Loading...
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/interrupt.h>
61#include <asm/io.h>
62#include <asm/nvram.h>
63#include <asm/cache.h>
64#include <asm/machdep.h>
65#include <linux/uaccess.h>
66#include <asm/time.h>
67#include <asm/prom.h>
68#include <asm/irq.h>
69#include <asm/div64.h>
70#include <asm/smp.h>
71#include <asm/vdso_datapage.h>
72#include <asm/firmware.h>
73#include <asm/asm-prototypes.h>
74
75/* powerpc clocksource/clockevent code */
76
77#include <linux/clockchips.h>
78#include <linux/timekeeper_internal.h>
79
80static u64 timebase_read(struct clocksource *);
81static struct clocksource clocksource_timebase = {
82 .name = "timebase",
83 .rating = 400,
84 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
85 .mask = CLOCKSOURCE_MASK(64),
86 .read = timebase_read,
87 .vdso_clock_mode = VDSO_CLOCKMODE_ARCHTIMER,
88};
89
90#define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
91u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
92
93static int decrementer_set_next_event(unsigned long evt,
94 struct clock_event_device *dev);
95static int decrementer_shutdown(struct clock_event_device *evt);
96
97struct 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};
108EXPORT_SYMBOL(decrementer_clockevent);
109
110DEFINE_PER_CPU(u64, decrementers_next_tb);
111static DEFINE_PER_CPU(struct clock_event_device, decrementers);
112
113#define XSEC_PER_SEC (1024*1024)
114
115#ifdef CONFIG_PPC64
116#define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
117#else
118/* compute ((xsec << 12) * max) >> 32 */
119#define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
120#endif
121
122unsigned long tb_ticks_per_jiffy;
123unsigned long tb_ticks_per_usec = 100; /* sane default */
124EXPORT_SYMBOL(tb_ticks_per_usec);
125unsigned long tb_ticks_per_sec;
126EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */
127
128DEFINE_SPINLOCK(rtc_lock);
129EXPORT_SYMBOL_GPL(rtc_lock);
130
131static u64 tb_to_ns_scale __read_mostly;
132static unsigned tb_to_ns_shift __read_mostly;
133static u64 boot_tb __read_mostly;
134
135extern struct timezone sys_tz;
136static long timezone_offset;
137
138unsigned long ppc_proc_freq;
139EXPORT_SYMBOL_GPL(ppc_proc_freq);
140unsigned long ppc_tb_freq;
141EXPORT_SYMBOL_GPL(ppc_tb_freq);
142
143bool tb_invalid;
144
145#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
146/*
147 * Factor for converting from cputime_t (timebase ticks) to
148 * microseconds. This is stored as 0.64 fixed-point binary fraction.
149 */
150u64 __cputime_usec_factor;
151EXPORT_SYMBOL(__cputime_usec_factor);
152
153#ifdef CONFIG_PPC_SPLPAR
154void (*dtl_consumer)(struct dtl_entry *, u64);
155#endif
156
157static void calc_cputime_factors(void)
158{
159 struct div_result res;
160
161 div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
162 __cputime_usec_factor = res.result_low;
163}
164
165/*
166 * Read the SPURR on systems that have it, otherwise the PURR,
167 * or if that doesn't exist return the timebase value passed in.
168 */
169static inline unsigned long read_spurr(unsigned long tb)
170{
171 if (cpu_has_feature(CPU_FTR_SPURR))
172 return mfspr(SPRN_SPURR);
173 if (cpu_has_feature(CPU_FTR_PURR))
174 return mfspr(SPRN_PURR);
175 return tb;
176}
177
178#ifdef CONFIG_PPC_SPLPAR
179
180#include <asm/dtl.h>
181
182/*
183 * Scan the dispatch trace log and count up the stolen time.
184 * Should be called with interrupts disabled.
185 */
186static u64 scan_dispatch_log(u64 stop_tb)
187{
188 u64 i = local_paca->dtl_ridx;
189 struct dtl_entry *dtl = local_paca->dtl_curr;
190 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
191 struct lppaca *vpa = local_paca->lppaca_ptr;
192 u64 tb_delta;
193 u64 stolen = 0;
194 u64 dtb;
195
196 if (!dtl)
197 return 0;
198
199 if (i == be64_to_cpu(vpa->dtl_idx))
200 return 0;
201 while (i < be64_to_cpu(vpa->dtl_idx)) {
202 dtb = be64_to_cpu(dtl->timebase);
203 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
204 be32_to_cpu(dtl->ready_to_enqueue_time);
205 barrier();
206 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
207 /* buffer has overflowed */
208 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
209 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
210 continue;
211 }
212 if (dtb > stop_tb)
213 break;
214 if (dtl_consumer)
215 dtl_consumer(dtl, i);
216 stolen += tb_delta;
217 ++i;
218 ++dtl;
219 if (dtl == dtl_end)
220 dtl = local_paca->dispatch_log;
221 }
222 local_paca->dtl_ridx = i;
223 local_paca->dtl_curr = dtl;
224 return stolen;
225}
226
227/*
228 * Accumulate stolen time by scanning the dispatch trace log.
229 * Called on entry from user mode.
230 */
231void notrace accumulate_stolen_time(void)
232{
233 u64 sst, ust;
234 struct cpu_accounting_data *acct = &local_paca->accounting;
235
236 sst = scan_dispatch_log(acct->starttime_user);
237 ust = scan_dispatch_log(acct->starttime);
238 acct->stime -= sst;
239 acct->utime -= ust;
240 acct->steal_time += ust + sst;
241}
242
243static inline u64 calculate_stolen_time(u64 stop_tb)
244{
245 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
246 return 0;
247
248 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
249 return scan_dispatch_log(stop_tb);
250
251 return 0;
252}
253
254#else /* CONFIG_PPC_SPLPAR */
255static inline u64 calculate_stolen_time(u64 stop_tb)
256{
257 return 0;
258}
259
260#endif /* CONFIG_PPC_SPLPAR */
261
262/*
263 * Account time for a transition between system, hard irq
264 * or soft irq state.
265 */
266static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
267 unsigned long now, unsigned long stime)
268{
269 unsigned long stime_scaled = 0;
270#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
271 unsigned long nowscaled, deltascaled;
272 unsigned long utime, utime_scaled;
273
274 nowscaled = read_spurr(now);
275 deltascaled = nowscaled - acct->startspurr;
276 acct->startspurr = nowscaled;
277 utime = acct->utime - acct->utime_sspurr;
278 acct->utime_sspurr = acct->utime;
279
280 /*
281 * Because we don't read the SPURR on every kernel entry/exit,
282 * deltascaled includes both user and system SPURR ticks.
283 * Apportion these ticks to system SPURR ticks and user
284 * SPURR ticks in the same ratio as the system time (delta)
285 * and user time (udelta) values obtained from the timebase
286 * over the same interval. The system ticks get accounted here;
287 * the user ticks get saved up in paca->user_time_scaled to be
288 * used by account_process_tick.
289 */
290 stime_scaled = stime;
291 utime_scaled = utime;
292 if (deltascaled != stime + utime) {
293 if (utime) {
294 stime_scaled = deltascaled * stime / (stime + utime);
295 utime_scaled = deltascaled - stime_scaled;
296 } else {
297 stime_scaled = deltascaled;
298 }
299 }
300 acct->utime_scaled += utime_scaled;
301#endif
302
303 return stime_scaled;
304}
305
306static unsigned long vtime_delta(struct cpu_accounting_data *acct,
307 unsigned long *stime_scaled,
308 unsigned long *steal_time)
309{
310 unsigned long now, stime;
311
312 WARN_ON_ONCE(!irqs_disabled());
313
314 now = mftb();
315 stime = now - acct->starttime;
316 acct->starttime = now;
317
318 *stime_scaled = vtime_delta_scaled(acct, now, stime);
319
320 *steal_time = calculate_stolen_time(now);
321
322 return stime;
323}
324
325static void vtime_delta_kernel(struct cpu_accounting_data *acct,
326 unsigned long *stime, unsigned long *stime_scaled)
327{
328 unsigned long steal_time;
329
330 *stime = vtime_delta(acct, stime_scaled, &steal_time);
331 *stime -= min(*stime, steal_time);
332 acct->steal_time += steal_time;
333}
334
335void vtime_account_kernel(struct task_struct *tsk)
336{
337 struct cpu_accounting_data *acct = get_accounting(tsk);
338 unsigned long stime, stime_scaled;
339
340 vtime_delta_kernel(acct, &stime, &stime_scaled);
341
342 if (tsk->flags & PF_VCPU) {
343 acct->gtime += stime;
344#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
345 acct->utime_scaled += stime_scaled;
346#endif
347 } else {
348 acct->stime += stime;
349#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
350 acct->stime_scaled += stime_scaled;
351#endif
352 }
353}
354EXPORT_SYMBOL_GPL(vtime_account_kernel);
355
356void vtime_account_idle(struct task_struct *tsk)
357{
358 unsigned long stime, stime_scaled, steal_time;
359 struct cpu_accounting_data *acct = get_accounting(tsk);
360
361 stime = vtime_delta(acct, &stime_scaled, &steal_time);
362 acct->idle_time += stime + steal_time;
363}
364
365static void vtime_account_irq_field(struct cpu_accounting_data *acct,
366 unsigned long *field)
367{
368 unsigned long stime, stime_scaled;
369
370 vtime_delta_kernel(acct, &stime, &stime_scaled);
371 *field += stime;
372#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
373 acct->stime_scaled += stime_scaled;
374#endif
375}
376
377void vtime_account_softirq(struct task_struct *tsk)
378{
379 struct cpu_accounting_data *acct = get_accounting(tsk);
380 vtime_account_irq_field(acct, &acct->softirq_time);
381}
382
383void vtime_account_hardirq(struct task_struct *tsk)
384{
385 struct cpu_accounting_data *acct = get_accounting(tsk);
386 vtime_account_irq_field(acct, &acct->hardirq_time);
387}
388
389static void vtime_flush_scaled(struct task_struct *tsk,
390 struct cpu_accounting_data *acct)
391{
392#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
393 if (acct->utime_scaled)
394 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
395 if (acct->stime_scaled)
396 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
397
398 acct->utime_scaled = 0;
399 acct->utime_sspurr = 0;
400 acct->stime_scaled = 0;
401#endif
402}
403
404/*
405 * Account the whole cputime accumulated in the paca
406 * Must be called with interrupts disabled.
407 * Assumes that vtime_account_kernel/idle() has been called
408 * recently (i.e. since the last entry from usermode) so that
409 * get_paca()->user_time_scaled is up to date.
410 */
411void vtime_flush(struct task_struct *tsk)
412{
413 struct cpu_accounting_data *acct = get_accounting(tsk);
414
415 if (acct->utime)
416 account_user_time(tsk, cputime_to_nsecs(acct->utime));
417
418 if (acct->gtime)
419 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
420
421 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
422 account_steal_time(cputime_to_nsecs(acct->steal_time));
423 acct->steal_time = 0;
424 }
425
426 if (acct->idle_time)
427 account_idle_time(cputime_to_nsecs(acct->idle_time));
428
429 if (acct->stime)
430 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
431 CPUTIME_SYSTEM);
432
433 if (acct->hardirq_time)
434 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
435 CPUTIME_IRQ);
436 if (acct->softirq_time)
437 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
438 CPUTIME_SOFTIRQ);
439
440 vtime_flush_scaled(tsk, acct);
441
442 acct->utime = 0;
443 acct->gtime = 0;
444 acct->idle_time = 0;
445 acct->stime = 0;
446 acct->hardirq_time = 0;
447 acct->softirq_time = 0;
448}
449
450#else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
451#define calc_cputime_factors()
452#endif
453
454void __delay(unsigned long loops)
455{
456 unsigned long start;
457
458 spin_begin();
459 if (tb_invalid) {
460 /*
461 * TB is in error state and isn't ticking anymore.
462 * HMI handler was unable to recover from TB error.
463 * Return immediately, so that kernel won't get stuck here.
464 */
465 spin_cpu_relax();
466 } else {
467 start = mftb();
468 while (mftb() - start < loops)
469 spin_cpu_relax();
470 }
471 spin_end();
472}
473EXPORT_SYMBOL(__delay);
474
475void udelay(unsigned long usecs)
476{
477 __delay(tb_ticks_per_usec * usecs);
478}
479EXPORT_SYMBOL(udelay);
480
481#ifdef CONFIG_SMP
482unsigned long profile_pc(struct pt_regs *regs)
483{
484 unsigned long pc = instruction_pointer(regs);
485
486 if (in_lock_functions(pc))
487 return regs->link;
488
489 return pc;
490}
491EXPORT_SYMBOL(profile_pc);
492#endif
493
494#ifdef CONFIG_IRQ_WORK
495
496/*
497 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
498 */
499#ifdef CONFIG_PPC64
500static inline void set_irq_work_pending_flag(void)
501{
502 asm volatile("stb %0,%1(13)" : :
503 "r" (1),
504 "i" (offsetof(struct paca_struct, irq_work_pending)));
505}
506
507static inline void clear_irq_work_pending(void)
508{
509 asm volatile("stb %0,%1(13)" : :
510 "r" (0),
511 "i" (offsetof(struct paca_struct, irq_work_pending)));
512}
513
514#else /* 32-bit */
515
516DEFINE_PER_CPU(u8, irq_work_pending);
517
518#define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
519#define test_irq_work_pending() __this_cpu_read(irq_work_pending)
520#define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
521
522#endif /* 32 vs 64 bit */
523
524void arch_irq_work_raise(void)
525{
526 /*
527 * 64-bit code that uses irq soft-mask can just cause an immediate
528 * interrupt here that gets soft masked, if this is called under
529 * local_irq_disable(). It might be possible to prevent that happening
530 * by noticing interrupts are disabled and setting decrementer pending
531 * to be replayed when irqs are enabled. The problem there is that
532 * tracing can call irq_work_raise, including in code that does low
533 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
534 * which could get tangled up if we're messing with the same state
535 * here.
536 */
537 preempt_disable();
538 set_irq_work_pending_flag();
539 set_dec(1);
540 preempt_enable();
541}
542
543#else /* CONFIG_IRQ_WORK */
544
545#define test_irq_work_pending() 0
546#define clear_irq_work_pending()
547
548#endif /* CONFIG_IRQ_WORK */
549
550/*
551 * timer_interrupt - gets called when the decrementer overflows,
552 * with interrupts disabled.
553 */
554DEFINE_INTERRUPT_HANDLER_ASYNC(timer_interrupt)
555{
556 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
557 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
558 struct pt_regs *old_regs;
559 u64 now;
560
561 /*
562 * Some implementations of hotplug will get timer interrupts while
563 * offline, just ignore these.
564 */
565 if (unlikely(!cpu_online(smp_processor_id()))) {
566 set_dec(decrementer_max);
567 return;
568 }
569
570 /* Ensure a positive value is written to the decrementer, or else
571 * some CPUs will continue to take decrementer exceptions. When the
572 * PPC_WATCHDOG (decrementer based) is configured, keep this at most
573 * 31 bits, which is about 4 seconds on most systems, which gives
574 * the watchdog a chance of catching timer interrupt hard lockups.
575 */
576 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
577 set_dec(0x7fffffff);
578 else
579 set_dec(decrementer_max);
580
581 /* Conditionally hard-enable interrupts now that the DEC has been
582 * bumped to its maximum value
583 */
584 may_hard_irq_enable();
585
586
587#if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
588 if (atomic_read(&ppc_n_lost_interrupts) != 0)
589 __do_IRQ(regs);
590#endif
591
592 old_regs = set_irq_regs(regs);
593
594 trace_timer_interrupt_entry(regs);
595
596 if (test_irq_work_pending()) {
597 clear_irq_work_pending();
598 irq_work_run();
599 }
600
601 now = get_tb();
602 if (now >= *next_tb) {
603 *next_tb = ~(u64)0;
604 if (evt->event_handler)
605 evt->event_handler(evt);
606 __this_cpu_inc(irq_stat.timer_irqs_event);
607 } else {
608 now = *next_tb - now;
609 if (now <= decrementer_max)
610 set_dec(now);
611 /* We may have raced with new irq work */
612 if (test_irq_work_pending())
613 set_dec(1);
614 __this_cpu_inc(irq_stat.timer_irqs_others);
615 }
616
617 trace_timer_interrupt_exit(regs);
618
619 set_irq_regs(old_regs);
620}
621EXPORT_SYMBOL(timer_interrupt);
622
623#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
624void timer_broadcast_interrupt(void)
625{
626 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
627
628 *next_tb = ~(u64)0;
629 tick_receive_broadcast();
630 __this_cpu_inc(irq_stat.broadcast_irqs_event);
631}
632#endif
633
634#ifdef CONFIG_SUSPEND
635static void generic_suspend_disable_irqs(void)
636{
637 /* Disable the decrementer, so that it doesn't interfere
638 * with suspending.
639 */
640
641 set_dec(decrementer_max);
642 local_irq_disable();
643 set_dec(decrementer_max);
644}
645
646static void generic_suspend_enable_irqs(void)
647{
648 local_irq_enable();
649}
650
651/* Overrides the weak version in kernel/power/main.c */
652void arch_suspend_disable_irqs(void)
653{
654 if (ppc_md.suspend_disable_irqs)
655 ppc_md.suspend_disable_irqs();
656 generic_suspend_disable_irqs();
657}
658
659/* Overrides the weak version in kernel/power/main.c */
660void arch_suspend_enable_irqs(void)
661{
662 generic_suspend_enable_irqs();
663 if (ppc_md.suspend_enable_irqs)
664 ppc_md.suspend_enable_irqs();
665}
666#endif
667
668unsigned long long tb_to_ns(unsigned long long ticks)
669{
670 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
671}
672EXPORT_SYMBOL_GPL(tb_to_ns);
673
674/*
675 * Scheduler clock - returns current time in nanosec units.
676 *
677 * Note: mulhdu(a, b) (multiply high double unsigned) returns
678 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
679 * are 64-bit unsigned numbers.
680 */
681notrace unsigned long long sched_clock(void)
682{
683 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
684}
685
686
687#ifdef CONFIG_PPC_PSERIES
688
689/*
690 * Running clock - attempts to give a view of time passing for a virtualised
691 * kernels.
692 * Uses the VTB register if available otherwise a next best guess.
693 */
694unsigned long long running_clock(void)
695{
696 /*
697 * Don't read the VTB as a host since KVM does not switch in host
698 * timebase into the VTB when it takes a guest off the CPU, reading the
699 * VTB would result in reading 'last switched out' guest VTB.
700 *
701 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
702 * would be unsafe to rely only on the #ifdef above.
703 */
704 if (firmware_has_feature(FW_FEATURE_LPAR) &&
705 cpu_has_feature(CPU_FTR_ARCH_207S))
706 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
707
708 /*
709 * This is a next best approximation without a VTB.
710 * On a host which is running bare metal there should never be any stolen
711 * time and on a host which doesn't do any virtualisation TB *should* equal
712 * VTB so it makes no difference anyway.
713 */
714 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
715}
716#endif
717
718static int __init get_freq(char *name, int cells, unsigned long *val)
719{
720 struct device_node *cpu;
721 const __be32 *fp;
722 int found = 0;
723
724 /* The cpu node should have timebase and clock frequency properties */
725 cpu = of_find_node_by_type(NULL, "cpu");
726
727 if (cpu) {
728 fp = of_get_property(cpu, name, NULL);
729 if (fp) {
730 found = 1;
731 *val = of_read_ulong(fp, cells);
732 }
733
734 of_node_put(cpu);
735 }
736
737 return found;
738}
739
740static void start_cpu_decrementer(void)
741{
742#if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
743 unsigned int tcr;
744
745 /* Clear any pending timer interrupts */
746 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
747
748 tcr = mfspr(SPRN_TCR);
749 /*
750 * The watchdog may have already been enabled by u-boot. So leave
751 * TRC[WP] (Watchdog Period) alone.
752 */
753 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
754 tcr |= TCR_DIE; /* Enable decrementer */
755 mtspr(SPRN_TCR, tcr);
756#endif
757}
758
759void __init generic_calibrate_decr(void)
760{
761 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
762
763 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
764 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
765
766 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
767 "(not found)\n");
768 }
769
770 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
771
772 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
773 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
774
775 printk(KERN_ERR "WARNING: Estimating processor frequency "
776 "(not found)\n");
777 }
778}
779
780int update_persistent_clock64(struct timespec64 now)
781{
782 struct rtc_time tm;
783
784 if (!ppc_md.set_rtc_time)
785 return -ENODEV;
786
787 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
788
789 return ppc_md.set_rtc_time(&tm);
790}
791
792static void __read_persistent_clock(struct timespec64 *ts)
793{
794 struct rtc_time tm;
795 static int first = 1;
796
797 ts->tv_nsec = 0;
798 /* XXX this is a litle fragile but will work okay in the short term */
799 if (first) {
800 first = 0;
801 if (ppc_md.time_init)
802 timezone_offset = ppc_md.time_init();
803
804 /* get_boot_time() isn't guaranteed to be safe to call late */
805 if (ppc_md.get_boot_time) {
806 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
807 return;
808 }
809 }
810 if (!ppc_md.get_rtc_time) {
811 ts->tv_sec = 0;
812 return;
813 }
814 ppc_md.get_rtc_time(&tm);
815
816 ts->tv_sec = rtc_tm_to_time64(&tm);
817}
818
819void read_persistent_clock64(struct timespec64 *ts)
820{
821 __read_persistent_clock(ts);
822
823 /* Sanitize it in case real time clock is set below EPOCH */
824 if (ts->tv_sec < 0) {
825 ts->tv_sec = 0;
826 ts->tv_nsec = 0;
827 }
828
829}
830
831/* clocksource code */
832static notrace u64 timebase_read(struct clocksource *cs)
833{
834 return (u64)get_tb();
835}
836
837static void __init clocksource_init(void)
838{
839 struct clocksource *clock = &clocksource_timebase;
840
841 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
842 printk(KERN_ERR "clocksource: %s is already registered\n",
843 clock->name);
844 return;
845 }
846
847 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
848 clock->name, clock->mult, clock->shift);
849}
850
851static int decrementer_set_next_event(unsigned long evt,
852 struct clock_event_device *dev)
853{
854 __this_cpu_write(decrementers_next_tb, get_tb() + evt);
855 set_dec(evt);
856
857 /* We may have raced with new irq work */
858 if (test_irq_work_pending())
859 set_dec(1);
860
861 return 0;
862}
863
864static int decrementer_shutdown(struct clock_event_device *dev)
865{
866 decrementer_set_next_event(decrementer_max, dev);
867 return 0;
868}
869
870static void register_decrementer_clockevent(int cpu)
871{
872 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
873
874 *dec = decrementer_clockevent;
875 dec->cpumask = cpumask_of(cpu);
876
877 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
878
879 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
880 dec->name, dec->mult, dec->shift, cpu);
881
882 /* Set values for KVM, see kvm_emulate_dec() */
883 decrementer_clockevent.mult = dec->mult;
884 decrementer_clockevent.shift = dec->shift;
885}
886
887static void enable_large_decrementer(void)
888{
889 if (!cpu_has_feature(CPU_FTR_ARCH_300))
890 return;
891
892 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
893 return;
894
895 /*
896 * If we're running as the hypervisor we need to enable the LD manually
897 * otherwise firmware should have done it for us.
898 */
899 if (cpu_has_feature(CPU_FTR_HVMODE))
900 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
901}
902
903static void __init set_decrementer_max(void)
904{
905 struct device_node *cpu;
906 u32 bits = 32;
907
908 /* Prior to ISAv3 the decrementer is always 32 bit */
909 if (!cpu_has_feature(CPU_FTR_ARCH_300))
910 return;
911
912 cpu = of_find_node_by_type(NULL, "cpu");
913
914 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
915 if (bits > 64 || bits < 32) {
916 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
917 bits = 32;
918 }
919
920 /* calculate the signed maximum given this many bits */
921 decrementer_max = (1ul << (bits - 1)) - 1;
922 }
923
924 of_node_put(cpu);
925
926 pr_info("time_init: %u bit decrementer (max: %llx)\n",
927 bits, decrementer_max);
928}
929
930static void __init init_decrementer_clockevent(void)
931{
932 register_decrementer_clockevent(smp_processor_id());
933}
934
935void secondary_cpu_time_init(void)
936{
937 /* Enable and test the large decrementer for this cpu */
938 enable_large_decrementer();
939
940 /* Start the decrementer on CPUs that have manual control
941 * such as BookE
942 */
943 start_cpu_decrementer();
944
945 /* FIME: Should make unrelatred change to move snapshot_timebase
946 * call here ! */
947 register_decrementer_clockevent(smp_processor_id());
948}
949
950/* This function is only called on the boot processor */
951void __init time_init(void)
952{
953 struct div_result res;
954 u64 scale;
955 unsigned shift;
956
957 /* Normal PowerPC with timebase register */
958 ppc_md.calibrate_decr();
959 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
960 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
961 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
962 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
963
964 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
965 tb_ticks_per_sec = ppc_tb_freq;
966 tb_ticks_per_usec = ppc_tb_freq / 1000000;
967 calc_cputime_factors();
968
969 /*
970 * Compute scale factor for sched_clock.
971 * The calibrate_decr() function has set tb_ticks_per_sec,
972 * which is the timebase frequency.
973 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
974 * the 128-bit result as a 64.64 fixed-point number.
975 * We then shift that number right until it is less than 1.0,
976 * giving us the scale factor and shift count to use in
977 * sched_clock().
978 */
979 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
980 scale = res.result_low;
981 for (shift = 0; res.result_high != 0; ++shift) {
982 scale = (scale >> 1) | (res.result_high << 63);
983 res.result_high >>= 1;
984 }
985 tb_to_ns_scale = scale;
986 tb_to_ns_shift = shift;
987 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
988 boot_tb = get_tb();
989
990 /* If platform provided a timezone (pmac), we correct the time */
991 if (timezone_offset) {
992 sys_tz.tz_minuteswest = -timezone_offset / 60;
993 sys_tz.tz_dsttime = 0;
994 }
995
996 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
997
998 /* initialise and enable the large decrementer (if we have one) */
999 set_decrementer_max();
1000 enable_large_decrementer();
1001
1002 /* Start the decrementer on CPUs that have manual control
1003 * such as BookE
1004 */
1005 start_cpu_decrementer();
1006
1007 /* Register the clocksource */
1008 clocksource_init();
1009
1010 init_decrementer_clockevent();
1011 tick_setup_hrtimer_broadcast();
1012
1013 of_clk_init(NULL);
1014 enable_sched_clock_irqtime();
1015}
1016
1017/*
1018 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1019 * result.
1020 */
1021void div128_by_32(u64 dividend_high, u64 dividend_low,
1022 unsigned divisor, struct div_result *dr)
1023{
1024 unsigned long a, b, c, d;
1025 unsigned long w, x, y, z;
1026 u64 ra, rb, rc;
1027
1028 a = dividend_high >> 32;
1029 b = dividend_high & 0xffffffff;
1030 c = dividend_low >> 32;
1031 d = dividend_low & 0xffffffff;
1032
1033 w = a / divisor;
1034 ra = ((u64)(a - (w * divisor)) << 32) + b;
1035
1036 rb = ((u64) do_div(ra, divisor) << 32) + c;
1037 x = ra;
1038
1039 rc = ((u64) do_div(rb, divisor) << 32) + d;
1040 y = rb;
1041
1042 do_div(rc, divisor);
1043 z = rc;
1044
1045 dr->result_high = ((u64)w << 32) + x;
1046 dr->result_low = ((u64)y << 32) + z;
1047
1048}
1049
1050/* We don't need to calibrate delay, we use the CPU timebase for that */
1051void calibrate_delay(void)
1052{
1053 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1054 * as the number of __delay(1) in a jiffy, so make it so
1055 */
1056 loops_per_jiffy = tb_ticks_per_jiffy;
1057}
1058
1059#if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1060static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1061{
1062 ppc_md.get_rtc_time(tm);
1063 return 0;
1064}
1065
1066static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1067{
1068 if (!ppc_md.set_rtc_time)
1069 return -EOPNOTSUPP;
1070
1071 if (ppc_md.set_rtc_time(tm) < 0)
1072 return -EOPNOTSUPP;
1073
1074 return 0;
1075}
1076
1077static const struct rtc_class_ops rtc_generic_ops = {
1078 .read_time = rtc_generic_get_time,
1079 .set_time = rtc_generic_set_time,
1080};
1081
1082static int __init rtc_init(void)
1083{
1084 struct platform_device *pdev;
1085
1086 if (!ppc_md.get_rtc_time)
1087 return -ENODEV;
1088
1089 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1090 &rtc_generic_ops,
1091 sizeof(rtc_generic_ops));
1092
1093 return PTR_ERR_OR_ZERO(pdev);
1094}
1095
1096device_initcall(rtc_init);
1097#endif
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 <linux/mc146818rtc.h>
58#include <linux/platform_device.h>
59
60#include <asm/trace.h>
61#include <asm/interrupt.h>
62#include <asm/io.h>
63#include <asm/nvram.h>
64#include <asm/cache.h>
65#include <asm/machdep.h>
66#include <linux/uaccess.h>
67#include <asm/time.h>
68#include <asm/irq.h>
69#include <asm/div64.h>
70#include <asm/smp.h>
71#include <asm/vdso_datapage.h>
72#include <asm/firmware.h>
73#include <asm/mce.h>
74#include <asm/systemcfg.h>
75
76/* powerpc clocksource/clockevent code */
77
78#include <linux/clockchips.h>
79
80static u64 timebase_read(struct clocksource *);
81static struct clocksource clocksource_timebase = {
82 .name = "timebase",
83 .rating = 400,
84 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
85 .mask = CLOCKSOURCE_MASK(64),
86 .read = timebase_read,
87 .vdso_clock_mode = VDSO_CLOCKMODE_ARCHTIMER,
88};
89
90#define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
91u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
92EXPORT_SYMBOL_GPL(decrementer_max); /* for KVM HDEC */
93
94static int decrementer_set_next_event(unsigned long evt,
95 struct clock_event_device *dev);
96static int decrementer_shutdown(struct clock_event_device *evt);
97
98struct clock_event_device decrementer_clockevent = {
99 .name = "decrementer",
100 .rating = 200,
101 .irq = 0,
102 .set_next_event = decrementer_set_next_event,
103 .set_state_oneshot_stopped = decrementer_shutdown,
104 .set_state_shutdown = decrementer_shutdown,
105 .tick_resume = decrementer_shutdown,
106 .features = CLOCK_EVT_FEAT_ONESHOT |
107 CLOCK_EVT_FEAT_C3STOP,
108};
109EXPORT_SYMBOL(decrementer_clockevent);
110
111/*
112 * This always puts next_tb beyond now, so the clock event will never fire
113 * with the usual comparison, no need for a separate test for stopped.
114 */
115#define DEC_CLOCKEVENT_STOPPED ~0ULL
116DEFINE_PER_CPU(u64, decrementers_next_tb) = DEC_CLOCKEVENT_STOPPED;
117EXPORT_SYMBOL_GPL(decrementers_next_tb);
118static DEFINE_PER_CPU(struct clock_event_device, decrementers);
119
120#define XSEC_PER_SEC (1024*1024)
121
122#ifdef CONFIG_PPC64
123#define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC)
124#else
125/* compute ((xsec << 12) * max) >> 32 */
126#define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max)
127#endif
128
129unsigned long tb_ticks_per_jiffy;
130unsigned long tb_ticks_per_usec = 100; /* sane default */
131EXPORT_SYMBOL(tb_ticks_per_usec);
132unsigned long tb_ticks_per_sec;
133EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime conversions */
134
135DEFINE_SPINLOCK(rtc_lock);
136EXPORT_SYMBOL_GPL(rtc_lock);
137
138static u64 tb_to_ns_scale __read_mostly;
139static unsigned tb_to_ns_shift __read_mostly;
140static u64 boot_tb __read_mostly;
141
142extern struct timezone sys_tz;
143static long timezone_offset;
144
145unsigned long ppc_proc_freq;
146EXPORT_SYMBOL_GPL(ppc_proc_freq);
147unsigned long ppc_tb_freq;
148EXPORT_SYMBOL_GPL(ppc_tb_freq);
149
150bool tb_invalid;
151
152#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
153/*
154 * Read the SPURR on systems that have it, otherwise the PURR,
155 * or if that doesn't exist return the timebase value passed in.
156 */
157static inline unsigned long read_spurr(unsigned long tb)
158{
159 if (cpu_has_feature(CPU_FTR_SPURR))
160 return mfspr(SPRN_SPURR);
161 if (cpu_has_feature(CPU_FTR_PURR))
162 return mfspr(SPRN_PURR);
163 return tb;
164}
165
166/*
167 * Account time for a transition between system, hard irq
168 * or soft irq state.
169 */
170static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
171 unsigned long now, unsigned long stime)
172{
173 unsigned long stime_scaled = 0;
174#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
175 unsigned long nowscaled, deltascaled;
176 unsigned long utime, utime_scaled;
177
178 nowscaled = read_spurr(now);
179 deltascaled = nowscaled - acct->startspurr;
180 acct->startspurr = nowscaled;
181 utime = acct->utime - acct->utime_sspurr;
182 acct->utime_sspurr = acct->utime;
183
184 /*
185 * Because we don't read the SPURR on every kernel entry/exit,
186 * deltascaled includes both user and system SPURR ticks.
187 * Apportion these ticks to system SPURR ticks and user
188 * SPURR ticks in the same ratio as the system time (delta)
189 * and user time (udelta) values obtained from the timebase
190 * over the same interval. The system ticks get accounted here;
191 * the user ticks get saved up in paca->user_time_scaled to be
192 * used by account_process_tick.
193 */
194 stime_scaled = stime;
195 utime_scaled = utime;
196 if (deltascaled != stime + utime) {
197 if (utime) {
198 stime_scaled = deltascaled * stime / (stime + utime);
199 utime_scaled = deltascaled - stime_scaled;
200 } else {
201 stime_scaled = deltascaled;
202 }
203 }
204 acct->utime_scaled += utime_scaled;
205#endif
206
207 return stime_scaled;
208}
209
210static unsigned long vtime_delta(struct cpu_accounting_data *acct,
211 unsigned long *stime_scaled,
212 unsigned long *steal_time)
213{
214 unsigned long now, stime;
215
216 WARN_ON_ONCE(!irqs_disabled());
217
218 now = mftb();
219 stime = now - acct->starttime;
220 acct->starttime = now;
221
222 *stime_scaled = vtime_delta_scaled(acct, now, stime);
223
224 if (IS_ENABLED(CONFIG_PPC_SPLPAR) &&
225 firmware_has_feature(FW_FEATURE_SPLPAR))
226 *steal_time = pseries_calculate_stolen_time(now);
227 else
228 *steal_time = 0;
229
230 return stime;
231}
232
233static void vtime_delta_kernel(struct cpu_accounting_data *acct,
234 unsigned long *stime, unsigned long *stime_scaled)
235{
236 unsigned long steal_time;
237
238 *stime = vtime_delta(acct, stime_scaled, &steal_time);
239 *stime -= min(*stime, steal_time);
240 acct->steal_time += steal_time;
241}
242
243void vtime_account_kernel(struct task_struct *tsk)
244{
245 struct cpu_accounting_data *acct = get_accounting(tsk);
246 unsigned long stime, stime_scaled;
247
248 vtime_delta_kernel(acct, &stime, &stime_scaled);
249
250 if (tsk->flags & PF_VCPU) {
251 acct->gtime += stime;
252#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
253 acct->utime_scaled += stime_scaled;
254#endif
255 } else {
256 acct->stime += stime;
257#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
258 acct->stime_scaled += stime_scaled;
259#endif
260 }
261}
262EXPORT_SYMBOL_GPL(vtime_account_kernel);
263
264void vtime_account_idle(struct task_struct *tsk)
265{
266 unsigned long stime, stime_scaled, steal_time;
267 struct cpu_accounting_data *acct = get_accounting(tsk);
268
269 stime = vtime_delta(acct, &stime_scaled, &steal_time);
270 acct->idle_time += stime + steal_time;
271}
272
273static void vtime_account_irq_field(struct cpu_accounting_data *acct,
274 unsigned long *field)
275{
276 unsigned long stime, stime_scaled;
277
278 vtime_delta_kernel(acct, &stime, &stime_scaled);
279 *field += stime;
280#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
281 acct->stime_scaled += stime_scaled;
282#endif
283}
284
285void vtime_account_softirq(struct task_struct *tsk)
286{
287 struct cpu_accounting_data *acct = get_accounting(tsk);
288 vtime_account_irq_field(acct, &acct->softirq_time);
289}
290
291void vtime_account_hardirq(struct task_struct *tsk)
292{
293 struct cpu_accounting_data *acct = get_accounting(tsk);
294 vtime_account_irq_field(acct, &acct->hardirq_time);
295}
296
297static void vtime_flush_scaled(struct task_struct *tsk,
298 struct cpu_accounting_data *acct)
299{
300#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
301 if (acct->utime_scaled)
302 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
303 if (acct->stime_scaled)
304 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
305
306 acct->utime_scaled = 0;
307 acct->utime_sspurr = 0;
308 acct->stime_scaled = 0;
309#endif
310}
311
312/*
313 * Account the whole cputime accumulated in the paca
314 * Must be called with interrupts disabled.
315 * Assumes that vtime_account_kernel/idle() has been called
316 * recently (i.e. since the last entry from usermode) so that
317 * get_paca()->user_time_scaled is up to date.
318 */
319void vtime_flush(struct task_struct *tsk)
320{
321 struct cpu_accounting_data *acct = get_accounting(tsk);
322
323 if (acct->utime)
324 account_user_time(tsk, cputime_to_nsecs(acct->utime));
325
326 if (acct->gtime)
327 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
328
329 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
330 account_steal_time(cputime_to_nsecs(acct->steal_time));
331 acct->steal_time = 0;
332 }
333
334 if (acct->idle_time)
335 account_idle_time(cputime_to_nsecs(acct->idle_time));
336
337 if (acct->stime)
338 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
339 CPUTIME_SYSTEM);
340
341 if (acct->hardirq_time)
342 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
343 CPUTIME_IRQ);
344 if (acct->softirq_time)
345 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
346 CPUTIME_SOFTIRQ);
347
348 vtime_flush_scaled(tsk, acct);
349
350 acct->utime = 0;
351 acct->gtime = 0;
352 acct->idle_time = 0;
353 acct->stime = 0;
354 acct->hardirq_time = 0;
355 acct->softirq_time = 0;
356}
357
358/*
359 * Called from the context switch with interrupts disabled, to charge all
360 * accumulated times to the current process, and to prepare accounting on
361 * the next process.
362 */
363void vtime_task_switch(struct task_struct *prev)
364{
365 if (is_idle_task(prev))
366 vtime_account_idle(prev);
367 else
368 vtime_account_kernel(prev);
369
370 vtime_flush(prev);
371
372 if (!IS_ENABLED(CONFIG_PPC64)) {
373 struct cpu_accounting_data *acct = get_accounting(current);
374 struct cpu_accounting_data *acct0 = get_accounting(prev);
375
376 acct->starttime = acct0->starttime;
377 }
378}
379#endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
380
381void __no_kcsan __delay(unsigned long loops)
382{
383 unsigned long start;
384
385 spin_begin();
386 if (tb_invalid) {
387 /*
388 * TB is in error state and isn't ticking anymore.
389 * HMI handler was unable to recover from TB error.
390 * Return immediately, so that kernel won't get stuck here.
391 */
392 spin_cpu_relax();
393 } else {
394 start = mftb();
395 while (mftb() - start < loops)
396 spin_cpu_relax();
397 }
398 spin_end();
399}
400EXPORT_SYMBOL(__delay);
401
402void __no_kcsan udelay(unsigned long usecs)
403{
404 __delay(tb_ticks_per_usec * usecs);
405}
406EXPORT_SYMBOL(udelay);
407
408#ifdef CONFIG_SMP
409unsigned long profile_pc(struct pt_regs *regs)
410{
411 unsigned long pc = instruction_pointer(regs);
412
413 if (in_lock_functions(pc))
414 return regs->link;
415
416 return pc;
417}
418EXPORT_SYMBOL(profile_pc);
419#endif
420
421#ifdef CONFIG_IRQ_WORK
422
423/*
424 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
425 */
426#ifdef CONFIG_PPC64
427static inline unsigned long test_irq_work_pending(void)
428{
429 unsigned long x;
430
431 asm volatile("lbz %0,%1(13)"
432 : "=r" (x)
433 : "i" (offsetof(struct paca_struct, irq_work_pending)));
434 return x;
435}
436
437static inline void set_irq_work_pending_flag(void)
438{
439 asm volatile("stb %0,%1(13)" : :
440 "r" (1),
441 "i" (offsetof(struct paca_struct, irq_work_pending)));
442}
443
444static inline void clear_irq_work_pending(void)
445{
446 asm volatile("stb %0,%1(13)" : :
447 "r" (0),
448 "i" (offsetof(struct paca_struct, irq_work_pending)));
449}
450
451#else /* 32-bit */
452
453DEFINE_PER_CPU(u8, irq_work_pending);
454
455#define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1)
456#define test_irq_work_pending() __this_cpu_read(irq_work_pending)
457#define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0)
458
459#endif /* 32 vs 64 bit */
460
461void arch_irq_work_raise(void)
462{
463 /*
464 * 64-bit code that uses irq soft-mask can just cause an immediate
465 * interrupt here that gets soft masked, if this is called under
466 * local_irq_disable(). It might be possible to prevent that happening
467 * by noticing interrupts are disabled and setting decrementer pending
468 * to be replayed when irqs are enabled. The problem there is that
469 * tracing can call irq_work_raise, including in code that does low
470 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
471 * which could get tangled up if we're messing with the same state
472 * here.
473 */
474 preempt_disable();
475 set_irq_work_pending_flag();
476 set_dec(1);
477 preempt_enable();
478}
479
480static void set_dec_or_work(u64 val)
481{
482 set_dec(val);
483 /* We may have raced with new irq work */
484 if (unlikely(test_irq_work_pending()))
485 set_dec(1);
486}
487
488#else /* CONFIG_IRQ_WORK */
489
490#define test_irq_work_pending() 0
491#define clear_irq_work_pending()
492
493static void set_dec_or_work(u64 val)
494{
495 set_dec(val);
496}
497#endif /* CONFIG_IRQ_WORK */
498
499#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
500void timer_rearm_host_dec(u64 now)
501{
502 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
503
504 WARN_ON_ONCE(!arch_irqs_disabled());
505 WARN_ON_ONCE(mfmsr() & MSR_EE);
506
507 if (now >= *next_tb) {
508 local_paca->irq_happened |= PACA_IRQ_DEC;
509 } else {
510 now = *next_tb - now;
511 if (now > decrementer_max)
512 now = decrementer_max;
513 set_dec_or_work(now);
514 }
515}
516EXPORT_SYMBOL_GPL(timer_rearm_host_dec);
517#endif
518
519/*
520 * timer_interrupt - gets called when the decrementer overflows,
521 * with interrupts disabled.
522 */
523DEFINE_INTERRUPT_HANDLER_ASYNC(timer_interrupt)
524{
525 struct clock_event_device *evt = this_cpu_ptr(&decrementers);
526 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
527 struct pt_regs *old_regs;
528 u64 now;
529
530 /*
531 * Some implementations of hotplug will get timer interrupts while
532 * offline, just ignore these.
533 */
534 if (unlikely(!cpu_online(smp_processor_id()))) {
535 set_dec(decrementer_max);
536 return;
537 }
538
539 /* Conditionally hard-enable interrupts. */
540 if (should_hard_irq_enable(regs)) {
541 /*
542 * Ensure a positive value is written to the decrementer, or
543 * else some CPUs will continue to take decrementer exceptions.
544 * When the PPC_WATCHDOG (decrementer based) is configured,
545 * keep this at most 31 bits, which is about 4 seconds on most
546 * systems, which gives the watchdog a chance of catching timer
547 * interrupt hard lockups.
548 */
549 if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
550 set_dec(0x7fffffff);
551 else
552 set_dec(decrementer_max);
553
554 do_hard_irq_enable();
555 }
556
557#if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
558 if (atomic_read(&ppc_n_lost_interrupts) != 0)
559 __do_IRQ(regs);
560#endif
561
562 old_regs = set_irq_regs(regs);
563
564 trace_timer_interrupt_entry(regs);
565
566 if (test_irq_work_pending()) {
567 clear_irq_work_pending();
568 mce_run_irq_context_handlers();
569 irq_work_run();
570 }
571
572 now = get_tb();
573 if (now >= *next_tb) {
574 evt->event_handler(evt);
575 __this_cpu_inc(irq_stat.timer_irqs_event);
576 } else {
577 now = *next_tb - now;
578 if (now > decrementer_max)
579 now = decrementer_max;
580 set_dec_or_work(now);
581 __this_cpu_inc(irq_stat.timer_irqs_others);
582 }
583
584 trace_timer_interrupt_exit(regs);
585
586 set_irq_regs(old_regs);
587}
588EXPORT_SYMBOL(timer_interrupt);
589
590#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
591void timer_broadcast_interrupt(void)
592{
593 tick_receive_broadcast();
594 __this_cpu_inc(irq_stat.broadcast_irqs_event);
595}
596#endif
597
598#ifdef CONFIG_SUSPEND
599/* Overrides the weak version in kernel/power/main.c */
600void arch_suspend_disable_irqs(void)
601{
602 if (ppc_md.suspend_disable_irqs)
603 ppc_md.suspend_disable_irqs();
604
605 /* Disable the decrementer, so that it doesn't interfere
606 * with suspending.
607 */
608
609 set_dec(decrementer_max);
610 local_irq_disable();
611 set_dec(decrementer_max);
612}
613
614/* Overrides the weak version in kernel/power/main.c */
615void arch_suspend_enable_irqs(void)
616{
617 local_irq_enable();
618
619 if (ppc_md.suspend_enable_irqs)
620 ppc_md.suspend_enable_irqs();
621}
622#endif
623
624unsigned long long tb_to_ns(unsigned long long ticks)
625{
626 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
627}
628EXPORT_SYMBOL_GPL(tb_to_ns);
629
630/*
631 * Scheduler clock - returns current time in nanosec units.
632 *
633 * Note: mulhdu(a, b) (multiply high double unsigned) returns
634 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
635 * are 64-bit unsigned numbers.
636 */
637notrace unsigned long long sched_clock(void)
638{
639 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
640}
641
642
643#ifdef CONFIG_PPC_PSERIES
644
645/*
646 * Running clock - attempts to give a view of time passing for a virtualised
647 * kernels.
648 * Uses the VTB register if available otherwise a next best guess.
649 */
650unsigned long long running_clock(void)
651{
652 /*
653 * Don't read the VTB as a host since KVM does not switch in host
654 * timebase into the VTB when it takes a guest off the CPU, reading the
655 * VTB would result in reading 'last switched out' guest VTB.
656 *
657 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
658 * would be unsafe to rely only on the #ifdef above.
659 */
660 if (firmware_has_feature(FW_FEATURE_LPAR) &&
661 cpu_has_feature(CPU_FTR_ARCH_207S))
662 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
663
664 /*
665 * This is a next best approximation without a VTB.
666 * On a host which is running bare metal there should never be any stolen
667 * time and on a host which doesn't do any virtualisation TB *should* equal
668 * VTB so it makes no difference anyway.
669 */
670 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
671}
672#endif
673
674static int __init get_freq(char *name, int cells, unsigned long *val)
675{
676 struct device_node *cpu;
677 const __be32 *fp;
678 int found = 0;
679
680 /* The cpu node should have timebase and clock frequency properties */
681 cpu = of_find_node_by_type(NULL, "cpu");
682
683 if (cpu) {
684 fp = of_get_property(cpu, name, NULL);
685 if (fp) {
686 found = 1;
687 *val = of_read_ulong(fp, cells);
688 }
689
690 of_node_put(cpu);
691 }
692
693 return found;
694}
695
696static void start_cpu_decrementer(void)
697{
698#ifdef CONFIG_BOOKE
699 unsigned int tcr;
700
701 /* Clear any pending timer interrupts */
702 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
703
704 tcr = mfspr(SPRN_TCR);
705 /*
706 * The watchdog may have already been enabled by u-boot. So leave
707 * TRC[WP] (Watchdog Period) alone.
708 */
709 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */
710 tcr |= TCR_DIE; /* Enable decrementer */
711 mtspr(SPRN_TCR, tcr);
712#endif
713}
714
715void __init generic_calibrate_decr(void)
716{
717 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */
718
719 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
720 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
721
722 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
723 "(not found)\n");
724 }
725
726 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */
727
728 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
729 !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
730
731 printk(KERN_ERR "WARNING: Estimating processor frequency "
732 "(not found)\n");
733 }
734}
735
736int update_persistent_clock64(struct timespec64 now)
737{
738 struct rtc_time tm;
739
740 if (!ppc_md.set_rtc_time)
741 return -ENODEV;
742
743 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
744
745 return ppc_md.set_rtc_time(&tm);
746}
747
748static void __read_persistent_clock(struct timespec64 *ts)
749{
750 struct rtc_time tm;
751 static int first = 1;
752
753 ts->tv_nsec = 0;
754 /* XXX this is a little fragile but will work okay in the short term */
755 if (first) {
756 first = 0;
757 if (ppc_md.time_init)
758 timezone_offset = ppc_md.time_init();
759
760 /* get_boot_time() isn't guaranteed to be safe to call late */
761 if (ppc_md.get_boot_time) {
762 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
763 return;
764 }
765 }
766 if (!ppc_md.get_rtc_time) {
767 ts->tv_sec = 0;
768 return;
769 }
770 ppc_md.get_rtc_time(&tm);
771
772 ts->tv_sec = rtc_tm_to_time64(&tm);
773}
774
775void read_persistent_clock64(struct timespec64 *ts)
776{
777 __read_persistent_clock(ts);
778
779 /* Sanitize it in case real time clock is set below EPOCH */
780 if (ts->tv_sec < 0) {
781 ts->tv_sec = 0;
782 ts->tv_nsec = 0;
783 }
784
785}
786
787/* clocksource code */
788static notrace u64 timebase_read(struct clocksource *cs)
789{
790 return (u64)get_tb();
791}
792
793static void __init clocksource_init(void)
794{
795 struct clocksource *clock = &clocksource_timebase;
796
797 if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
798 printk(KERN_ERR "clocksource: %s is already registered\n",
799 clock->name);
800 return;
801 }
802
803 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
804 clock->name, clock->mult, clock->shift);
805}
806
807static int decrementer_set_next_event(unsigned long evt,
808 struct clock_event_device *dev)
809{
810 __this_cpu_write(decrementers_next_tb, get_tb() + evt);
811 set_dec_or_work(evt);
812
813 return 0;
814}
815
816static int decrementer_shutdown(struct clock_event_device *dev)
817{
818 __this_cpu_write(decrementers_next_tb, DEC_CLOCKEVENT_STOPPED);
819 set_dec_or_work(decrementer_max);
820
821 return 0;
822}
823
824static void register_decrementer_clockevent(int cpu)
825{
826 struct clock_event_device *dec = &per_cpu(decrementers, cpu);
827
828 *dec = decrementer_clockevent;
829 dec->cpumask = cpumask_of(cpu);
830
831 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
832
833 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
834 dec->name, dec->mult, dec->shift, cpu);
835
836 /* Set values for KVM, see kvm_emulate_dec() */
837 decrementer_clockevent.mult = dec->mult;
838 decrementer_clockevent.shift = dec->shift;
839}
840
841static void enable_large_decrementer(void)
842{
843 if (!cpu_has_feature(CPU_FTR_ARCH_300))
844 return;
845
846 if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
847 return;
848
849 /*
850 * If we're running as the hypervisor we need to enable the LD manually
851 * otherwise firmware should have done it for us.
852 */
853 if (cpu_has_feature(CPU_FTR_HVMODE))
854 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
855}
856
857static void __init set_decrementer_max(void)
858{
859 struct device_node *cpu;
860 u32 bits = 32;
861
862 /* Prior to ISAv3 the decrementer is always 32 bit */
863 if (!cpu_has_feature(CPU_FTR_ARCH_300))
864 return;
865
866 cpu = of_find_node_by_type(NULL, "cpu");
867
868 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
869 if (bits > 64 || bits < 32) {
870 pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
871 bits = 32;
872 }
873
874 /* calculate the signed maximum given this many bits */
875 decrementer_max = (1ul << (bits - 1)) - 1;
876 }
877
878 of_node_put(cpu);
879
880 pr_info("time_init: %u bit decrementer (max: %llx)\n",
881 bits, decrementer_max);
882}
883
884static void __init init_decrementer_clockevent(void)
885{
886 register_decrementer_clockevent(smp_processor_id());
887}
888
889void secondary_cpu_time_init(void)
890{
891 /* Enable and test the large decrementer for this cpu */
892 enable_large_decrementer();
893
894 /* Start the decrementer on CPUs that have manual control
895 * such as BookE
896 */
897 start_cpu_decrementer();
898
899 /* FIME: Should make unrelated change to move snapshot_timebase
900 * call here ! */
901 register_decrementer_clockevent(smp_processor_id());
902}
903
904/* This function is only called on the boot processor */
905void __init time_init(void)
906{
907 struct div_result res;
908 u64 scale;
909 unsigned shift;
910
911 /* Normal PowerPC with timebase register */
912 if (ppc_md.calibrate_decr)
913 ppc_md.calibrate_decr();
914 else
915 generic_calibrate_decr();
916
917 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
918 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
919 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n",
920 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
921
922 tb_ticks_per_jiffy = ppc_tb_freq / HZ;
923 tb_ticks_per_sec = ppc_tb_freq;
924 tb_ticks_per_usec = ppc_tb_freq / 1000000;
925
926 /*
927 * Compute scale factor for sched_clock.
928 * The calibrate_decr() function has set tb_ticks_per_sec,
929 * which is the timebase frequency.
930 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
931 * the 128-bit result as a 64.64 fixed-point number.
932 * We then shift that number right until it is less than 1.0,
933 * giving us the scale factor and shift count to use in
934 * sched_clock().
935 */
936 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
937 scale = res.result_low;
938 for (shift = 0; res.result_high != 0; ++shift) {
939 scale = (scale >> 1) | (res.result_high << 63);
940 res.result_high >>= 1;
941 }
942 tb_to_ns_scale = scale;
943 tb_to_ns_shift = shift;
944 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
945 boot_tb = get_tb();
946
947 /* If platform provided a timezone (pmac), we correct the time */
948 if (timezone_offset) {
949 sys_tz.tz_minuteswest = -timezone_offset / 60;
950 sys_tz.tz_dsttime = 0;
951 }
952
953 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
954#ifdef CONFIG_PPC64_PROC_SYSTEMCFG
955 systemcfg->tb_ticks_per_sec = tb_ticks_per_sec;
956#endif
957
958 /* initialise and enable the large decrementer (if we have one) */
959 set_decrementer_max();
960 enable_large_decrementer();
961
962 /* Start the decrementer on CPUs that have manual control
963 * such as BookE
964 */
965 start_cpu_decrementer();
966
967 /* Register the clocksource */
968 clocksource_init();
969
970 init_decrementer_clockevent();
971 tick_setup_hrtimer_broadcast();
972
973 of_clk_init(NULL);
974 enable_sched_clock_irqtime();
975}
976
977/*
978 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
979 * result.
980 */
981void div128_by_32(u64 dividend_high, u64 dividend_low,
982 unsigned divisor, struct div_result *dr)
983{
984 unsigned long a, b, c, d;
985 unsigned long w, x, y, z;
986 u64 ra, rb, rc;
987
988 a = dividend_high >> 32;
989 b = dividend_high & 0xffffffff;
990 c = dividend_low >> 32;
991 d = dividend_low & 0xffffffff;
992
993 w = a / divisor;
994 ra = ((u64)(a - (w * divisor)) << 32) + b;
995
996 rb = ((u64) do_div(ra, divisor) << 32) + c;
997 x = ra;
998
999 rc = ((u64) do_div(rb, divisor) << 32) + d;
1000 y = rb;
1001
1002 do_div(rc, divisor);
1003 z = rc;
1004
1005 dr->result_high = ((u64)w << 32) + x;
1006 dr->result_low = ((u64)y << 32) + z;
1007
1008}
1009
1010/* We don't need to calibrate delay, we use the CPU timebase for that */
1011void calibrate_delay(void)
1012{
1013 /* Some generic code (such as spinlock debug) use loops_per_jiffy
1014 * as the number of __delay(1) in a jiffy, so make it so
1015 */
1016 loops_per_jiffy = tb_ticks_per_jiffy;
1017}
1018
1019#if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1020static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1021{
1022 ppc_md.get_rtc_time(tm);
1023 return 0;
1024}
1025
1026static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1027{
1028 if (!ppc_md.set_rtc_time)
1029 return -EOPNOTSUPP;
1030
1031 if (ppc_md.set_rtc_time(tm) < 0)
1032 return -EOPNOTSUPP;
1033
1034 return 0;
1035}
1036
1037static const struct rtc_class_ops rtc_generic_ops = {
1038 .read_time = rtc_generic_get_time,
1039 .set_time = rtc_generic_set_time,
1040};
1041
1042static int __init rtc_init(void)
1043{
1044 struct platform_device *pdev;
1045
1046 if (!ppc_md.get_rtc_time)
1047 return -ENODEV;
1048
1049 pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1050 &rtc_generic_ops,
1051 sizeof(rtc_generic_ops));
1052
1053 return PTR_ERR_OR_ZERO(pdev);
1054}
1055
1056device_initcall(rtc_init);
1057#endif