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/kernel.h>
35 #include <linux/param.h>
36 #include <linux/string.h>
37 #include <linux/mm.h>
38 #include <linux/interrupt.h>
39 #include <linux/timex.h>
40 #include <linux/kernel_stat.h>
41 #include <linux/time.h>
42 #include <linux/init.h>
43 #include <linux/profile.h>
44 #include <linux/cpu.h>
45 #include <linux/security.h>
46 #include <linux/percpu.h>
47 #include <linux/rtc.h>
48 #include <linux/jiffies.h>
49 #include <linux/posix-timers.h>
50 #include <linux/irq.h>
51 #include <linux/delay.h>
52 #include <linux/irq_work.h>
53 #include <linux/of_clk.h>
54 #include <linux/suspend.h>
55 #include <linux/sched/cputime.h>
56 #include <linux/sched/clock.h>
57 #include <linux/processor.h>
58 #include <asm/trace.h>
59
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 };
87
88 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
89 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
90
91 static int decrementer_set_next_event(unsigned long evt,
92 struct clock_event_device *dev);
93 static int decrementer_shutdown(struct clock_event_device *evt);
94
95 struct clock_event_device decrementer_clockevent = {
96 .name = "decrementer",
97 .rating = 200,
98 .irq = 0,
99 .set_next_event = decrementer_set_next_event,
100 .set_state_oneshot_stopped = decrementer_shutdown,
101 .set_state_shutdown = decrementer_shutdown,
102 .tick_resume = decrementer_shutdown,
103 .features = CLOCK_EVT_FEAT_ONESHOT |
104 CLOCK_EVT_FEAT_C3STOP,
105 };
106 EXPORT_SYMBOL(decrementer_clockevent);
107
108 DEFINE_PER_CPU(u64, decrementers_next_tb);
109 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
110
111 #define XSEC_PER_SEC (1024*1024)
112
113 #ifdef CONFIG_PPC64
114 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
115 #else
116 /* compute ((xsec << 12) * max) >> 32 */
117 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
118 #endif
119
120 unsigned long tb_ticks_per_jiffy;
121 unsigned long tb_ticks_per_usec = 100; /* sane default */
122 EXPORT_SYMBOL(tb_ticks_per_usec);
123 unsigned long tb_ticks_per_sec;
124 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
125
126 DEFINE_SPINLOCK(rtc_lock);
127 EXPORT_SYMBOL_GPL(rtc_lock);
128
129 static u64 tb_to_ns_scale __read_mostly;
130 static unsigned tb_to_ns_shift __read_mostly;
131 static u64 boot_tb __read_mostly;
132
133 extern struct timezone sys_tz;
134 static long timezone_offset;
135
136 unsigned long ppc_proc_freq;
137 EXPORT_SYMBOL_GPL(ppc_proc_freq);
138 unsigned long ppc_tb_freq;
139 EXPORT_SYMBOL_GPL(ppc_tb_freq);
140
141 bool tb_invalid;
142
143 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
144 /*
145 * Factor for converting from cputime_t (timebase ticks) to
146 * microseconds. This is stored as 0.64 fixed-point binary fraction.
147 */
148 u64 __cputime_usec_factor;
149 EXPORT_SYMBOL(__cputime_usec_factor);
150
151 #ifdef CONFIG_PPC_SPLPAR
152 void (*dtl_consumer)(struct dtl_entry *, u64);
153 #endif
154
calc_cputime_factors(void)155 static void calc_cputime_factors(void)
156 {
157 struct div_result res;
158
159 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
160 __cputime_usec_factor = res.result_low;
161 }
162
163 /*
164 * Read the SPURR on systems that have it, otherwise the PURR,
165 * or if that doesn't exist return the timebase value passed in.
166 */
read_spurr(unsigned long tb)167 static inline unsigned long read_spurr(unsigned long tb)
168 {
169 if (cpu_has_feature(CPU_FTR_SPURR))
170 return mfspr(SPRN_SPURR);
171 if (cpu_has_feature(CPU_FTR_PURR))
172 return mfspr(SPRN_PURR);
173 return tb;
174 }
175
176 #ifdef CONFIG_PPC_SPLPAR
177
178 #include <asm/dtl.h>
179
180 /*
181 * Scan the dispatch trace log and count up the stolen time.
182 * Should be called with interrupts disabled.
183 */
scan_dispatch_log(u64 stop_tb)184 static u64 scan_dispatch_log(u64 stop_tb)
185 {
186 u64 i = local_paca->dtl_ridx;
187 struct dtl_entry *dtl = local_paca->dtl_curr;
188 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
189 struct lppaca *vpa = local_paca->lppaca_ptr;
190 u64 tb_delta;
191 u64 stolen = 0;
192 u64 dtb;
193
194 if (!dtl)
195 return 0;
196
197 if (i == be64_to_cpu(vpa->dtl_idx))
198 return 0;
199 while (i < be64_to_cpu(vpa->dtl_idx)) {
200 dtb = be64_to_cpu(dtl->timebase);
201 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
202 be32_to_cpu(dtl->ready_to_enqueue_time);
203 barrier();
204 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
205 /* buffer has overflowed */
206 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
207 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
208 continue;
209 }
210 if (dtb > stop_tb)
211 break;
212 if (dtl_consumer)
213 dtl_consumer(dtl, i);
214 stolen += tb_delta;
215 ++i;
216 ++dtl;
217 if (dtl == dtl_end)
218 dtl = local_paca->dispatch_log;
219 }
220 local_paca->dtl_ridx = i;
221 local_paca->dtl_curr = dtl;
222 return stolen;
223 }
224
225 /*
226 * Accumulate stolen time by scanning the dispatch trace log.
227 * Called on entry from user mode.
228 */
accumulate_stolen_time(void)229 void notrace accumulate_stolen_time(void)
230 {
231 u64 sst, ust;
232 unsigned long save_irq_soft_mask = irq_soft_mask_return();
233 struct cpu_accounting_data *acct = &local_paca->accounting;
234
235 /* We are called early in the exception entry, before
236 * soft/hard_enabled are sync'ed to the expected state
237 * for the exception. We are hard disabled but the PACA
238 * needs to reflect that so various debug stuff doesn't
239 * complain
240 */
241 irq_soft_mask_set(IRQS_DISABLED);
242
243 sst = scan_dispatch_log(acct->starttime_user);
244 ust = scan_dispatch_log(acct->starttime);
245 acct->stime -= sst;
246 acct->utime -= ust;
247 acct->steal_time += ust + sst;
248
249 irq_soft_mask_set(save_irq_soft_mask);
250 }
251
calculate_stolen_time(u64 stop_tb)252 static inline u64 calculate_stolen_time(u64 stop_tb)
253 {
254 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
255 return 0;
256
257 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
258 return scan_dispatch_log(stop_tb);
259
260 return 0;
261 }
262
263 #else /* CONFIG_PPC_SPLPAR */
calculate_stolen_time(u64 stop_tb)264 static inline u64 calculate_stolen_time(u64 stop_tb)
265 {
266 return 0;
267 }
268
269 #endif /* CONFIG_PPC_SPLPAR */
270
271 /*
272 * Account time for a transition between system, hard irq
273 * or soft irq state.
274 */
vtime_delta_scaled(struct cpu_accounting_data * acct,unsigned long now,unsigned long stime)275 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
276 unsigned long now, unsigned long stime)
277 {
278 unsigned long stime_scaled = 0;
279 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
280 unsigned long nowscaled, deltascaled;
281 unsigned long utime, utime_scaled;
282
283 nowscaled = read_spurr(now);
284 deltascaled = nowscaled - acct->startspurr;
285 acct->startspurr = nowscaled;
286 utime = acct->utime - acct->utime_sspurr;
287 acct->utime_sspurr = acct->utime;
288
289 /*
290 * Because we don't read the SPURR on every kernel entry/exit,
291 * deltascaled includes both user and system SPURR ticks.
292 * Apportion these ticks to system SPURR ticks and user
293 * SPURR ticks in the same ratio as the system time (delta)
294 * and user time (udelta) values obtained from the timebase
295 * over the same interval. The system ticks get accounted here;
296 * the user ticks get saved up in paca->user_time_scaled to be
297 * used by account_process_tick.
298 */
299 stime_scaled = stime;
300 utime_scaled = utime;
301 if (deltascaled != stime + utime) {
302 if (utime) {
303 stime_scaled = deltascaled * stime / (stime + utime);
304 utime_scaled = deltascaled - stime_scaled;
305 } else {
306 stime_scaled = deltascaled;
307 }
308 }
309 acct->utime_scaled += utime_scaled;
310 #endif
311
312 return stime_scaled;
313 }
314
vtime_delta(struct task_struct * tsk,unsigned long * stime_scaled,unsigned long * steal_time)315 static unsigned long vtime_delta(struct task_struct *tsk,
316 unsigned long *stime_scaled,
317 unsigned long *steal_time)
318 {
319 unsigned long now, stime;
320 struct cpu_accounting_data *acct = get_accounting(tsk);
321
322 WARN_ON_ONCE(!irqs_disabled());
323
324 now = mftb();
325 stime = now - acct->starttime;
326 acct->starttime = now;
327
328 *stime_scaled = vtime_delta_scaled(acct, now, stime);
329
330 *steal_time = calculate_stolen_time(now);
331
332 return stime;
333 }
334
vtime_account_kernel(struct task_struct * tsk)335 void vtime_account_kernel(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 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
348 acct->utime_scaled += stime_scaled;
349 #endif
350 } else {
351 if (hardirq_count())
352 acct->hardirq_time += stime;
353 else if (in_serving_softirq())
354 acct->softirq_time += stime;
355 else
356 acct->stime += stime;
357
358 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
359 acct->stime_scaled += stime_scaled;
360 #endif
361 }
362 }
363 EXPORT_SYMBOL_GPL(vtime_account_kernel);
364
vtime_account_idle(struct task_struct * tsk)365 void vtime_account_idle(struct task_struct *tsk)
366 {
367 unsigned long stime, stime_scaled, steal_time;
368 struct cpu_accounting_data *acct = get_accounting(tsk);
369
370 stime = vtime_delta(tsk, &stime_scaled, &steal_time);
371 acct->idle_time += stime + steal_time;
372 }
373
vtime_flush_scaled(struct task_struct * tsk,struct cpu_accounting_data * acct)374 static void vtime_flush_scaled(struct task_struct *tsk,
375 struct cpu_accounting_data *acct)
376 {
377 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
378 if (acct->utime_scaled)
379 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
380 if (acct->stime_scaled)
381 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
382
383 acct->utime_scaled = 0;
384 acct->utime_sspurr = 0;
385 acct->stime_scaled = 0;
386 #endif
387 }
388
389 /*
390 * Account the whole cputime accumulated in the paca
391 * Must be called with interrupts disabled.
392 * Assumes that vtime_account_kernel/idle() has been called
393 * recently (i.e. since the last entry from usermode) so that
394 * get_paca()->user_time_scaled is up to date.
395 */
vtime_flush(struct task_struct * tsk)396 void vtime_flush(struct task_struct *tsk)
397 {
398 struct cpu_accounting_data *acct = get_accounting(tsk);
399
400 if (acct->utime)
401 account_user_time(tsk, cputime_to_nsecs(acct->utime));
402
403 if (acct->gtime)
404 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
405
406 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
407 account_steal_time(cputime_to_nsecs(acct->steal_time));
408 acct->steal_time = 0;
409 }
410
411 if (acct->idle_time)
412 account_idle_time(cputime_to_nsecs(acct->idle_time));
413
414 if (acct->stime)
415 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
416 CPUTIME_SYSTEM);
417
418 if (acct->hardirq_time)
419 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
420 CPUTIME_IRQ);
421 if (acct->softirq_time)
422 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
423 CPUTIME_SOFTIRQ);
424
425 vtime_flush_scaled(tsk, acct);
426
427 acct->utime = 0;
428 acct->gtime = 0;
429 acct->idle_time = 0;
430 acct->stime = 0;
431 acct->hardirq_time = 0;
432 acct->softirq_time = 0;
433 }
434
435 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
436 #define calc_cputime_factors()
437 #endif
438
__delay(unsigned long loops)439 void __delay(unsigned long loops)
440 {
441 unsigned long start;
442
443 spin_begin();
444 if (tb_invalid) {
445 /*
446 * TB is in error state and isn't ticking anymore.
447 * HMI handler was unable to recover from TB error.
448 * Return immediately, so that kernel won't get stuck here.
449 */
450 spin_cpu_relax();
451 } else {
452 start = mftb();
453 while (mftb() - start < loops)
454 spin_cpu_relax();
455 }
456 spin_end();
457 }
458 EXPORT_SYMBOL(__delay);
459
udelay(unsigned long usecs)460 void udelay(unsigned long usecs)
461 {
462 __delay(tb_ticks_per_usec * usecs);
463 }
464 EXPORT_SYMBOL(udelay);
465
466 #ifdef CONFIG_SMP
profile_pc(struct pt_regs * regs)467 unsigned long profile_pc(struct pt_regs *regs)
468 {
469 unsigned long pc = instruction_pointer(regs);
470
471 if (in_lock_functions(pc))
472 return regs->link;
473
474 return pc;
475 }
476 EXPORT_SYMBOL(profile_pc);
477 #endif
478
479 #ifdef CONFIG_IRQ_WORK
480
481 /*
482 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
483 */
484 #ifdef CONFIG_PPC64
test_irq_work_pending(void)485 static inline unsigned long test_irq_work_pending(void)
486 {
487 unsigned long x;
488
489 asm volatile("lbz %0,%1(13)"
490 : "=r" (x)
491 : "i" (offsetof(struct paca_struct, irq_work_pending)));
492 return x;
493 }
494
set_irq_work_pending_flag(void)495 static inline void set_irq_work_pending_flag(void)
496 {
497 asm volatile("stb %0,%1(13)" : :
498 "r" (1),
499 "i" (offsetof(struct paca_struct, irq_work_pending)));
500 }
501
clear_irq_work_pending(void)502 static inline void clear_irq_work_pending(void)
503 {
504 asm volatile("stb %0,%1(13)" : :
505 "r" (0),
506 "i" (offsetof(struct paca_struct, irq_work_pending)));
507 }
508
509 #else /* 32-bit */
510
511 DEFINE_PER_CPU(u8, irq_work_pending);
512
513 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
514 #define test_irq_work_pending() __this_cpu_read(irq_work_pending)
515 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
516
517 #endif /* 32 vs 64 bit */
518
arch_irq_work_raise(void)519 void arch_irq_work_raise(void)
520 {
521 /*
522 * 64-bit code that uses irq soft-mask can just cause an immediate
523 * interrupt here that gets soft masked, if this is called under
524 * local_irq_disable(). It might be possible to prevent that happening
525 * by noticing interrupts are disabled and setting decrementer pending
526 * to be replayed when irqs are enabled. The problem there is that
527 * tracing can call irq_work_raise, including in code that does low
528 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
529 * which could get tangled up if we're messing with the same state
530 * here.
531 */
532 preempt_disable();
533 set_irq_work_pending_flag();
534 set_dec(1);
535 preempt_enable();
536 }
537
538 #else /* CONFIG_IRQ_WORK */
539
540 #define test_irq_work_pending() 0
541 #define clear_irq_work_pending()
542
543 #endif /* CONFIG_IRQ_WORK */
544
545 /*
546 * timer_interrupt - gets called when the decrementer overflows,
547 * with interrupts disabled.
548 */
timer_interrupt(struct pt_regs * regs)549 void timer_interrupt(struct pt_regs *regs)
550 {
551 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
552 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
553 struct pt_regs *old_regs;
554 u64 now;
555
556 /*
557 * Some implementations of hotplug will get timer interrupts while
558 * offline, just ignore these.
559 */
560 if (unlikely(!cpu_online(smp_processor_id()))) {
561 set_dec(decrementer_max);
562 return;
563 }
564
565 /* Ensure a positive value is written to the decrementer, or else
566 * some CPUs will continue to take decrementer exceptions. When the
567 * PPC_WATCHDOG (decrementer based) is configured, keep this at most
568 * 31 bits, which is about 4 seconds on most systems, which gives
569 * the watchdog a chance of catching timer interrupt hard lockups.
570 */
571 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
572 set_dec(0x7fffffff);
573 else
574 set_dec(decrementer_max);
575
576 /* Conditionally hard-enable interrupts now that the DEC has been
577 * bumped to its maximum value
578 */
579 may_hard_irq_enable();
580
581
582 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
583 if (atomic_read(&ppc_n_lost_interrupts) != 0)
584 do_IRQ(regs);
585 #endif
586
587 old_regs = set_irq_regs(regs);
588 irq_enter();
589 trace_timer_interrupt_entry(regs);
590
591 if (test_irq_work_pending()) {
592 clear_irq_work_pending();
593 irq_work_run();
594 }
595
596 now = get_tb();
597 if (now >= *next_tb) {
598 *next_tb = ~(u64)0;
599 if (evt->event_handler)
600 evt->event_handler(evt);
601 __this_cpu_inc(irq_stat.timer_irqs_event);
602 } else {
603 now = *next_tb - now;
604 if (now <= decrementer_max)
605 set_dec(now);
606 /* We may have raced with new irq work */
607 if (test_irq_work_pending())
608 set_dec(1);
609 __this_cpu_inc(irq_stat.timer_irqs_others);
610 }
611
612 trace_timer_interrupt_exit(regs);
613 irq_exit();
614 set_irq_regs(old_regs);
615 }
616 EXPORT_SYMBOL(timer_interrupt);
617
618 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
timer_broadcast_interrupt(void)619 void timer_broadcast_interrupt(void)
620 {
621 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
622
623 *next_tb = ~(u64)0;
624 tick_receive_broadcast();
625 __this_cpu_inc(irq_stat.broadcast_irqs_event);
626 }
627 #endif
628
629 #ifdef CONFIG_SUSPEND
generic_suspend_disable_irqs(void)630 static void generic_suspend_disable_irqs(void)
631 {
632 /* Disable the decrementer, so that it doesn't interfere
633 * with suspending.
634 */
635
636 set_dec(decrementer_max);
637 local_irq_disable();
638 set_dec(decrementer_max);
639 }
640
generic_suspend_enable_irqs(void)641 static void generic_suspend_enable_irqs(void)
642 {
643 local_irq_enable();
644 }
645
646 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_disable_irqs(void)647 void arch_suspend_disable_irqs(void)
648 {
649 if (ppc_md.suspend_disable_irqs)
650 ppc_md.suspend_disable_irqs();
651 generic_suspend_disable_irqs();
652 }
653
654 /* Overrides the weak version in kernel/power/main.c */
arch_suspend_enable_irqs(void)655 void arch_suspend_enable_irqs(void)
656 {
657 generic_suspend_enable_irqs();
658 if (ppc_md.suspend_enable_irqs)
659 ppc_md.suspend_enable_irqs();
660 }
661 #endif
662
tb_to_ns(unsigned long long ticks)663 unsigned long long tb_to_ns(unsigned long long ticks)
664 {
665 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
666 }
667 EXPORT_SYMBOL_GPL(tb_to_ns);
668
669 /*
670 * Scheduler clock - returns current time in nanosec units.
671 *
672 * Note: mulhdu(a, b) (multiply high double unsigned) returns
673 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
674 * are 64-bit unsigned numbers.
675 */
sched_clock(void)676 notrace unsigned long long sched_clock(void)
677 {
678 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
679 }
680
681
682 #ifdef CONFIG_PPC_PSERIES
683
684 /*
685 * Running clock - attempts to give a view of time passing for a virtualised
686 * kernels.
687 * Uses the VTB register if available otherwise a next best guess.
688 */
running_clock(void)689 unsigned long long running_clock(void)
690 {
691 /*
692 * Don't read the VTB as a host since KVM does not switch in host
693 * timebase into the VTB when it takes a guest off the CPU, reading the
694 * VTB would result in reading 'last switched out' guest VTB.
695 *
696 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
697 * would be unsafe to rely only on the #ifdef above.
698 */
699 if (firmware_has_feature(FW_FEATURE_LPAR) &&
700 cpu_has_feature(CPU_FTR_ARCH_207S))
701 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
702
703 /*
704 * This is a next best approximation without a VTB.
705 * On a host which is running bare metal there should never be any stolen
706 * time and on a host which doesn't do any virtualisation TB *should* equal
707 * VTB so it makes no difference anyway.
708 */
709 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
710 }
711 #endif
712
get_freq(char * name,int cells,unsigned long * val)713 static int __init get_freq(char *name, int cells, unsigned long *val)
714 {
715 struct device_node *cpu;
716 const __be32 *fp;
717 int found = 0;
718
719 /* The cpu node should have timebase and clock frequency properties */
720 cpu = of_find_node_by_type(NULL, "cpu");
721
722 if (cpu) {
723 fp = of_get_property(cpu, name, NULL);
724 if (fp) {
725 found = 1;
726 *val = of_read_ulong(fp, cells);
727 }
728
729 of_node_put(cpu);
730 }
731
732 return found;
733 }
734
start_cpu_decrementer(void)735 static void start_cpu_decrementer(void)
736 {
737 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
738 unsigned int tcr;
739
740 /* Clear any pending timer interrupts */
741 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
742
743 tcr = mfspr(SPRN_TCR);
744 /*
745 * The watchdog may have already been enabled by u-boot. So leave
746 * TRC[WP] (Watchdog Period) alone.
747 */
748 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
749 tcr |= TCR_DIE; /* Enable decrementer */
750 mtspr(SPRN_TCR, tcr);
751 #endif
752 }
753
generic_calibrate_decr(void)754 void __init generic_calibrate_decr(void)
755 {
756 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
757
758 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
759 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
760
761 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
762 "(not found)\n");
763 }
764
765 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
766
767 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
768 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
769
770 printk(KERN_ERR "WARNING: Estimating processor frequency "
771 "(not found)\n");
772 }
773 }
774
update_persistent_clock64(struct timespec64 now)775 int update_persistent_clock64(struct timespec64 now)
776 {
777 struct rtc_time tm;
778
779 if (!ppc_md.set_rtc_time)
780 return -ENODEV;
781
782 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
783
784 return ppc_md.set_rtc_time(&tm);
785 }
786
__read_persistent_clock(struct timespec64 * ts)787 static void __read_persistent_clock(struct timespec64 *ts)
788 {
789 struct rtc_time tm;
790 static int first = 1;
791
792 ts->tv_nsec = 0;
793 /* XXX this is a litle fragile but will work okay in the short term */
794 if (first) {
795 first = 0;
796 if (ppc_md.time_init)
797 timezone_offset = ppc_md.time_init();
798
799 /* get_boot_time() isn't guaranteed to be safe to call late */
800 if (ppc_md.get_boot_time) {
801 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
802 return;
803 }
804 }
805 if (!ppc_md.get_rtc_time) {
806 ts->tv_sec = 0;
807 return;
808 }
809 ppc_md.get_rtc_time(&tm);
810
811 ts->tv_sec = rtc_tm_to_time64(&tm);
812 }
813
read_persistent_clock64(struct timespec64 * ts)814 void read_persistent_clock64(struct timespec64 *ts)
815 {
816 __read_persistent_clock(ts);
817
818 /* Sanitize it in case real time clock is set below EPOCH */
819 if (ts->tv_sec < 0) {
820 ts->tv_sec = 0;
821 ts->tv_nsec = 0;
822 }
823
824 }
825
826 /* clocksource code */
timebase_read(struct clocksource * cs)827 static notrace u64 timebase_read(struct clocksource *cs)
828 {
829 return (u64)get_tb();
830 }
831
832
update_vsyscall(struct timekeeper * tk)833 void update_vsyscall(struct timekeeper *tk)
834 {
835 struct timespec64 xt;
836 struct clocksource *clock = tk->tkr_mono.clock;
837 u32 mult = tk->tkr_mono.mult;
838 u32 shift = tk->tkr_mono.shift;
839 u64 cycle_last = tk->tkr_mono.cycle_last;
840 u64 new_tb_to_xs, new_stamp_xsec;
841 u64 frac_sec;
842
843 if (clock != &clocksource_timebase)
844 return;
845
846 xt.tv_sec = tk->xtime_sec;
847 xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
848
849 /* Make userspace gettimeofday spin until we're done. */
850 ++vdso_data->tb_update_count;
851 smp_mb();
852
853 /*
854 * This computes ((2^20 / 1e9) * mult) >> shift as a
855 * 0.64 fixed-point fraction.
856 * The computation in the else clause below won't overflow
857 * (as long as the timebase frequency is >= 1.049 MHz)
858 * but loses precision because we lose the low bits of the constant
859 * in the shift. Note that 19342813113834067 ~= 2^(20+64) / 1e9.
860 * For a shift of 24 the error is about 0.5e-9, or about 0.5ns
861 * over a second. (Shift values are usually 22, 23 or 24.)
862 * For high frequency clocks such as the 512MHz timebase clock
863 * on POWER[6789], the mult value is small (e.g. 32768000)
864 * and so we can shift the constant by 16 initially
865 * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the
866 * remaining shifts after the multiplication, which gives a
867 * more accurate result (e.g. with mult = 32768000, shift = 24,
868 * the error is only about 1.2e-12, or 0.7ns over 10 minutes).
869 */
870 if (mult <= 62500000 && clock->shift >= 16)
871 new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16);
872 else
873 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
874
875 /*
876 * Compute the fractional second in units of 2^-32 seconds.
877 * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift
878 * in nanoseconds, so multiplying that by 2^32 / 1e9 gives
879 * it in units of 2^-32 seconds.
880 * We assume shift <= 32 because clocks_calc_mult_shift()
881 * generates shift values in the range 0 - 32.
882 */
883 frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift);
884 do_div(frac_sec, NSEC_PER_SEC);
885
886 /*
887 * Work out new stamp_xsec value for any legacy users of systemcfg.
888 * stamp_xsec is in units of 2^-20 seconds.
889 */
890 new_stamp_xsec = frac_sec >> 12;
891 new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC;
892
893 /*
894 * tb_update_count is used to allow the userspace gettimeofday code
895 * to assure itself that it sees a consistent view of the tb_to_xs and
896 * stamp_xsec variables. It reads the tb_update_count, then reads
897 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If
898 * the two values of tb_update_count match and are even then the
899 * tb_to_xs and stamp_xsec values are consistent. If not, then it
900 * loops back and reads them again until this criteria is met.
901 */
902 vdso_data->tb_orig_stamp = cycle_last;
903 vdso_data->stamp_xsec = new_stamp_xsec;
904 vdso_data->tb_to_xs = new_tb_to_xs;
905 vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec;
906 vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec;
907 vdso_data->stamp_xtime_sec = xt.tv_sec;
908 vdso_data->stamp_xtime_nsec = xt.tv_nsec;
909 vdso_data->stamp_sec_fraction = frac_sec;
910 vdso_data->hrtimer_res = hrtimer_resolution;
911 smp_wmb();
912 ++(vdso_data->tb_update_count);
913 }
914
update_vsyscall_tz(void)915 void update_vsyscall_tz(void)
916 {
917 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
918 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
919 }
920
clocksource_init(void)921 static void __init clocksource_init(void)
922 {
923 struct clocksource *clock = &clocksource_timebase;
924
925 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
926 printk(KERN_ERR "clocksource: %s is already registered\n",
927 clock->name);
928 return;
929 }
930
931 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
932 clock->name, clock->mult, clock->shift);
933 }
934
decrementer_set_next_event(unsigned long evt,struct clock_event_device * dev)935 static int decrementer_set_next_event(unsigned long evt,
936 struct clock_event_device *dev)
937 {
938 __this_cpu_write(decrementers_next_tb, get_tb() + evt);
939 set_dec(evt);
940
941 /* We may have raced with new irq work */
942 if (test_irq_work_pending())
943 set_dec(1);
944
945 return 0;
946 }
947
decrementer_shutdown(struct clock_event_device * dev)948 static int decrementer_shutdown(struct clock_event_device *dev)
949 {
950 decrementer_set_next_event(decrementer_max, dev);
951 return 0;
952 }
953
register_decrementer_clockevent(int cpu)954 static void register_decrementer_clockevent(int cpu)
955 {
956 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
957
958 *dec = decrementer_clockevent;
959 dec->cpumask = cpumask_of(cpu);
960
961 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
962
963 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
964 dec->name, dec->mult, dec->shift, cpu);
965
966 /* Set values for KVM, see kvm_emulate_dec() */
967 decrementer_clockevent.mult = dec->mult;
968 decrementer_clockevent.shift = dec->shift;
969 }
970
enable_large_decrementer(void)971 static void enable_large_decrementer(void)
972 {
973 if (!cpu_has_feature(CPU_FTR_ARCH_300))
974 return;
975
976 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
977 return;
978
979 /*
980 * If we're running as the hypervisor we need to enable the LD manually
981 * otherwise firmware should have done it for us.
982 */
983 if (cpu_has_feature(CPU_FTR_HVMODE))
984 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
985 }
986
set_decrementer_max(void)987 static void __init set_decrementer_max(void)
988 {
989 struct device_node *cpu;
990 u32 bits = 32;
991
992 /* Prior to ISAv3 the decrementer is always 32 bit */
993 if (!cpu_has_feature(CPU_FTR_ARCH_300))
994 return;
995
996 cpu = of_find_node_by_type(NULL, "cpu");
997
998 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
999 if (bits > 64 || bits < 32) {
1000 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
1001 bits = 32;
1002 }
1003
1004 /* calculate the signed maximum given this many bits */
1005 decrementer_max = (1ul << (bits - 1)) - 1;
1006 }
1007
1008 of_node_put(cpu);
1009
1010 pr_info("time_init: %u bit decrementer (max: %llx)\n",
1011 bits, decrementer_max);
1012 }
1013
init_decrementer_clockevent(void)1014 static void __init init_decrementer_clockevent(void)
1015 {
1016 register_decrementer_clockevent(smp_processor_id());
1017 }
1018
secondary_cpu_time_init(void)1019 void secondary_cpu_time_init(void)
1020 {
1021 /* Enable and test the large decrementer for this cpu */
1022 enable_large_decrementer();
1023
1024 /* Start the decrementer on CPUs that have manual control
1025 * such as BookE
1026 */
1027 start_cpu_decrementer();
1028
1029 /* FIME: Should make unrelatred change to move snapshot_timebase
1030 * call here ! */
1031 register_decrementer_clockevent(smp_processor_id());
1032 }
1033
1034 /* This function is only called on the boot processor */
time_init(void)1035 void __init time_init(void)
1036 {
1037 struct div_result res;
1038 u64 scale;
1039 unsigned shift;
1040
1041 /* Normal PowerPC with timebase register */
1042 ppc_md.calibrate_decr();
1043 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1044 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1045 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
1046 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1047
1048 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1049 tb_ticks_per_sec = ppc_tb_freq;
1050 tb_ticks_per_usec = ppc_tb_freq / 1000000;
1051 calc_cputime_factors();
1052
1053 /*
1054 * Compute scale factor for sched_clock.
1055 * The calibrate_decr() function has set tb_ticks_per_sec,
1056 * which is the timebase frequency.
1057 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1058 * the 128-bit result as a 64.64 fixed-point number.
1059 * We then shift that number right until it is less than 1.0,
1060 * giving us the scale factor and shift count to use in
1061 * sched_clock().
1062 */
1063 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1064 scale = res.result_low;
1065 for (shift = 0; res.result_high != 0; ++shift) {
1066 scale = (scale >> 1) | (res.result_high << 63);
1067 res.result_high >>= 1;
1068 }
1069 tb_to_ns_scale = scale;
1070 tb_to_ns_shift = shift;
1071 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1072 boot_tb = get_tb();
1073
1074 /* If platform provided a timezone (pmac), we correct the time */
1075 if (timezone_offset) {
1076 sys_tz.tz_minuteswest = -timezone_offset / 60;
1077 sys_tz.tz_dsttime = 0;
1078 }
1079
1080 vdso_data->tb_update_count = 0;
1081 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1082
1083 /* initialise and enable the large decrementer (if we have one) */
1084 set_decrementer_max();
1085 enable_large_decrementer();
1086
1087 /* Start the decrementer on CPUs that have manual control
1088 * such as BookE
1089 */
1090 start_cpu_decrementer();
1091
1092 /* Register the clocksource */
1093 clocksource_init();
1094
1095 init_decrementer_clockevent();
1096 tick_setup_hrtimer_broadcast();
1097
1098 of_clk_init(NULL);
1099 enable_sched_clock_irqtime();
1100 }
1101
1102 /*
1103 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1104 * result.
1105 */
div128_by_32(u64 dividend_high,u64 dividend_low,unsigned divisor,struct div_result * dr)1106 void div128_by_32(u64 dividend_high, u64 dividend_low,
1107 unsigned divisor, struct div_result *dr)
1108 {
1109 unsigned long a, b, c, d;
1110 unsigned long w, x, y, z;
1111 u64 ra, rb, rc;
1112
1113 a = dividend_high >> 32;
1114 b = dividend_high & 0xffffffff;
1115 c = dividend_low >> 32;
1116 d = dividend_low & 0xffffffff;
1117
1118 w = a / divisor;
1119 ra = ((u64)(a - (w * divisor)) << 32) + b;
1120
1121 rb = ((u64) do_div(ra, divisor) << 32) + c;
1122 x = ra;
1123
1124 rc = ((u64) do_div(rb, divisor) << 32) + d;
1125 y = rb;
1126
1127 do_div(rc, divisor);
1128 z = rc;
1129
1130 dr->result_high = ((u64)w << 32) + x;
1131 dr->result_low = ((u64)y << 32) + z;
1132
1133 }
1134
1135 /* We don't need to calibrate delay, we use the CPU timebase for that */
calibrate_delay(void)1136 void calibrate_delay(void)
1137 {
1138 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1139 * as the number of __delay(1) in a jiffy, so make it so
1140 */
1141 loops_per_jiffy = tb_ticks_per_jiffy;
1142 }
1143
1144 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
rtc_generic_get_time(struct device * dev,struct rtc_time * tm)1145 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1146 {
1147 ppc_md.get_rtc_time(tm);
1148 return 0;
1149 }
1150
rtc_generic_set_time(struct device * dev,struct rtc_time * tm)1151 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1152 {
1153 if (!ppc_md.set_rtc_time)
1154 return -EOPNOTSUPP;
1155
1156 if (ppc_md.set_rtc_time(tm) < 0)
1157 return -EOPNOTSUPP;
1158
1159 return 0;
1160 }
1161
1162 static const struct rtc_class_ops rtc_generic_ops = {
1163 .read_time = rtc_generic_get_time,
1164 .set_time = rtc_generic_set_time,
1165 };
1166
rtc_init(void)1167 static int __init rtc_init(void)
1168 {
1169 struct platform_device *pdev;
1170
1171 if (!ppc_md.get_rtc_time)
1172 return -ENODEV;
1173
1174 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1175 &rtc_generic_ops,
1176 sizeof(rtc_generic_ops));
1177
1178 return PTR_ERR_OR_ZERO(pdev);
1179 }
1180
1181 device_initcall(rtc_init);
1182 #endif
1183