~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/powerpc/kernel/time.c

Version: ~ [ linux-6.1-rc7 ] ~ [ linux-6.0.10 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.80 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.156 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.225 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.267 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.300 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.334 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Common time routines among all ppc machines.
  3  *
  4  * Written by Cort Dougan (cort@cs.nmt.edu) to merge
  5  * Paul Mackerras' version and mine for PReP and Pmac.
  6  * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net).
  7  * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com)
  8  *
  9  * First round of bugfixes by Gabriel Paubert (paubert@iram.es)
 10  * to make clock more stable (2.4.0-test5). The only thing
 11  * that this code assumes is that the timebases have been synchronized
 12  * by firmware on SMP and are never stopped (never do sleep
 13  * on SMP then, nap and doze are OK).
 14  * 
 15  * Speeded up do_gettimeofday by getting rid of references to
 16  * xtime (which required locks for consistency). (mikejc@us.ibm.com)
 17  *
 18  * TODO (not necessarily in this file):
 19  * - improve precision and reproducibility of timebase frequency
 20  * measurement at boot time.
 21  * - for astronomical applications: add a new function to get
 22  * non ambiguous timestamps even around leap seconds. This needs
 23  * a new timestamp format and a good name.
 24  *
 25  * 1997-09-10  Updated NTP code according to technical memorandum Jan '96
 26  *             "A Kernel Model for Precision Timekeeping" by Dave Mills
 27  *
 28  *      This program is free software; you can redistribute it and/or
 29  *      modify it under the terms of the GNU General Public License
 30  *      as published by the Free Software Foundation; either version
 31  *      2 of the License, or (at your option) any later version.
 32  */
 33 
 34 #include <linux/errno.h>
 35 #include <linux/export.h>
 36 #include <linux/sched.h>
 37 #include <linux/sched/clock.h>
 38 #include <linux/kernel.h>
 39 #include <linux/param.h>
 40 #include <linux/string.h>
 41 #include <linux/mm.h>
 42 #include <linux/interrupt.h>
 43 #include <linux/timex.h>
 44 #include <linux/kernel_stat.h>
 45 #include <linux/time.h>
 46 #include <linux/clockchips.h>
 47 #include <linux/init.h>
 48 #include <linux/profile.h>
 49 #include <linux/cpu.h>
 50 #include <linux/security.h>
 51 #include <linux/percpu.h>
 52 #include <linux/rtc.h>
 53 #include <linux/jiffies.h>
 54 #include <linux/posix-timers.h>
 55 #include <linux/irq.h>
 56 #include <linux/delay.h>
 57 #include <linux/irq_work.h>
 58 #include <linux/clk-provider.h>
 59 #include <linux/suspend.h>
 60 #include <linux/rtc.h>
 61 #include <linux/sched/cputime.h>
 62 #include <linux/processor.h>
 63 #include <asm/trace.h>
 64 
 65 #include <asm/io.h>
 66 #include <asm/nvram.h>
 67 #include <asm/cache.h>
 68 #include <asm/machdep.h>
 69 #include <linux/uaccess.h>
 70 #include <asm/time.h>
 71 #include <asm/prom.h>
 72 #include <asm/irq.h>
 73 #include <asm/div64.h>
 74 #include <asm/smp.h>
 75 #include <asm/vdso_datapage.h>
 76 #include <asm/firmware.h>
 77 #include <asm/asm-prototypes.h>
 78 
 79 /* powerpc clocksource/clockevent code */
 80 
 81 #include <linux/clockchips.h>
 82 #include <linux/timekeeper_internal.h>
 83 
 84 static u64 rtc_read(struct clocksource *);
 85 static struct clocksource clocksource_rtc = {
 86         .name         = "rtc",
 87         .rating       = 400,
 88         .flags        = CLOCK_SOURCE_IS_CONTINUOUS,
 89         .mask         = CLOCKSOURCE_MASK(64),
 90         .read         = rtc_read,
 91 };
 92 
 93 static u64 timebase_read(struct clocksource *);
 94 static struct clocksource clocksource_timebase = {
 95         .name         = "timebase",
 96         .rating       = 400,
 97         .flags        = CLOCK_SOURCE_IS_CONTINUOUS,
 98         .mask         = CLOCKSOURCE_MASK(64),
 99         .read         = timebase_read,
100 };
101 
102 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
103 u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
104 
105 static int decrementer_set_next_event(unsigned long evt,
106                                       struct clock_event_device *dev);
107 static int decrementer_shutdown(struct clock_event_device *evt);
108 
109 struct clock_event_device decrementer_clockevent = {
110         .name                   = "decrementer",
111         .rating                 = 200,
112         .irq                    = 0,
113         .set_next_event         = decrementer_set_next_event,
114         .set_state_oneshot_stopped = decrementer_shutdown,
115         .set_state_shutdown     = decrementer_shutdown,
116         .tick_resume            = decrementer_shutdown,
117         .features               = CLOCK_EVT_FEAT_ONESHOT |
118                                   CLOCK_EVT_FEAT_C3STOP,
119 };
120 EXPORT_SYMBOL(decrementer_clockevent);
121 
122 DEFINE_PER_CPU(u64, decrementers_next_tb);
123 static DEFINE_PER_CPU(struct clock_event_device, decrementers);
124 
125 #define XSEC_PER_SEC (1024*1024)
126 
127 #ifdef CONFIG_PPC64
128 #define SCALE_XSEC(xsec, max)   (((xsec) * max) / XSEC_PER_SEC)
129 #else
130 /* compute ((xsec << 12) * max) >> 32 */
131 #define SCALE_XSEC(xsec, max)   mulhwu((xsec) << 12, max)
132 #endif
133 
134 unsigned long tb_ticks_per_jiffy;
135 unsigned long tb_ticks_per_usec = 100; /* sane default */
136 EXPORT_SYMBOL(tb_ticks_per_usec);
137 unsigned long tb_ticks_per_sec;
138 EXPORT_SYMBOL(tb_ticks_per_sec);        /* for cputime_t conversions */
139 
140 DEFINE_SPINLOCK(rtc_lock);
141 EXPORT_SYMBOL_GPL(rtc_lock);
142 
143 static u64 tb_to_ns_scale __read_mostly;
144 static unsigned tb_to_ns_shift __read_mostly;
145 static u64 boot_tb __read_mostly;
146 
147 extern struct timezone sys_tz;
148 static long timezone_offset;
149 
150 unsigned long ppc_proc_freq;
151 EXPORT_SYMBOL_GPL(ppc_proc_freq);
152 unsigned long ppc_tb_freq;
153 EXPORT_SYMBOL_GPL(ppc_tb_freq);
154 
155 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
156 /*
157  * Factor for converting from cputime_t (timebase ticks) to
158  * microseconds. This is stored as 0.64 fixed-point binary fraction.
159  */
160 u64 __cputime_usec_factor;
161 EXPORT_SYMBOL(__cputime_usec_factor);
162 
163 #ifdef CONFIG_PPC_SPLPAR
164 void (*dtl_consumer)(struct dtl_entry *, u64);
165 #endif
166 
167 static void calc_cputime_factors(void)
168 {
169         struct div_result res;
170 
171         div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
172         __cputime_usec_factor = res.result_low;
173 }
174 
175 /*
176  * Read the SPURR on systems that have it, otherwise the PURR,
177  * or if that doesn't exist return the timebase value passed in.
178  */
179 static inline unsigned long read_spurr(unsigned long tb)
180 {
181         if (cpu_has_feature(CPU_FTR_SPURR))
182                 return mfspr(SPRN_SPURR);
183         if (cpu_has_feature(CPU_FTR_PURR))
184                 return mfspr(SPRN_PURR);
185         return tb;
186 }
187 
188 #ifdef CONFIG_PPC_SPLPAR
189 
190 /*
191  * Scan the dispatch trace log and count up the stolen time.
192  * Should be called with interrupts disabled.
193  */
194 static u64 scan_dispatch_log(u64 stop_tb)
195 {
196         u64 i = local_paca->dtl_ridx;
197         struct dtl_entry *dtl = local_paca->dtl_curr;
198         struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
199         struct lppaca *vpa = local_paca->lppaca_ptr;
200         u64 tb_delta;
201         u64 stolen = 0;
202         u64 dtb;
203 
204         if (!dtl)
205                 return 0;
206 
207         if (i == be64_to_cpu(vpa->dtl_idx))
208                 return 0;
209         while (i < be64_to_cpu(vpa->dtl_idx)) {
210                 dtb = be64_to_cpu(dtl->timebase);
211                 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
212                         be32_to_cpu(dtl->ready_to_enqueue_time);
213                 barrier();
214                 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
215                         /* buffer has overflowed */
216                         i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
217                         dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
218                         continue;
219                 }
220                 if (dtb > stop_tb)
221                         break;
222                 if (dtl_consumer)
223                         dtl_consumer(dtl, i);
224                 stolen += tb_delta;
225                 ++i;
226                 ++dtl;
227                 if (dtl == dtl_end)
228                         dtl = local_paca->dispatch_log;
229         }
230         local_paca->dtl_ridx = i;
231         local_paca->dtl_curr = dtl;
232         return stolen;
233 }
234 
235 /*
236  * Accumulate stolen time by scanning the dispatch trace log.
237  * Called on entry from user mode.
238  */
239 void accumulate_stolen_time(void)
240 {
241         u64 sst, ust;
242         unsigned long save_irq_soft_mask = irq_soft_mask_return();
243         struct cpu_accounting_data *acct = &local_paca->accounting;
244 
245         /* We are called early in the exception entry, before
246          * soft/hard_enabled are sync'ed to the expected state
247          * for the exception. We are hard disabled but the PACA
248          * needs to reflect that so various debug stuff doesn't
249          * complain
250          */
251         irq_soft_mask_set(IRQS_DISABLED);
252 
253         sst = scan_dispatch_log(acct->starttime_user);
254         ust = scan_dispatch_log(acct->starttime);
255         acct->stime -= sst;
256         acct->utime -= ust;
257         acct->steal_time += ust + sst;
258 
259         irq_soft_mask_set(save_irq_soft_mask);
260 }
261 
262 static inline u64 calculate_stolen_time(u64 stop_tb)
263 {
264         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
265                 return 0;
266 
267         if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx))
268                 return scan_dispatch_log(stop_tb);
269 
270         return 0;
271 }
272 
273 #else /* CONFIG_PPC_SPLPAR */
274 static inline u64 calculate_stolen_time(u64 stop_tb)
275 {
276         return 0;
277 }
278 
279 #endif /* CONFIG_PPC_SPLPAR */
280 
281 /*
282  * Account time for a transition between system, hard irq
283  * or soft irq state.
284  */
285 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
286                                         unsigned long now, unsigned long stime)
287 {
288         unsigned long stime_scaled = 0;
289 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
290         unsigned long nowscaled, deltascaled;
291         unsigned long utime, utime_scaled;
292 
293         nowscaled = read_spurr(now);
294         deltascaled = nowscaled - acct->startspurr;
295         acct->startspurr = nowscaled;
296         utime = acct->utime - acct->utime_sspurr;
297         acct->utime_sspurr = acct->utime;
298 
299         /*
300          * Because we don't read the SPURR on every kernel entry/exit,
301          * deltascaled includes both user and system SPURR ticks.
302          * Apportion these ticks to system SPURR ticks and user
303          * SPURR ticks in the same ratio as the system time (delta)
304          * and user time (udelta) values obtained from the timebase
305          * over the same interval.  The system ticks get accounted here;
306          * the user ticks get saved up in paca->user_time_scaled to be
307          * used by account_process_tick.
308          */
309         stime_scaled = stime;
310         utime_scaled = utime;
311         if (deltascaled != stime + utime) {
312                 if (utime) {
313                         stime_scaled = deltascaled * stime / (stime + utime);
314                         utime_scaled = deltascaled - stime_scaled;
315                 } else {
316                         stime_scaled = deltascaled;
317                 }
318         }
319         acct->utime_scaled += utime_scaled;
320 #endif
321 
322         return stime_scaled;
323 }
324 
325 static unsigned long vtime_delta(struct task_struct *tsk,
326                                  unsigned long *stime_scaled,
327                                  unsigned long *steal_time)
328 {
329         unsigned long now, stime;
330         struct cpu_accounting_data *acct = get_accounting(tsk);
331 
332         WARN_ON_ONCE(!irqs_disabled());
333 
334         now = mftb();
335         stime = now - acct->starttime;
336         acct->starttime = now;
337 
338         *stime_scaled = vtime_delta_scaled(acct, now, stime);
339 
340         *steal_time = calculate_stolen_time(now);
341 
342         return stime;
343 }
344 
345 void vtime_account_system(struct task_struct *tsk)
346 {
347         unsigned long stime, stime_scaled, steal_time;
348         struct cpu_accounting_data *acct = get_accounting(tsk);
349 
350         stime = vtime_delta(tsk, &stime_scaled, &steal_time);
351 
352         stime -= min(stime, steal_time);
353         acct->steal_time += steal_time;
354 
355         if ((tsk->flags & PF_VCPU) && !irq_count()) {
356                 acct->gtime += stime;
357 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
358                 acct->utime_scaled += stime_scaled;
359 #endif
360         } else {
361                 if (hardirq_count())
362                         acct->hardirq_time += stime;
363                 else if (in_serving_softirq())
364                         acct->softirq_time += stime;
365                 else
366                         acct->stime += stime;
367 
368 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
369                 acct->stime_scaled += stime_scaled;
370 #endif
371         }
372 }
373 EXPORT_SYMBOL_GPL(vtime_account_system);
374 
375 void vtime_account_idle(struct task_struct *tsk)
376 {
377         unsigned long stime, stime_scaled, steal_time;
378         struct cpu_accounting_data *acct = get_accounting(tsk);
379 
380         stime = vtime_delta(tsk, &stime_scaled, &steal_time);
381         acct->idle_time += stime + steal_time;
382 }
383 
384 static void vtime_flush_scaled(struct task_struct *tsk,
385                                struct cpu_accounting_data *acct)
386 {
387 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
388         if (acct->utime_scaled)
389                 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
390         if (acct->stime_scaled)
391                 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
392 
393         acct->utime_scaled = 0;
394         acct->utime_sspurr = 0;
395         acct->stime_scaled = 0;
396 #endif
397 }
398 
399 /*
400  * Account the whole cputime accumulated in the paca
401  * Must be called with interrupts disabled.
402  * Assumes that vtime_account_system/idle() has been called
403  * recently (i.e. since the last entry from usermode) so that
404  * get_paca()->user_time_scaled is up to date.
405  */
406 void vtime_flush(struct task_struct *tsk)
407 {
408         struct cpu_accounting_data *acct = get_accounting(tsk);
409 
410         if (acct->utime)
411                 account_user_time(tsk, cputime_to_nsecs(acct->utime));
412 
413         if (acct->gtime)
414                 account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
415 
416         if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
417                 account_steal_time(cputime_to_nsecs(acct->steal_time));
418                 acct->steal_time = 0;
419         }
420 
421         if (acct->idle_time)
422                 account_idle_time(cputime_to_nsecs(acct->idle_time));
423 
424         if (acct->stime)
425                 account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
426                                           CPUTIME_SYSTEM);
427 
428         if (acct->hardirq_time)
429                 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
430                                           CPUTIME_IRQ);
431         if (acct->softirq_time)
432                 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
433                                           CPUTIME_SOFTIRQ);
434 
435         vtime_flush_scaled(tsk, acct);
436 
437         acct->utime = 0;
438         acct->gtime = 0;
439         acct->idle_time = 0;
440         acct->stime = 0;
441         acct->hardirq_time = 0;
442         acct->softirq_time = 0;
443 }
444 
445 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
446 #define calc_cputime_factors()
447 #endif
448 
449 void __delay(unsigned long loops)
450 {
451         unsigned long start;
452         int diff;
453 
454         spin_begin();
455         if (__USE_RTC()) {
456                 start = get_rtcl();
457                 do {
458                         /* the RTCL register wraps at 1000000000 */
459                         diff = get_rtcl() - start;
460                         if (diff < 0)
461                                 diff += 1000000000;
462                         spin_cpu_relax();
463                 } while (diff < loops);
464         } else {
465                 start = get_tbl();
466                 while (get_tbl() - start < loops)
467                         spin_cpu_relax();
468         }
469         spin_end();
470 }
471 EXPORT_SYMBOL(__delay);
472 
473 void udelay(unsigned long usecs)
474 {
475         __delay(tb_ticks_per_usec * usecs);
476 }
477 EXPORT_SYMBOL(udelay);
478 
479 #ifdef CONFIG_SMP
480 unsigned long profile_pc(struct pt_regs *regs)
481 {
482         unsigned long pc = instruction_pointer(regs);
483 
484         if (in_lock_functions(pc))
485                 return regs->link;
486 
487         return pc;
488 }
489 EXPORT_SYMBOL(profile_pc);
490 #endif
491 
492 #ifdef CONFIG_IRQ_WORK
493 
494 /*
495  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
496  */
497 #ifdef CONFIG_PPC64
498 static inline unsigned long test_irq_work_pending(void)
499 {
500         unsigned long x;
501 
502         asm volatile("lbz %0,%1(13)"
503                 : "=r" (x)
504                 : "i" (offsetof(struct paca_struct, irq_work_pending)));
505         return x;
506 }
507 
508 static inline void set_irq_work_pending_flag(void)
509 {
510         asm volatile("stb %0,%1(13)" : :
511                 "r" (1),
512                 "i" (offsetof(struct paca_struct, irq_work_pending)));
513 }
514 
515 static inline void clear_irq_work_pending(void)
516 {
517         asm volatile("stb %0,%1(13)" : :
518                 "r" (0),
519                 "i" (offsetof(struct paca_struct, irq_work_pending)));
520 }
521 
522 void arch_irq_work_raise(void)
523 {
524         preempt_disable();
525         set_irq_work_pending_flag();
526         /*
527          * Non-nmi code running with interrupts disabled will replay
528          * irq_happened before it re-enables interrupts, so setthe
529          * decrementer there instead of causing a hardware exception
530          * which would immediately hit the masked interrupt handler
531          * and have the net effect of setting the decrementer in
532          * irq_happened.
533          *
534          * NMI interrupts can not check this when they return, so the
535          * decrementer hardware exception is raised, which will fire
536          * when interrupts are next enabled.
537          *
538          * BookE does not support this yet, it must audit all NMI
539          * interrupt handlers to ensure they call nmi_enter() so this
540          * check would be correct.
541          */
542         if (IS_ENABLED(CONFIG_BOOKE) || !irqs_disabled() || in_nmi()) {
543                 set_dec(1);
544         } else {
545                 hard_irq_disable();
546                 local_paca->irq_happened |= PACA_IRQ_DEC;
547         }
548         preempt_enable();
549 }
550 
551 #else /* 32-bit */
552 
553 DEFINE_PER_CPU(u8, irq_work_pending);
554 
555 #define set_irq_work_pending_flag()     __this_cpu_write(irq_work_pending, 1)
556 #define test_irq_work_pending()         __this_cpu_read(irq_work_pending)
557 #define clear_irq_work_pending()        __this_cpu_write(irq_work_pending, 0)
558 
559 void arch_irq_work_raise(void)
560 {
561         preempt_disable();
562         set_irq_work_pending_flag();
563         set_dec(1);
564         preempt_enable();
565 }
566 
567 #endif /* 32 vs 64 bit */
568 
569 #else  /* CONFIG_IRQ_WORK */
570 
571 #define test_irq_work_pending() 0
572 #define clear_irq_work_pending()
573 
574 #endif /* CONFIG_IRQ_WORK */
575 
576 /*
577  * timer_interrupt - gets called when the decrementer overflows,
578  * with interrupts disabled.
579  */
580 void timer_interrupt(struct pt_regs *regs)
581 {
582         struct clock_event_device *evt = this_cpu_ptr(&decrementers);
583         u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
584         struct pt_regs *old_regs;
585         u64 now;
586 
587         /* Some implementations of hotplug will get timer interrupts while
588          * offline, just ignore these and we also need to set
589          * decrementers_next_tb as MAX to make sure __check_irq_replay
590          * don't replay timer interrupt when return, otherwise we'll trap
591          * here infinitely :(
592          */
593         if (unlikely(!cpu_online(smp_processor_id()))) {
594                 *next_tb = ~(u64)0;
595                 set_dec(decrementer_max);
596                 return;
597         }
598 
599         /* Ensure a positive value is written to the decrementer, or else
600          * some CPUs will continue to take decrementer exceptions. When the
601          * PPC_WATCHDOG (decrementer based) is configured, keep this at most
602          * 31 bits, which is about 4 seconds on most systems, which gives
603          * the watchdog a chance of catching timer interrupt hard lockups.
604          */
605         if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
606                 set_dec(0x7fffffff);
607         else
608                 set_dec(decrementer_max);
609 
610         /* Conditionally hard-enable interrupts now that the DEC has been
611          * bumped to its maximum value
612          */
613         may_hard_irq_enable();
614 
615 
616 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
617         if (atomic_read(&ppc_n_lost_interrupts) != 0)
618                 do_IRQ(regs);
619 #endif
620 
621         old_regs = set_irq_regs(regs);
622         irq_enter();
623         trace_timer_interrupt_entry(regs);
624 
625         if (test_irq_work_pending()) {
626                 clear_irq_work_pending();
627                 irq_work_run();
628         }
629 
630         now = get_tb_or_rtc();
631         if (now >= *next_tb) {
632                 *next_tb = ~(u64)0;
633                 if (evt->event_handler)
634                         evt->event_handler(evt);
635                 __this_cpu_inc(irq_stat.timer_irqs_event);
636         } else {
637                 now = *next_tb - now;
638                 if (now <= decrementer_max)
639                         set_dec(now);
640                 /* We may have raced with new irq work */
641                 if (test_irq_work_pending())
642                         set_dec(1);
643                 __this_cpu_inc(irq_stat.timer_irqs_others);
644         }
645 
646         trace_timer_interrupt_exit(regs);
647         irq_exit();
648         set_irq_regs(old_regs);
649 }
650 EXPORT_SYMBOL(timer_interrupt);
651 
652 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
653 void timer_broadcast_interrupt(void)
654 {
655         u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
656 
657         *next_tb = ~(u64)0;
658         tick_receive_broadcast();
659         __this_cpu_inc(irq_stat.broadcast_irqs_event);
660 }
661 #endif
662 
663 /*
664  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
665  * left pending on exit from a KVM guest.  We don't need to do anything
666  * to clear them, as they are edge-triggered.
667  */
668 void hdec_interrupt(struct pt_regs *regs)
669 {
670 }
671 
672 #ifdef CONFIG_SUSPEND
673 static void generic_suspend_disable_irqs(void)
674 {
675         /* Disable the decrementer, so that it doesn't interfere
676          * with suspending.
677          */
678 
679         set_dec(decrementer_max);
680         local_irq_disable();
681         set_dec(decrementer_max);
682 }
683 
684 static void generic_suspend_enable_irqs(void)
685 {
686         local_irq_enable();
687 }
688 
689 /* Overrides the weak version in kernel/power/main.c */
690 void arch_suspend_disable_irqs(void)
691 {
692         if (ppc_md.suspend_disable_irqs)
693                 ppc_md.suspend_disable_irqs();
694         generic_suspend_disable_irqs();
695 }
696 
697 /* Overrides the weak version in kernel/power/main.c */
698 void arch_suspend_enable_irqs(void)
699 {
700         generic_suspend_enable_irqs();
701         if (ppc_md.suspend_enable_irqs)
702                 ppc_md.suspend_enable_irqs();
703 }
704 #endif
705 
706 unsigned long long tb_to_ns(unsigned long long ticks)
707 {
708         return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
709 }
710 EXPORT_SYMBOL_GPL(tb_to_ns);
711 
712 /*
713  * Scheduler clock - returns current time in nanosec units.
714  *
715  * Note: mulhdu(a, b) (multiply high double unsigned) returns
716  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
717  * are 64-bit unsigned numbers.
718  */
719 notrace unsigned long long sched_clock(void)
720 {
721         if (__USE_RTC())
722                 return get_rtc();
723         return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
724 }
725 
726 
727 #ifdef CONFIG_PPC_PSERIES
728 
729 /*
730  * Running clock - attempts to give a view of time passing for a virtualised
731  * kernels.
732  * Uses the VTB register if available otherwise a next best guess.
733  */
734 unsigned long long running_clock(void)
735 {
736         /*
737          * Don't read the VTB as a host since KVM does not switch in host
738          * timebase into the VTB when it takes a guest off the CPU, reading the
739          * VTB would result in reading 'last switched out' guest VTB.
740          *
741          * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
742          * would be unsafe to rely only on the #ifdef above.
743          */
744         if (firmware_has_feature(FW_FEATURE_LPAR) &&
745             cpu_has_feature(CPU_FTR_ARCH_207S))
746                 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
747 
748         /*
749          * This is a next best approximation without a VTB.
750          * On a host which is running bare metal there should never be any stolen
751          * time and on a host which doesn't do any virtualisation TB *should* equal
752          * VTB so it makes no difference anyway.
753          */
754         return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
755 }
756 #endif
757 
758 static int __init get_freq(char *name, int cells, unsigned long *val)
759 {
760         struct device_node *cpu;
761         const __be32 *fp;
762         int found = 0;
763 
764         /* The cpu node should have timebase and clock frequency properties */
765         cpu = of_find_node_by_type(NULL, "cpu");
766 
767         if (cpu) {
768                 fp = of_get_property(cpu, name, NULL);
769                 if (fp) {
770                         found = 1;
771                         *val = of_read_ulong(fp, cells);
772                 }
773 
774                 of_node_put(cpu);
775         }
776 
777         return found;
778 }
779 
780 static void start_cpu_decrementer(void)
781 {
782 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
783         unsigned int tcr;
784 
785         /* Clear any pending timer interrupts */
786         mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
787 
788         tcr = mfspr(SPRN_TCR);
789         /*
790          * The watchdog may have already been enabled by u-boot. So leave
791          * TRC[WP] (Watchdog Period) alone.
792          */
793         tcr &= TCR_WP_MASK;     /* Clear all bits except for TCR[WP] */
794         tcr |= TCR_DIE;         /* Enable decrementer */
795         mtspr(SPRN_TCR, tcr);
796 #endif
797 }
798 
799 void __init generic_calibrate_decr(void)
800 {
801         ppc_tb_freq = DEFAULT_TB_FREQ;          /* hardcoded default */
802 
803         if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
804             !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
805 
806                 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
807                                 "(not found)\n");
808         }
809 
810         ppc_proc_freq = DEFAULT_PROC_FREQ;      /* hardcoded default */
811 
812         if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
813             !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
814 
815                 printk(KERN_ERR "WARNING: Estimating processor frequency "
816                                 "(not found)\n");
817         }
818 }
819 
820 int update_persistent_clock64(struct timespec64 now)
821 {
822         struct rtc_time tm;
823 
824         if (!ppc_md.set_rtc_time)
825                 return -ENODEV;
826 
827         rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
828 
829         return ppc_md.set_rtc_time(&tm);
830 }
831 
832 static void __read_persistent_clock(struct timespec64 *ts)
833 {
834         struct rtc_time tm;
835         static int first = 1;
836 
837         ts->tv_nsec = 0;
838         /* XXX this is a litle fragile but will work okay in the short term */
839         if (first) {
840                 first = 0;
841                 if (ppc_md.time_init)
842                         timezone_offset = ppc_md.time_init();
843 
844                 /* get_boot_time() isn't guaranteed to be safe to call late */
845                 if (ppc_md.get_boot_time) {
846                         ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
847                         return;
848                 }
849         }
850         if (!ppc_md.get_rtc_time) {
851                 ts->tv_sec = 0;
852                 return;
853         }
854         ppc_md.get_rtc_time(&tm);
855 
856         ts->tv_sec = rtc_tm_to_time64(&tm);
857 }
858 
859 void read_persistent_clock64(struct timespec64 *ts)
860 {
861         __read_persistent_clock(ts);
862 
863         /* Sanitize it in case real time clock is set below EPOCH */
864         if (ts->tv_sec < 0) {
865                 ts->tv_sec = 0;
866                 ts->tv_nsec = 0;
867         }
868                 
869 }
870 
871 /* clocksource code */
872 static notrace u64 rtc_read(struct clocksource *cs)
873 {
874         return (u64)get_rtc();
875 }
876 
877 static notrace u64 timebase_read(struct clocksource *cs)
878 {
879         return (u64)get_tb();
880 }
881 
882 
883 void update_vsyscall(struct timekeeper *tk)
884 {
885         struct timespec xt;
886         struct clocksource *clock = tk->tkr_mono.clock;
887         u32 mult = tk->tkr_mono.mult;
888         u32 shift = tk->tkr_mono.shift;
889         u64 cycle_last = tk->tkr_mono.cycle_last;
890         u64 new_tb_to_xs, new_stamp_xsec;
891         u64 frac_sec;
892 
893         if (clock != &clocksource_timebase)
894                 return;
895 
896         xt.tv_sec = tk->xtime_sec;
897         xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
898 
899         /* Make userspace gettimeofday spin until we're done. */
900         ++vdso_data->tb_update_count;
901         smp_mb();
902 
903         /*
904          * This computes ((2^20 / 1e9) * mult) >> shift as a
905          * 0.64 fixed-point fraction.
906          * The computation in the else clause below won't overflow
907          * (as long as the timebase frequency is >= 1.049 MHz)
908          * but loses precision because we lose the low bits of the constant
909          * in the shift.  Note that 19342813113834067 ~= 2^(20+64) / 1e9.
910          * For a shift of 24 the error is about 0.5e-9, or about 0.5ns
911          * over a second.  (Shift values are usually 22, 23 or 24.)
912          * For high frequency clocks such as the 512MHz timebase clock
913          * on POWER[6789], the mult value is small (e.g. 32768000)
914          * and so we can shift the constant by 16 initially
915          * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the
916          * remaining shifts after the multiplication, which gives a
917          * more accurate result (e.g. with mult = 32768000, shift = 24,
918          * the error is only about 1.2e-12, or 0.7ns over 10 minutes).
919          */
920         if (mult <= 62500000 && clock->shift >= 16)
921                 new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16);
922         else
923                 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
924 
925         /*
926          * Compute the fractional second in units of 2^-32 seconds.
927          * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift
928          * in nanoseconds, so multiplying that by 2^32 / 1e9 gives
929          * it in units of 2^-32 seconds.
930          * We assume shift <= 32 because clocks_calc_mult_shift()
931          * generates shift values in the range 0 - 32.
932          */
933         frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift);
934         do_div(frac_sec, NSEC_PER_SEC);
935 
936         /*
937          * Work out new stamp_xsec value for any legacy users of systemcfg.
938          * stamp_xsec is in units of 2^-20 seconds.
939          */
940         new_stamp_xsec = frac_sec >> 12;
941         new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC;
942 
943         /*
944          * tb_update_count is used to allow the userspace gettimeofday code
945          * to assure itself that it sees a consistent view of the tb_to_xs and
946          * stamp_xsec variables.  It reads the tb_update_count, then reads
947          * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
948          * the two values of tb_update_count match and are even then the
949          * tb_to_xs and stamp_xsec values are consistent.  If not, then it
950          * loops back and reads them again until this criteria is met.
951          */
952         vdso_data->tb_orig_stamp = cycle_last;
953         vdso_data->stamp_xsec = new_stamp_xsec;
954         vdso_data->tb_to_xs = new_tb_to_xs;
955         vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec;
956         vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec;
957         vdso_data->stamp_xtime = xt;
958         vdso_data->stamp_sec_fraction = frac_sec;
959         smp_wmb();
960         ++(vdso_data->tb_update_count);
961 }
962 
963 void update_vsyscall_tz(void)
964 {
965         vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
966         vdso_data->tz_dsttime = sys_tz.tz_dsttime;
967 }
968 
969 static void __init clocksource_init(void)
970 {
971         struct clocksource *clock;
972 
973         if (__USE_RTC())
974                 clock = &clocksource_rtc;
975         else
976                 clock = &clocksource_timebase;
977 
978         if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
979                 printk(KERN_ERR "clocksource: %s is already registered\n",
980                        clock->name);
981                 return;
982         }
983 
984         printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
985                clock->name, clock->mult, clock->shift);
986 }
987 
988 static int decrementer_set_next_event(unsigned long evt,
989                                       struct clock_event_device *dev)
990 {
991         __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
992         set_dec(evt);
993 
994         /* We may have raced with new irq work */
995         if (test_irq_work_pending())
996                 set_dec(1);
997 
998         return 0;
999 }
1000 
1001 static int decrementer_shutdown(struct clock_event_device *dev)
1002 {
1003         decrementer_set_next_event(decrementer_max, dev);
1004         return 0;
1005 }
1006 
1007 static void register_decrementer_clockevent(int cpu)
1008 {
1009         struct clock_event_device *dec = &per_cpu(decrementers, cpu);
1010 
1011         *dec = decrementer_clockevent;
1012         dec->cpumask = cpumask_of(cpu);
1013 
1014         clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
1015 
1016         printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
1017                     dec->name, dec->mult, dec->shift, cpu);
1018 
1019         /* Set values for KVM, see kvm_emulate_dec() */
1020         decrementer_clockevent.mult = dec->mult;
1021         decrementer_clockevent.shift = dec->shift;
1022 }
1023 
1024 static void enable_large_decrementer(void)
1025 {
1026         if (!cpu_has_feature(CPU_FTR_ARCH_300))
1027                 return;
1028 
1029         if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
1030                 return;
1031 
1032         /*
1033          * If we're running as the hypervisor we need to enable the LD manually
1034          * otherwise firmware should have done it for us.
1035          */
1036         if (cpu_has_feature(CPU_FTR_HVMODE))
1037                 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
1038 }
1039 
1040 static void __init set_decrementer_max(void)
1041 {
1042         struct device_node *cpu;
1043         u32 bits = 32;
1044 
1045         /* Prior to ISAv3 the decrementer is always 32 bit */
1046         if (!cpu_has_feature(CPU_FTR_ARCH_300))
1047                 return;
1048 
1049         cpu = of_find_node_by_type(NULL, "cpu");
1050 
1051         if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
1052                 if (bits > 64 || bits < 32) {
1053                         pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
1054                         bits = 32;
1055                 }
1056 
1057                 /* calculate the signed maximum given this many bits */
1058                 decrementer_max = (1ul << (bits - 1)) - 1;
1059         }
1060 
1061         of_node_put(cpu);
1062 
1063         pr_info("time_init: %u bit decrementer (max: %llx)\n",
1064                 bits, decrementer_max);
1065 }
1066 
1067 static void __init init_decrementer_clockevent(void)
1068 {
1069         register_decrementer_clockevent(smp_processor_id());
1070 }
1071 
1072 void secondary_cpu_time_init(void)
1073 {
1074         /* Enable and test the large decrementer for this cpu */
1075         enable_large_decrementer();
1076 
1077         /* Start the decrementer on CPUs that have manual control
1078          * such as BookE
1079          */
1080         start_cpu_decrementer();
1081 
1082         /* FIME: Should make unrelatred change to move snapshot_timebase
1083          * call here ! */
1084         register_decrementer_clockevent(smp_processor_id());
1085 }
1086 
1087 /* This function is only called on the boot processor */
1088 void __init time_init(void)
1089 {
1090         struct div_result res;
1091         u64 scale;
1092         unsigned shift;
1093 
1094         if (__USE_RTC()) {
1095                 /* 601 processor: dec counts down by 128 every 128ns */
1096                 ppc_tb_freq = 1000000000;
1097         } else {
1098                 /* Normal PowerPC with timebase register */
1099                 ppc_md.calibrate_decr();
1100                 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
1101                        ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
1102                 printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
1103                        ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
1104         }
1105 
1106         tb_ticks_per_jiffy = ppc_tb_freq / HZ;
1107         tb_ticks_per_sec = ppc_tb_freq;
1108         tb_ticks_per_usec = ppc_tb_freq / 1000000;
1109         calc_cputime_factors();
1110 
1111         /*
1112          * Compute scale factor for sched_clock.
1113          * The calibrate_decr() function has set tb_ticks_per_sec,
1114          * which is the timebase frequency.
1115          * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
1116          * the 128-bit result as a 64.64 fixed-point number.
1117          * We then shift that number right until it is less than 1.0,
1118          * giving us the scale factor and shift count to use in
1119          * sched_clock().
1120          */
1121         div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
1122         scale = res.result_low;
1123         for (shift = 0; res.result_high != 0; ++shift) {
1124                 scale = (scale >> 1) | (res.result_high << 63);
1125                 res.result_high >>= 1;
1126         }
1127         tb_to_ns_scale = scale;
1128         tb_to_ns_shift = shift;
1129         /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
1130         boot_tb = get_tb_or_rtc();
1131 
1132         /* If platform provided a timezone (pmac), we correct the time */
1133         if (timezone_offset) {
1134                 sys_tz.tz_minuteswest = -timezone_offset / 60;
1135                 sys_tz.tz_dsttime = 0;
1136         }
1137 
1138         vdso_data->tb_update_count = 0;
1139         vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
1140 
1141         /* initialise and enable the large decrementer (if we have one) */
1142         set_decrementer_max();
1143         enable_large_decrementer();
1144 
1145         /* Start the decrementer on CPUs that have manual control
1146          * such as BookE
1147          */
1148         start_cpu_decrementer();
1149 
1150         /* Register the clocksource */
1151         clocksource_init();
1152 
1153         init_decrementer_clockevent();
1154         tick_setup_hrtimer_broadcast();
1155 
1156 #ifdef CONFIG_COMMON_CLK
1157         of_clk_init(NULL);
1158 #endif
1159 }
1160 
1161 /*
1162  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1163  * result.
1164  */
1165 void div128_by_32(u64 dividend_high, u64 dividend_low,
1166                   unsigned divisor, struct div_result *dr)
1167 {
1168         unsigned long a, b, c, d;
1169         unsigned long w, x, y, z;
1170         u64 ra, rb, rc;
1171 
1172         a = dividend_high >> 32;
1173         b = dividend_high & 0xffffffff;
1174         c = dividend_low >> 32;
1175         d = dividend_low & 0xffffffff;
1176 
1177         w = a / divisor;
1178         ra = ((u64)(a - (w * divisor)) << 32) + b;
1179 
1180         rb = ((u64) do_div(ra, divisor) << 32) + c;
1181         x = ra;
1182 
1183         rc = ((u64) do_div(rb, divisor) << 32) + d;
1184         y = rb;
1185 
1186         do_div(rc, divisor);
1187         z = rc;
1188 
1189         dr->result_high = ((u64)w << 32) + x;
1190         dr->result_low  = ((u64)y << 32) + z;
1191 
1192 }
1193 
1194 /* We don't need to calibrate delay, we use the CPU timebase for that */
1195 void calibrate_delay(void)
1196 {
1197         /* Some generic code (such as spinlock debug) use loops_per_jiffy
1198          * as the number of __delay(1) in a jiffy, so make it so
1199          */
1200         loops_per_jiffy = tb_ticks_per_jiffy;
1201 }
1202 
1203 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1204 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1205 {
1206         ppc_md.get_rtc_time(tm);
1207         return 0;
1208 }
1209 
1210 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1211 {
1212         if (!ppc_md.set_rtc_time)
1213                 return -EOPNOTSUPP;
1214 
1215         if (ppc_md.set_rtc_time(tm) < 0)
1216                 return -EOPNOTSUPP;
1217 
1218         return 0;
1219 }
1220 
1221 static const struct rtc_class_ops rtc_generic_ops = {
1222         .read_time = rtc_generic_get_time,
1223         .set_time = rtc_generic_set_time,
1224 };
1225 
1226 static int __init rtc_init(void)
1227 {
1228         struct platform_device *pdev;
1229 
1230         if (!ppc_md.get_rtc_time)
1231                 return -ENODEV;
1232 
1233         pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1234                                              &rtc_generic_ops,
1235                                              sizeof(rtc_generic_ops));
1236 
1237         return PTR_ERR_OR_ZERO(pdev);
1238 }
1239 
1240 device_initcall(rtc_init);
1241 #endif
1242 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp