xref: /openbmc/linux/arch/powerpc/kernel/time.c (revision 0faf20a1)
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/asm-prototypes.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 DEFINE_PER_CPU(u64, decrementers_next_tb);
111 EXPORT_SYMBOL_GPL(decrementers_next_tb);
112 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
113 
114 #define XSEC_PER_SEC (1024*1024)
115 
116 #ifdef CONFIG_PPC64
117 #define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
118 #else
119 /* compute ((xsec << 12) * max) >> 32 */
120 #define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
121 #endif
122 
123 unsigned long tb_ticks_per_jiffy;
124 unsigned long tb_ticks_per_usec = 100; /* sane default */
125 EXPORT_SYMBOL(tb_ticks_per_usec);
126 unsigned long tb_ticks_per_sec;
127 EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime_t conversions */
128 
129 DEFINE_SPINLOCK(rtc_lock);
130 EXPORT_SYMBOL_GPL(rtc_lock);
131 
132 static u64 tb_to_ns_scale __read_mostly;
133 static unsigned tb_to_ns_shift __read_mostly;
134 static u64 boot_tb __read_mostly;
135 
136 extern struct timezone sys_tz;
137 static long timezone_offset;
138 
139 unsigned long ppc_proc_freq;
140 EXPORT_SYMBOL_GPL(ppc_proc_freq);
141 unsigned long ppc_tb_freq;
142 EXPORT_SYMBOL_GPL(ppc_tb_freq);
143 
144 bool tb_invalid;
145 
146 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
147 /*
148  * Factor for converting from cputime_t (timebase ticks) to
149  * microseconds. This is stored as 0.64 fixed-point binary fraction.
150  */
151 u64 __cputime_usec_factor;
152 EXPORT_SYMBOL(__cputime_usec_factor);
153 
154 #ifdef CONFIG_PPC_SPLPAR
155 void (*dtl_consumer)(struct dtl_entry *, u64);
156 #endif
157 
158 static void calc_cputime_factors(void)
159 {
160 	struct div_result res;
161 
162 	div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
163 	__cputime_usec_factor = res.result_low;
164 }
165 
166 /*
167  * Read the SPURR on systems that have it, otherwise the PURR,
168  * or if that doesn't exist return the timebase value passed in.
169  */
170 static inline unsigned long read_spurr(unsigned long tb)
171 {
172 	if (cpu_has_feature(CPU_FTR_SPURR))
173 		return mfspr(SPRN_SPURR);
174 	if (cpu_has_feature(CPU_FTR_PURR))
175 		return mfspr(SPRN_PURR);
176 	return tb;
177 }
178 
179 #ifdef CONFIG_PPC_SPLPAR
180 
181 #include <asm/dtl.h>
182 
183 /*
184  * Scan the dispatch trace log and count up the stolen time.
185  * Should be called with interrupts disabled.
186  */
187 static u64 scan_dispatch_log(u64 stop_tb)
188 {
189 	u64 i = local_paca->dtl_ridx;
190 	struct dtl_entry *dtl = local_paca->dtl_curr;
191 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
192 	struct lppaca *vpa = local_paca->lppaca_ptr;
193 	u64 tb_delta;
194 	u64 stolen = 0;
195 	u64 dtb;
196 
197 	if (!dtl)
198 		return 0;
199 
200 	if (i == be64_to_cpu(vpa->dtl_idx))
201 		return 0;
202 	while (i < be64_to_cpu(vpa->dtl_idx)) {
203 		dtb = be64_to_cpu(dtl->timebase);
204 		tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
205 			be32_to_cpu(dtl->ready_to_enqueue_time);
206 		barrier();
207 		if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
208 			/* buffer has overflowed */
209 			i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
210 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
211 			continue;
212 		}
213 		if (dtb > stop_tb)
214 			break;
215 		if (dtl_consumer)
216 			dtl_consumer(dtl, i);
217 		stolen += tb_delta;
218 		++i;
219 		++dtl;
220 		if (dtl == dtl_end)
221 			dtl = local_paca->dispatch_log;
222 	}
223 	local_paca->dtl_ridx = i;
224 	local_paca->dtl_curr = dtl;
225 	return stolen;
226 }
227 
228 /*
229  * Accumulate stolen time by scanning the dispatch trace log.
230  * Called on entry from user mode.
231  */
232 void notrace accumulate_stolen_time(void)
233 {
234 	u64 sst, ust;
235 	struct cpu_accounting_data *acct = &local_paca->accounting;
236 
237 	sst = scan_dispatch_log(acct->starttime_user);
238 	ust = scan_dispatch_log(acct->starttime);
239 	acct->stime -= sst;
240 	acct->utime -= ust;
241 	acct->steal_time += ust + sst;
242 }
243 
244 static inline u64 calculate_stolen_time(u64 stop_tb)
245 {
246 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
247 		return 0;
248 
249 	if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
250 		return scan_dispatch_log(stop_tb);
251 
252 	return 0;
253 }
254 
255 #else /* CONFIG_PPC_SPLPAR */
256 static inline u64 calculate_stolen_time(u64 stop_tb)
257 {
258 	return 0;
259 }
260 
261 #endif /* CONFIG_PPC_SPLPAR */
262 
263 /*
264  * Account time for a transition between system, hard irq
265  * or soft irq state.
266  */
267 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
268 					unsigned long now, unsigned long stime)
269 {
270 	unsigned long stime_scaled = 0;
271 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
272 	unsigned long nowscaled, deltascaled;
273 	unsigned long utime, utime_scaled;
274 
275 	nowscaled = read_spurr(now);
276 	deltascaled = nowscaled - acct->startspurr;
277 	acct->startspurr = nowscaled;
278 	utime = acct->utime - acct->utime_sspurr;
279 	acct->utime_sspurr = acct->utime;
280 
281 	/*
282 	 * Because we don't read the SPURR on every kernel entry/exit,
283 	 * deltascaled includes both user and system SPURR ticks.
284 	 * Apportion these ticks to system SPURR ticks and user
285 	 * SPURR ticks in the same ratio as the system time (delta)
286 	 * and user time (udelta) values obtained from the timebase
287 	 * over the same interval.  The system ticks get accounted here;
288 	 * the user ticks get saved up in paca->user_time_scaled to be
289 	 * used by account_process_tick.
290 	 */
291 	stime_scaled = stime;
292 	utime_scaled = utime;
293 	if (deltascaled != stime + utime) {
294 		if (utime) {
295 			stime_scaled = deltascaled * stime / (stime + utime);
296 			utime_scaled = deltascaled - stime_scaled;
297 		} else {
298 			stime_scaled = deltascaled;
299 		}
300 	}
301 	acct->utime_scaled += utime_scaled;
302 #endif
303 
304 	return stime_scaled;
305 }
306 
307 static unsigned long vtime_delta(struct cpu_accounting_data *acct,
308 				 unsigned long *stime_scaled,
309 				 unsigned long *steal_time)
310 {
311 	unsigned long now, stime;
312 
313 	WARN_ON_ONCE(!irqs_disabled());
314 
315 	now = mftb();
316 	stime = now - acct->starttime;
317 	acct->starttime = now;
318 
319 	*stime_scaled = vtime_delta_scaled(acct, now, stime);
320 
321 	*steal_time = calculate_stolen_time(now);
322 
323 	return stime;
324 }
325 
326 static void vtime_delta_kernel(struct cpu_accounting_data *acct,
327 			       unsigned long *stime, unsigned long *stime_scaled)
328 {
329 	unsigned long steal_time;
330 
331 	*stime = vtime_delta(acct, stime_scaled, &steal_time);
332 	*stime -= min(*stime, steal_time);
333 	acct->steal_time += steal_time;
334 }
335 
336 void vtime_account_kernel(struct task_struct *tsk)
337 {
338 	struct cpu_accounting_data *acct = get_accounting(tsk);
339 	unsigned long stime, stime_scaled;
340 
341 	vtime_delta_kernel(acct, &stime, &stime_scaled);
342 
343 	if (tsk->flags & PF_VCPU) {
344 		acct->gtime += stime;
345 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
346 		acct->utime_scaled += stime_scaled;
347 #endif
348 	} else {
349 		acct->stime += stime;
350 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
351 		acct->stime_scaled += stime_scaled;
352 #endif
353 	}
354 }
355 EXPORT_SYMBOL_GPL(vtime_account_kernel);
356 
357 void vtime_account_idle(struct task_struct *tsk)
358 {
359 	unsigned long stime, stime_scaled, steal_time;
360 	struct cpu_accounting_data *acct = get_accounting(tsk);
361 
362 	stime = vtime_delta(acct, &stime_scaled, &steal_time);
363 	acct->idle_time += stime + steal_time;
364 }
365 
366 static void vtime_account_irq_field(struct cpu_accounting_data *acct,
367 				    unsigned long *field)
368 {
369 	unsigned long stime, stime_scaled;
370 
371 	vtime_delta_kernel(acct, &stime, &stime_scaled);
372 	*field += stime;
373 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
374 	acct->stime_scaled += stime_scaled;
375 #endif
376 }
377 
378 void vtime_account_softirq(struct task_struct *tsk)
379 {
380 	struct cpu_accounting_data *acct = get_accounting(tsk);
381 	vtime_account_irq_field(acct, &acct->softirq_time);
382 }
383 
384 void vtime_account_hardirq(struct task_struct *tsk)
385 {
386 	struct cpu_accounting_data *acct = get_accounting(tsk);
387 	vtime_account_irq_field(acct, &acct->hardirq_time);
388 }
389 
390 static void vtime_flush_scaled(struct task_struct *tsk,
391 			       struct cpu_accounting_data *acct)
392 {
393 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
394 	if (acct->utime_scaled)
395 		tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
396 	if (acct->stime_scaled)
397 		tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
398 
399 	acct->utime_scaled = 0;
400 	acct->utime_sspurr = 0;
401 	acct->stime_scaled = 0;
402 #endif
403 }
404 
405 /*
406  * Account the whole cputime accumulated in the paca
407  * Must be called with interrupts disabled.
408  * Assumes that vtime_account_kernel/idle() has been called
409  * recently (i.e. since the last entry from usermode) so that
410  * get_paca()->user_time_scaled is up to date.
411  */
412 void vtime_flush(struct task_struct *tsk)
413 {
414 	struct cpu_accounting_data *acct = get_accounting(tsk);
415 
416 	if (acct->utime)
417 		account_user_time(tsk, cputime_to_nsecs(acct->utime));
418 
419 	if (acct->gtime)
420 		account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
421 
422 	if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
423 		account_steal_time(cputime_to_nsecs(acct->steal_time));
424 		acct->steal_time = 0;
425 	}
426 
427 	if (acct->idle_time)
428 		account_idle_time(cputime_to_nsecs(acct->idle_time));
429 
430 	if (acct->stime)
431 		account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
432 					  CPUTIME_SYSTEM);
433 
434 	if (acct->hardirq_time)
435 		account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
436 					  CPUTIME_IRQ);
437 	if (acct->softirq_time)
438 		account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
439 					  CPUTIME_SOFTIRQ);
440 
441 	vtime_flush_scaled(tsk, acct);
442 
443 	acct->utime = 0;
444 	acct->gtime = 0;
445 	acct->idle_time = 0;
446 	acct->stime = 0;
447 	acct->hardirq_time = 0;
448 	acct->softirq_time = 0;
449 }
450 
451 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
452 #define calc_cputime_factors()
453 #endif
454 
455 void __delay(unsigned long loops)
456 {
457 	unsigned long start;
458 
459 	spin_begin();
460 	if (tb_invalid) {
461 		/*
462 		 * TB is in error state and isn't ticking anymore.
463 		 * HMI handler was unable to recover from TB error.
464 		 * Return immediately, so that kernel won't get stuck here.
465 		 */
466 		spin_cpu_relax();
467 	} else {
468 		start = mftb();
469 		while (mftb() - start < loops)
470 			spin_cpu_relax();
471 	}
472 	spin_end();
473 }
474 EXPORT_SYMBOL(__delay);
475 
476 void udelay(unsigned long usecs)
477 {
478 	__delay(tb_ticks_per_usec * usecs);
479 }
480 EXPORT_SYMBOL(udelay);
481 
482 #ifdef CONFIG_SMP
483 unsigned long profile_pc(struct pt_regs *regs)
484 {
485 	unsigned long pc = instruction_pointer(regs);
486 
487 	if (in_lock_functions(pc))
488 		return regs->link;
489 
490 	return pc;
491 }
492 EXPORT_SYMBOL(profile_pc);
493 #endif
494 
495 #ifdef CONFIG_IRQ_WORK
496 
497 /*
498  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
499  */
500 #ifdef CONFIG_PPC64
501 static inline unsigned long test_irq_work_pending(void)
502 {
503 	unsigned long x;
504 
505 	asm volatile("lbz %0,%1(13)"
506 		: "=r" (x)
507 		: "i" (offsetof(struct paca_struct, irq_work_pending)));
508 	return x;
509 }
510 
511 static inline void set_irq_work_pending_flag(void)
512 {
513 	asm volatile("stb %0,%1(13)" : :
514 		"r" (1),
515 		"i" (offsetof(struct paca_struct, irq_work_pending)));
516 }
517 
518 static inline void clear_irq_work_pending(void)
519 {
520 	asm volatile("stb %0,%1(13)" : :
521 		"r" (0),
522 		"i" (offsetof(struct paca_struct, irq_work_pending)));
523 }
524 
525 #else /* 32-bit */
526 
527 DEFINE_PER_CPU(u8, irq_work_pending);
528 
529 #define set_irq_work_pending_flag()	__this_cpu_write(irq_work_pending, 1)
530 #define test_irq_work_pending()		__this_cpu_read(irq_work_pending)
531 #define clear_irq_work_pending()	__this_cpu_write(irq_work_pending, 0)
532 
533 #endif /* 32 vs 64 bit */
534 
535 void arch_irq_work_raise(void)
536 {
537 	/*
538 	 * 64-bit code that uses irq soft-mask can just cause an immediate
539 	 * interrupt here that gets soft masked, if this is called under
540 	 * local_irq_disable(). It might be possible to prevent that happening
541 	 * by noticing interrupts are disabled and setting decrementer pending
542 	 * to be replayed when irqs are enabled. The problem there is that
543 	 * tracing can call irq_work_raise, including in code that does low
544 	 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
545 	 * which could get tangled up if we're messing with the same state
546 	 * here.
547 	 */
548 	preempt_disable();
549 	set_irq_work_pending_flag();
550 	set_dec(1);
551 	preempt_enable();
552 }
553 
554 static void set_dec_or_work(u64 val)
555 {
556 	set_dec(val);
557 	/* We may have raced with new irq work */
558 	if (unlikely(test_irq_work_pending()))
559 		set_dec(1);
560 }
561 
562 #else  /* CONFIG_IRQ_WORK */
563 
564 #define test_irq_work_pending()	0
565 #define clear_irq_work_pending()
566 
567 static void set_dec_or_work(u64 val)
568 {
569 	set_dec(val);
570 }
571 #endif /* CONFIG_IRQ_WORK */
572 
573 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
574 void timer_rearm_host_dec(u64 now)
575 {
576 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
577 
578 	WARN_ON_ONCE(!arch_irqs_disabled());
579 	WARN_ON_ONCE(mfmsr() & MSR_EE);
580 
581 	if (now >= *next_tb) {
582 		local_paca->irq_happened |= PACA_IRQ_DEC;
583 	} else {
584 		now = *next_tb - now;
585 		if (now <= decrementer_max)
586 			set_dec_or_work(now);
587 	}
588 }
589 EXPORT_SYMBOL_GPL(timer_rearm_host_dec);
590 #endif
591 
592 /*
593  * timer_interrupt - gets called when the decrementer overflows,
594  * with interrupts disabled.
595  */
596 DEFINE_INTERRUPT_HANDLER_ASYNC(timer_interrupt)
597 {
598 	struct clock_event_device *evt = this_cpu_ptr(&decrementers);
599 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
600 	struct pt_regs *old_regs;
601 	u64 now;
602 
603 	/*
604 	 * Some implementations of hotplug will get timer interrupts while
605 	 * offline, just ignore these.
606 	 */
607 	if (unlikely(!cpu_online(smp_processor_id()))) {
608 		set_dec(decrementer_max);
609 		return;
610 	}
611 
612 	/* Conditionally hard-enable interrupts. */
613 	if (should_hard_irq_enable()) {
614 		/*
615 		 * Ensure a positive value is written to the decrementer, or
616 		 * else some CPUs will continue to take decrementer exceptions.
617 		 * When the PPC_WATCHDOG (decrementer based) is configured,
618 		 * keep this at most 31 bits, which is about 4 seconds on most
619 		 * systems, which gives the watchdog a chance of catching timer
620 		 * interrupt hard lockups.
621 		 */
622 		if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
623 			set_dec(0x7fffffff);
624 		else
625 			set_dec(decrementer_max);
626 
627 		do_hard_irq_enable();
628 	}
629 
630 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
631 	if (atomic_read(&ppc_n_lost_interrupts) != 0)
632 		__do_IRQ(regs);
633 #endif
634 
635 	old_regs = set_irq_regs(regs);
636 
637 	trace_timer_interrupt_entry(regs);
638 
639 	if (test_irq_work_pending()) {
640 		clear_irq_work_pending();
641 		irq_work_run();
642 	}
643 
644 	now = get_tb();
645 	if (now >= *next_tb) {
646 		*next_tb = ~(u64)0;
647 		if (evt->event_handler)
648 			evt->event_handler(evt);
649 		__this_cpu_inc(irq_stat.timer_irqs_event);
650 	} else {
651 		now = *next_tb - now;
652 		if (now <= decrementer_max)
653 			set_dec_or_work(now);
654 		__this_cpu_inc(irq_stat.timer_irqs_others);
655 	}
656 
657 	trace_timer_interrupt_exit(regs);
658 
659 	set_irq_regs(old_regs);
660 }
661 EXPORT_SYMBOL(timer_interrupt);
662 
663 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
664 void timer_broadcast_interrupt(void)
665 {
666 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
667 
668 	*next_tb = ~(u64)0;
669 	tick_receive_broadcast();
670 	__this_cpu_inc(irq_stat.broadcast_irqs_event);
671 }
672 #endif
673 
674 #ifdef CONFIG_SUSPEND
675 /* Overrides the weak version in kernel/power/main.c */
676 void arch_suspend_disable_irqs(void)
677 {
678 	if (ppc_md.suspend_disable_irqs)
679 		ppc_md.suspend_disable_irqs();
680 
681 	/* Disable the decrementer, so that it doesn't interfere
682 	 * with suspending.
683 	 */
684 
685 	set_dec(decrementer_max);
686 	local_irq_disable();
687 	set_dec(decrementer_max);
688 }
689 
690 /* Overrides the weak version in kernel/power/main.c */
691 void arch_suspend_enable_irqs(void)
692 {
693 	local_irq_enable();
694 
695 	if (ppc_md.suspend_enable_irqs)
696 		ppc_md.suspend_enable_irqs();
697 }
698 #endif
699 
700 unsigned long long tb_to_ns(unsigned long long ticks)
701 {
702 	return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
703 }
704 EXPORT_SYMBOL_GPL(tb_to_ns);
705 
706 /*
707  * Scheduler clock - returns current time in nanosec units.
708  *
709  * Note: mulhdu(a, b) (multiply high double unsigned) returns
710  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
711  * are 64-bit unsigned numbers.
712  */
713 notrace unsigned long long sched_clock(void)
714 {
715 	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
716 }
717 
718 
719 #ifdef CONFIG_PPC_PSERIES
720 
721 /*
722  * Running clock - attempts to give a view of time passing for a virtualised
723  * kernels.
724  * Uses the VTB register if available otherwise a next best guess.
725  */
726 unsigned long long running_clock(void)
727 {
728 	/*
729 	 * Don't read the VTB as a host since KVM does not switch in host
730 	 * timebase into the VTB when it takes a guest off the CPU, reading the
731 	 * VTB would result in reading 'last switched out' guest VTB.
732 	 *
733 	 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
734 	 * would be unsafe to rely only on the #ifdef above.
735 	 */
736 	if (firmware_has_feature(FW_FEATURE_LPAR) &&
737 	    cpu_has_feature(CPU_FTR_ARCH_207S))
738 		return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
739 
740 	/*
741 	 * This is a next best approximation without a VTB.
742 	 * On a host which is running bare metal there should never be any stolen
743 	 * time and on a host which doesn't do any virtualisation TB *should* equal
744 	 * VTB so it makes no difference anyway.
745 	 */
746 	return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
747 }
748 #endif
749 
750 static int __init get_freq(char *name, int cells, unsigned long *val)
751 {
752 	struct device_node *cpu;
753 	const __be32 *fp;
754 	int found = 0;
755 
756 	/* The cpu node should have timebase and clock frequency properties */
757 	cpu = of_find_node_by_type(NULL, "cpu");
758 
759 	if (cpu) {
760 		fp = of_get_property(cpu, name, NULL);
761 		if (fp) {
762 			found = 1;
763 			*val = of_read_ulong(fp, cells);
764 		}
765 
766 		of_node_put(cpu);
767 	}
768 
769 	return found;
770 }
771 
772 static void start_cpu_decrementer(void)
773 {
774 #ifdef CONFIG_BOOKE_OR_40x
775 	unsigned int tcr;
776 
777 	/* Clear any pending timer interrupts */
778 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
779 
780 	tcr = mfspr(SPRN_TCR);
781 	/*
782 	 * The watchdog may have already been enabled by u-boot. So leave
783 	 * TRC[WP] (Watchdog Period) alone.
784 	 */
785 	tcr &= TCR_WP_MASK;	/* Clear all bits except for TCR[WP] */
786 	tcr |= TCR_DIE;		/* Enable decrementer */
787 	mtspr(SPRN_TCR, tcr);
788 #endif
789 }
790 
791 void __init generic_calibrate_decr(void)
792 {
793 	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
794 
795 	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
796 	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
797 
798 		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
799 				"(not found)\n");
800 	}
801 
802 	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
803 
804 	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
805 	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
806 
807 		printk(KERN_ERR "WARNING: Estimating processor frequency "
808 				"(not found)\n");
809 	}
810 }
811 
812 int update_persistent_clock64(struct timespec64 now)
813 {
814 	struct rtc_time tm;
815 
816 	if (!ppc_md.set_rtc_time)
817 		return -ENODEV;
818 
819 	rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
820 
821 	return ppc_md.set_rtc_time(&tm);
822 }
823 
824 static void __read_persistent_clock(struct timespec64 *ts)
825 {
826 	struct rtc_time tm;
827 	static int first = 1;
828 
829 	ts->tv_nsec = 0;
830 	/* XXX this is a litle fragile but will work okay in the short term */
831 	if (first) {
832 		first = 0;
833 		if (ppc_md.time_init)
834 			timezone_offset = ppc_md.time_init();
835 
836 		/* get_boot_time() isn't guaranteed to be safe to call late */
837 		if (ppc_md.get_boot_time) {
838 			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
839 			return;
840 		}
841 	}
842 	if (!ppc_md.get_rtc_time) {
843 		ts->tv_sec = 0;
844 		return;
845 	}
846 	ppc_md.get_rtc_time(&tm);
847 
848 	ts->tv_sec = rtc_tm_to_time64(&tm);
849 }
850 
851 void read_persistent_clock64(struct timespec64 *ts)
852 {
853 	__read_persistent_clock(ts);
854 
855 	/* Sanitize it in case real time clock is set below EPOCH */
856 	if (ts->tv_sec < 0) {
857 		ts->tv_sec = 0;
858 		ts->tv_nsec = 0;
859 	}
860 
861 }
862 
863 /* clocksource code */
864 static notrace u64 timebase_read(struct clocksource *cs)
865 {
866 	return (u64)get_tb();
867 }
868 
869 static void __init clocksource_init(void)
870 {
871 	struct clocksource *clock = &clocksource_timebase;
872 
873 	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
874 		printk(KERN_ERR "clocksource: %s is already registered\n",
875 		       clock->name);
876 		return;
877 	}
878 
879 	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
880 	       clock->name, clock->mult, clock->shift);
881 }
882 
883 static int decrementer_set_next_event(unsigned long evt,
884 				      struct clock_event_device *dev)
885 {
886 	__this_cpu_write(decrementers_next_tb, get_tb() + evt);
887 	set_dec_or_work(evt);
888 
889 	return 0;
890 }
891 
892 static int decrementer_shutdown(struct clock_event_device *dev)
893 {
894 	decrementer_set_next_event(decrementer_max, dev);
895 	return 0;
896 }
897 
898 static void register_decrementer_clockevent(int cpu)
899 {
900 	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
901 
902 	*dec = decrementer_clockevent;
903 	dec->cpumask = cpumask_of(cpu);
904 
905 	clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
906 
907 	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
908 		    dec->name, dec->mult, dec->shift, cpu);
909 
910 	/* Set values for KVM, see kvm_emulate_dec() */
911 	decrementer_clockevent.mult = dec->mult;
912 	decrementer_clockevent.shift = dec->shift;
913 }
914 
915 static void enable_large_decrementer(void)
916 {
917 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
918 		return;
919 
920 	if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
921 		return;
922 
923 	/*
924 	 * If we're running as the hypervisor we need to enable the LD manually
925 	 * otherwise firmware should have done it for us.
926 	 */
927 	if (cpu_has_feature(CPU_FTR_HVMODE))
928 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
929 }
930 
931 static void __init set_decrementer_max(void)
932 {
933 	struct device_node *cpu;
934 	u32 bits = 32;
935 
936 	/* Prior to ISAv3 the decrementer is always 32 bit */
937 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
938 		return;
939 
940 	cpu = of_find_node_by_type(NULL, "cpu");
941 
942 	if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
943 		if (bits > 64 || bits < 32) {
944 			pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
945 			bits = 32;
946 		}
947 
948 		/* calculate the signed maximum given this many bits */
949 		decrementer_max = (1ul << (bits - 1)) - 1;
950 	}
951 
952 	of_node_put(cpu);
953 
954 	pr_info("time_init: %u bit decrementer (max: %llx)\n",
955 		bits, decrementer_max);
956 }
957 
958 static void __init init_decrementer_clockevent(void)
959 {
960 	register_decrementer_clockevent(smp_processor_id());
961 }
962 
963 void secondary_cpu_time_init(void)
964 {
965 	/* Enable and test the large decrementer for this cpu */
966 	enable_large_decrementer();
967 
968 	/* Start the decrementer on CPUs that have manual control
969 	 * such as BookE
970 	 */
971 	start_cpu_decrementer();
972 
973 	/* FIME: Should make unrelatred change to move snapshot_timebase
974 	 * call here ! */
975 	register_decrementer_clockevent(smp_processor_id());
976 }
977 
978 /* This function is only called on the boot processor */
979 void __init time_init(void)
980 {
981 	struct div_result res;
982 	u64 scale;
983 	unsigned shift;
984 
985 	/* Normal PowerPC with timebase register */
986 	ppc_md.calibrate_decr();
987 	printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
988 	       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
989 	printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
990 	       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
991 
992 	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
993 	tb_ticks_per_sec = ppc_tb_freq;
994 	tb_ticks_per_usec = ppc_tb_freq / 1000000;
995 	calc_cputime_factors();
996 
997 	/*
998 	 * Compute scale factor for sched_clock.
999 	 * The calibrate_decr() function has set tb_ticks_per_sec,
1000 	 * which is the timebase frequency.
1001 	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1002 	 * the 128-bit result as a 64.64 fixed-point number.
1003 	 * We then shift that number right until it is less than 1.0,
1004 	 * giving us the scale factor and shift count to use in
1005 	 * sched_clock().
1006 	 */
1007 	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1008 	scale = res.result_low;
1009 	for (shift = 0; res.result_high != 0; ++shift) {
1010 		scale = (scale >> 1) | (res.result_high << 63);
1011 		res.result_high >>= 1;
1012 	}
1013 	tb_to_ns_scale = scale;
1014 	tb_to_ns_shift = shift;
1015 	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1016 	boot_tb = get_tb();
1017 
1018 	/* If platform provided a timezone (pmac), we correct the time */
1019 	if (timezone_offset) {
1020 		sys_tz.tz_minuteswest = -timezone_offset / 60;
1021 		sys_tz.tz_dsttime = 0;
1022 	}
1023 
1024 	vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1025 
1026 	/* initialise and enable the large decrementer (if we have one) */
1027 	set_decrementer_max();
1028 	enable_large_decrementer();
1029 
1030 	/* Start the decrementer on CPUs that have manual control
1031 	 * such as BookE
1032 	 */
1033 	start_cpu_decrementer();
1034 
1035 	/* Register the clocksource */
1036 	clocksource_init();
1037 
1038 	init_decrementer_clockevent();
1039 	tick_setup_hrtimer_broadcast();
1040 
1041 	of_clk_init(NULL);
1042 	enable_sched_clock_irqtime();
1043 }
1044 
1045 /*
1046  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1047  * result.
1048  */
1049 void div128_by_32(u64 dividend_high, u64 dividend_low,
1050 		  unsigned divisor, struct div_result *dr)
1051 {
1052 	unsigned long a, b, c, d;
1053 	unsigned long w, x, y, z;
1054 	u64 ra, rb, rc;
1055 
1056 	a = dividend_high >> 32;
1057 	b = dividend_high & 0xffffffff;
1058 	c = dividend_low >> 32;
1059 	d = dividend_low & 0xffffffff;
1060 
1061 	w = a / divisor;
1062 	ra = ((u64)(a - (w * divisor)) << 32) + b;
1063 
1064 	rb = ((u64) do_div(ra, divisor) << 32) + c;
1065 	x = ra;
1066 
1067 	rc = ((u64) do_div(rb, divisor) << 32) + d;
1068 	y = rb;
1069 
1070 	do_div(rc, divisor);
1071 	z = rc;
1072 
1073 	dr->result_high = ((u64)w << 32) + x;
1074 	dr->result_low  = ((u64)y << 32) + z;
1075 
1076 }
1077 
1078 /* We don't need to calibrate delay, we use the CPU timebase for that */
1079 void calibrate_delay(void)
1080 {
1081 	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1082 	 * as the number of __delay(1) in a jiffy, so make it so
1083 	 */
1084 	loops_per_jiffy = tb_ticks_per_jiffy;
1085 }
1086 
1087 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1088 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1089 {
1090 	ppc_md.get_rtc_time(tm);
1091 	return 0;
1092 }
1093 
1094 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1095 {
1096 	if (!ppc_md.set_rtc_time)
1097 		return -EOPNOTSUPP;
1098 
1099 	if (ppc_md.set_rtc_time(tm) < 0)
1100 		return -EOPNOTSUPP;
1101 
1102 	return 0;
1103 }
1104 
1105 static const struct rtc_class_ops rtc_generic_ops = {
1106 	.read_time = rtc_generic_get_time,
1107 	.set_time = rtc_generic_set_time,
1108 };
1109 
1110 static int __init rtc_init(void)
1111 {
1112 	struct platform_device *pdev;
1113 
1114 	if (!ppc_md.get_rtc_time)
1115 		return -ENODEV;
1116 
1117 	pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1118 					     &rtc_generic_ops,
1119 					     sizeof(rtc_generic_ops));
1120 
1121 	return PTR_ERR_OR_ZERO(pdev);
1122 }
1123 
1124 device_initcall(rtc_init);
1125 #endif
1126