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

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

Version: ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ 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/kernel.h>
 38 #include <linux/param.h>
 39 #include <linux/string.h>
 40 #include <linux/mm.h>
 41 #include <linux/interrupt.h>
 42 #include <linux/timex.h>
 43 #include <linux/kernel_stat.h>
 44 #include <linux/time.h>
 45 #include <linux/clockchips.h>
 46 #include <linux/init.h>
 47 #include <linux/profile.h>
 48 #include <linux/cpu.h>
 49 #include <linux/security.h>
 50 #include <linux/percpu.h>
 51 #include <linux/rtc.h>
 52 #include <linux/jiffies.h>
 53 #include <linux/posix-timers.h>
 54 #include <linux/irq.h>
 55 #include <linux/delay.h>
 56 #include <linux/irq_work.h>
 57 #include <linux/clk-provider.h>
 58 #include <linux/suspend.h>
 59 #include <asm/trace.h>
 60 
 61 #include <asm/io.h>
 62 #include <asm/processor.h>
 63 #include <asm/nvram.h>
 64 #include <asm/cache.h>
 65 #include <asm/machdep.h>
 66 #include <asm/uaccess.h>
 67 #include <asm/time.h>
 68 #include <asm/prom.h>
 69 #include <asm/irq.h>
 70 #include <asm/div64.h>
 71 #include <asm/smp.h>
 72 #include <asm/vdso_datapage.h>
 73 #include <asm/firmware.h>
 74 #include <asm/cputime.h>
 75 
 76 /* powerpc clocksource/clockevent code */
 77 
 78 #include <linux/clockchips.h>
 79 #include <linux/timekeeper_internal.h>
 80 
 81 static cycle_t rtc_read(struct clocksource *);
 82 static struct clocksource clocksource_rtc = {
 83         .name         = "rtc",
 84         .rating       = 400,
 85         .flags        = CLOCK_SOURCE_IS_CONTINUOUS,
 86         .mask         = CLOCKSOURCE_MASK(64),
 87         .read         = rtc_read,
 88 };
 89 
 90 static cycle_t timebase_read(struct clocksource *);
 91 static struct clocksource clocksource_timebase = {
 92         .name         = "timebase",
 93         .rating       = 400,
 94         .flags        = CLOCK_SOURCE_IS_CONTINUOUS,
 95         .mask         = CLOCKSOURCE_MASK(64),
 96         .read         = timebase_read,
 97 };
 98 
 99 #define DECREMENTER_MAX 0x7fffffff
100 
101 static int decrementer_set_next_event(unsigned long evt,
102                                       struct clock_event_device *dev);
103 static int decrementer_shutdown(struct clock_event_device *evt);
104 
105 struct clock_event_device decrementer_clockevent = {
106         .name                   = "decrementer",
107         .rating                 = 200,
108         .irq                    = 0,
109         .set_next_event         = decrementer_set_next_event,
110         .set_state_shutdown     = decrementer_shutdown,
111         .tick_resume            = decrementer_shutdown,
112         .features               = CLOCK_EVT_FEAT_ONESHOT |
113                                   CLOCK_EVT_FEAT_C3STOP,
114 };
115 EXPORT_SYMBOL(decrementer_clockevent);
116 
117 DEFINE_PER_CPU(u64, decrementers_next_tb);
118 static 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 
129 unsigned long tb_ticks_per_jiffy;
130 unsigned long tb_ticks_per_usec = 100; /* sane default */
131 EXPORT_SYMBOL(tb_ticks_per_usec);
132 unsigned long tb_ticks_per_sec;
133 EXPORT_SYMBOL(tb_ticks_per_sec);        /* for cputime_t conversions */
134 
135 DEFINE_SPINLOCK(rtc_lock);
136 EXPORT_SYMBOL_GPL(rtc_lock);
137 
138 static u64 tb_to_ns_scale __read_mostly;
139 static unsigned tb_to_ns_shift __read_mostly;
140 static u64 boot_tb __read_mostly;
141 
142 extern struct timezone sys_tz;
143 static long timezone_offset;
144 
145 unsigned long ppc_proc_freq;
146 EXPORT_SYMBOL_GPL(ppc_proc_freq);
147 unsigned long ppc_tb_freq;
148 EXPORT_SYMBOL_GPL(ppc_tb_freq);
149 
150 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
151 /*
152  * Factors for converting from cputime_t (timebase ticks) to
153  * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds).
154  * These are all stored as 0.64 fixed-point binary fractions.
155  */
156 u64 __cputime_jiffies_factor;
157 EXPORT_SYMBOL(__cputime_jiffies_factor);
158 u64 __cputime_usec_factor;
159 EXPORT_SYMBOL(__cputime_usec_factor);
160 u64 __cputime_sec_factor;
161 EXPORT_SYMBOL(__cputime_sec_factor);
162 u64 __cputime_clockt_factor;
163 EXPORT_SYMBOL(__cputime_clockt_factor);
164 DEFINE_PER_CPU(unsigned long, cputime_last_delta);
165 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta);
166 
167 cputime_t cputime_one_jiffy;
168 
169 void (*dtl_consumer)(struct dtl_entry *, u64);
170 
171 static void calc_cputime_factors(void)
172 {
173         struct div_result res;
174 
175         div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
176         __cputime_jiffies_factor = res.result_low;
177         div128_by_32(1000000, 0, tb_ticks_per_sec, &res);
178         __cputime_usec_factor = res.result_low;
179         div128_by_32(1, 0, tb_ticks_per_sec, &res);
180         __cputime_sec_factor = res.result_low;
181         div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
182         __cputime_clockt_factor = res.result_low;
183 }
184 
185 /*
186  * Read the SPURR on systems that have it, otherwise the PURR,
187  * or if that doesn't exist return the timebase value passed in.
188  */
189 static u64 read_spurr(u64 tb)
190 {
191         if (cpu_has_feature(CPU_FTR_SPURR))
192                 return mfspr(SPRN_SPURR);
193         if (cpu_has_feature(CPU_FTR_PURR))
194                 return mfspr(SPRN_PURR);
195         return tb;
196 }
197 
198 #ifdef CONFIG_PPC_SPLPAR
199 
200 /*
201  * Scan the dispatch trace log and count up the stolen time.
202  * Should be called with interrupts disabled.
203  */
204 static u64 scan_dispatch_log(u64 stop_tb)
205 {
206         u64 i = local_paca->dtl_ridx;
207         struct dtl_entry *dtl = local_paca->dtl_curr;
208         struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
209         struct lppaca *vpa = local_paca->lppaca_ptr;
210         u64 tb_delta;
211         u64 stolen = 0;
212         u64 dtb;
213 
214         if (!dtl)
215                 return 0;
216 
217         if (i == be64_to_cpu(vpa->dtl_idx))
218                 return 0;
219         while (i < be64_to_cpu(vpa->dtl_idx)) {
220                 dtb = be64_to_cpu(dtl->timebase);
221                 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) +
222                         be32_to_cpu(dtl->ready_to_enqueue_time);
223                 barrier();
224                 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
225                         /* buffer has overflowed */
226                         i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
227                         dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
228                         continue;
229                 }
230                 if (dtb > stop_tb)
231                         break;
232                 if (dtl_consumer)
233                         dtl_consumer(dtl, i);
234                 stolen += tb_delta;
235                 ++i;
236                 ++dtl;
237                 if (dtl == dtl_end)
238                         dtl = local_paca->dispatch_log;
239         }
240         local_paca->dtl_ridx = i;
241         local_paca->dtl_curr = dtl;
242         return stolen;
243 }
244 
245 /*
246  * Accumulate stolen time by scanning the dispatch trace log.
247  * Called on entry from user mode.
248  */
249 void accumulate_stolen_time(void)
250 {
251         u64 sst, ust;
252 
253         u8 save_soft_enabled = local_paca->soft_enabled;
254 
255         /* We are called early in the exception entry, before
256          * soft/hard_enabled are sync'ed to the expected state
257          * for the exception. We are hard disabled but the PACA
258          * needs to reflect that so various debug stuff doesn't
259          * complain
260          */
261         local_paca->soft_enabled = 0;
262 
263         sst = scan_dispatch_log(local_paca->starttime_user);
264         ust = scan_dispatch_log(local_paca->starttime);
265         local_paca->system_time -= sst;
266         local_paca->user_time -= ust;
267         local_paca->stolen_time += ust + sst;
268 
269         local_paca->soft_enabled = save_soft_enabled;
270 }
271 
272 static inline u64 calculate_stolen_time(u64 stop_tb)
273 {
274         u64 stolen = 0;
275 
276         if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) {
277                 stolen = scan_dispatch_log(stop_tb);
278                 get_paca()->system_time -= stolen;
279         }
280 
281         stolen += get_paca()->stolen_time;
282         get_paca()->stolen_time = 0;
283         return stolen;
284 }
285 
286 #else /* CONFIG_PPC_SPLPAR */
287 static inline u64 calculate_stolen_time(u64 stop_tb)
288 {
289         return 0;
290 }
291 
292 #endif /* CONFIG_PPC_SPLPAR */
293 
294 /*
295  * Account time for a transition between system, hard irq
296  * or soft irq state.
297  */
298 static u64 vtime_delta(struct task_struct *tsk,
299                         u64 *sys_scaled, u64 *stolen)
300 {
301         u64 now, nowscaled, deltascaled;
302         u64 udelta, delta, user_scaled;
303 
304         WARN_ON_ONCE(!irqs_disabled());
305 
306         now = mftb();
307         nowscaled = read_spurr(now);
308         get_paca()->system_time += now - get_paca()->starttime;
309         get_paca()->starttime = now;
310         deltascaled = nowscaled - get_paca()->startspurr;
311         get_paca()->startspurr = nowscaled;
312 
313         *stolen = calculate_stolen_time(now);
314 
315         delta = get_paca()->system_time;
316         get_paca()->system_time = 0;
317         udelta = get_paca()->user_time - get_paca()->utime_sspurr;
318         get_paca()->utime_sspurr = get_paca()->user_time;
319 
320         /*
321          * Because we don't read the SPURR on every kernel entry/exit,
322          * deltascaled includes both user and system SPURR ticks.
323          * Apportion these ticks to system SPURR ticks and user
324          * SPURR ticks in the same ratio as the system time (delta)
325          * and user time (udelta) values obtained from the timebase
326          * over the same interval.  The system ticks get accounted here;
327          * the user ticks get saved up in paca->user_time_scaled to be
328          * used by account_process_tick.
329          */
330         *sys_scaled = delta;
331         user_scaled = udelta;
332         if (deltascaled != delta + udelta) {
333                 if (udelta) {
334                         *sys_scaled = deltascaled * delta / (delta + udelta);
335                         user_scaled = deltascaled - *sys_scaled;
336                 } else {
337                         *sys_scaled = deltascaled;
338                 }
339         }
340         get_paca()->user_time_scaled += user_scaled;
341 
342         return delta;
343 }
344 
345 void vtime_account_system(struct task_struct *tsk)
346 {
347         u64 delta, sys_scaled, stolen;
348 
349         delta = vtime_delta(tsk, &sys_scaled, &stolen);
350         account_system_time(tsk, 0, delta, sys_scaled);
351         if (stolen)
352                 account_steal_time(stolen);
353 }
354 EXPORT_SYMBOL_GPL(vtime_account_system);
355 
356 void vtime_account_idle(struct task_struct *tsk)
357 {
358         u64 delta, sys_scaled, stolen;
359 
360         delta = vtime_delta(tsk, &sys_scaled, &stolen);
361         account_idle_time(delta + stolen);
362 }
363 
364 /*
365  * Transfer the user time accumulated in the paca
366  * by the exception entry and exit code to the generic
367  * process user time records.
368  * Must be called with interrupts disabled.
369  * Assumes that vtime_account_system/idle() has been called
370  * recently (i.e. since the last entry from usermode) so that
371  * get_paca()->user_time_scaled is up to date.
372  */
373 void vtime_account_user(struct task_struct *tsk)
374 {
375         cputime_t utime, utimescaled;
376 
377         utime = get_paca()->user_time;
378         utimescaled = get_paca()->user_time_scaled;
379         get_paca()->user_time = 0;
380         get_paca()->user_time_scaled = 0;
381         get_paca()->utime_sspurr = 0;
382         account_user_time(tsk, utime, utimescaled);
383 }
384 
385 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
386 #define calc_cputime_factors()
387 #endif
388 
389 void __delay(unsigned long loops)
390 {
391         unsigned long start;
392         int diff;
393 
394         if (__USE_RTC()) {
395                 start = get_rtcl();
396                 do {
397                         /* the RTCL register wraps at 1000000000 */
398                         diff = get_rtcl() - start;
399                         if (diff < 0)
400                                 diff += 1000000000;
401                 } while (diff < loops);
402         } else {
403                 start = get_tbl();
404                 while (get_tbl() - start < loops)
405                         HMT_low();
406                 HMT_medium();
407         }
408 }
409 EXPORT_SYMBOL(__delay);
410 
411 void udelay(unsigned long usecs)
412 {
413         __delay(tb_ticks_per_usec * usecs);
414 }
415 EXPORT_SYMBOL(udelay);
416 
417 #ifdef CONFIG_SMP
418 unsigned long profile_pc(struct pt_regs *regs)
419 {
420         unsigned long pc = instruction_pointer(regs);
421 
422         if (in_lock_functions(pc))
423                 return regs->link;
424 
425         return pc;
426 }
427 EXPORT_SYMBOL(profile_pc);
428 #endif
429 
430 #ifdef CONFIG_IRQ_WORK
431 
432 /*
433  * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
434  */
435 #ifdef CONFIG_PPC64
436 static inline unsigned long test_irq_work_pending(void)
437 {
438         unsigned long x;
439 
440         asm volatile("lbz %0,%1(13)"
441                 : "=r" (x)
442                 : "i" (offsetof(struct paca_struct, irq_work_pending)));
443         return x;
444 }
445 
446 static inline void set_irq_work_pending_flag(void)
447 {
448         asm volatile("stb %0,%1(13)" : :
449                 "r" (1),
450                 "i" (offsetof(struct paca_struct, irq_work_pending)));
451 }
452 
453 static inline void clear_irq_work_pending(void)
454 {
455         asm volatile("stb %0,%1(13)" : :
456                 "r" (0),
457                 "i" (offsetof(struct paca_struct, irq_work_pending)));
458 }
459 
460 #else /* 32-bit */
461 
462 DEFINE_PER_CPU(u8, irq_work_pending);
463 
464 #define set_irq_work_pending_flag()     __this_cpu_write(irq_work_pending, 1)
465 #define test_irq_work_pending()         __this_cpu_read(irq_work_pending)
466 #define clear_irq_work_pending()        __this_cpu_write(irq_work_pending, 0)
467 
468 #endif /* 32 vs 64 bit */
469 
470 void arch_irq_work_raise(void)
471 {
472         preempt_disable();
473         set_irq_work_pending_flag();
474         set_dec(1);
475         preempt_enable();
476 }
477 
478 #else  /* CONFIG_IRQ_WORK */
479 
480 #define test_irq_work_pending() 0
481 #define clear_irq_work_pending()
482 
483 #endif /* CONFIG_IRQ_WORK */
484 
485 static void __timer_interrupt(void)
486 {
487         struct pt_regs *regs = get_irq_regs();
488         u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
489         struct clock_event_device *evt = this_cpu_ptr(&decrementers);
490         u64 now;
491 
492         trace_timer_interrupt_entry(regs);
493 
494         if (test_irq_work_pending()) {
495                 clear_irq_work_pending();
496                 irq_work_run();
497         }
498 
499         now = get_tb_or_rtc();
500         if (now >= *next_tb) {
501                 *next_tb = ~(u64)0;
502                 if (evt->event_handler)
503                         evt->event_handler(evt);
504                 __this_cpu_inc(irq_stat.timer_irqs_event);
505         } else {
506                 now = *next_tb - now;
507                 if (now <= DECREMENTER_MAX)
508                         set_dec((int)now);
509                 /* We may have raced with new irq work */
510                 if (test_irq_work_pending())
511                         set_dec(1);
512                 __this_cpu_inc(irq_stat.timer_irqs_others);
513         }
514 
515 #ifdef CONFIG_PPC64
516         /* collect purr register values often, for accurate calculations */
517         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
518                 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
519                 cu->current_tb = mfspr(SPRN_PURR);
520         }
521 #endif
522 
523         trace_timer_interrupt_exit(regs);
524 }
525 
526 /*
527  * timer_interrupt - gets called when the decrementer overflows,
528  * with interrupts disabled.
529  */
530 void timer_interrupt(struct pt_regs * regs)
531 {
532         struct pt_regs *old_regs;
533         u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
534 
535         /* Ensure a positive value is written to the decrementer, or else
536          * some CPUs will continue to take decrementer exceptions.
537          */
538         set_dec(DECREMENTER_MAX);
539 
540         /* Some implementations of hotplug will get timer interrupts while
541          * offline, just ignore these and we also need to set
542          * decrementers_next_tb as MAX to make sure __check_irq_replay
543          * don't replay timer interrupt when return, otherwise we'll trap
544          * here infinitely :(
545          */
546         if (!cpu_online(smp_processor_id())) {
547                 *next_tb = ~(u64)0;
548                 return;
549         }
550 
551         /* Conditionally hard-enable interrupts now that the DEC has been
552          * bumped to its maximum value
553          */
554         may_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         irq_enter();
564 
565         __timer_interrupt();
566         irq_exit();
567         set_irq_regs(old_regs);
568 }
569 
570 /*
571  * Hypervisor decrementer interrupts shouldn't occur but are sometimes
572  * left pending on exit from a KVM guest.  We don't need to do anything
573  * to clear them, as they are edge-triggered.
574  */
575 void hdec_interrupt(struct pt_regs *regs)
576 {
577 }
578 
579 #ifdef CONFIG_SUSPEND
580 static void generic_suspend_disable_irqs(void)
581 {
582         /* Disable the decrementer, so that it doesn't interfere
583          * with suspending.
584          */
585 
586         set_dec(DECREMENTER_MAX);
587         local_irq_disable();
588         set_dec(DECREMENTER_MAX);
589 }
590 
591 static void generic_suspend_enable_irqs(void)
592 {
593         local_irq_enable();
594 }
595 
596 /* Overrides the weak version in kernel/power/main.c */
597 void arch_suspend_disable_irqs(void)
598 {
599         if (ppc_md.suspend_disable_irqs)
600                 ppc_md.suspend_disable_irqs();
601         generic_suspend_disable_irqs();
602 }
603 
604 /* Overrides the weak version in kernel/power/main.c */
605 void arch_suspend_enable_irqs(void)
606 {
607         generic_suspend_enable_irqs();
608         if (ppc_md.suspend_enable_irqs)
609                 ppc_md.suspend_enable_irqs();
610 }
611 #endif
612 
613 unsigned long long tb_to_ns(unsigned long long ticks)
614 {
615         return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
616 }
617 EXPORT_SYMBOL_GPL(tb_to_ns);
618 
619 /*
620  * Scheduler clock - returns current time in nanosec units.
621  *
622  * Note: mulhdu(a, b) (multiply high double unsigned) returns
623  * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
624  * are 64-bit unsigned numbers.
625  */
626 unsigned long long sched_clock(void)
627 {
628         if (__USE_RTC())
629                 return get_rtc();
630         return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
631 }
632 
633 
634 #ifdef CONFIG_PPC_PSERIES
635 
636 /*
637  * Running clock - attempts to give a view of time passing for a virtualised
638  * kernels.
639  * Uses the VTB register if available otherwise a next best guess.
640  */
641 unsigned long long running_clock(void)
642 {
643         /*
644          * Don't read the VTB as a host since KVM does not switch in host
645          * timebase into the VTB when it takes a guest off the CPU, reading the
646          * VTB would result in reading 'last switched out' guest VTB.
647          *
648          * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
649          * would be unsafe to rely only on the #ifdef above.
650          */
651         if (firmware_has_feature(FW_FEATURE_LPAR) &&
652             cpu_has_feature(CPU_FTR_ARCH_207S))
653                 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
654 
655         /*
656          * This is a next best approximation without a VTB.
657          * On a host which is running bare metal there should never be any stolen
658          * time and on a host which doesn't do any virtualisation TB *should* equal
659          * VTB so it makes no difference anyway.
660          */
661         return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]);
662 }
663 #endif
664 
665 static int __init get_freq(char *name, int cells, unsigned long *val)
666 {
667         struct device_node *cpu;
668         const __be32 *fp;
669         int found = 0;
670 
671         /* The cpu node should have timebase and clock frequency properties */
672         cpu = of_find_node_by_type(NULL, "cpu");
673 
674         if (cpu) {
675                 fp = of_get_property(cpu, name, NULL);
676                 if (fp) {
677                         found = 1;
678                         *val = of_read_ulong(fp, cells);
679                 }
680 
681                 of_node_put(cpu);
682         }
683 
684         return found;
685 }
686 
687 static void start_cpu_decrementer(void)
688 {
689 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
690         /* Clear any pending timer interrupts */
691         mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
692 
693         /* Enable decrementer interrupt */
694         mtspr(SPRN_TCR, TCR_DIE);
695 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
696 }
697 
698 void __init generic_calibrate_decr(void)
699 {
700         ppc_tb_freq = DEFAULT_TB_FREQ;          /* hardcoded default */
701 
702         if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
703             !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
704 
705                 printk(KERN_ERR "WARNING: Estimating decrementer frequency "
706                                 "(not found)\n");
707         }
708 
709         ppc_proc_freq = DEFAULT_PROC_FREQ;      /* hardcoded default */
710 
711         if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
712             !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
713 
714                 printk(KERN_ERR "WARNING: Estimating processor frequency "
715                                 "(not found)\n");
716         }
717 }
718 
719 int update_persistent_clock(struct timespec now)
720 {
721         struct rtc_time tm;
722 
723         if (!ppc_md.set_rtc_time)
724                 return -ENODEV;
725 
726         to_tm(now.tv_sec + 1 + timezone_offset, &tm);
727         tm.tm_year -= 1900;
728         tm.tm_mon -= 1;
729 
730         return ppc_md.set_rtc_time(&tm);
731 }
732 
733 static void __read_persistent_clock(struct timespec *ts)
734 {
735         struct rtc_time tm;
736         static int first = 1;
737 
738         ts->tv_nsec = 0;
739         /* XXX this is a litle fragile but will work okay in the short term */
740         if (first) {
741                 first = 0;
742                 if (ppc_md.time_init)
743                         timezone_offset = ppc_md.time_init();
744 
745                 /* get_boot_time() isn't guaranteed to be safe to call late */
746                 if (ppc_md.get_boot_time) {
747                         ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
748                         return;
749                 }
750         }
751         if (!ppc_md.get_rtc_time) {
752                 ts->tv_sec = 0;
753                 return;
754         }
755         ppc_md.get_rtc_time(&tm);
756 
757         ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday,
758                             tm.tm_hour, tm.tm_min, tm.tm_sec);
759 }
760 
761 void read_persistent_clock(struct timespec *ts)
762 {
763         __read_persistent_clock(ts);
764 
765         /* Sanitize it in case real time clock is set below EPOCH */
766         if (ts->tv_sec < 0) {
767                 ts->tv_sec = 0;
768                 ts->tv_nsec = 0;
769         }
770                 
771 }
772 
773 /* clocksource code */
774 static cycle_t rtc_read(struct clocksource *cs)
775 {
776         return (cycle_t)get_rtc();
777 }
778 
779 static cycle_t timebase_read(struct clocksource *cs)
780 {
781         return (cycle_t)get_tb();
782 }
783 
784 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm,
785                          struct clocksource *clock, u32 mult, cycle_t cycle_last)
786 {
787         u64 new_tb_to_xs, new_stamp_xsec;
788         u32 frac_sec;
789 
790         if (clock != &clocksource_timebase)
791                 return;
792 
793         /* Make userspace gettimeofday spin until we're done. */
794         ++vdso_data->tb_update_count;
795         smp_mb();
796 
797         /* 19342813113834067 ~= 2^(20+64) / 1e9 */
798         new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift);
799         new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC;
800         do_div(new_stamp_xsec, 1000000000);
801         new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC;
802 
803         BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC);
804         /* this is tv_nsec / 1e9 as a 0.32 fraction */
805         frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32;
806 
807         /*
808          * tb_update_count is used to allow the userspace gettimeofday code
809          * to assure itself that it sees a consistent view of the tb_to_xs and
810          * stamp_xsec variables.  It reads the tb_update_count, then reads
811          * tb_to_xs and stamp_xsec and then reads tb_update_count again.  If
812          * the two values of tb_update_count match and are even then the
813          * tb_to_xs and stamp_xsec values are consistent.  If not, then it
814          * loops back and reads them again until this criteria is met.
815          * We expect the caller to have done the first increment of
816          * vdso_data->tb_update_count already.
817          */
818         vdso_data->tb_orig_stamp = cycle_last;
819         vdso_data->stamp_xsec = new_stamp_xsec;
820         vdso_data->tb_to_xs = new_tb_to_xs;
821         vdso_data->wtom_clock_sec = wtm->tv_sec;
822         vdso_data->wtom_clock_nsec = wtm->tv_nsec;
823         vdso_data->stamp_xtime = *wall_time;
824         vdso_data->stamp_sec_fraction = frac_sec;
825         smp_wmb();
826         ++(vdso_data->tb_update_count);
827 }
828 
829 void update_vsyscall_tz(void)
830 {
831         vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
832         vdso_data->tz_dsttime = sys_tz.tz_dsttime;
833 }
834 
835 static void __init clocksource_init(void)
836 {
837         struct clocksource *clock;
838 
839         if (__USE_RTC())
840                 clock = &clocksource_rtc;
841         else
842                 clock = &clocksource_timebase;
843 
844         if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
845                 printk(KERN_ERR "clocksource: %s is already registered\n",
846                        clock->name);
847                 return;
848         }
849 
850         printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
851                clock->name, clock->mult, clock->shift);
852 }
853 
854 static int decrementer_set_next_event(unsigned long evt,
855                                       struct clock_event_device *dev)
856 {
857         __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt);
858         set_dec(evt);
859 
860         /* We may have raced with new irq work */
861         if (test_irq_work_pending())
862                 set_dec(1);
863 
864         return 0;
865 }
866 
867 static int decrementer_shutdown(struct clock_event_device *dev)
868 {
869         decrementer_set_next_event(DECREMENTER_MAX, dev);
870         return 0;
871 }
872 
873 /* Interrupt handler for the timer broadcast IPI */
874 void tick_broadcast_ipi_handler(void)
875 {
876         u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
877 
878         *next_tb = get_tb_or_rtc();
879         __timer_interrupt();
880 }
881 
882 static void register_decrementer_clockevent(int cpu)
883 {
884         struct clock_event_device *dec = &per_cpu(decrementers, cpu);
885 
886         *dec = decrementer_clockevent;
887         dec->cpumask = cpumask_of(cpu);
888 
889         printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
890                     dec->name, dec->mult, dec->shift, cpu);
891 
892         clockevents_register_device(dec);
893 }
894 
895 static void __init init_decrementer_clockevent(void)
896 {
897         int cpu = smp_processor_id();
898 
899         clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4);
900 
901         decrementer_clockevent.max_delta_ns =
902                 clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent);
903         decrementer_clockevent.min_delta_ns =
904                 clockevent_delta2ns(2, &decrementer_clockevent);
905 
906         register_decrementer_clockevent(cpu);
907 }
908 
909 void secondary_cpu_time_init(void)
910 {
911         /* Start the decrementer on CPUs that have manual control
912          * such as BookE
913          */
914         start_cpu_decrementer();
915 
916         /* FIME: Should make unrelatred change to move snapshot_timebase
917          * call here ! */
918         register_decrementer_clockevent(smp_processor_id());
919 }
920 
921 /* This function is only called on the boot processor */
922 void __init time_init(void)
923 {
924         struct div_result res;
925         u64 scale;
926         unsigned shift;
927 
928         if (__USE_RTC()) {
929                 /* 601 processor: dec counts down by 128 every 128ns */
930                 ppc_tb_freq = 1000000000;
931         } else {
932                 /* Normal PowerPC with timebase register */
933                 ppc_md.calibrate_decr();
934                 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
935                        ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
936                 printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
937                        ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
938         }
939 
940         tb_ticks_per_jiffy = ppc_tb_freq / HZ;
941         tb_ticks_per_sec = ppc_tb_freq;
942         tb_ticks_per_usec = ppc_tb_freq / 1000000;
943         calc_cputime_factors();
944         setup_cputime_one_jiffy();
945 
946         /*
947          * Compute scale factor for sched_clock.
948          * The calibrate_decr() function has set tb_ticks_per_sec,
949          * which is the timebase frequency.
950          * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
951          * the 128-bit result as a 64.64 fixed-point number.
952          * We then shift that number right until it is less than 1.0,
953          * giving us the scale factor and shift count to use in
954          * sched_clock().
955          */
956         div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
957         scale = res.result_low;
958         for (shift = 0; res.result_high != 0; ++shift) {
959                 scale = (scale >> 1) | (res.result_high << 63);
960                 res.result_high >>= 1;
961         }
962         tb_to_ns_scale = scale;
963         tb_to_ns_shift = shift;
964         /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
965         boot_tb = get_tb_or_rtc();
966 
967         /* If platform provided a timezone (pmac), we correct the time */
968         if (timezone_offset) {
969                 sys_tz.tz_minuteswest = -timezone_offset / 60;
970                 sys_tz.tz_dsttime = 0;
971         }
972 
973         vdso_data->tb_update_count = 0;
974         vdso_data->tb_ticks_per_sec = tb_ticks_per_sec;
975 
976         /* Start the decrementer on CPUs that have manual control
977          * such as BookE
978          */
979         start_cpu_decrementer();
980 
981         /* Register the clocksource */
982         clocksource_init();
983 
984         init_decrementer_clockevent();
985         tick_setup_hrtimer_broadcast();
986 
987 #ifdef CONFIG_COMMON_CLK
988         of_clk_init(NULL);
989 #endif
990 }
991 
992 
993 #define FEBRUARY        2
994 #define STARTOFTIME     1970
995 #define SECDAY          86400L
996 #define SECYR           (SECDAY * 365)
997 #define leapyear(year)          ((year) % 4 == 0 && \
998                                  ((year) % 100 != 0 || (year) % 400 == 0))
999 #define days_in_year(a)         (leapyear(a) ? 366 : 365)
1000 #define days_in_month(a)        (month_days[(a) - 1])
1001 
1002 static int month_days[12] = {
1003         31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31
1004 };
1005 
1006 void to_tm(int tim, struct rtc_time * tm)
1007 {
1008         register int    i;
1009         register long   hms, day;
1010 
1011         day = tim / SECDAY;
1012         hms = tim % SECDAY;
1013 
1014         /* Hours, minutes, seconds are easy */
1015         tm->tm_hour = hms / 3600;
1016         tm->tm_min = (hms % 3600) / 60;
1017         tm->tm_sec = (hms % 3600) % 60;
1018 
1019         /* Number of years in days */
1020         for (i = STARTOFTIME; day >= days_in_year(i); i++)
1021                 day -= days_in_year(i);
1022         tm->tm_year = i;
1023 
1024         /* Number of months in days left */
1025         if (leapyear(tm->tm_year))
1026                 days_in_month(FEBRUARY) = 29;
1027         for (i = 1; day >= days_in_month(i); i++)
1028                 day -= days_in_month(i);
1029         days_in_month(FEBRUARY) = 28;
1030         tm->tm_mon = i;
1031 
1032         /* Days are what is left over (+1) from all that. */
1033         tm->tm_mday = day + 1;
1034 
1035         /*
1036          * No-one uses the day of the week.
1037          */
1038         tm->tm_wday = -1;
1039 }
1040 EXPORT_SYMBOL(to_tm);
1041 
1042 /*
1043  * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
1044  * result.
1045  */
1046 void div128_by_32(u64 dividend_high, u64 dividend_low,
1047                   unsigned divisor, struct div_result *dr)
1048 {
1049         unsigned long a, b, c, d;
1050         unsigned long w, x, y, z;
1051         u64 ra, rb, rc;
1052 
1053         a = dividend_high >> 32;
1054         b = dividend_high & 0xffffffff;
1055         c = dividend_low >> 32;
1056         d = dividend_low & 0xffffffff;
1057 
1058         w = a / divisor;
1059         ra = ((u64)(a - (w * divisor)) << 32) + b;
1060 
1061         rb = ((u64) do_div(ra, divisor) << 32) + c;
1062         x = ra;
1063 
1064         rc = ((u64) do_div(rb, divisor) << 32) + d;
1065         y = rb;
1066 
1067         do_div(rc, divisor);
1068         z = rc;
1069 
1070         dr->result_high = ((u64)w << 32) + x;
1071         dr->result_low  = ((u64)y << 32) + z;
1072 
1073 }
1074 
1075 /* We don't need to calibrate delay, we use the CPU timebase for that */
1076 void calibrate_delay(void)
1077 {
1078         /* Some generic code (such as spinlock debug) use loops_per_jiffy
1079          * as the number of __delay(1) in a jiffy, so make it so
1080          */
1081         loops_per_jiffy = tb_ticks_per_jiffy;
1082 }
1083 
1084 static int __init rtc_init(void)
1085 {
1086         struct platform_device *pdev;
1087 
1088         if (!ppc_md.get_rtc_time)
1089                 return -ENODEV;
1090 
1091         pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0);
1092 
1093         return PTR_ERR_OR_ZERO(pdev);
1094 }
1095 
1096 device_initcall(rtc_init);
1097 

~ [ 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