tm_sec = (hms % 3600) % 60;
1098 
1099 	/* Number of years in days */
1100 	for (i = STARTOFTIME; day >= days_in_year(i); i++)
1101 		day -= days_in_year(i);
1102 	tm->tm_year = i;
1103 
1104 	/* Number of months in days left */
1105 	if (leapyear(tm->tm_year))
1106 		days_in_month(FEBRUARY) = 29;
1107 	for (i = 1; day >= days_in_month(i); i++)
1108 		day -= days_in_month(i);
1109 	days_in_month(FEBRUARY) = 28;
1110 	tm->tm_mon = i;
1111 
1112 	/* Days are what is left over (+1) from all that. */
1113 	tm->tm_mday = day + 1;
1114 
1115 	/*
1116 	 * No-one uses the day of the week.
1117 	 */
1118 	tm->tm_wday = -1;
1119 }
1120 EXPORT_SYMBOL(to_tm);
1121 
1122 /*
1123  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1124  * result.
1125  */
1126 void div128_by_32(u64 dividend_high, u64 dividend_low,
1127 		  unsigned divisor, struct div_result *dr)
1128 {
1129 	unsigned long a, b, c, d;
1130 	unsigned long w, x, y, z;
1131 	u64 ra, rb, rc;
1132 
1133 	a = dividend_high >> 32;
1134 	b = dividend_high & 0xffffffff;
1135 	c = dividend_low >> 32;
1136 	d = dividend_low & 0xffffffff;
1137 
1138 	w = a / divisor;
1139 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1140 
1141 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1142 	x = ra;
1143 
1144 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1145 	y = rb;
1146 
1147 	do_div(rc, divisor);
1148 	z = rc;
1149 
1150 	dr->result_high = ((u64)w << 32) + x;
1151 	dr->result_low  = ((u64)y << 32) + z;
1152 
1153 }
1154 
1155 /* We don't need to calibrate delay, we use the CPU timebase for that */
1156 void calibrate_delay(void)
1157 {
1158 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1159 	 * as the number of __delay(1) in a jiffy, so make it so
1160 	 */
1161 	loops_per_jiffy = tb_ticks_per_jiffy;
1162 }
1163 
1164 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1165 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1166 {
1167 	ppc_md.get_rtc_time(tm);
1168 	return rtc_valid_tm(tm);
1169 }
1170 
1171 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1172 {
1173 	if (!ppc_md.set_rtc_time)
1174 		return -EOPNOTSUPP;
1175 
1176 	if (ppc_md.set_rtc_time(tm) < 0)
1177 		return -EOPNOTSUPP;
1178 
1179 	return 0;
1180 }
1181 
1182 static const struct rtc_class_ops rtc_generic_ops = {
1183 	.read_time = rtc_generic_get_time,
1184 	.set_time = rtc_generic_set_time,
1185 };
1186 
1187 static int __init rtc_init(void)
1188 {
1189 	struct platform_device *pdev;
1190 
1191 	if (!ppc_md.get_rtc_time)
1192 		return -ENODEV;
1193 
1194 	pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1195 					     &rtc_generic_ops,
1196 					     sizeof(rtc_generic_ops));
1197 
1198 	return PTR_ERR_OR_ZERO(pdev);
1199 }
1200 
1201 device_initcall(rtc_init);
1202 #endif
1203