xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision b9b77222)
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/sched/clock.h>
38 #include <linux/kernel.h>
39 #include <linux/param.h>
40 #include <linux/string.h>
41 #include <linux/mm.h>
42 #include <linux/interrupt.h>
43 #include <linux/timex.h>
44 #include <linux/kernel_stat.h>
45 #include <linux/time.h>
46 #include <linux/clockchips.h>
47 #include <linux/init.h>
48 #include <linux/profile.h>
49 #include <linux/cpu.h>
50 #include <linux/security.h>
51 #include <linux/percpu.h>
52 #include <linux/rtc.h>
53 #include <linux/jiffies.h>
54 #include <linux/posix-timers.h>
55 #include <linux/irq.h>
56 #include <linux/delay.h>
57 #include <linux/irq_work.h>
58 #include <linux/clk-provider.h>
59 #include <linux/suspend.h>
60 #include <linux/rtc.h>
61 #include <linux/sched/cputime.h>
62 #include <linux/processor.h>
63 #include <asm/trace.h>
64 
65 #include <asm/io.h>
66 #include <asm/nvram.h>
67 #include <asm/cache.h>
68 #include <asm/machdep.h>
69 #include <linux/uaccess.h>
70 #include <asm/time.h>
71 #include <asm/prom.h>
72 #include <asm/irq.h>
73 #include <asm/div64.h>
74 #include <asm/smp.h>
75 #include <asm/vdso_datapage.h>
76 #include <asm/firmware.h>
77 #include <asm/asm-prototypes.h>
78 
79 /* powerpc clocksource/clockevent code */
80 
81 #include <linux/clockchips.h>
82 #include <linux/timekeeper_internal.h>
83 
84 static u64 rtc_read(struct clocksource *);
85 static struct clocksource clocksource_rtc = {
86 	.name         = "rtc",
87 	.rating       = 400,
88 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
89 	.mask         = CLOCKSOURCE_MASK(64),
90 	.read         = rtc_read,
91 };
92 
93 static u64 timebase_read(struct clocksource *);
94 static struct clocksource clocksource_timebase = {
95 	.name         = "timebase",
96 	.rating       = 400,
97 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
98 	.mask         = CLOCKSOURCE_MASK(64),
99 	.read         = timebase_read,
100 };
101 
102 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
103 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
104 
105 static int decrementer_set_next_event(unsigned long evt,
106 				      struct clock_event_device *dev);
107 static int decrementer_shutdown(struct clock_event_device *evt);
108 
109 struct clock_event_device decrementer_clockevent = {
110 	.name			= "decrementer",
111 	.rating			= 200,
112 	.irq			= 0,
113 	.set_next_event		= decrementer_set_next_event,
114 	.set_state_shutdown	= decrementer_shutdown,
115 	.tick_resume		= decrementer_shutdown,
116 	.features		= CLOCK_EVT_FEAT_ONESHOT |
117 				  CLOCK_EVT_FEAT_C3STOP,
118 };
119 EXPORT_SYMBOL(decrementer_clockevent);
120 
121 DEFINE_PER_CPU(u64, decrementers_next_tb);
122 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
123 
124 #define XSEC_PER_SEC (1024*1024)
125 
126 #ifdef CONFIG_PPC64
127 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
128 #else
129 /* compute ((xsec << 12) * max) >> 32 */
130 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
131 #endif
132 
133 unsigned long tb_ticks_per_jiffy;
134 unsigned long tb_ticks_per_usec = 100; /* sane default */
135 EXPORT_SYMBOL(tb_ticks_per_usec);
136 unsigned long tb_ticks_per_sec;
137 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
138 
139 DEFINE_SPINLOCK(rtc_lock);
140 EXPORT_SYMBOL_GPL(rtc_lock);
141 
142 static u64 tb_to_ns_scale __read_mostly;
143 static unsigned tb_to_ns_shift __read_mostly;
144 static u64 boot_tb __read_mostly;
145 
146 extern struct timezone sys_tz;
147 static long timezone_offset;
148 
149 unsigned long ppc_proc_freq;
150 EXPORT_SYMBOL_GPL(ppc_proc_freq);
151 unsigned long ppc_tb_freq;
152 EXPORT_SYMBOL_GPL(ppc_tb_freq);
153 
154 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
155 /*
156  * Factor for converting from cputime_t (timebase ticks) to
157  * microseconds. This is stored as 0.64 fixed-point binary fraction.
158  */
159 u64 __cputime_usec_factor;
160 EXPORT_SYMBOL(__cputime_usec_factor);
161 
162 #ifdef CONFIG_PPC_SPLPAR
163 void (*dtl_consumer)(struct dtl_entry *, u64);
164 #endif
165 
166 static void calc_cputime_factors(void)
167 {
168 	struct div_result res;
169 
170 	div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
171 	__cputime_usec_factor = res.result_low;
172 }
173 
174 /*
175  * Read the SPURR on systems that have it, otherwise the PURR,
176  * or if that doesn't exist return the timebase value passed in.
177  */
178 static unsigned long read_spurr(unsigned long tb)
179 {
180 	if (cpu_has_feature(CPU_FTR_SPURR))
181 		return mfspr(SPRN_SPURR);
182 	if (cpu_has_feature(CPU_FTR_PURR))
183 		return mfspr(SPRN_PURR);
184 	return tb;
185 }
186 
187 #ifdef CONFIG_PPC_SPLPAR
188 
189 /*
190  * Scan the dispatch trace log and count up the stolen time.
191  * Should be called with interrupts disabled.
192  */
193 static u64 scan_dispatch_log(u64 stop_tb)
194 {
195 	u64 i = local_paca->dtl_ridx;
196 	struct dtl_entry *dtl = local_paca->dtl_curr;
197 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
198 	struct lppaca *vpa = local_paca->lppaca_ptr;
199 	u64 tb_delta;
200 	u64 stolen = 0;
201 	u64 dtb;
202 
203 	if (!dtl)
204 		return 0;
205 
206 	if (i == be64_to_cpu(vpa->dtl_idx))
207 		return 0;
208 	while (i < be64_to_cpu(vpa->dtl_idx)) {
209 		dtb = be64_to_cpu(dtl->timebase);
210 		tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
211 			be32_to_cpu(dtl->ready_to_enqueue_time);
212 		barrier();
213 		if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
214 			/* buffer has overflowed */
215 			i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
216 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
217 			continue;
218 		}
219 		if (dtb > stop_tb)
220 			break;
221 		if (dtl_consumer)
222 			dtl_consumer(dtl, i);
223 		stolen += tb_delta;
224 		++i;
225 		++dtl;
226 		if (dtl == dtl_end)
227 			dtl = local_paca->dispatch_log;
228 	}
229 	local_paca->dtl_ridx = i;
230 	local_paca->dtl_curr = dtl;
231 	return stolen;
232 }
233 
234 /*
235  * Accumulate stolen time by scanning the dispatch trace log.
236  * Called on entry from user mode.
237  */
238 void accumulate_stolen_time(void)
239 {
240 	u64 sst, ust;
241 	unsigned long save_irq_soft_mask = irq_soft_mask_return();
242 	struct cpu_accounting_data *acct = &local_paca->accounting;
243 
244 	/* We are called early in the exception entry, before
245 	 * soft/hard_enabled are sync'ed to the expected state
246 	 * for the exception. We are hard disabled but the PACA
247 	 * needs to reflect that so various debug stuff doesn't
248 	 * complain
249 	 */
250 	irq_soft_mask_set(IRQS_DISABLED);
251 
252 	sst = scan_dispatch_log(acct->starttime_user);
253 	ust = scan_dispatch_log(acct->starttime);
254 	acct->stime -= sst;
255 	acct->utime -= ust;
256 	acct->steal_time += ust + sst;
257 
258 	irq_soft_mask_set(save_irq_soft_mask);
259 }
260 
261 static inline u64 calculate_stolen_time(u64 stop_tb)
262 {
263 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
264 		return 0;
265 
266 	if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
267 		return scan_dispatch_log(stop_tb);
268 
269 	return 0;
270 }
271 
272 #else /* CONFIG_PPC_SPLPAR */
273 static inline u64 calculate_stolen_time(u64 stop_tb)
274 {
275 	return 0;
276 }
277 
278 #endif /* CONFIG_PPC_SPLPAR */
279 
280 /*
281  * Account time for a transition between system, hard irq
282  * or soft irq state.
283  */
284 static unsigned long vtime_delta(struct task_struct *tsk,
285 				 unsigned long *stime_scaled,
286 				 unsigned long *steal_time)
287 {
288 	unsigned long now, nowscaled, deltascaled;
289 	unsigned long stime;
290 	unsigned long utime, utime_scaled;
291 	struct cpu_accounting_data *acct = get_accounting(tsk);
292 
293 	WARN_ON_ONCE(!irqs_disabled());
294 
295 	now = mftb();
296 	nowscaled = read_spurr(now);
297 	stime = now - acct->starttime;
298 	acct->starttime = now;
299 	deltascaled = nowscaled - acct->startspurr;
300 	acct->startspurr = nowscaled;
301 
302 	*steal_time = calculate_stolen_time(now);
303 
304 	utime = acct->utime - acct->utime_sspurr;
305 	acct->utime_sspurr = acct->utime;
306 
307 	/*
308 	 * Because we don't read the SPURR on every kernel entry/exit,
309 	 * deltascaled includes both user and system SPURR ticks.
310 	 * Apportion these ticks to system SPURR ticks and user
311 	 * SPURR ticks in the same ratio as the system time (delta)
312 	 * and user time (udelta) values obtained from the timebase
313 	 * over the same interval.  The system ticks get accounted here;
314 	 * the user ticks get saved up in paca->user_time_scaled to be
315 	 * used by account_process_tick.
316 	 */
317 	*stime_scaled = stime;
318 	utime_scaled = utime;
319 	if (deltascaled != stime + utime) {
320 		if (utime) {
321 			*stime_scaled = deltascaled * stime / (stime + utime);
322 			utime_scaled = deltascaled - *stime_scaled;
323 		} else {
324 			*stime_scaled = deltascaled;
325 		}
326 	}
327 	acct->utime_scaled += utime_scaled;
328 
329 	return stime;
330 }
331 
332 void vtime_account_system(struct task_struct *tsk)
333 {
334 	unsigned long stime, stime_scaled, steal_time;
335 	struct cpu_accounting_data *acct = get_accounting(tsk);
336 
337 	stime = vtime_delta(tsk, &stime_scaled, &steal_time);
338 
339 	stime -= min(stime, steal_time);
340 	acct->steal_time += steal_time;
341 
342 	if ((tsk->flags & PF_VCPU) && !irq_count()) {
343 		acct->gtime += stime;
344 		acct->utime_scaled += stime_scaled;
345 	} else {
346 		if (hardirq_count())
347 			acct->hardirq_time += stime;
348 		else if (in_serving_softirq())
349 			acct->softirq_time += stime;
350 		else
351 			acct->stime += stime;
352 
353 		acct->stime_scaled += stime_scaled;
354 	}
355 }
356 EXPORT_SYMBOL_GPL(vtime_account_system);
357 
358 void vtime_account_idle(struct task_struct *tsk)
359 {
360 	unsigned long stime, stime_scaled, steal_time;
361 	struct cpu_accounting_data *acct = get_accounting(tsk);
362 
363 	stime = vtime_delta(tsk, &stime_scaled, &steal_time);
364 	acct->idle_time += stime + steal_time;
365 }
366 
367 /*
368  * Account the whole cputime accumulated in the paca
369  * Must be called with interrupts disabled.
370  * Assumes that vtime_account_system/idle() has been called
371  * recently (i.e. since the last entry from usermode) so that
372  * get_paca()->user_time_scaled is up to date.
373  */
374 void vtime_flush(struct task_struct *tsk)
375 {
376 	struct cpu_accounting_data *acct = get_accounting(tsk);
377 
378 	if (acct->utime)
379 		account_user_time(tsk, cputime_to_nsecs(acct->utime));
380 
381 	if (acct->utime_scaled)
382 		tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
383 
384 	if (acct->gtime)
385 		account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
386 
387 	if (acct->steal_time)
388 		account_steal_time(cputime_to_nsecs(acct->steal_time));
389 
390 	if (acct->idle_time)
391 		account_idle_time(cputime_to_nsecs(acct->idle_time));
392 
393 	if (acct->stime)
394 		account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
395 					  CPUTIME_SYSTEM);
396 	if (acct->stime_scaled)
397 		tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
398 
399 	if (acct->hardirq_time)
400 		account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
401 					  CPUTIME_IRQ);
402 	if (acct->softirq_time)
403 		account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
404 					  CPUTIME_SOFTIRQ);
405 
406 	acct->utime = 0;
407 	acct->utime_scaled = 0;
408 	acct->utime_sspurr = 0;
409 	acct->gtime = 0;
410 	acct->steal_time = 0;
411 	acct->idle_time = 0;
412 	acct->stime = 0;
413 	acct->stime_scaled = 0;
414 	acct->hardirq_time = 0;
415 	acct->softirq_time = 0;
416 }
417 
418 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
419 #define calc_cputime_factors()
420 #endif
421 
422 void __delay(unsigned long loops)
423 {
424 	unsigned long start;
425 	int diff;
426 
427 	spin_begin();
428 	if (__USE_RTC()) {
429 		start = get_rtcl();
430 		do {
431 			/* the RTCL register wraps at 1000000000 */
432 			diff = get_rtcl() - start;
433 			if (diff < 0)
434 				diff += 1000000000;
435 			spin_cpu_relax();
436 		} while (diff < loops);
437 	} else {
438 		start = get_tbl();
439 		while (get_tbl() - start < loops)
440 			spin_cpu_relax();
441 	}
442 	spin_end();
443 }
444 EXPORT_SYMBOL(__delay);
445 
446 void udelay(unsigned long usecs)
447 {
448 	__delay(tb_ticks_per_usec * usecs);
449 }
450 EXPORT_SYMBOL(udelay);
451 
452 #ifdef CONFIG_SMP
453 unsigned long profile_pc(struct pt_regs *regs)
454 {
455 	unsigned long pc = instruction_pointer(regs);
456 
457 	if (in_lock_functions(pc))
458 		return regs->link;
459 
460 	return pc;
461 }
462 EXPORT_SYMBOL(profile_pc);
463 #endif
464 
465 #ifdef CONFIG_IRQ_WORK
466 
467 /*
468  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
469  */
470 #ifdef CONFIG_PPC64
471 static inline unsigned long test_irq_work_pending(void)
472 {
473 	unsigned long x;
474 
475 	asm volatile("lbz %0,%1(13)"
476 		: "=r" (x)
477 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
478 	return x;
479 }
480 
481 static inline void set_irq_work_pending_flag(void)
482 {
483 	asm volatile("stb %0,%1(13)" : :
484 		"r" (1),
485 		"i" (offsetof(struct paca_struct, irq_work_pending)));
486 }
487 
488 static inline void clear_irq_work_pending(void)
489 {
490 	asm volatile("stb %0,%1(13)" : :
491 		"r" (0),
492 		"i" (offsetof(struct paca_struct, irq_work_pending)));
493 }
494 
495 void arch_irq_work_raise(void)
496 {
497 	preempt_disable();
498 	set_irq_work_pending_flag();
499 	/*
500 	 * Non-nmi code running with interrupts disabled will replay
501 	 * irq_happened before it re-enables interrupts, so setthe
502 	 * decrementer there instead of causing a hardware exception
503 	 * which would immediately hit the masked interrupt handler
504 	 * and have the net effect of setting the decrementer in
505 	 * irq_happened.
506 	 *
507 	 * NMI interrupts can not check this when they return, so the
508 	 * decrementer hardware exception is raised, which will fire
509 	 * when interrupts are next enabled.
510 	 *
511 	 * BookE does not support this yet, it must audit all NMI
512 	 * interrupt handlers to ensure they call nmi_enter() so this
513 	 * check would be correct.
514 	 */
515 	if (IS_ENABLED(CONFIG_BOOKE) || !irqs_disabled() || in_nmi()) {
516 		set_dec(1);
517 	} else {
518 		hard_irq_disable();
519 		local_paca->irq_happened |= PACA_IRQ_DEC;
520 	}
521 	preempt_enable();
522 }
523 
524 #else /* 32-bit */
525 
526 DEFINE_PER_CPU(u8, irq_work_pending);
527 
528 #define set_irq_work_pending_flag()	__this_cpu_write(irq_work_pending, 1)
529 #define test_irq_work_pending()		__this_cpu_read(irq_work_pending)
530 #define clear_irq_work_pending()	__this_cpu_write(irq_work_pending, 0)
531 
532 void arch_irq_work_raise(void)
533 {
534 	preempt_disable();
535 	set_irq_work_pending_flag();
536 	set_dec(1);
537 	preempt_enable();
538 }
539 
540 #endif /* 32 vs 64 bit */
541 
542 #else  /* CONFIG_IRQ_WORK */
543 
544 #define test_irq_work_pending()	0
545 #define clear_irq_work_pending()
546 
547 #endif /* CONFIG_IRQ_WORK */
548 
549 /*
550  * timer_interrupt - gets called when the decrementer overflows,
551  * with interrupts disabled.
552  */
553 void timer_interrupt(struct pt_regs *regs)
554 {
555 	struct clock_event_device *evt = this_cpu_ptr(&decrementers);
556 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
557 	struct pt_regs *old_regs;
558 	u64 now;
559 
560 	/* Some implementations of hotplug will get timer interrupts while
561 	 * offline, just ignore these and we also need to set
562 	 * decrementers_next_tb as MAX to make sure __check_irq_replay
563 	 * don't replay timer interrupt when return, otherwise we'll trap
564 	 * here infinitely :(
565 	 */
566 	if (unlikely(!cpu_online(smp_processor_id()))) {
567 		*next_tb = ~(u64)0;
568 		set_dec(decrementer_max);
569 		return;
570 	}
571 
572 	/* Ensure a positive value is written to the decrementer, or else
573 	 * some CPUs will continue to take decrementer exceptions. When the
574 	 * PPC_WATCHDOG (decrementer based) is configured, keep this at most
575 	 * 31 bits, which is about 4 seconds on most systems, which gives
576 	 * the watchdog a chance of catching timer interrupt hard lockups.
577 	 */
578 	if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
579 		set_dec(0x7fffffff);
580 	else
581 		set_dec(decrementer_max);
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 	trace_timer_interrupt_entry(regs);
597 
598 	if (test_irq_work_pending()) {
599 		clear_irq_work_pending();
600 		irq_work_run();
601 	}
602 
603 	now = get_tb_or_rtc();
604 	if (now >= *next_tb) {
605 		*next_tb = ~(u64)0;
606 		if (evt->event_handler)
607 			evt->event_handler(evt);
608 		__this_cpu_inc(irq_stat.timer_irqs_event);
609 	} else {
610 		now = *next_tb - now;
611 		if (now <= decrementer_max)
612 			set_dec(now);
613 		/* We may have raced with new irq work */
614 		if (test_irq_work_pending())
615 			set_dec(1);
616 		__this_cpu_inc(irq_stat.timer_irqs_others);
617 	}
618 
619 	trace_timer_interrupt_exit(regs);
620 	irq_exit();
621 	set_irq_regs(old_regs);
622 }
623 EXPORT_SYMBOL(timer_interrupt);
624 
625 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
626 void timer_broadcast_interrupt(void)
627 {
628 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
629 
630 	*next_tb = ~(u64)0;
631 	tick_receive_broadcast();
632 	__this_cpu_inc(irq_stat.broadcast_irqs_event);
633 }
634 #endif
635 
636 /*
637  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
638  * left pending on exit from a KVM guest.  We don't need to do anything
639  * to clear them, as they are edge-triggered.
640  */
641 void hdec_interrupt(struct pt_regs *regs)
642 {
643 }
644 
645 #ifdef CONFIG_SUSPEND
646 static void generic_suspend_disable_irqs(void)
647 {
648 	/* Disable the decrementer, so that it doesn't interfere
649 	 * with suspending.
650 	 */
651 
652 	set_dec(decrementer_max);
653 	local_irq_disable();
654 	set_dec(decrementer_max);
655 }
656 
657 static void generic_suspend_enable_irqs(void)
658 {
659 	local_irq_enable();
660 }
661 
662 /* Overrides the weak version in kernel/power/main.c */
663 void arch_suspend_disable_irqs(void)
664 {
665 	if (ppc_md.suspend_disable_irqs)
666 		ppc_md.suspend_disable_irqs();
667 	generic_suspend_disable_irqs();
668 }
669 
670 /* Overrides the weak version in kernel/power/main.c */
671 void arch_suspend_enable_irqs(void)
672 {
673 	generic_suspend_enable_irqs();
674 	if (ppc_md.suspend_enable_irqs)
675 		ppc_md.suspend_enable_irqs();
676 }
677 #endif
678 
679 unsigned long long tb_to_ns(unsigned long long ticks)
680 {
681 	return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
682 }
683 EXPORT_SYMBOL_GPL(tb_to_ns);
684 
685 /*
686  * Scheduler clock - returns current time in nanosec units.
687  *
688  * Note: mulhdu(a, b) (multiply high double unsigned) returns
689  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
690  * are 64-bit unsigned numbers.
691  */
692 notrace unsigned long long sched_clock(void)
693 {
694 	if (__USE_RTC())
695 		return get_rtc();
696 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
697 }
698 
699 
700 #ifdef CONFIG_PPC_PSERIES
701 
702 /*
703  * Running clock - attempts to give a view of time passing for a virtualised
704  * kernels.
705  * Uses the VTB register if available otherwise a next best guess.
706  */
707 unsigned long long running_clock(void)
708 {
709 	/*
710 	 * Don't read the VTB as a host since KVM does not switch in host
711 	 * timebase into the VTB when it takes a guest off the CPU, reading the
712 	 * VTB would result in reading 'last switched out' guest VTB.
713 	 *
714 	 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
715 	 * would be unsafe to rely only on the #ifdef above.
716 	 */
717 	if (firmware_has_feature(FW_FEATURE_LPAR) &&
718 	    cpu_has_feature(CPU_FTR_ARCH_207S))
719 		return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
720 
721 	/*
722 	 * This is a next best approximation without a VTB.
723 	 * On a host which is running bare metal there should never be any stolen
724 	 * time and on a host which doesn't do any virtualisation TB *should* equal
725 	 * VTB so it makes no difference anyway.
726 	 */
727 	return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
728 }
729 #endif
730 
731 static int __init get_freq(char *name, int cells, unsigned long *val)
732 {
733 	struct device_node *cpu;
734 	const __be32 *fp;
735 	int found = 0;
736 
737 	/* The cpu node should have timebase and clock frequency properties */
738 	cpu = of_find_node_by_type(NULL, "cpu");
739 
740 	if (cpu) {
741 		fp = of_get_property(cpu, name, NULL);
742 		if (fp) {
743 			found = 1;
744 			*val = of_read_ulong(fp, cells);
745 		}
746 
747 		of_node_put(cpu);
748 	}
749 
750 	return found;
751 }
752 
753 static void start_cpu_decrementer(void)
754 {
755 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
756 	unsigned int tcr;
757 
758 	/* Clear any pending timer interrupts */
759 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
760 
761 	tcr = mfspr(SPRN_TCR);
762 	/*
763 	 * The watchdog may have already been enabled by u-boot. So leave
764 	 * TRC[WP] (Watchdog Period) alone.
765 	 */
766 	tcr &= TCR_WP_MASK;	/* Clear all bits except for TCR[WP] */
767 	tcr |= TCR_DIE;		/* Enable decrementer */
768 	mtspr(SPRN_TCR, tcr);
769 #endif
770 }
771 
772 void __init generic_calibrate_decr(void)
773 {
774 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
775 
776 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
777 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
778 
779 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
780 				"(not found)\n");
781 	}
782 
783 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
784 
785 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
786 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
787 
788 		printk(KERN_ERR "WARNING: Estimating processor frequency "
789 				"(not found)\n");
790 	}
791 }
792 
793 int update_persistent_clock64(struct timespec64 now)
794 {
795 	struct rtc_time tm;
796 
797 	if (!ppc_md.set_rtc_time)
798 		return -ENODEV;
799 
800 	rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
801 
802 	return ppc_md.set_rtc_time(&tm);
803 }
804 
805 static void __read_persistent_clock(struct timespec64 *ts)
806 {
807 	struct rtc_time tm;
808 	static int first = 1;
809 
810 	ts->tv_nsec = 0;
811 	/* XXX this is a litle fragile but will work okay in the short term */
812 	if (first) {
813 		first = 0;
814 		if (ppc_md.time_init)
815 			timezone_offset = ppc_md.time_init();
816 
817 		/* get_boot_time() isn't guaranteed to be safe to call late */
818 		if (ppc_md.get_boot_time) {
819 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
820 			return;
821 		}
822 	}
823 	if (!ppc_md.get_rtc_time) {
824 		ts->tv_sec = 0;
825 		return;
826 	}
827 	ppc_md.get_rtc_time(&tm);
828 
829 	ts->tv_sec = rtc_tm_to_time64(&tm);
830 }
831 
832 void read_persistent_clock64(struct timespec64 *ts)
833 {
834 	__read_persistent_clock(ts);
835 
836 	/* Sanitize it in case real time clock is set below EPOCH */
837 	if (ts->tv_sec < 0) {
838 		ts->tv_sec = 0;
839 		ts->tv_nsec = 0;
840 	}
841 
842 }
843 
844 /* clocksource code */
845 static notrace u64 rtc_read(struct clocksource *cs)
846 {
847 	return (u64)get_rtc();
848 }
849 
850 static notrace u64 timebase_read(struct clocksource *cs)
851 {
852 	return (u64)get_tb();
853 }
854 
855 
856 void update_vsyscall(struct timekeeper *tk)
857 {
858 	struct timespec xt;
859 	struct clocksource *clock = tk->tkr_mono.clock;
860 	u32 mult = tk->tkr_mono.mult;
861 	u32 shift = tk->tkr_mono.shift;
862 	u64 cycle_last = tk->tkr_mono.cycle_last;
863 	u64 new_tb_to_xs, new_stamp_xsec;
864 	u64 frac_sec;
865 
866 	if (clock != &clocksource_timebase)
867 		return;
868 
869 	xt.tv_sec = tk->xtime_sec;
870 	xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
871 
872 	/* Make userspace gettimeofday spin until we're done. */
873 	++vdso_data->tb_update_count;
874 	smp_mb();
875 
876 	/*
877 	 * This computes ((2^20 / 1e9) * mult) >> shift as a
878 	 * 0.64 fixed-point fraction.
879 	 * The computation in the else clause below won't overflow
880 	 * (as long as the timebase frequency is >= 1.049 MHz)
881 	 * but loses precision because we lose the low bits of the constant
882 	 * in the shift.  Note that 19342813113834067 ~= 2^(20+64) / 1e9.
883 	 * For a shift of 24 the error is about 0.5e-9, or about 0.5ns
884 	 * over a second.  (Shift values are usually 22, 23 or 24.)
885 	 * For high frequency clocks such as the 512MHz timebase clock
886 	 * on POWER[6789], the mult value is small (e.g. 32768000)
887 	 * and so we can shift the constant by 16 initially
888 	 * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the
889 	 * remaining shifts after the multiplication, which gives a
890 	 * more accurate result (e.g. with mult = 32768000, shift = 24,
891 	 * the error is only about 1.2e-12, or 0.7ns over 10 minutes).
892 	 */
893 	if (mult <= 62500000 && clock->shift >= 16)
894 		new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16);
895 	else
896 		new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
897 
898 	/*
899 	 * Compute the fractional second in units of 2^-32 seconds.
900 	 * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift
901 	 * in nanoseconds, so multiplying that by 2^32 / 1e9 gives
902 	 * it in units of 2^-32 seconds.
903 	 * We assume shift <= 32 because clocks_calc_mult_shift()
904 	 * generates shift values in the range 0 - 32.
905 	 */
906 	frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift);
907 	do_div(frac_sec, NSEC_PER_SEC);
908 
909 	/*
910 	 * Work out new stamp_xsec value for any legacy users of systemcfg.
911 	 * stamp_xsec is in units of 2^-20 seconds.
912 	 */
913 	new_stamp_xsec = frac_sec >> 12;
914 	new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC;
915 
916 	/*
917 	 * tb_update_count is used to allow the userspace gettimeofday code
918 	 * to assure itself that it sees a consistent view of the tb_to_xs and
919 	 * stamp_xsec variables.  It reads the tb_update_count, then reads
920 	 * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
921 	 * the two values of tb_update_count match and are even then the
922 	 * tb_to_xs and stamp_xsec values are consistent.  If not, then it
923 	 * loops back and reads them again until this criteria is met.
924 	 */
925 	vdso_data->tb_orig_stamp = cycle_last;
926 	vdso_data->stamp_xsec = new_stamp_xsec;
927 	vdso_data->tb_to_xs = new_tb_to_xs;
928 	vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec;
929 	vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec;
930 	vdso_data->stamp_xtime = xt;
931 	vdso_data->stamp_sec_fraction = frac_sec;
932 	smp_wmb();
933 	++(vdso_data->tb_update_count);
934 }
935 
936 void update_vsyscall_tz(void)
937 {
938 	vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
939 	vdso_data->tz_dsttime = sys_tz.tz_dsttime;
940 }
941 
942 static void __init clocksource_init(void)
943 {
944 	struct clocksource *clock;
945 
946 	if (__USE_RTC())
947 		clock = &clocksource_rtc;
948 	else
949 		clock = &clocksource_timebase;
950 
951 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
952 		printk(KERN_ERR "clocksource: %s is already registered\n",
953 		       clock->name);
954 		return;
955 	}
956 
957 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
958 	       clock->name, clock->mult, clock->shift);
959 }
960 
961 static int decrementer_set_next_event(unsigned long evt,
962 				      struct clock_event_device *dev)
963 {
964 	__this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
965 	set_dec(evt);
966 
967 	/* We may have raced with new irq work */
968 	if (test_irq_work_pending())
969 		set_dec(1);
970 
971 	return 0;
972 }
973 
974 static int decrementer_shutdown(struct clock_event_device *dev)
975 {
976 	decrementer_set_next_event(decrementer_max, dev);
977 	return 0;
978 }
979 
980 static void register_decrementer_clockevent(int cpu)
981 {
982 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
983 
984 	*dec = decrementer_clockevent;
985 	dec->cpumask = cpumask_of(cpu);
986 
987 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
988 		    dec->name, dec->mult, dec->shift, cpu);
989 
990 	clockevents_register_device(dec);
991 }
992 
993 static void enable_large_decrementer(void)
994 {
995 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
996 		return;
997 
998 	if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
999 		return;
1000 
1001 	/*
1002 	 * If we're running as the hypervisor we need to enable the LD manually
1003 	 * otherwise firmware should have done it for us.
1004 	 */
1005 	if (cpu_has_feature(CPU_FTR_HVMODE))
1006 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
1007 }
1008 
1009 static void __init set_decrementer_max(void)
1010 {
1011 	struct device_node *cpu;
1012 	u32 bits = 32;
1013 
1014 	/* Prior to ISAv3 the decrementer is always 32 bit */
1015 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1016 		return;
1017 
1018 	cpu = of_find_node_by_type(NULL, "cpu");
1019 
1020 	if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
1021 		if (bits > 64 || bits < 32) {
1022 			pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
1023 			bits = 32;
1024 		}
1025 
1026 		/* calculate the signed maximum given this many bits */
1027 		decrementer_max = (1ul << (bits - 1)) - 1;
1028 	}
1029 
1030 	of_node_put(cpu);
1031 
1032 	pr_info("time_init: %u bit decrementer (max: %llx)\n",
1033 		bits, decrementer_max);
1034 }
1035 
1036 static void __init init_decrementer_clockevent(void)
1037 {
1038 	int cpu = smp_processor_id();
1039 
1040 	clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
1041 
1042 	decrementer_clockevent.max_delta_ns =
1043 		clockevent_delta2ns(decrementer_max, &decrementer_clockevent);
1044 	decrementer_clockevent.max_delta_ticks = decrementer_max;
1045 	decrementer_clockevent.min_delta_ns =
1046 		clockevent_delta2ns(2, &decrementer_clockevent);
1047 	decrementer_clockevent.min_delta_ticks = 2;
1048 
1049 	register_decrementer_clockevent(cpu);
1050 }
1051 
1052 void secondary_cpu_time_init(void)
1053 {
1054 	/* Enable and test the large decrementer for this cpu */
1055 	enable_large_decrementer();
1056 
1057 	/* Start the decrementer on CPUs that have manual control
1058 	 * such as BookE
1059 	 */
1060 	start_cpu_decrementer();
1061 
1062 	/* FIME: Should make unrelatred change to move snapshot_timebase
1063 	 * call here ! */
1064 	register_decrementer_clockevent(smp_processor_id());
1065 }
1066 
1067 /* This function is only called on the boot processor */
1068 void __init time_init(void)
1069 {
1070 	struct div_result res;
1071 	u64 scale;
1072 	unsigned shift;
1073 
1074 	if (__USE_RTC()) {
1075 		/* 601 processor: dec counts down by 128 every 128ns */
1076 		ppc_tb_freq = 1000000000;
1077 	} else {
1078 		/* Normal PowerPC with timebase register */
1079 		ppc_md.calibrate_decr();
1080 		printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1081 		       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1082 		printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
1083 		       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1084 	}
1085 
1086 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1087 	tb_ticks_per_sec = ppc_tb_freq;
1088 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
1089 	calc_cputime_factors();
1090 
1091 	/*
1092 	 * Compute scale factor for sched_clock.
1093 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1094 	 * which is the timebase frequency.
1095 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1096 	 * the 128-bit result as a 64.64 fixed-point number.
1097 	 * We then shift that number right until it is less than 1.0,
1098 	 * giving us the scale factor and shift count to use in
1099 	 * sched_clock().
1100 	 */
1101 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1102 	scale = res.result_low;
1103 	for (shift = 0; res.result_high != 0; ++shift) {
1104 		scale = (scale >> 1) | (res.result_high << 63);
1105 		res.result_high >>= 1;
1106 	}
1107 	tb_to_ns_scale = scale;
1108 	tb_to_ns_shift = shift;
1109 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1110 	boot_tb = get_tb_or_rtc();
1111 
1112 	/* If platform provided a timezone (pmac), we correct the time */
1113 	if (timezone_offset) {
1114 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1115 		sys_tz.tz_dsttime = 0;
1116 	}
1117 
1118 	vdso_data->tb_update_count = 0;
1119 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1120 
1121 	/* initialise and enable the large decrementer (if we have one) */
1122 	set_decrementer_max();
1123 	enable_large_decrementer();
1124 
1125 	/* Start the decrementer on CPUs that have manual control
1126 	 * such as BookE
1127 	 */
1128 	start_cpu_decrementer();
1129 
1130 	/* Register the clocksource */
1131 	clocksource_init();
1132 
1133 	init_decrementer_clockevent();
1134 	tick_setup_hrtimer_broadcast();
1135 
1136 #ifdef CONFIG_COMMON_CLK
1137 	of_clk_init(NULL);
1138 #endif
1139 }
1140 
1141 /*
1142  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1143  * result.
1144  */
1145 void div128_by_32(u64 dividend_high, u64 dividend_low,
1146 		  unsigned divisor, struct div_result *dr)
1147 {
1148 	unsigned long a, b, c, d;
1149 	unsigned long w, x, y, z;
1150 	u64 ra, rb, rc;
1151 
1152 	a = dividend_high >> 32;
1153 	b = dividend_high & 0xffffffff;
1154 	c = dividend_low >> 32;
1155 	d = dividend_low & 0xffffffff;
1156 
1157 	w = a / divisor;
1158 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1159 
1160 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1161 	x = ra;
1162 
1163 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1164 	y = rb;
1165 
1166 	do_div(rc, divisor);
1167 	z = rc;
1168 
1169 	dr->result_high = ((u64)w << 32) + x;
1170 	dr->result_low  = ((u64)y << 32) + z;
1171 
1172 }
1173 
1174 /* We don't need to calibrate delay, we use the CPU timebase for that */
1175 void calibrate_delay(void)
1176 {
1177 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1178 	 * as the number of __delay(1) in a jiffy, so make it so
1179 	 */
1180 	loops_per_jiffy = tb_ticks_per_jiffy;
1181 }
1182 
1183 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1184 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1185 {
1186 	ppc_md.get_rtc_time(tm);
1187 	return 0;
1188 }
1189 
1190 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1191 {
1192 	if (!ppc_md.set_rtc_time)
1193 		return -EOPNOTSUPP;
1194 
1195 	if (ppc_md.set_rtc_time(tm) < 0)
1196 		return -EOPNOTSUPP;
1197 
1198 	return 0;
1199 }
1200 
1201 static const struct rtc_class_ops rtc_generic_ops = {
1202 	.read_time = rtc_generic_get_time,
1203 	.set_time = rtc_generic_set_time,
1204 };
1205 
1206 static int __init rtc_init(void)
1207 {
1208 	struct platform_device *pdev;
1209 
1210 	if (!ppc_md.get_rtc_time)
1211 		return -ENODEV;
1212 
1213 	pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1214 					     &rtc_generic_ops,
1215 					     sizeof(rtc_generic_ops));
1216 
1217 	return PTR_ERR_OR_ZERO(pdev);
1218 }
1219 
1220 device_initcall(rtc_init);
1221 #endif
1222