xref: /linux/arch/powerpc/kernel/time.c (revision 34f7c6e7d4396090692a09789db231e12cb4762b)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Common time routines among all ppc machines.
4  *
5  * Written by Cort Dougan (cort@cs.nmt.edu) to merge
6  * Paul Mackerras' version and mine for PReP and Pmac.
7  * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net).
8  * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com)
9  *
10  * First round of bugfixes by Gabriel Paubert (paubert@iram.es)
11  * to make clock more stable (2.4.0-test5). The only thing
12  * that this code assumes is that the timebases have been synchronized
13  * by firmware on SMP and are never stopped (never do sleep
14  * on SMP then, nap and doze are OK).
15  *
16  * Speeded up do_gettimeofday by getting rid of references to
17  * xtime (which required locks for consistency). (mikejc@us.ibm.com)
18  *
19  * TODO (not necessarily in this file):
20  * - improve precision and reproducibility of timebase frequency
21  * measurement at boot time.
22  * - for astronomical applications: add a new function to get
23  * non ambiguous timestamps even around leap seconds. This needs
24  * a new timestamp format and a good name.
25  *
26  * 1997-09-10  Updated NTP code according to technical memorandum Jan '96
27  *             "A Kernel Model for Precision Timekeeping" by Dave Mills
28  */
29 
30 #include <linux/errno.h>
31 #include <linux/export.h>
32 #include <linux/sched.h>
33 #include <linux/sched/clock.h>
34 #include <linux/sched/cputime.h>
35 #include <linux/kernel.h>
36 #include <linux/param.h>
37 #include <linux/string.h>
38 #include <linux/mm.h>
39 #include <linux/interrupt.h>
40 #include <linux/timex.h>
41 #include <linux/kernel_stat.h>
42 #include <linux/time.h>
43 #include <linux/init.h>
44 #include <linux/profile.h>
45 #include <linux/cpu.h>
46 #include <linux/security.h>
47 #include <linux/percpu.h>
48 #include <linux/rtc.h>
49 #include <linux/jiffies.h>
50 #include <linux/posix-timers.h>
51 #include <linux/irq.h>
52 #include <linux/delay.h>
53 #include <linux/irq_work.h>
54 #include <linux/of_clk.h>
55 #include <linux/suspend.h>
56 #include <linux/processor.h>
57 #include <asm/trace.h>
58 
59 #include <asm/interrupt.h>
60 #include <asm/io.h>
61 #include <asm/nvram.h>
62 #include <asm/cache.h>
63 #include <asm/machdep.h>
64 #include <linux/uaccess.h>
65 #include <asm/time.h>
66 #include <asm/prom.h>
67 #include <asm/irq.h>
68 #include <asm/div64.h>
69 #include <asm/smp.h>
70 #include <asm/vdso_datapage.h>
71 #include <asm/firmware.h>
72 #include <asm/mce.h>
73 
74 /* powerpc clocksource/clockevent code */
75 
76 #include <linux/clockchips.h>
77 #include <linux/timekeeper_internal.h>
78 
79 static u64 timebase_read(struct clocksource *);
80 static struct clocksource clocksource_timebase = {
81 	.name         = "timebase",
82 	.rating       = 400,
83 	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
84 	.mask         = CLOCKSOURCE_MASK(64),
85 	.read         = timebase_read,
86 	.vdso_clock_mode	= VDSO_CLOCKMODE_ARCHTIMER,
87 };
88 
89 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
90 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
91 EXPORT_SYMBOL_GPL(decrementer_max); /* for KVM HDEC */
92 
93 static int decrementer_set_next_event(unsigned long evt,
94 				      struct clock_event_device *dev);
95 static int decrementer_shutdown(struct clock_event_device *evt);
96 
97 struct clock_event_device decrementer_clockevent = {
98 	.name			= "decrementer",
99 	.rating			= 200,
100 	.irq			= 0,
101 	.set_next_event		= decrementer_set_next_event,
102 	.set_state_oneshot_stopped = decrementer_shutdown,
103 	.set_state_shutdown	= decrementer_shutdown,
104 	.tick_resume		= decrementer_shutdown,
105 	.features		= CLOCK_EVT_FEAT_ONESHOT |
106 				  CLOCK_EVT_FEAT_C3STOP,
107 };
108 EXPORT_SYMBOL(decrementer_clockevent);
109 
110 /*
111  * This always puts next_tb beyond now, so the clock event will never fire
112  * with the usual comparison, no need for a separate test for stopped.
113  */
114 #define DEC_CLOCKEVENT_STOPPED ~0ULL
115 DEFINE_PER_CPU(u64, decrementers_next_tb) = DEC_CLOCKEVENT_STOPPED;
116 EXPORT_SYMBOL_GPL(decrementers_next_tb);
117 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
118 
119 #define XSEC_PER_SEC (1024*1024)
120 
121 #ifdef CONFIG_PPC64
122 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
123 #else
124 /* compute ((xsec << 12) * max) >> 32 */
125 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
126 #endif
127 
128 unsigned long tb_ticks_per_jiffy;
129 unsigned long tb_ticks_per_usec = 100; /* sane default */
130 EXPORT_SYMBOL(tb_ticks_per_usec);
131 unsigned long tb_ticks_per_sec;
132 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
133 
134 DEFINE_SPINLOCK(rtc_lock);
135 EXPORT_SYMBOL_GPL(rtc_lock);
136 
137 static u64 tb_to_ns_scale __read_mostly;
138 static unsigned tb_to_ns_shift __read_mostly;
139 static u64 boot_tb __read_mostly;
140 
141 extern struct timezone sys_tz;
142 static long timezone_offset;
143 
144 unsigned long ppc_proc_freq;
145 EXPORT_SYMBOL_GPL(ppc_proc_freq);
146 unsigned long ppc_tb_freq;
147 EXPORT_SYMBOL_GPL(ppc_tb_freq);
148 
149 bool tb_invalid;
150 
151 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
152 /*
153  * Factor for converting from cputime_t (timebase ticks) to
154  * microseconds. This is stored as 0.64 fixed-point binary fraction.
155  */
156 u64 __cputime_usec_factor;
157 EXPORT_SYMBOL(__cputime_usec_factor);
158 
159 #ifdef CONFIG_PPC_SPLPAR
160 void (*dtl_consumer)(struct dtl_entry *, u64);
161 #endif
162 
163 static void calc_cputime_factors(void)
164 {
165 	struct div_result res;
166 
167 	div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
168 	__cputime_usec_factor = res.result_low;
169 }
170 
171 /*
172  * Read the SPURR on systems that have it, otherwise the PURR,
173  * or if that doesn't exist return the timebase value passed in.
174  */
175 static inline unsigned long read_spurr(unsigned long tb)
176 {
177 	if (cpu_has_feature(CPU_FTR_SPURR))
178 		return mfspr(SPRN_SPURR);
179 	if (cpu_has_feature(CPU_FTR_PURR))
180 		return mfspr(SPRN_PURR);
181 	return tb;
182 }
183 
184 #ifdef CONFIG_PPC_SPLPAR
185 
186 #include <asm/dtl.h>
187 
188 /*
189  * Scan the dispatch trace log and count up the stolen time.
190  * Should be called with interrupts disabled.
191  */
192 static u64 scan_dispatch_log(u64 stop_tb)
193 {
194 	u64 i = local_paca->dtl_ridx;
195 	struct dtl_entry *dtl = local_paca->dtl_curr;
196 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
197 	struct lppaca *vpa = local_paca->lppaca_ptr;
198 	u64 tb_delta;
199 	u64 stolen = 0;
200 	u64 dtb;
201 
202 	if (!dtl)
203 		return 0;
204 
205 	if (i == be64_to_cpu(vpa->dtl_idx))
206 		return 0;
207 	while (i < be64_to_cpu(vpa->dtl_idx)) {
208 		dtb = be64_to_cpu(dtl->timebase);
209 		tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
210 			be32_to_cpu(dtl->ready_to_enqueue_time);
211 		barrier();
212 		if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
213 			/* buffer has overflowed */
214 			i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
215 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
216 			continue;
217 		}
218 		if (dtb > stop_tb)
219 			break;
220 		if (dtl_consumer)
221 			dtl_consumer(dtl, i);
222 		stolen += tb_delta;
223 		++i;
224 		++dtl;
225 		if (dtl == dtl_end)
226 			dtl = local_paca->dispatch_log;
227 	}
228 	local_paca->dtl_ridx = i;
229 	local_paca->dtl_curr = dtl;
230 	return stolen;
231 }
232 
233 /*
234  * Accumulate stolen time by scanning the dispatch trace log.
235  * Called on entry from user mode.
236  */
237 void notrace accumulate_stolen_time(void)
238 {
239 	u64 sst, ust;
240 	struct cpu_accounting_data *acct = &local_paca->accounting;
241 
242 	sst = scan_dispatch_log(acct->starttime_user);
243 	ust = scan_dispatch_log(acct->starttime);
244 	acct->stime -= sst;
245 	acct->utime -= ust;
246 	acct->steal_time += ust + sst;
247 }
248 
249 static inline u64 calculate_stolen_time(u64 stop_tb)
250 {
251 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
252 		return 0;
253 
254 	if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
255 		return scan_dispatch_log(stop_tb);
256 
257 	return 0;
258 }
259 
260 #else /* CONFIG_PPC_SPLPAR */
261 static inline u64 calculate_stolen_time(u64 stop_tb)
262 {
263 	return 0;
264 }
265 
266 #endif /* CONFIG_PPC_SPLPAR */
267 
268 /*
269  * Account time for a transition between system, hard irq
270  * or soft irq state.
271  */
272 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
273 					unsigned long now, unsigned long stime)
274 {
275 	unsigned long stime_scaled = 0;
276 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
277 	unsigned long nowscaled, deltascaled;
278 	unsigned long utime, utime_scaled;
279 
280 	nowscaled = read_spurr(now);
281 	deltascaled = nowscaled - acct->startspurr;
282 	acct->startspurr = nowscaled;
283 	utime = acct->utime - acct->utime_sspurr;
284 	acct->utime_sspurr = acct->utime;
285 
286 	/*
287 	 * Because we don't read the SPURR on every kernel entry/exit,
288 	 * deltascaled includes both user and system SPURR ticks.
289 	 * Apportion these ticks to system SPURR ticks and user
290 	 * SPURR ticks in the same ratio as the system time (delta)
291 	 * and user time (udelta) values obtained from the timebase
292 	 * over the same interval.  The system ticks get accounted here;
293 	 * the user ticks get saved up in paca->user_time_scaled to be
294 	 * used by account_process_tick.
295 	 */
296 	stime_scaled = stime;
297 	utime_scaled = utime;
298 	if (deltascaled != stime + utime) {
299 		if (utime) {
300 			stime_scaled = deltascaled * stime / (stime + utime);
301 			utime_scaled = deltascaled - stime_scaled;
302 		} else {
303 			stime_scaled = deltascaled;
304 		}
305 	}
306 	acct->utime_scaled += utime_scaled;
307 #endif
308 
309 	return stime_scaled;
310 }
311 
312 static unsigned long vtime_delta(struct cpu_accounting_data *acct,
313 				 unsigned long *stime_scaled,
314 				 unsigned long *steal_time)
315 {
316 	unsigned long now, stime;
317 
318 	WARN_ON_ONCE(!irqs_disabled());
319 
320 	now = mftb();
321 	stime = now - acct->starttime;
322 	acct->starttime = now;
323 
324 	*stime_scaled = vtime_delta_scaled(acct, now, stime);
325 
326 	*steal_time = calculate_stolen_time(now);
327 
328 	return stime;
329 }
330 
331 static void vtime_delta_kernel(struct cpu_accounting_data *acct,
332 			       unsigned long *stime, unsigned long *stime_scaled)
333 {
334 	unsigned long steal_time;
335 
336 	*stime = vtime_delta(acct, stime_scaled, &steal_time);
337 	*stime -= min(*stime, steal_time);
338 	acct->steal_time += steal_time;
339 }
340 
341 void vtime_account_kernel(struct task_struct *tsk)
342 {
343 	struct cpu_accounting_data *acct = get_accounting(tsk);
344 	unsigned long stime, stime_scaled;
345 
346 	vtime_delta_kernel(acct, &stime, &stime_scaled);
347 
348 	if (tsk->flags & PF_VCPU) {
349 		acct->gtime += stime;
350 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
351 		acct->utime_scaled += stime_scaled;
352 #endif
353 	} else {
354 		acct->stime += stime;
355 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
356 		acct->stime_scaled += stime_scaled;
357 #endif
358 	}
359 }
360 EXPORT_SYMBOL_GPL(vtime_account_kernel);
361 
362 void vtime_account_idle(struct task_struct *tsk)
363 {
364 	unsigned long stime, stime_scaled, steal_time;
365 	struct cpu_accounting_data *acct = get_accounting(tsk);
366 
367 	stime = vtime_delta(acct, &stime_scaled, &steal_time);
368 	acct->idle_time += stime + steal_time;
369 }
370 
371 static void vtime_account_irq_field(struct cpu_accounting_data *acct,
372 				    unsigned long *field)
373 {
374 	unsigned long stime, stime_scaled;
375 
376 	vtime_delta_kernel(acct, &stime, &stime_scaled);
377 	*field += stime;
378 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
379 	acct->stime_scaled += stime_scaled;
380 #endif
381 }
382 
383 void vtime_account_softirq(struct task_struct *tsk)
384 {
385 	struct cpu_accounting_data *acct = get_accounting(tsk);
386 	vtime_account_irq_field(acct, &acct->softirq_time);
387 }
388 
389 void vtime_account_hardirq(struct task_struct *tsk)
390 {
391 	struct cpu_accounting_data *acct = get_accounting(tsk);
392 	vtime_account_irq_field(acct, &acct->hardirq_time);
393 }
394 
395 static void vtime_flush_scaled(struct task_struct *tsk,
396 			       struct cpu_accounting_data *acct)
397 {
398 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
399 	if (acct->utime_scaled)
400 		tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
401 	if (acct->stime_scaled)
402 		tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
403 
404 	acct->utime_scaled = 0;
405 	acct->utime_sspurr = 0;
406 	acct->stime_scaled = 0;
407 #endif
408 }
409 
410 /*
411  * Account the whole cputime accumulated in the paca
412  * Must be called with interrupts disabled.
413  * Assumes that vtime_account_kernel/idle() has been called
414  * recently (i.e. since the last entry from usermode) so that
415  * get_paca()->user_time_scaled is up to date.
416  */
417 void vtime_flush(struct task_struct *tsk)
418 {
419 	struct cpu_accounting_data *acct = get_accounting(tsk);
420 
421 	if (acct->utime)
422 		account_user_time(tsk, cputime_to_nsecs(acct->utime));
423 
424 	if (acct->gtime)
425 		account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
426 
427 	if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
428 		account_steal_time(cputime_to_nsecs(acct->steal_time));
429 		acct->steal_time = 0;
430 	}
431 
432 	if (acct->idle_time)
433 		account_idle_time(cputime_to_nsecs(acct->idle_time));
434 
435 	if (acct->stime)
436 		account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
437 					  CPUTIME_SYSTEM);
438 
439 	if (acct->hardirq_time)
440 		account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
441 					  CPUTIME_IRQ);
442 	if (acct->softirq_time)
443 		account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
444 					  CPUTIME_SOFTIRQ);
445 
446 	vtime_flush_scaled(tsk, acct);
447 
448 	acct->utime = 0;
449 	acct->gtime = 0;
450 	acct->idle_time = 0;
451 	acct->stime = 0;
452 	acct->hardirq_time = 0;
453 	acct->softirq_time = 0;
454 }
455 
456 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
457 #define calc_cputime_factors()
458 #endif
459 
460 void __delay(unsigned long loops)
461 {
462 	unsigned long start;
463 
464 	spin_begin();
465 	if (tb_invalid) {
466 		/*
467 		 * TB is in error state and isn't ticking anymore.
468 		 * HMI handler was unable to recover from TB error.
469 		 * Return immediately, so that kernel won't get stuck here.
470 		 */
471 		spin_cpu_relax();
472 	} else {
473 		start = mftb();
474 		while (mftb() - start < loops)
475 			spin_cpu_relax();
476 	}
477 	spin_end();
478 }
479 EXPORT_SYMBOL(__delay);
480 
481 void udelay(unsigned long usecs)
482 {
483 	__delay(tb_ticks_per_usec * usecs);
484 }
485 EXPORT_SYMBOL(udelay);
486 
487 #ifdef CONFIG_SMP
488 unsigned long profile_pc(struct pt_regs *regs)
489 {
490 	unsigned long pc = instruction_pointer(regs);
491 
492 	if (in_lock_functions(pc))
493 		return regs->link;
494 
495 	return pc;
496 }
497 EXPORT_SYMBOL(profile_pc);
498 #endif
499 
500 #ifdef CONFIG_IRQ_WORK
501 
502 /*
503  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
504  */
505 #ifdef CONFIG_PPC64
506 static inline unsigned long test_irq_work_pending(void)
507 {
508 	unsigned long x;
509 
510 	asm volatile("lbz %0,%1(13)"
511 		: "=r" (x)
512 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
513 	return x;
514 }
515 
516 static inline void set_irq_work_pending_flag(void)
517 {
518 	asm volatile("stb %0,%1(13)" : :
519 		"r" (1),
520 		"i" (offsetof(struct paca_struct, irq_work_pending)));
521 }
522 
523 static inline void clear_irq_work_pending(void)
524 {
525 	asm volatile("stb %0,%1(13)" : :
526 		"r" (0),
527 		"i" (offsetof(struct paca_struct, irq_work_pending)));
528 }
529 
530 #else /* 32-bit */
531 
532 DEFINE_PER_CPU(u8, irq_work_pending);
533 
534 #define set_irq_work_pending_flag()	__this_cpu_write(irq_work_pending, 1)
535 #define test_irq_work_pending()		__this_cpu_read(irq_work_pending)
536 #define clear_irq_work_pending()	__this_cpu_write(irq_work_pending, 0)
537 
538 #endif /* 32 vs 64 bit */
539 
540 void arch_irq_work_raise(void)
541 {
542 	/*
543 	 * 64-bit code that uses irq soft-mask can just cause an immediate
544 	 * interrupt here that gets soft masked, if this is called under
545 	 * local_irq_disable(). It might be possible to prevent that happening
546 	 * by noticing interrupts are disabled and setting decrementer pending
547 	 * to be replayed when irqs are enabled. The problem there is that
548 	 * tracing can call irq_work_raise, including in code that does low
549 	 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
550 	 * which could get tangled up if we're messing with the same state
551 	 * here.
552 	 */
553 	preempt_disable();
554 	set_irq_work_pending_flag();
555 	set_dec(1);
556 	preempt_enable();
557 }
558 
559 static void set_dec_or_work(u64 val)
560 {
561 	set_dec(val);
562 	/* We may have raced with new irq work */
563 	if (unlikely(test_irq_work_pending()))
564 		set_dec(1);
565 }
566 
567 #else  /* CONFIG_IRQ_WORK */
568 
569 #define test_irq_work_pending()	0
570 #define clear_irq_work_pending()
571 
572 static void set_dec_or_work(u64 val)
573 {
574 	set_dec(val);
575 }
576 #endif /* CONFIG_IRQ_WORK */
577 
578 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
579 void timer_rearm_host_dec(u64 now)
580 {
581 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
582 
583 	WARN_ON_ONCE(!arch_irqs_disabled());
584 	WARN_ON_ONCE(mfmsr() & MSR_EE);
585 
586 	if (now >= *next_tb) {
587 		local_paca->irq_happened |= PACA_IRQ_DEC;
588 	} else {
589 		now = *next_tb - now;
590 		if (now > decrementer_max)
591 			now = decrementer_max;
592 		set_dec_or_work(now);
593 	}
594 }
595 EXPORT_SYMBOL_GPL(timer_rearm_host_dec);
596 #endif
597 
598 /*
599  * timer_interrupt - gets called when the decrementer overflows,
600  * with interrupts disabled.
601  */
602 DEFINE_INTERRUPT_HANDLER_ASYNC(timer_interrupt)
603 {
604 	struct clock_event_device *evt = this_cpu_ptr(&decrementers);
605 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
606 	struct pt_regs *old_regs;
607 	u64 now;
608 
609 	/*
610 	 * Some implementations of hotplug will get timer interrupts while
611 	 * offline, just ignore these.
612 	 */
613 	if (unlikely(!cpu_online(smp_processor_id()))) {
614 		set_dec(decrementer_max);
615 		return;
616 	}
617 
618 	/* Conditionally hard-enable interrupts. */
619 	if (should_hard_irq_enable()) {
620 		/*
621 		 * Ensure a positive value is written to the decrementer, or
622 		 * else some CPUs will continue to take decrementer exceptions.
623 		 * When the PPC_WATCHDOG (decrementer based) is configured,
624 		 * keep this at most 31 bits, which is about 4 seconds on most
625 		 * systems, which gives the watchdog a chance of catching timer
626 		 * interrupt hard lockups.
627 		 */
628 		if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
629 			set_dec(0x7fffffff);
630 		else
631 			set_dec(decrementer_max);
632 
633 		do_hard_irq_enable();
634 	}
635 
636 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
637 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
638 		__do_IRQ(regs);
639 #endif
640 
641 	old_regs = set_irq_regs(regs);
642 
643 	trace_timer_interrupt_entry(regs);
644 
645 	if (test_irq_work_pending()) {
646 		clear_irq_work_pending();
647 		mce_run_irq_context_handlers();
648 		irq_work_run();
649 	}
650 
651 	now = get_tb();
652 	if (now >= *next_tb) {
653 		evt->event_handler(evt);
654 		__this_cpu_inc(irq_stat.timer_irqs_event);
655 	} else {
656 		now = *next_tb - now;
657 		if (now > decrementer_max)
658 			now = decrementer_max;
659 		set_dec_or_work(now);
660 		__this_cpu_inc(irq_stat.timer_irqs_others);
661 	}
662 
663 	trace_timer_interrupt_exit(regs);
664 
665 	set_irq_regs(old_regs);
666 }
667 EXPORT_SYMBOL(timer_interrupt);
668 
669 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
670 void timer_broadcast_interrupt(void)
671 {
672 	tick_receive_broadcast();
673 	__this_cpu_inc(irq_stat.broadcast_irqs_event);
674 }
675 #endif
676 
677 #ifdef CONFIG_SUSPEND
678 /* Overrides the weak version in kernel/power/main.c */
679 void arch_suspend_disable_irqs(void)
680 {
681 	if (ppc_md.suspend_disable_irqs)
682 		ppc_md.suspend_disable_irqs();
683 
684 	/* Disable the decrementer, so that it doesn't interfere
685 	 * with suspending.
686 	 */
687 
688 	set_dec(decrementer_max);
689 	local_irq_disable();
690 	set_dec(decrementer_max);
691 }
692 
693 /* Overrides the weak version in kernel/power/main.c */
694 void arch_suspend_enable_irqs(void)
695 {
696 	local_irq_enable();
697 
698 	if (ppc_md.suspend_enable_irqs)
699 		ppc_md.suspend_enable_irqs();
700 }
701 #endif
702 
703 unsigned long long tb_to_ns(unsigned long long ticks)
704 {
705 	return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
706 }
707 EXPORT_SYMBOL_GPL(tb_to_ns);
708 
709 /*
710  * Scheduler clock - returns current time in nanosec units.
711  *
712  * Note: mulhdu(a, b) (multiply high double unsigned) returns
713  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
714  * are 64-bit unsigned numbers.
715  */
716 notrace unsigned long long sched_clock(void)
717 {
718 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
719 }
720 
721 
722 #ifdef CONFIG_PPC_PSERIES
723 
724 /*
725  * Running clock - attempts to give a view of time passing for a virtualised
726  * kernels.
727  * Uses the VTB register if available otherwise a next best guess.
728  */
729 unsigned long long running_clock(void)
730 {
731 	/*
732 	 * Don't read the VTB as a host since KVM does not switch in host
733 	 * timebase into the VTB when it takes a guest off the CPU, reading the
734 	 * VTB would result in reading 'last switched out' guest VTB.
735 	 *
736 	 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
737 	 * would be unsafe to rely only on the #ifdef above.
738 	 */
739 	if (firmware_has_feature(FW_FEATURE_LPAR) &&
740 	    cpu_has_feature(CPU_FTR_ARCH_207S))
741 		return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
742 
743 	/*
744 	 * This is a next best approximation without a VTB.
745 	 * On a host which is running bare metal there should never be any stolen
746 	 * time and on a host which doesn't do any virtualisation TB *should* equal
747 	 * VTB so it makes no difference anyway.
748 	 */
749 	return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
750 }
751 #endif
752 
753 static int __init get_freq(char *name, int cells, unsigned long *val)
754 {
755 	struct device_node *cpu;
756 	const __be32 *fp;
757 	int found = 0;
758 
759 	/* The cpu node should have timebase and clock frequency properties */
760 	cpu = of_find_node_by_type(NULL, "cpu");
761 
762 	if (cpu) {
763 		fp = of_get_property(cpu, name, NULL);
764 		if (fp) {
765 			found = 1;
766 			*val = of_read_ulong(fp, cells);
767 		}
768 
769 		of_node_put(cpu);
770 	}
771 
772 	return found;
773 }
774 
775 static void start_cpu_decrementer(void)
776 {
777 #ifdef CONFIG_BOOKE_OR_40x
778 	unsigned int tcr;
779 
780 	/* Clear any pending timer interrupts */
781 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
782 
783 	tcr = mfspr(SPRN_TCR);
784 	/*
785 	 * The watchdog may have already been enabled by u-boot. So leave
786 	 * TRC[WP] (Watchdog Period) alone.
787 	 */
788 	tcr &= TCR_WP_MASK;	/* Clear all bits except for TCR[WP] */
789 	tcr |= TCR_DIE;		/* Enable decrementer */
790 	mtspr(SPRN_TCR, tcr);
791 #endif
792 }
793 
794 void __init generic_calibrate_decr(void)
795 {
796 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
797 
798 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
799 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
800 
801 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
802 				"(not found)\n");
803 	}
804 
805 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
806 
807 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
808 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
809 
810 		printk(KERN_ERR "WARNING: Estimating processor frequency "
811 				"(not found)\n");
812 	}
813 }
814 
815 int update_persistent_clock64(struct timespec64 now)
816 {
817 	struct rtc_time tm;
818 
819 	if (!ppc_md.set_rtc_time)
820 		return -ENODEV;
821 
822 	rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
823 
824 	return ppc_md.set_rtc_time(&tm);
825 }
826 
827 static void __read_persistent_clock(struct timespec64 *ts)
828 {
829 	struct rtc_time tm;
830 	static int first = 1;
831 
832 	ts->tv_nsec = 0;
833 	/* XXX this is a litle fragile but will work okay in the short term */
834 	if (first) {
835 		first = 0;
836 		if (ppc_md.time_init)
837 			timezone_offset = ppc_md.time_init();
838 
839 		/* get_boot_time() isn't guaranteed to be safe to call late */
840 		if (ppc_md.get_boot_time) {
841 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
842 			return;
843 		}
844 	}
845 	if (!ppc_md.get_rtc_time) {
846 		ts->tv_sec = 0;
847 		return;
848 	}
849 	ppc_md.get_rtc_time(&tm);
850 
851 	ts->tv_sec = rtc_tm_to_time64(&tm);
852 }
853 
854 void read_persistent_clock64(struct timespec64 *ts)
855 {
856 	__read_persistent_clock(ts);
857 
858 	/* Sanitize it in case real time clock is set below EPOCH */
859 	if (ts->tv_sec < 0) {
860 		ts->tv_sec = 0;
861 		ts->tv_nsec = 0;
862 	}
863 
864 }
865 
866 /* clocksource code */
867 static notrace u64 timebase_read(struct clocksource *cs)
868 {
869 	return (u64)get_tb();
870 }
871 
872 static void __init clocksource_init(void)
873 {
874 	struct clocksource *clock = &clocksource_timebase;
875 
876 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
877 		printk(KERN_ERR "clocksource: %s is already registered\n",
878 		       clock->name);
879 		return;
880 	}
881 
882 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
883 	       clock->name, clock->mult, clock->shift);
884 }
885 
886 static int decrementer_set_next_event(unsigned long evt,
887 				      struct clock_event_device *dev)
888 {
889 	__this_cpu_write(decrementers_next_tb, get_tb() + evt);
890 	set_dec_or_work(evt);
891 
892 	return 0;
893 }
894 
895 static int decrementer_shutdown(struct clock_event_device *dev)
896 {
897 	__this_cpu_write(decrementers_next_tb, DEC_CLOCKEVENT_STOPPED);
898 	set_dec_or_work(decrementer_max);
899 
900 	return 0;
901 }
902 
903 static void register_decrementer_clockevent(int cpu)
904 {
905 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
906 
907 	*dec = decrementer_clockevent;
908 	dec->cpumask = cpumask_of(cpu);
909 
910 	clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
911 
912 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
913 		    dec->name, dec->mult, dec->shift, cpu);
914 
915 	/* Set values for KVM, see kvm_emulate_dec() */
916 	decrementer_clockevent.mult = dec->mult;
917 	decrementer_clockevent.shift = dec->shift;
918 }
919 
920 static void enable_large_decrementer(void)
921 {
922 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
923 		return;
924 
925 	if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
926 		return;
927 
928 	/*
929 	 * If we're running as the hypervisor we need to enable the LD manually
930 	 * otherwise firmware should have done it for us.
931 	 */
932 	if (cpu_has_feature(CPU_FTR_HVMODE))
933 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
934 }
935 
936 static void __init set_decrementer_max(void)
937 {
938 	struct device_node *cpu;
939 	u32 bits = 32;
940 
941 	/* Prior to ISAv3 the decrementer is always 32 bit */
942 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
943 		return;
944 
945 	cpu = of_find_node_by_type(NULL, "cpu");
946 
947 	if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
948 		if (bits > 64 || bits < 32) {
949 			pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
950 			bits = 32;
951 		}
952 
953 		/* calculate the signed maximum given this many bits */
954 		decrementer_max = (1ul << (bits - 1)) - 1;
955 	}
956 
957 	of_node_put(cpu);
958 
959 	pr_info("time_init: %u bit decrementer (max: %llx)\n",
960 		bits, decrementer_max);
961 }
962 
963 static void __init init_decrementer_clockevent(void)
964 {
965 	register_decrementer_clockevent(smp_processor_id());
966 }
967 
968 void secondary_cpu_time_init(void)
969 {
970 	/* Enable and test the large decrementer for this cpu */
971 	enable_large_decrementer();
972 
973 	/* Start the decrementer on CPUs that have manual control
974 	 * such as BookE
975 	 */
976 	start_cpu_decrementer();
977 
978 	/* FIME: Should make unrelatred change to move snapshot_timebase
979 	 * call here ! */
980 	register_decrementer_clockevent(smp_processor_id());
981 }
982 
983 /* This function is only called on the boot processor */
984 void __init time_init(void)
985 {
986 	struct div_result res;
987 	u64 scale;
988 	unsigned shift;
989 
990 	/* Normal PowerPC with timebase register */
991 	ppc_md.calibrate_decr();
992 	printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
993 	       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
994 	printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
995 	       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
996 
997 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
998 	tb_ticks_per_sec = ppc_tb_freq;
999 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
1000 	calc_cputime_factors();
1001 
1002 	/*
1003 	 * Compute scale factor for sched_clock.
1004 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1005 	 * which is the timebase frequency.
1006 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1007 	 * the 128-bit result as a 64.64 fixed-point number.
1008 	 * We then shift that number right until it is less than 1.0,
1009 	 * giving us the scale factor and shift count to use in
1010 	 * sched_clock().
1011 	 */
1012 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1013 	scale = res.result_low;
1014 	for (shift = 0; res.result_high != 0; ++shift) {
1015 		scale = (scale >> 1) | (res.result_high << 63);
1016 		res.result_high >>= 1;
1017 	}
1018 	tb_to_ns_scale = scale;
1019 	tb_to_ns_shift = shift;
1020 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1021 	boot_tb = get_tb();
1022 
1023 	/* If platform provided a timezone (pmac), we correct the time */
1024 	if (timezone_offset) {
1025 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1026 		sys_tz.tz_dsttime = 0;
1027 	}
1028 
1029 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1030 
1031 	/* initialise and enable the large decrementer (if we have one) */
1032 	set_decrementer_max();
1033 	enable_large_decrementer();
1034 
1035 	/* Start the decrementer on CPUs that have manual control
1036 	 * such as BookE
1037 	 */
1038 	start_cpu_decrementer();
1039 
1040 	/* Register the clocksource */
1041 	clocksource_init();
1042 
1043 	init_decrementer_clockevent();
1044 	tick_setup_hrtimer_broadcast();
1045 
1046 	of_clk_init(NULL);
1047 	enable_sched_clock_irqtime();
1048 }
1049 
1050 /*
1051  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1052  * result.
1053  */
1054 void div128_by_32(u64 dividend_high, u64 dividend_low,
1055 		  unsigned divisor, struct div_result *dr)
1056 {
1057 	unsigned long a, b, c, d;
1058 	unsigned long w, x, y, z;
1059 	u64 ra, rb, rc;
1060 
1061 	a = dividend_high >> 32;
1062 	b = dividend_high & 0xffffffff;
1063 	c = dividend_low >> 32;
1064 	d = dividend_low & 0xffffffff;
1065 
1066 	w = a / divisor;
1067 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1068 
1069 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1070 	x = ra;
1071 
1072 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1073 	y = rb;
1074 
1075 	do_div(rc, divisor);
1076 	z = rc;
1077 
1078 	dr->result_high = ((u64)w << 32) + x;
1079 	dr->result_low  = ((u64)y << 32) + z;
1080 
1081 }
1082 
1083 /* We don't need to calibrate delay, we use the CPU timebase for that */
1084 void calibrate_delay(void)
1085 {
1086 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1087 	 * as the number of __delay(1) in a jiffy, so make it so
1088 	 */
1089 	loops_per_jiffy = tb_ticks_per_jiffy;
1090 }
1091 
1092 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1093 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1094 {
1095 	ppc_md.get_rtc_time(tm);
1096 	return 0;
1097 }
1098 
1099 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1100 {
1101 	if (!ppc_md.set_rtc_time)
1102 		return -EOPNOTSUPP;
1103 
1104 	if (ppc_md.set_rtc_time(tm) < 0)
1105 		return -EOPNOTSUPP;
1106 
1107 	return 0;
1108 }
1109 
1110 static const struct rtc_class_ops rtc_generic_ops = {
1111 	.read_time = rtc_generic_get_time,
1112 	.set_time = rtc_generic_set_time,
1113 };
1114 
1115 static int __init rtc_init(void)
1116 {
1117 	struct platform_device *pdev;
1118 
1119 	if (!ppc_md.get_rtc_time)
1120 		return -ENODEV;
1121 
1122 	pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1123 					     &rtc_generic_ops,
1124 					     sizeof(rtc_generic_ops));
1125 
1126 	return PTR_ERR_OR_ZERO(pdev);
1127 }
1128 
1129 device_initcall(rtc_init);
1130 #endif
1131