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 #ifdef CONFIG_PPC64
167 #define get_accounting(tsk) (&get_paca()->accounting)
168 #else
169 #define get_accounting(tsk) (&task_thread_info(tsk)->accounting)
170 #endif
171
calc_cputime_factors(void)172 static void calc_cputime_factors(void)
173 {
174 struct div_result res;
175
176 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
177 __cputime_usec_factor = res.result_low;
178 }
179
180 /*
181 * Read the SPURR on systems that have it, otherwise the PURR,
182 * or if that doesn't exist return the timebase value passed in.
183 */
read_spurr(unsigned long tb)184 static unsigned long read_spurr(unsigned long tb)
185 {
186 if (cpu_has_feature(CPU_FTR_SPURR))
187 return mfspr(SPRN_SPURR);
188 if (cpu_has_feature(CPU_FTR_PURR))
189 return mfspr(SPRN_PURR);
190 return tb;
191 }
192
193 #ifdef CONFIG_PPC_SPLPAR
194
195 /*
196 * Scan the dispatch trace log and count up the stolen time.
197 * Should be called with interrupts disabled.
198 */
scan_dispatch_log(u64 stop_tb)199 static u64 scan_dispatch_log(u64 stop_tb)
200 {
201 u64 i = local_paca->dtl_ridx;
202 struct dtl_entry *dtl = local_paca->dtl_curr;
203 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
204 struct lppaca *vpa = local_paca->lppaca_ptr;
205 u64 tb_delta;
206 u64 stolen = 0;
207 u64 dtb;
208
209 if (!dtl)
210 return 0;
211
212 if (i == be64_to_cpu(vpa->dtl_idx))
213 return 0;
214 while (i < be64_to_cpu(vpa->dtl_idx)) {
215 dtb = be64_to_cpu(dtl->timebase);
216 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
217 be32_to_cpu(dtl->ready_to_enqueue_time);
218 barrier();
219 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
220 /* buffer has overflowed */
221 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
222 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
223 continue;
224 }
225 if (dtb > stop_tb)
226 break;
227 if (dtl_consumer)
228 dtl_consumer(dtl, i);
229 stolen += tb_delta;
230 ++i;
231 ++dtl;
232 if (dtl == dtl_end)
233 dtl = local_paca->dispatch_log;
234 }
235 local_paca->dtl_ridx = i;
236 local_paca->dtl_curr = dtl;
237 return stolen;
238 }
239
240 /*
241 * Accumulate stolen time by scanning the dispatch trace log.
242 * Called on entry from user mode.
243 */
accumulate_stolen_time(void)244 void notrace accumulate_stolen_time(void)
245 {
246 u64 sst, ust;
247 u8 save_soft_enabled = local_paca->soft_enabled;
248 struct cpu_accounting_data *acct = &local_paca->accounting;
249
250 /* We are called early in the exception entry, before
251 * soft/hard_enabled are sync'ed to the expected state
252 * for the exception. We are hard disabled but the PACA
253 * needs to reflect that so various debug stuff doesn't
254 * complain
255 */
256 local_paca->soft_enabled = 0;
257
258 sst = scan_dispatch_log(acct->starttime_user);
259 ust = scan_dispatch_log(acct->starttime);
260 acct->stime -= sst;
261 acct->utime -= ust;
262 acct->steal_time += ust + sst;
263
264 local_paca->soft_enabled = save_soft_enabled;
265 }
266
calculate_stolen_time(u64 stop_tb)267 static inline u64 calculate_stolen_time(u64 stop_tb)
268 {
269 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
270 return scan_dispatch_log(stop_tb);
271
272 return 0;
273 }
274
275 #else /* CONFIG_PPC_SPLPAR */
calculate_stolen_time(u64 stop_tb)276 static inline u64 calculate_stolen_time(u64 stop_tb)
277 {
278 return 0;
279 }
280
281 #endif /* CONFIG_PPC_SPLPAR */
282
283 /*
284 * Account time for a transition between system, hard irq
285 * or soft irq state.
286 */
vtime_delta(struct task_struct * tsk,unsigned long * stime_scaled,unsigned long * steal_time)287 static unsigned long vtime_delta(struct task_struct *tsk,
288 unsigned long *stime_scaled,
289 unsigned long *steal_time)
290 {
291 unsigned long now, nowscaled, deltascaled;
292 unsigned long stime;
293 unsigned long utime, utime_scaled;
294 struct cpu_accounting_data *acct = get_accounting(tsk);
295
296 WARN_ON_ONCE(!irqs_disabled());
297
298 now = mftb();
299 nowscaled = read_spurr(now);
300 stime = now - acct->starttime;
301 acct->starttime = now;
302 deltascaled = nowscaled - acct->startspurr;
303 acct->startspurr = nowscaled;
304
305 *steal_time = calculate_stolen_time(now);
306
307 utime = acct->utime - acct->utime_sspurr;
308 acct->utime_sspurr = acct->utime;
309
310 /*
311 * Because we don't read the SPURR on every kernel entry/exit,
312 * deltascaled includes both user and system SPURR ticks.
313 * Apportion these ticks to system SPURR ticks and user
314 * SPURR ticks in the same ratio as the system time (delta)
315 * and user time (udelta) values obtained from the timebase
316 * over the same interval. The system ticks get accounted here;
317 * the user ticks get saved up in paca->user_time_scaled to be
318 * used by account_process_tick.
319 */
320 *stime_scaled = stime;
321 utime_scaled = utime;
322 if (deltascaled != stime + utime) {
323 if (utime) {
324 *stime_scaled = deltascaled * stime / (stime + utime);
325 utime_scaled = deltascaled - *stime_scaled;
326 } else {
327 *stime_scaled = deltascaled;
328 }
329 }
330 acct->utime_scaled += utime_scaled;
331
332 return stime;
333 }
334
vtime_account_system(struct task_struct * tsk)335 void vtime_account_system(struct task_struct *tsk)
336 {
337 unsigned long stime, stime_scaled, steal_time;
338 struct cpu_accounting_data *acct = get_accounting(tsk);
339
340 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
341
342 stime -= min(stime, steal_time);
343 acct->steal_time += steal_time;
344
345 if ((tsk->flags & PF_VCPU) && !irq_count()) {
346 acct->gtime += stime;
347 acct->utime_scaled += stime_scaled;
348 } else {
349 if (hardirq_count())
350 acct->hardirq_time += stime;
351 else if (in_serving_softirq())
352 acct->softirq_time += stime;
353 else
354 acct->stime += stime;
355
356 acct->stime_scaled += stime_scaled;
357 }
358 }
359 EXPORT_SYMBOL_GPL(vtime_account_system);
360
vtime_account_idle(struct task_struct * tsk)361 void vtime_account_idle(struct task_struct *tsk)
362 {
363 unsigned long stime, stime_scaled, steal_time;
364 struct cpu_accounting_data *acct = get_accounting(tsk);
365
366 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
367 acct->idle_time += stime + steal_time;
368 }
369
370 /*
371 * Account the whole cputime accumulated in the paca
372 * Must be called with interrupts disabled.
373 * Assumes that vtime_account_system/idle() has been called
374 * recently (i.e. since the last entry from usermode) so that
375 * get_paca()->user_time_scaled is up to date.
376 */
vtime_flush(struct task_struct * tsk)377 void vtime_flush(struct task_struct *tsk)
378 {
379 struct cpu_accounting_data *acct = get_accounting(tsk);
380
381 if (acct->utime)
382 account_user_time(tsk, cputime_to_nsecs(acct->utime));
383
384 if (acct->utime_scaled)
385 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
386
387 if (acct->gtime)
388 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
389
390 if (acct->steal_time)
391 account_steal_time(cputime_to_nsecs(acct->steal_time));
392
393 if (acct->idle_time)
394 account_idle_time(cputime_to_nsecs(acct->idle_time));
395
396 if (acct->stime)
397 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
398 CPUTIME_SYSTEM);
399 if (acct->stime_scaled)
400 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
401
402 if (acct->hardirq_time)
403 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
404 CPUTIME_IRQ);
405 if (acct->softirq_time)
406 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
407 CPUTIME_SOFTIRQ);
408
409 acct->utime = 0;
410 acct->utime_scaled = 0;
411 acct->utime_sspurr = 0;
412 acct->gtime = 0;
413 acct->steal_time = 0;
414 acct->idle_time = 0;
415 acct->stime = 0;
416 acct->stime_scaled = 0;
417 acct->hardirq_time = 0;
418 acct->softirq_time = 0;
419 }
420
421 #ifdef CONFIG_PPC32
422 /*
423 * Called from the context switch with interrupts disabled, to charge all
424 * accumulated times to the current process, and to prepare accounting on
425 * the next process.
426 */
arch_vtime_task_switch(struct task_struct * prev)427 void arch_vtime_task_switch(struct task_struct *prev)
428 {
429 struct cpu_accounting_data *acct = get_accounting(current);
430
431 acct->starttime = get_accounting(prev)->starttime;
432 acct->startspurr = get_accounting(prev)->startspurr;
433 }
434 #endif /* CONFIG_PPC32 */
435
436 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
437 #define calc_cputime_factors()
438 #endif
439
__delay(unsigned long loops)440 void __delay(unsigned long loops)
441 {
442 unsigned long start;
443 int diff;
444
445 spin_begin();
446 if (__USE_RTC()) {
447 start = get_rtcl();
448 do {
449 /* the RTCL register wraps at 1000000000 */
450 diff = get_rtcl() - start;
451 if (diff < 0)
452 diff += 1000000000;
453 spin_cpu_relax();
454 } while (diff < loops);
455 } else {
456 start = get_tbl();
457 while (get_tbl() - start < loops)
458 spin_cpu_relax();
459 }
460 spin_end();
461 }
462 EXPORT_SYMBOL(__delay);
463
udelay(unsigned long usecs)464 void udelay(unsigned long usecs)
465 {
466 __delay(tb_ticks_per_usec * usecs);
467 }
468 EXPORT_SYMBOL(udelay);
469
470 #ifdef CONFIG_SMP
profile_pc(struct pt_regs * regs)471 unsigned long profile_pc(struct pt_regs *regs)
472 {
473 unsigned long pc = instruction_pointer(regs);
474
475 if (in_lock_functions(pc))
476 return regs->link;
477
478 return pc;
479 }
480 EXPORT_SYMBOL(profile_pc);
481 #endif
482
483 #ifdef CONFIG_IRQ_WORK
484
485 /*
486 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
487 */
488 #ifdef CONFIG_PPC64
test_irq_work_pending(void)489 static inline unsigned long test_irq_work_pending(void)
490 {
491 unsigned long x;
492
493 asm volatile("lbz %0,%1(13)"
494 : "=r" (x)
495 : "i" (offsetof(struct paca_struct, irq_work_pending)));
496 return x;
497 }
498
set_irq_work_pending_flag(void)499 static inline void set_irq_work_pending_flag(void)
500 {
501 asm volatile("stb %0,%1(13)" : :
502 "r" (1),
503 "i" (offsetof(struct paca_struct, irq_work_pending)));
504 }
505
clear_irq_work_pending(void)506 static inline void clear_irq_work_pending(void)
507 {
508 asm volatile("stb %0,%1(13)" : :
509 "r" (0),
510 "i" (offsetof(struct paca_struct, irq_work_pending)));
511 }
512
513 #else /* 32-bit */
514
515 DEFINE_PER_CPU(u8, irq_work_pending);
516
517 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
518 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
519 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
520
521 #endif /* 32 vs 64 bit */
522
arch_irq_work_raise(void)523 void arch_irq_work_raise(void)
524 {
525 preempt_disable();
526 set_irq_work_pending_flag();
527 set_dec(1);
528 preempt_enable();
529 }
530
531 #else /* CONFIG_IRQ_WORK */
532
533 #define test_irq_work_pending() 0
534 #define clear_irq_work_pending()
535
536 #endif /* CONFIG_IRQ_WORK */
537
__timer_interrupt(void)538 static void __timer_interrupt(void)
539 {
540 struct pt_regs *regs = get_irq_regs();
541 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
542 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
543 u64 now;
544
545 trace_timer_interrupt_entry(regs);
546
547 if (test_irq_work_pending()) {
548 clear_irq_work_pending();
549 irq_work_run();
550 }
551
552 now = get_tb_or_rtc();
553 if (now >= *next_tb) {
554 *next_tb = ~(u64)0;
555 if (evt->event_handler)
556 evt->event_handler(evt);
557 __this_cpu_inc(irq_stat.timer_irqs_event);
558 } else {
559 now = *next_tb - now;
560 if (now <= decrementer_max)
561 set_dec(now);
562 /* We may have raced with new irq work */
563 if (test_irq_work_pending())
564 set_dec(1);
565 __this_cpu_inc(irq_stat.timer_irqs_others);
566 }
567
568 #ifdef CONFIG_PPC64
569 /* collect purr register values often, for accurate calculations */
570 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
571 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
572 cu->current_tb = mfspr(SPRN_PURR);
573 }
574 #endif
575
576 trace_timer_interrupt_exit(regs);
577 }
578
579 /*
580 * timer_interrupt - gets called when the decrementer overflows,
581 * with interrupts disabled.
582 */
timer_interrupt(struct pt_regs * regs)583 void timer_interrupt(struct pt_regs * regs)
584 {
585 struct pt_regs *old_regs;
586 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
587
588 /* Ensure a positive value is written to the decrementer, or else
589 * some CPUs will continue to take decrementer exceptions.
590 */
591 set_dec(decrementer_max);
592
593 /* Some implementations of hotplug will get timer interrupts while
594 * offline, just ignore these and we also need to set
595 * decrementers_next_tb as MAX to make sure __check_irq_replay
596 * don't replay timer interrupt when return, otherwise we'll trap
597 * here infinitely :(
598 */
599 if (!cpu_online(smp_processor_id())) {
600 *next_tb = ~(u64)0;
601 return;
602 }
603
604 /* Conditionally hard-enable interrupts now that the DEC has been
605 * bumped to its maximum value
606 */
607 may_hard_irq_enable();
608
609
610 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
611 if (atomic_read(&ppc_n_lost_interrupts) != 0)
612 do_IRQ(regs);
613 #endif
614
615 old_regs = set_irq_regs(regs);
616 irq_enter();
617
618 __timer_interrupt();
619 irq_exit();
620 set_irq_regs(old_regs);
621 }
622 EXPORT_SYMBOL(timer_interrupt);
623
624 /*
625 * Hypervisor decrementer interrupts shouldn't occur but are sometimes
626 * left pending on exit from a KVM guest. We don't need to do anything
627 * to clear them, as they are edge-triggered.
628 */
hdec_interrupt(struct pt_regs * regs)629 void hdec_interrupt(struct pt_regs *regs)
630 {
631 }
632
633 #ifdef CONFIG_SUSPEND
generic_suspend_disable_irqs(void)634 static void generic_suspend_disable_irqs(void)
635 {
636 /* Disable the decrementer, so that it doesn't interfere
637 * with suspending.
638 */
639
640 set_dec(decrementer_max);
641 local_irq_disable();
642 set_dec(decrementer_max);
643 }
644
generic_suspend_enable_irqs(void)645 static void generic_suspend_enable_irqs(void)
646 {
647 local_irq_enable();
648 }
649
650 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_disable_irqs(void)651 void arch_suspend_disable_irqs(void)
652 {
653 if (ppc_md.suspend_disable_irqs)
654 ppc_md.suspend_disable_irqs();
655 generic_suspend_disable_irqs();
656 }
657
658 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_enable_irqs(void)659 void arch_suspend_enable_irqs(void)
660 {
661 generic_suspend_enable_irqs();
662 if (ppc_md.suspend_enable_irqs)
663 ppc_md.suspend_enable_irqs();
664 }
665 #endif
666
tb_to_ns(unsigned long long ticks)667 unsigned long long tb_to_ns(unsigned long long ticks)
668 {
669 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
670 }
671 EXPORT_SYMBOL_GPL(tb_to_ns);
672
673 /*
674 * Scheduler clock - returns current time in nanosec units.
675 *
676 * Note: mulhdu(a, b) (multiply high double unsigned) returns
677 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
678 * are 64-bit unsigned numbers.
679 */
sched_clock(void)680 notrace unsigned long long sched_clock(void)
681 {
682 if (__USE_RTC())
683 return get_rtc();
684 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
685 }
686
687
688 #ifdef CONFIG_PPC_PSERIES
689
690 /*
691 * Running clock - attempts to give a view of time passing for a virtualised
692 * kernels.
693 * Uses the VTB register if available otherwise a next best guess.
694 */
running_clock(void)695 unsigned long long running_clock(void)
696 {
697 /*
698 * Don't read the VTB as a host since KVM does not switch in host
699 * timebase into the VTB when it takes a guest off the CPU, reading the
700 * VTB would result in reading 'last switched out' guest VTB.
701 *
702 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
703 * would be unsafe to rely only on the #ifdef above.
704 */
705 if (firmware_has_feature(FW_FEATURE_LPAR) &&
706 cpu_has_feature(CPU_FTR_ARCH_207S))
707 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
708
709 /*
710 * This is a next best approximation without a VTB.
711 * On a host which is running bare metal there should never be any stolen
712 * time and on a host which doesn't do any virtualisation TB *should* equal
713 * VTB so it makes no difference anyway.
714 */
715 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
716 }
717 #endif
718
get_freq(char * name,int cells,unsigned long * val)719 static int __init get_freq(char *name, int cells, unsigned long *val)
720 {
721 struct device_node *cpu;
722 const __be32 *fp;
723 int found = 0;
724
725 /* The cpu node should have timebase and clock frequency properties */
726 cpu = of_find_node_by_type(NULL, "cpu");
727
728 if (cpu) {
729 fp = of_get_property(cpu, name, NULL);
730 if (fp) {
731 found = 1;
732 *val = of_read_ulong(fp, cells);
733 }
734
735 of_node_put(cpu);
736 }
737
738 return found;
739 }
740
start_cpu_decrementer(void)741 static void start_cpu_decrementer(void)
742 {
743 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
744 unsigned int tcr;
745
746 /* Clear any pending timer interrupts */
747 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
748
749 tcr = mfspr(SPRN_TCR);
750 /*
751 * The watchdog may have already been enabled by u-boot. So leave
752 * TRC[WP] (Watchdog Period) alone.
753 */
754 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
755 tcr |= TCR_DIE; /* Enable decrementer */
756 mtspr(SPRN_TCR, tcr);
757 #endif
758 }
759
generic_calibrate_decr(void)760 void __init generic_calibrate_decr(void)
761 {
762 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
763
764 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
765 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
766
767 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
768 "(not found)\n");
769 }
770
771 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
772
773 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
774 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
775
776 printk(KERN_ERR "WARNING: Estimating processor frequency "
777 "(not found)\n");
778 }
779 }
780
update_persistent_clock(struct timespec now)781 int update_persistent_clock(struct timespec now)
782 {
783 struct rtc_time tm;
784
785 if (!ppc_md.set_rtc_time)
786 return -ENODEV;
787
788 to_tm(now.tv_sec + 1 + timezone_offset, &tm);
789 tm.tm_year -= 1900;
790 tm.tm_mon -= 1;
791
792 return ppc_md.set_rtc_time(&tm);
793 }
794
__read_persistent_clock(struct timespec * ts)795 static void __read_persistent_clock(struct timespec *ts)
796 {
797 struct rtc_time tm;
798 static int first = 1;
799
800 ts->tv_nsec = 0;
801 /* XXX this is a litle fragile but will work okay in the short term */
802 if (first) {
803 first = 0;
804 if (ppc_md.time_init)
805 timezone_offset = ppc_md.time_init();
806
807 /* get_boot_time() isn't guaranteed to be safe to call late */
808 if (ppc_md.get_boot_time) {
809 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
810 return;
811 }
812 }
813 if (!ppc_md.get_rtc_time) {
814 ts->tv_sec = 0;
815 return;
816 }
817 ppc_md.get_rtc_time(&tm);
818
819 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
820 tm.tm_hour, tm.tm_min, tm.tm_sec);
821 }
822
read_persistent_clock(struct timespec * ts)823 void read_persistent_clock(struct timespec *ts)
824 {
825 __read_persistent_clock(ts);
826
827 /* Sanitize it in case real time clock is set below EPOCH */
828 if (ts->tv_sec < 0) {
829 ts->tv_sec = 0;
830 ts->tv_nsec = 0;
831 }
832
833 }
834
835 /* clocksource code */
rtc_read(struct clocksource * cs)836 static notrace u64 rtc_read(struct clocksource *cs)
837 {
838 return (u64)get_rtc();
839 }
840
timebase_read(struct clocksource * cs)841 static notrace u64 timebase_read(struct clocksource *cs)
842 {
843 return (u64)get_tb();
844 }
845
846
update_vsyscall(struct timekeeper * tk)847 void update_vsyscall(struct timekeeper *tk)
848 {
849 struct timespec xt;
850 struct clocksource *clock = tk->tkr_mono.clock;
851 u32 mult = tk->tkr_mono.mult;
852 u32 shift = tk->tkr_mono.shift;
853 u64 cycle_last = tk->tkr_mono.cycle_last;
854 u64 new_tb_to_xs, new_stamp_xsec;
855 u64 frac_sec;
856
857 if (clock != &clocksource_timebase)
858 return;
859
860 xt.tv_sec = tk->xtime_sec;
861 xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
862
863 /* Make userspace gettimeofday spin until we're done. */
864 ++vdso_data->tb_update_count;
865 smp_mb();
866
867 /*
868 * This computes ((2^20 / 1e9) * mult) >> shift as a
869 * 0.64 fixed-point fraction.
870 * The computation in the else clause below won't overflow
871 * (as long as the timebase frequency is >= 1.049 MHz)
872 * but loses precision because we lose the low bits of the constant
873 * in the shift. Note that 19342813113834067 ~= 2^(20+64) / 1e9.
874 * For a shift of 24 the error is about 0.5e-9, or about 0.5ns
875 * over a second. (Shift values are usually 22, 23 or 24.)
876 * For high frequency clocks such as the 512MHz timebase clock
877 * on POWER[6789], the mult value is small (e.g. 32768000)
878 * and so we can shift the constant by 16 initially
879 * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the
880 * remaining shifts after the multiplication, which gives a
881 * more accurate result (e.g. with mult = 32768000, shift = 24,
882 * the error is only about 1.2e-12, or 0.7ns over 10 minutes).
883 */
884 if (mult <= 62500000 && clock->shift >= 16)
885 new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16);
886 else
887 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
888
889 /*
890 * Compute the fractional second in units of 2^-32 seconds.
891 * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift
892 * in nanoseconds, so multiplying that by 2^32 / 1e9 gives
893 * it in units of 2^-32 seconds.
894 * We assume shift <= 32 because clocks_calc_mult_shift()
895 * generates shift values in the range 0 - 32.
896 */
897 frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift);
898 do_div(frac_sec, NSEC_PER_SEC);
899
900 /*
901 * Work out new stamp_xsec value for any legacy users of systemcfg.
902 * stamp_xsec is in units of 2^-20 seconds.
903 */
904 new_stamp_xsec = frac_sec >> 12;
905 new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC;
906
907 /*
908 * tb_update_count is used to allow the userspace gettimeofday code
909 * to assure itself that it sees a consistent view of the tb_to_xs and
910 * stamp_xsec variables. It reads the tb_update_count, then reads
911 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
912 * the two values of tb_update_count match and are even then the
913 * tb_to_xs and stamp_xsec values are consistent. If not, then it
914 * loops back and reads them again until this criteria is met.
915 */
916 vdso_data->tb_orig_stamp = cycle_last;
917 vdso_data->stamp_xsec = new_stamp_xsec;
918 vdso_data->tb_to_xs = new_tb_to_xs;
919 vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec;
920 vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec;
921 vdso_data->stamp_xtime = xt;
922 vdso_data->stamp_sec_fraction = frac_sec;
923 vdso_data->hrtimer_res = hrtimer_resolution;
924 smp_wmb();
925 ++(vdso_data->tb_update_count);
926 }
927
update_vsyscall_tz(void)928 void update_vsyscall_tz(void)
929 {
930 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
931 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
932 }
933
clocksource_init(void)934 static void __init clocksource_init(void)
935 {
936 struct clocksource *clock;
937
938 if (__USE_RTC())
939 clock = &clocksource_rtc;
940 else
941 clock = &clocksource_timebase;
942
943 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
944 printk(KERN_ERR "clocksource: %s is already registered\n",
945 clock->name);
946 return;
947 }
948
949 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
950 clock->name, clock->mult, clock->shift);
951 }
952
decrementer_set_next_event(unsigned long evt,struct clock_event_device * dev)953 static int decrementer_set_next_event(unsigned long evt,
954 struct clock_event_device *dev)
955 {
956 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
957 set_dec(evt);
958
959 /* We may have raced with new irq work */
960 if (test_irq_work_pending())
961 set_dec(1);
962
963 return 0;
964 }
965
decrementer_shutdown(struct clock_event_device * dev)966 static int decrementer_shutdown(struct clock_event_device *dev)
967 {
968 decrementer_set_next_event(decrementer_max, dev);
969 return 0;
970 }
971
972 /* Interrupt handler for the timer broadcast IPI */
tick_broadcast_ipi_handler(void)973 void tick_broadcast_ipi_handler(void)
974 {
975 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
976
977 *next_tb = get_tb_or_rtc();
978 __timer_interrupt();
979 }
980
register_decrementer_clockevent(int cpu)981 static void register_decrementer_clockevent(int cpu)
982 {
983 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
984
985 *dec = decrementer_clockevent;
986 dec->cpumask = cpumask_of(cpu);
987
988 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
989
990 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
991 dec->name, dec->mult, dec->shift, cpu);
992
993 /* Set values for KVM, see kvm_emulate_dec() */
994 decrementer_clockevent.mult = dec->mult;
995 decrementer_clockevent.shift = dec->shift;
996 }
997
enable_large_decrementer(void)998 static void enable_large_decrementer(void)
999 {
1000 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1001 return;
1002
1003 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
1004 return;
1005
1006 /*
1007 * If we're running as the hypervisor we need to enable the LD manually
1008 * otherwise firmware should have done it for us.
1009 */
1010 if (cpu_has_feature(CPU_FTR_HVMODE))
1011 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
1012 }
1013
set_decrementer_max(void)1014 static void __init set_decrementer_max(void)
1015 {
1016 struct device_node *cpu;
1017 u32 bits = 32;
1018
1019 /* Prior to ISAv3 the decrementer is always 32 bit */
1020 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1021 return;
1022
1023 cpu = of_find_node_by_type(NULL, "cpu");
1024
1025 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
1026 if (bits > 64 || bits < 32) {
1027 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
1028 bits = 32;
1029 }
1030
1031 /* calculate the signed maximum given this many bits */
1032 decrementer_max = (1ul << (bits - 1)) - 1;
1033 }
1034
1035 of_node_put(cpu);
1036
1037 pr_info("time_init: %u bit decrementer (max: %llx)\n",
1038 bits, decrementer_max);
1039 }
1040
init_decrementer_clockevent(void)1041 static void __init init_decrementer_clockevent(void)
1042 {
1043 register_decrementer_clockevent(smp_processor_id());
1044 }
1045
secondary_cpu_time_init(void)1046 void secondary_cpu_time_init(void)
1047 {
1048 /* Enable and test the large decrementer for this cpu */
1049 enable_large_decrementer();
1050
1051 /* Start the decrementer on CPUs that have manual control
1052 * such as BookE
1053 */
1054 start_cpu_decrementer();
1055
1056 /* FIME: Should make unrelatred change to move snapshot_timebase
1057 * call here ! */
1058 register_decrementer_clockevent(smp_processor_id());
1059 }
1060
1061 /* This function is only called on the boot processor */
time_init(void)1062 void __init time_init(void)
1063 {
1064 struct div_result res;
1065 u64 scale;
1066 unsigned shift;
1067
1068 if (__USE_RTC()) {
1069 /* 601 processor: dec counts down by 128 every 128ns */
1070 ppc_tb_freq = 1000000000;
1071 } else {
1072 /* Normal PowerPC with timebase register */
1073 ppc_md.calibrate_decr();
1074 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1075 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1076 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1077 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1078 }
1079
1080 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1081 tb_ticks_per_sec = ppc_tb_freq;
1082 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1083 calc_cputime_factors();
1084
1085 /*
1086 * Compute scale factor for sched_clock.
1087 * The calibrate_decr() function has set tb_ticks_per_sec,
1088 * which is the timebase frequency.
1089 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1090 * the 128-bit result as a 64.64 fixed-point number.
1091 * We then shift that number right until it is less than 1.0,
1092 * giving us the scale factor and shift count to use in
1093 * sched_clock().
1094 */
1095 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1096 scale = res.result_low;
1097 for (shift = 0; res.result_high != 0; ++shift) {
1098 scale = (scale >> 1) | (res.result_high << 63);
1099 res.result_high >>= 1;
1100 }
1101 tb_to_ns_scale = scale;
1102 tb_to_ns_shift = shift;
1103 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1104 boot_tb = get_tb_or_rtc();
1105
1106 /* If platform provided a timezone (pmac), we correct the time */
1107 if (timezone_offset) {
1108 sys_tz.tz_minuteswest = -timezone_offset / 60;
1109 sys_tz.tz_dsttime = 0;
1110 }
1111
1112 vdso_data->tb_update_count = 0;
1113 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1114
1115 /* initialise and enable the large decrementer (if we have one) */
1116 set_decrementer_max();
1117 enable_large_decrementer();
1118
1119 /* Start the decrementer on CPUs that have manual control
1120 * such as BookE
1121 */
1122 start_cpu_decrementer();
1123
1124 /* Register the clocksource */
1125 clocksource_init();
1126
1127 init_decrementer_clockevent();
1128 tick_setup_hrtimer_broadcast();
1129
1130 #ifdef CONFIG_COMMON_CLK
1131 of_clk_init(NULL);
1132 #endif
1133 }
1134
1135
1136 #define FEBRUARY 2
1137 #define STARTOFTIME 1970
1138 #define SECDAY 86400L
1139 #define SECYR (SECDAY * 365)
1140 #define leapyear(year) ((year) % 4 == 0 && \
1141 ((year) % 100 != 0 || (year) % 400 == 0))
1142 #define days_in_year(a) (leapyear(a) ? 366 : 365)
1143 #define days_in_month(a) (month_days[(a) - 1])
1144
1145 static int month_days[12] = {
1146 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1147 };
1148
to_tm(int tim,struct rtc_time * tm)1149 void to_tm(int tim, struct rtc_time * tm)
1150 {
1151 register int i;
1152 register long hms, day;
1153
1154 day = tim / SECDAY;
1155 hms = tim % SECDAY;
1156
1157 /* Hours, minutes, seconds are easy */
1158 tm->tm_hour = hms / 3600;
1159 tm->tm_min = (hms % 3600) / 60;
1160 tm->tm_sec = (hms % 3600) % 60;
1161
1162 /* Number of years in days */
1163 for (i = STARTOFTIME; day >= days_in_year(i); i++)
1164 day -= days_in_year(i);
1165 tm->tm_year = i;
1166
1167 /* Number of months in days left */
1168 if (leapyear(tm->tm_year))
1169 days_in_month(FEBRUARY) = 29;
1170 for (i = 1; day >= days_in_month(i); i++)
1171 day -= days_in_month(i);
1172 days_in_month(FEBRUARY) = 28;
1173 tm->tm_mon = i;
1174
1175 /* Days are what is left over (+1) from all that. */
1176 tm->tm_mday = day + 1;
1177
1178 /*
1179 * No-one uses the day of the week.
1180 */
1181 tm->tm_wday = -1;
1182 }
1183 EXPORT_SYMBOL(to_tm);
1184
1185 /*
1186 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1187 * result.
1188 */
div128_by_32(u64 dividend_high,u64 dividend_low,unsigned divisor,struct div_result * dr)1189 void div128_by_32(u64 dividend_high, u64 dividend_low,
1190 unsigned divisor, struct div_result *dr)
1191 {
1192 unsigned long a, b, c, d;
1193 unsigned long w, x, y, z;
1194 u64 ra, rb, rc;
1195
1196 a = dividend_high >> 32;
1197 b = dividend_high & 0xffffffff;
1198 c = dividend_low >> 32;
1199 d = dividend_low & 0xffffffff;
1200
1201 w = a / divisor;
1202 ra = ((u64)(a - (w * divisor)) << 32) + b;
1203
1204 rb = ((u64) do_div(ra, divisor) << 32) + c;
1205 x = ra;
1206
1207 rc = ((u64) do_div(rb, divisor) << 32) + d;
1208 y = rb;
1209
1210 do_div(rc, divisor);
1211 z = rc;
1212
1213 dr->result_high = ((u64)w << 32) + x;
1214 dr->result_low = ((u64)y << 32) + z;
1215
1216 }
1217
1218 /* We don't need to calibrate delay, we use the CPU timebase for that */
calibrate_delay(void)1219 void calibrate_delay(void)
1220 {
1221 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1222 * as the number of __delay(1) in a jiffy, so make it so
1223 */
1224 loops_per_jiffy = tb_ticks_per_jiffy;
1225 }
1226
1227 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
rtc_generic_get_time(struct device * dev,struct rtc_time * tm)1228 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1229 {
1230 ppc_md.get_rtc_time(tm);
1231 return rtc_valid_tm(tm);
1232 }
1233
rtc_generic_set_time(struct device * dev,struct rtc_time * tm)1234 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1235 {
1236 if (!ppc_md.set_rtc_time)
1237 return -EOPNOTSUPP;
1238
1239 if (ppc_md.set_rtc_time(tm) < 0)
1240 return -EOPNOTSUPP;
1241
1242 return 0;
1243 }
1244
1245 static const struct rtc_class_ops rtc_generic_ops = {
1246 .read_time = rtc_generic_get_time,
1247 .set_time = rtc_generic_set_time,
1248 };
1249
rtc_init(void)1250 static int __init rtc_init(void)
1251 {
1252 struct platform_device *pdev;
1253
1254 if (!ppc_md.get_rtc_time)
1255 return -ENODEV;
1256
1257 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1258 &rtc_generic_ops,
1259 sizeof(rtc_generic_ops));
1260
1261 return PTR_ERR_OR_ZERO(pdev);
1262 }
1263
1264 device_initcall(rtc_init);
1265 #endif
1266